# The Rundown University: Complete Public Content > Practical AI education for professionals, with step-by-step guides, expert-led courses, live workshops, and member resources. Source: https://app.therundown.ai This generated document contains the full editorial text of current public guides and the public catalog description of current courses. Course lesson video, transcript, enrollment, progress, and completion data are intentionally excluded. Consult the linked canonical page for current access terms. ## Guides ### Guide: 4 Prompts That Turn Any CSV Into an Excel Dashboard (Claude in Excel Tutorial) > Learn four copy-paste prompts that turn a raw CSV into a clean, visual Excel dashboard using Claude in Excel. - Canonical page: https://app.therundown.ai/guides/4-prompts-that-turn-any-csv-into-an-excel-dashboard-claude-in-excel-tutorial - Machine-readable page: https://app.therundown.ai/guides/4-prompts-that-turn-any-csv-into-an-excel-dashboard-claude-in-excel-tutorial.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Data, Research & Analysis, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-03-04 - Updated: 2026-07-30 ### Overview #### The Rundown Learn four copy-paste prompts that turn a raw CSV into a clean, visual Excel dashboard using Claude in Excel. You will import a dataset, have Claude inspect and clean it, then generate summary tables, conditional formatting, and charts. The cleaning pass can take a minute or two depending on the data, and the dashboard-building step takes about five minutes. #### Who This Is Useful For - Analysts and operations teams who spend the first hour of a project cleaning data - Founders and managers who receive CSV exports but are not Excel power users - Students and researchers working with messy public datasets #### What You Will Build You will build a polished Excel dashboard from a raw CSV, with reusable prompts you can apply to other datasets. - Cleaned and structured workbook data - Summary tables based on the most useful metrics - Conditional formatting for top and bottom performers - Two to three labeled charts on a Dashboard tab - A reusable set of four prompts for future CSV files The workflow starts from messy CSV-style data like this and turns it into a dashboard. #### What You Need - Microsoft 365 with Excel on desktop, web, or Mac - A Claude Pro, Max, Team, or Enterprise plan for the Excel add-in - A CSV file, such as your own data, a client export, or a public dataset #### Going Further - Try the same four prompts with your own data from a CRM, accounting tool, ad platform, or another source. - Build template workbooks for recurring team use, such as monthly reports, client dashboards, or inventory tracking. - Have Claude convert a finished dashboard into a template so you can reuse it every cycle with fresh data. ### Step 1 Install the add-in and import your data Open Excel, click Add-ins in the top ribbon, then Get Add-ins. Search for Claude and install it. The Claude button will appear in your ribbon. Click it and log in. Open your CSV in Excel. If it opens as a flat file, save it as a workbook (.xlsx). > Pro tip: Always duplicate your data into a second tab or save a backup copy before you start. Claude can modify cells directly and occasionally overwrite things, so it is better to have a safety net. ### Step 2 Explore the data Paste this prompt into Claude: ```text Look at all the data in this workbook. Tell me what this data is, what each column represents, and flag any problems with the structure or formatting. List the issues. ``` Claude will scan the workbook and return a plain-English breakdown of what the data is, what each column means, and what is wrong with the structure or formatting. This helps Claude understand the dataset before it starts building anything. - If Claude gets confused about something, such as abbreviations or date formats, answer its questions in the chat. A one-line clarification goes a long way. - You do not need to tell Claude what kind of data this is. It figures that out on its own. ### Step 3 Clean and structure the workbook Paste the second prompt: ```text Fix every issue you just identified. Remove any junk rows or metadata. Format numbers so they're human-readable. Create proper tables with clear headers. Sort by whatever the most important metric is. ``` Claude will remove junk rows, reformat numbers, build a proper table with headers, and sort by the metric that matters most. - This step can take a minute or two depending on how much data you have. - You can watch Claude work in real time as it edits cells. - If it flags something you disagree with, such as renaming a column, tell it to skip that one and fix the rest. > Pro tip: Turn on session logging in Claude’s settings. It tracks every change Claude makes in a separate sheet, which is helpful if you need to undo something or explain the process to a teammate. ### Step 4 Build the dashboard Paste the third prompt: ```text Create a new "Dashboard" tab. Identify the top 3-5 most useful metrics from this data and build a summary table for each. Rank the entries. Use conditional formatting to highlight the top performers in green and the bottom performers in red. ``` Claude creates a new tab and builds summary tables based on the metrics it identifies as most useful. In the source examples, it chose rankings, growth rates, and year-over-year change for GDP data, and scoring leaders, assist leaders, and efficiency for NBA stats. - Claude sometimes puts dashboard content in the wrong tab. If that happens, tell it: “I noticed some dashboard tables ended up in the raw data sheet. Can you move everything to the Dashboard tab?” - This step takes about five minutes. You can let it run in the background while you do other things. ### Step 5 Add charts Paste the final dashboard prompt: ```text Add 2-3 charts to the Dashboard tab that tell the story of this data. Pick the chart types that make the most sense (bar, line, scatter, etc). Make them clean and labeled. ``` Claude picks chart types based on the data. For GDP data, it built a line chart showing GDP over time and a bar chart comparing countries. For NBA stats, it made a bar chart for top scorers and a scatter plot for points vs. assists. - Charts are a newer feature in Claude for Excel. They work, but Claude sometimes hard-codes source data and then hides the rows, which breaks the chart. If your charts look blank, unhide the rows above or below them. - You may need to drag charts around or resize them. The data is right, but the layout may need a nudge. ### Step 6 Turn the dashboard into a reusable template If you want to keep using the dashboard every month or quarter, paste this optional prompt: ```text Turn this dashboard into a reusable template. Replace the data with placeholder labels so I can swap in new data next month. ``` Claude will turn the dashboard into a template workbook so you can drop in fresh data and run the prompts again. ### Pro tip > Always duplicate your data into a second tab or save a backup copy before you start. Claude can modify cells directly and may overwrite things. ### Guide: Add CGI To Any Video With Higgsfield > In this guide, you will learn how to add a believable AI VFX element to a short video clip with Higgsfield. - Canonical page: https://app.therundown.ai/guides/add-cgi-to-any-video-with-higgsfield - Machine-readable page: https://app.therundown.ai/guides/add-cgi-to-any-video-with-higgsfield.md - Difficulty: beginner - Featured tool or company: Higgsfield - Categories: Content & Creative - Instructors: Billy Howell - Published: 2026-07-08 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to add a believable AI VFX element to a short video clip with Higgsfield. You will pick footage the model can preserve, use ChatGPT to tighten the effect prompt before spending video credits, then generate and review the edit in Gemini Omni Flash. #### Who This Is Useful For - Creators who want to add sci-fi, monster, portal, or stylized VFX moments to short social clips. - Editors who need a fast way to test AI video effects without burning credits on weak prompts. - Anyone turning horizontal, real-world footage into more cinematic or vertical-friendly AI edits. #### What You Will Build You will learn how to take a short video shot and add one controlled AI VFX element into it, such as a creature, portal, spaceship, explosion, or stylized background effect. In the demo, we added a sci-fi spaceship into the background of a plain parking-garage clip: ![Higgsfield spaceship output](https://tru-images.b-cdn.net/guide-assets/local-312/add-cgi-to-any-video-with-higgsfield/01-higgsfield-spaceship-output.png) The reusable part is the workflow: - A source-clip checklist for deciding whether a shot is worth editing. - A prompt structure that separates the new VFX element from the parts of the video that must stay unchanged. - A visual QA pass for checking subject integrity, camera motion, lighting, scale, and unwanted changes. - One targeted revision prompt instead of starting over from scratch. #### What You Need To Get Started - A Higgsfield (https://higgsfield.ai/) account. - A short video clip. - Optional: ChatGPT or Claude for improving the effect prompt before generating. #### Going Further Once you have one believable output, save the source clip, prompt, generated result, and QA notes together. That gives you a repeatable prompt pattern for future edits: ```text Add [one specific VFX object or effect] to [specific location in the shot]. Keep everything else the same: same [subject], same [camera motion], same [framing], same [timing], same [background/environment], and same [lighting]. Single continuous shot. No scene cuts. No unrelated changes. ``` You can also browse Higgsfield's Visual Effects collection (https://higgsfield.ai/collection/effects) for effect ideas, or use Mixed Media (https://higgsfield.ai/mixed-media-intro) when you want a stylized preset look instead of precise object insertion. Treat those as alternate paths, not replacements for the source-video edit workflow. The main takeaway is simple: pick footage the model can preserve, make the prompt specific before spending video credits, and review the result against the original before trusting it. ### Step 1 Pick Footage The Model Can Preserve Start with the source clip, not the prompt. When you are ready to generate, use the exact Gemini Omni Flash (https://higgsfield.ai/gemini-omni-flash) link because this workflow can be hard to find from the main Higgsfield interface. The best AI VFX results come from footage with enough open space for the model to add something without rewriting the whole scene. Use clips with: - One readable shot and no cuts. - A clear subject or simple empty scene. - Enough background room for the VFX element. - Foreground and background that are both reasonably visible. - Simple camera motion and stable lighting. Avoid tight, cluttered shots for the first pass. A tight shot with an out-of-focus fence can make an effect like a background portal harder to solve because the model may not cleanly understand the depth of the scene. A quick prompt like "put me in the jungle" can create the same problem from another angle: vague prompts and cramped source footage invite the model to remake the scene instead of editing it. > Pro tip: If you have a 10-second clip, trim a 2-second test first. AI video generation is usually billed by duration, so short tests are cheaper while you are still finding the right prompt. ### Step 2 Improve The Prompt Before Spending Video Credits Do not start with a tiny vibe prompt and hope the model guesses correctly. In the demo, the weak first prompt was: ```text Giant donut monster running past the guy in the warehouse, acid trip style. ``` That prompt produced a fun result, but it also changed too much of the scene. It added rainbow lighting and extra style changes that were not part of a controlled edit. ![Overchanged donut output](https://tru-images.b-cdn.net/guide-assets/local-312/add-cgi-to-any-video-with-higgsfield/02-higgsfield-donut-overchanged-output.png) Before generating, use ChatGPT to expand the idea into a more specific video-edit prompt. The useful instruction is: ```text Help me improve this prompt for an AI special effects generator. I want to be more specific about exactly what to add into the scene and what not to change. Source clip: [describe the shot] Effect idea: [describe the effect] Return a concise prompt that adds one effect and preserves the original subject, camera movement, framing, timing, background, and lighting. ``` The stronger spaceship prompt worked because it named the object, the quantity, the position, the motion, and the preservation rules: ```text Add one large photorealistic sci-fi craft hovering far in the background outside the parking garage. It should move slowly and feel distant, heavy, and physically present in the scene. Keep everything else the same: same person, same camera movement, same framing, same timing, same background/environment, and same lighting direction. Single continuous shot. No scene cuts. No new characters. No face changes. No body warping. No extra explosions or camera moves. ``` That specificity is the difference between "make this epic" and "add one thing without breaking the shot." ### Step 3 Generate The Edit In Higgsfield Open Higgsfield Gemini Omni Flash (https://higgsfield.ai/gemini-omni-flash), then click Start generating. In the logged-in workspace, use the video workflow that accepts source media. The visible route may be https://higgsfield.ai/ai/video?model=gemini-omni or it may normalize to https://higgsfield.ai/ai/video. You are in the right place when the panel has an upload area for media, a prompt box, and the model set to Gemini Omni Flash. Upload the source clip, paste the improved prompt, and generate one pass. Higgsfield may take a few minutes to generate the edit. If live generation stalls, use a prepared output and continue the walkthrough from the review step. If Gemini Omni Flash is not available, use Higgsfield AI Video Editor (https://higgsfield.ai/ai-video-editor) or https://higgsfield.ai/ai/video/edit, but only if the editor accepts an uploaded video plus a text prompt. A fallback model such as Kling 3.0 Omni Edit or Kling O1 Video Edit is acceptable. Do not switch to a generic text-to-video or image-to-video flow and pretend it is the same workflow. ### Step 4 Review The Result Against The Original After the generation finishes, compare the AI edit against the original clip. Do not only ask whether it looks cool. Ask whether it still behaves like the same shot. Check: - Subject integrity: did the person, face, hands, or body change? - Camera continuity: did the camera motion, framing, or timing drift? - Effect integration: does the object sit in the background instead of floating on top? - Lighting match: do the shadows, glow, color temperature, and haze fit the source? - Source preservation: did Higgsfield change anything unrelated? In the strong spaceship example, the prompt kept the main shot intact because it clearly said the object should be one distant sci-fi craft outside the parking garage and repeated the preservation constraints. ![Spaceship prompt details](https://tru-images.b-cdn.net/guide-assets/local-312/add-cgi-to-any-video-with-higgsfield/03-higgsfield-spaceship-prompt-details.png) In the weaker examples, the model either added too much style, struggled with depth, or stacked a new spaceship in front of the old one. Those failures are useful because they tell you exactly what to fix in the next prompt. ### Step 5 Make One Targeted Revision Use the review pass to fix one issue at a time. Do not rewrite the whole concept unless the first result is unusable. Good revision targets: - Make the object smaller or farther away. - Remove unwanted color effects. - Preserve the subject more strictly. - Match the lighting better. - Reduce edge flicker. - Reframe for vertical video. Use a revision prompt like: ```text Revise the previous edit. Keep the same sci-fi craft idea, but make the craft feel farther away and more physically integrated into the background. Preserve the original person, camera motion, framing, timing, background, and lighting. Keep everything else the same. ``` One useful workaround is to download the generated clip and upload that result back into Gemini Omni Flash as the new source if an edit gets stuck or a direct revision does not behave. Then ask for a narrower correction, such as removing unwanted rainbow light effects while keeping the main character or object. ### Step 6 Change The Aspect Ratio With Ratio And Crop Adjuster A useful tool we found while recording was Higgsfield's Ratio and Crop Adjuster. When you edit a video, it can automatically crop and change the aspect ratio of your video. ![Ratio and Crop Adjuster result](https://tru-images.b-cdn.net/guide-assets/local-312/add-cgi-to-any-video-with-higgsfield/05-higgsfield-vertical-reframe-result.png) Use this as a separate workflow. First get the VFX edit looking right. Then use Ratio and Crop Adjuster to test vertical versions for TikTok, Shorts, or Reels. If you ask for too many creative changes at the same time, it becomes harder to know which instruction caused the failure. ### Guide: AEO/GEO Part 1: Find Out Where Your Brand Appears in AI Search > In this guide, you will learn how to run an "AI SEO" audit on a website. - Canonical page: https://app.therundown.ai/guides/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search - Machine-readable page: https://app.therundown.ai/guides/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search.md - Difficulty: beginner - Categories: AI - Instructors: Billy Howell - Published: 2026-09-09 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to run an "AI SEO" audit on a website. You will learn where your site appears in AI answers and identify opportunities to improve its visibility. #### Who This Is Useful For - Business owners checking whether AI can find and accurately describe their business. - Marketers comparing audit findings before choosing what to improve on a website. - Anyone planning content around the questions their customers ask Google and AI assistants. #### What You Will Build You will build a baseline for your website's search visibility and a short list of questions you want it to answer. We use The Arlington Bagel, a weekly newsletter covering events and local business in Arlington, Virginia. You can follow the same steps with your own website. ![Codex table of suggested Arlington newsletter queries and target pages](https://tru-images.b-cdn.net/guide-assets/local-337/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search/01-codex-priority-query-table.png) Use the audit findings to choose queries and pages to work on. These are suggested targets, not measured search volumes or rankings. #### What You Need To Get Started - A public website URL and a short description of your business. - An email address for the audit tools' report or signup forms. - A Google account and permission to verify your website in Search Console, or access to an existing property. - Codex with browser access if you want to compile the reports in Going Further. #### Going Further Have Codex run the free audits and compile the findings into a one-page report. If you already have the reports open, point it to those browser tabs or provide the exports so it can start with your actual results. ![The prompt asking Codex to review open audit reports and suggest priorities](https://tru-images.b-cdn.net/guide-assets/local-337/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search/05-codex-audit-summary-prompt.png) Give Codex the reports, your website, and the decision you want help making. Here is a reusable version: ```text Run these free audits for [YOUR WEBSITE]: - Peekaboo: https://www.aipeekaboo.com/ - HubSpot AI Search Grader: https://www.hubspot.com/ai-search-grader - Semrush SEO Checker: https://www.semrush.com/free-tools/seo-checker/ - Optional Semrush AI visibility check: https://www.semrush.com/free-tools/ai-search-visibility-checker/ Reuse my completed reports if they are already open in [BROWSER] or attached here. Check the site's Google Search Console property if I have given you access. Create a one-page PDF report with: - The main findings from each report, including source links and dates. - My top three priorities and why they matter. - Five customer queries we should target and the best page for each. Link an existing page when available; otherwise label it a proposed new page. - A quick checklist of next actions. Keep SEO scores, AI mentions, and Google AI impressions separate. Use actual results and label recommendations as recommendations. If a report is inaccessible, say what is missing. Ask me to handle login or verification when needed, and don't start paid plans or change the website. ``` Use that page to choose the first improvements you want to make. Keep the original reports linked so you can check the evidence and compare the next audit with this baseline. Part 2 covers the website improvements you can make from these findings. ### Step 1: Run a Peekaboo Audit AEO stands for Answer Engine Optimization: making your information easy to find and use in direct answers. GEO stands for Generative Engine Optimization: improving how your business and content appear in AI-generated answers. The terms overlap. For this exercise, the useful question is whether someone looking for what you offer can discover your business and understand it correctly. Open Peekaboo (https://www.aipeekaboo.com/), enter your website URL, and select Check my AI Visibility. Review the business name and description it pulls from your website. Correct the location or category if needed before starting the report. ![Peekaboo detecting The Arlington Bagel and its newsletter description](https://tru-images.b-cdn.net/guide-assets/local-337/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search/02-peekaboo-brand-setup.png) Confirm the detected description. The 72/100 card on the right is labeled “Example”; it is not the Bagel's result. Choose Quick Start for the suggested setup or Customize to review the choices. Complete the free-report flow, then check which questions and AI products it tested. Read the actual answers and cited sources alongside the score. Use the Brand mentioned filter to inspect the individual answers. In our check, the Bagel appeared in one result while other results omitted it. Open the citations to see which pages supported those answers. Keep the report open for comparison with the next tools. Note whether your brand appears, whether the description is accurate, and which other businesses are mentioned. Peekaboo's free report and ongoing paid monitoring are separate options. ### Step 2: Check How HubSpot Describes Your Brand Open HubSpot AI Search Grader (https://www.hubspot.com/ai-search-grader). Enter your brand name, location, industry, and what you offer. Use the same description you checked in Peekaboo so the tools are evaluating the same business. HubSpot offers a free one-time check of how ChatGPT, Perplexity, and Gemini describe your brand based on their training data. Complete the report form and read the written assessment, including brand recognition and sentiment. We got a HubSpot report back, but the useful part is checking what supports its assessment. Does it know what your business does, associate it with the right location, and describe its offering accurately? Keep those findings separate from live web-search recommendations. A tool that recognizes your brand when you supply its name answers a different question from one that recommends it without being prompted. ### Step 3: Run Semrush's SEO Checker Open the Semrush SEO Checker (https://www.semrush.com/free-tools/seo-checker/), enter your domain, and choose Analyze Website. Once the report finishes, start with Overview, then open Issues and Recommendations. Our Arlington Bagel result showed 61/100, with six critical issues and two minor issues flagged by Semrush. Treat those as the tool's findings to investigate, rather than eight independently confirmed defects. ![Semrush SEO Checker showing a 61 out of 100 score for The Arlington Bagel](https://tru-images.b-cdn.net/guide-assets/local-337/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search/03-semrush-seo-audit-results.png) The SEO report gives you a starting list of website improvements to inspect. Semrush also has a separate AI Visibility Checker (https://www.semrush.com/free-tools/ai-search-visibility-checker/). Use it if you want another AI-search snapshot; keep its results separate from this SEO report. Read the explanation behind each recommendation and choose the items relevant to your site. The SEO score measures different things from the AI-brand reports above, so keep the scores in separate columns rather than averaging them together. ### Step 4: Set Up and Check Google Search Console Open Google Search Console (https://search.google.com/search-console) and sign in with your Google account. If your site is already listed, select its property. Otherwise, choose Add property and follow Google's verification instructions. A Domain property covers subdomains and protocols and normally requires DNS verification; a URL-prefix property offers other verification methods. Google setup instructions (https://support.google.com/webmasters/answer/34592) Once you have access, open Performance → Generative AI. Look at impressions and the pages appearing in Google's AI features, with the date range visible. Google documents this report for AI Overviews and AI Mode, with a separate Discover report. Google report details (https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports) ![Search Console navigation showing Performance and Generative AI](https://tru-images.b-cdn.net/guide-assets/local-337/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search/04-search-console-generative-ai-menu.png) Find the Google AI report under Performance. Also check Search results → Queries for the search terms already bringing your site into Google results. Newly added properties can take a few days to show data. Keep each property's results with its own website. The Generative AI view reports Google visibility, not ChatGPT or Claude mentions. ### Step 5: Choose the Questions You Want to Answer Think about what a potential customer would ask before knowing your brand name. For the Bagel, useful examples include: - What newsletters cover events and local business in Arlington, Virginia? - What is happening in Arlington this weekend? - Where can I learn about new restaurants opening in Arlington? For a garden store in Houston, “Where can I buy fertilizer in Houston?” is a useful question because it connects a product to the area the business serves. Think about who you want arriving at your site and what they need. Choose a small set that matches what you actually offer. Beside each question, identify the existing page that answers it, or note where a useful page is missing. The query table above suggests newsletter landing pages, an events hub, and restaurant-opening coverage as possible targets. Try the questions in ChatGPT and Google, then compare the answers and citations. You can repeat them in Claude, Perplexity, or Gemini for another view. Use the audits to prioritize this list. An inaccurate brand description suggests a different next action from a relevant page that is difficult to find. Save the exact questions so you can repeat the same checks after making improvements. ### Guide: AEO/GEO Part 2: Make Your Website Easier for AI to Find and Cite > In this guide, you will learn how to turn your AEO and GEO audit results into a plan for improving your website. - Canonical page: https://app.therundown.ai/guides/aeogeo-part-2-make-your-website-easier-for-ai-to-find-and-cite - Machine-readable page: https://app.therundown.ai/guides/aeogeo-part-2-make-your-website-easier-for-ai-to-find-and-cite.md - Difficulty: intermediate - Categories: AI, seo, geo, aeo, codex - Instructors: Billy Howell - Published: 2026-09-09 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to turn your AEO and GEO audit results into a plan for improving your website. You will build a Google Sheets workbook with Codex so you can work through the fixes and track what still needs doing. #### Who This Is Useful For - Publishers who want to make their newsletter's audience and coverage easier for AI systems to understand. - Business owners who want to turn a website audit into a practical homepage and identity cleanup. - Marketers who need a repeatable way to prioritize AEO/GEO fixes and keep future page ideas organized. #### What you will build You will build a Google Sheets audit workbook. It'll give you a source of truth for fixes and tasks for your AEO and GEO optimization. #### What you need to get started - A website or publication you want to improve. - The audit results from Part 1 (https://app.therundown.ai/guides/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search), or a fresh public-site audit. - An AI harness with access to the website information and reports you want it to review. We use Codex for this, but you can use Claude Code or another AI harness. - Optional: a Google Sheets account if you want a shared online workbook. Excel or a CSV works too. #### Going further After publishing a change, repeat the same nonbranded questions from Part 1 with the same wording. Record the date, tool or model, sources, mentions, citations, and accuracy. Preserve "no change" results; a content edit is not proof of improved AI visibility. For pages that change regularly, such as events this weekend or a quarterly market update, draft the page once and create a monthly review task for new facts and links. Keep evergreen pages stable. If your hosted platform limits the edits Codex can make, ask for a platform-specific task list or a safe subdomain plan, then review live changes yourself before publishing. ### Step 1: Turn audit findings into fixes Take the Part 1 reports (https://app.therundown.ai/guides/aeogeo-part-1-find-out-where-your-brand-appears-in-ai-search), the URL you want to improve, and the customer question you care about, and put them in the same chat. We use Codex for this, but you can use Claude Code or another AI harness. For example, the target question for The Arlington Bagel is: "What local newsletters should I subscribe to for events and local business news in Arlington, Virginia?" That question gives the audit a real job to do. Quick refresher: AEO is about making information easy to find, understand, and use in an answer. GEO is about how that information shows up in AI responses, including mentions and citations. ![AEO and GEO explained as two terms with one useful goal](https://tru-images.b-cdn.net/guide-assets/local-479/aeogeo-part-2-make-your-website-easier-for-ai-to-find-and-cite/01-aeo-geo-primer.png) Ask the AI harness to inspect the public homepage, About or Authors page, recent editions, and any available robots.txt or sitemap. Have it list the URL and the specific thing it found for every recommendation, separate actual problems from optional improvements, and avoid inventing rankings, traffic, visibility scores, or citations. > Pro tip: If you skipped Part 1, have your AI harness run the audit first. You can also give it a public GEO audit as an input and ask it to turn the result into website-specific tasks. ![Audit findings becoming an actionable task list](https://tru-images.b-cdn.net/guide-assets/local-479/aeogeo-part-2-make-your-website-easier-for-ai-to-find-and-cite/02-audit-findings-task-list.png) The output should answer three questions: - Can the page be reached and read? - Can a reader quickly understand what the business or publication is, where it operates, and what it offers? - Can someone verify who publishes it and which sources support its claims? The score is just a diagnostic. The thing you need is a list of specific changes you can make and check. ### Step 2: Build the audit workbook Ask your AI harness to turn the audit findings into a task list in Google Sheets. You can use our workbook (https://docs.google.com/spreadsheets/d/1QKnJDXG_2LMRRnSAlzgsMwnN2foSGrMFmy0M2TgGWsE/edit?gid=984290698#gid=984290698) as an example. Excel or a CSV works too. At minimum, include columns for: - Task ID and priority - Status and owner - Workstream or page location - Proposed change or acceptance criteria - Effort and verification link Keep the sheet simple. Each row should tell you what needs to change, where the change belongs, how important it is, and what still needs to be checked. ![A prioritized AEO/GEO task tracker](https://tru-images.b-cdn.net/guide-assets/local-479/aeogeo-part-2-make-your-website-easier-for-ai-to-find-and-cite/03-aeo-geo-task-tracker.png) ![Workbook summary with task counts and priority breakdown](https://tru-images.b-cdn.net/guide-assets/local-479/aeogeo-part-2-make-your-website-easier-for-ai-to-find-and-cite/04-aeo-geo-workbook-summary.png) ### Step 3: Review the workbook and start the fixes Go through the workbook and separate immediate fixes from longer-term plans. Check that each task actually fits your website, then fill in the owner, due date, and status. For The Arlington Bagel, one immediate fix was clearer homepage copy. The existing copy said: > Your weekly rundown on events and local business in Arlington, VA! The proposed revision made the format, location, coverage, and next action clearer: H1: The Arlington Bagel: Your Weekly Arlington, VA Newsletter Introduction: The Arlington Bagel is a weekly newsletter covering events and local business in Arlington, Virginia. Browse past editions to see what we cover, then subscribe for future updates. CTA: Subscribe to The Arlington Bagel ![The real homepage and its visible subscription path](https://tru-images.b-cdn.net/guide-assets/local-479/aeogeo-part-2-make-your-website-easier-for-ai-to-find-and-cite/05-homepage-proof.png) Keep proposed changes tied to facts you can check. Do not add subscriber counts, delivery days, reviews, prices, or other claims that the audit did not confirm. For publisher information, link to the existing Authors page, public archive, and subscription form instead of creating a second identity story. "Bagel" in a name does not justify inventing restaurant hours, a storefront, or bakery details. ### Step 4: Add evergreen page ideas If the workbook is missing evergreen content ideas, ask your AI harness for the five pages that would best answer questions your customers ask. Have it add those ideas to the Google Sheet, then fill in the page purpose, owner, status, and next step. For example, a realtor might add pages about neighborhoods, buying or renting in the area, and the questions clients ask before a move. A local newsletter might add evergreen pages about the area it covers, how to find local events, and what a new subscriber can expect. Keep these pages separate from time-sensitive updates. A page about events this weekend needs regular updates. A page explaining the area or your service should stay useful between updates. ### Guide: AI-Powered Movie Production With Claude > In this guide, you will learn how to use a free AI-enhanced video editor with Claude through an Anthropic API key or MCP connection. - Canonical page: https://app.therundown.ai/guides/ai-powered-movie-production-with-claude - Machine-readable page: https://app.therundown.ai/guides/ai-powered-movie-production-with-claude.md - Difficulty: beginner - Featured tool or company: Claude + Palmier Pro - Categories: Content & Creative - Instructors: Billy Howell - Published: 2026-06-27 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to use a free AI-enhanced video editor with Claude through an Anthropic API key or MCP connection. You can use it to have AI automatically transcribe, caption, cut, and color grade your own raw footage. #### Who This Is Useful For - Creators who want to turn long recordings into short-form videos faster. - Claude users who want AI editing help inside a real video editor instead of only getting editing advice in a chat window. - Anyone who wants a simple Mac video editor with an optional MCP connection for more automated workflows. #### What You Will Build ![palmier edited timeline](https://tru-images.b-cdn.net/guide-assets/local-293/ai-powered-movie-production-with-claude/01-palmier-edited-timeline.png) By the end, you will have a Palmier Pro project with raw footage imported, an AI agent connected, and a repeatable workflow for turning longer clips into short-form edits. The finished workflow has three practical pieces. - Claude makes the first cut from longer footage. - Palmier adds captions, pacing fixes, and light visual polish. - You review the final edit and export a publishable short-form video. You will also have the optional MCP command ready if you want Claude Code or Codex to work with the open Palmier project from outside the in-app chat. #### What You Need To Get Started You only need Palmier Pro, an Anthropic API key, and a video file you have permission to edit. Start with a normal raw clip before trying to automate a complicated project. - The free Palmier Pro Mac app (https://github.com/palmier-io/palmier-pro/releases/latest/download/PalmierPro.dmg). - An Anthropic API key (https://platform.claude.com/settings/keys) for the in-app agent. - Claude Console billing or API credits may be required, and API usage is separate from a Claude Pro or Max subscription. - A 3-5 minute raw clip, interview, tutorial, product video, or other file you have permission to edit. - Optional Palmier Pro docs (https://www.palmier.io/docs) if you want to review the current feature list before you start. #### Going Further ![codex mcp install command](https://tru-images.b-cdn.net/guide-assets/local-293/ai-powered-movie-production-with-claude/03-codex-mcp-install-command.png) Palmier Pro can also run as an MCP server for external agents. In Palmier Pro, open Help > MCP Instructions, or run one of these commands while Palmier is open. ```text claude mcp add --transport http palmier-pro http://127.0.0.1:19789/mcp codex mcp add palmier-pro --url http://127.0.0.1:19789/mcp ``` Use MCP when you want Claude Code or Codex to inspect a project, create an edit plan, organize clips, add captions, or build repeatable short-form workflows outside the in-app chat. ### Step 1 Download Palmier Pro Download Palmier Pro for macOS (https://github.com/palmier-io/palmier-pro/releases/latest/download/PalmierPro.dmg), install it, and open a new project. Palmier's docs say the editor and MCP server are free to use, while generative AI features use paid-plan credits. For this guide, focus on editing your own footage with Claude instead of generating new video from scratch. > Pro tip Keep the first project simple. A single five-minute clip is enough to test whether the AI edit workflow is useful. ### Step 2 Get Your Anthropic API Key ![create anthropic api key](https://tru-images.b-cdn.net/guide-assets/local-293/ai-powered-movie-production-with-claude/02-create-anthropic-api-key.png) Go to Anthropic's API key page (https://platform.claude.com/settings/keys) and create a new key. Name it something obvious like Palmier Pro. If Anthropic asks you to set up billing or credits, complete that before using the key in Palmier. Copy the key when Anthropic shows it, because you may not be able to reveal it again later. > Pro tip Treat the key like a password. Do not show it in a screen recording, screenshot, public doc, or shared project file. ### Step 3 Connect Claude Inside Palmier Pro In Palmier Pro, open Settings > Agent and paste your Anthropic API key into the agent settings. This connects Palmier's in-app chat to your Anthropic account so you can ask Claude to edit the project. If the chat does not respond, recheck the key and make sure Palmier Pro is open to the project you want to edit. ### Step 4 Import A Video Drag a video into Palmier Pro, or import a file from your Mac. Start with something simple, such as a five-minute raw clip, interview, product video, tutorial, or talking-head recording. You do not need a prepared editing project or a complicated timeline. The goal is to give Claude normal raw footage and ask it to make a useful first pass. ### Step 5 Ask Claude For The First Cut Open the Palmier chat and ask Claude to turn the raw clip into a short-form version. Use this prompt. ```text Edit this down to about one minute. Keep the strongest moments, remove dead air, and preserve the main point. ``` Review the first cut before asking for polish. Claude can move quickly, so check that the shortened video still makes sense before continuing. > Pro tip If the first cut removes something important, ask Claude to restore that section before you ask for captions or visual polish. ### Step 6 Add Captions And Light Polish After the first cut, ask Claude to add captions and make the video easier to watch. Use this prompt. ```text Add captions, improve the pacing, and apply light color or crop/keyframe polish where it helps. Keep the edit natural and make sure the main point stays clear. ``` Review the captions for timing and readability. Keep color grading, crops, and keyframes light unless the video clearly needs a stronger style. ### Step 7 Run The Three-Pass Production Cycle Use the same three-pass cycle every time you make short-form content in Palmier Pro. - Pass 1 asks Claude for the first cut from long footage to a focused short. - Pass 2 asks Claude for captions, pacing improvements, and light visual polish. - Pass 3 is your manual review before export. Watch the video yourself before posting it. Fix aggressive cuts, awkward caption timing, and color changes that look too heavy. > Pro tip Save the prompts that work best for your content. Over time, those prompts become your short-form editing template. ### Step 8 Export The Finished Video When the edit is ready, use Palmier Pro's export option. For short-form publishing, export a finished video file unless you specifically need to continue editing somewhere else. Palmier's docs also list NLE XML export options for editor handoff. Watch the exported file once before posting it so you catch caption mistakes, awkward cuts, or color changes. ### Guide: Analyze Top-Performing Ad Creatives in Minutes > This guide shows you how to use Gumloop to quickly analyze Facebook ad creatives with AI. - Canonical page: https://app.therundown.ai/guides/analyze-top-performing-ad-creatives-in-minutes - Machine-readable page: https://app.therundown.ai/guides/analyze-top-performing-ad-creatives-in-minutes.md - Difficulty: beginner - Featured tool or company: Gumloop - Categories: Marketing & Growth, Data, Research & Analysis - Instructors: Billy Howell - Published: 2025-12-05 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to use Gumloop to quickly analyze Facebook ad creatives with AI. You’ll build a flow that scrapes static ads from a Facebook Ad Library profile, sends the creative image to AI, and writes the resulting reverse-engineered creative prompt into a Google Sheet. The workflow is designed for fast content and ad creative analysis, helping you analyze and ideate on creative ideas in minutes. #### Who This Is Useful For - People interested in Gumloop who have not taken the plunge yet - People who want to learn how to set up automated content analyzers - Anyone running Facebook ads #### What You Will Build You’ll build a reusable Gumloop flow that analyzes a Facebook Ad Library account and reverse engineers a prompt to recreate each static ad creative. - A Gumloop scraper flow for static Facebook ads - A Google Sheet that stores the scraped ad data - AI-generated prompts that describe how to recreate each ad creative - An optional weekly automation that pulls ads from a list of competitors ![5_gumloop_flow](https://tru-images.b-cdn.net/guide-assets/69/693309c5bfd6fb75639aed87/5-gumloop-flow.png) #### What You Need - A Gumloop account; the source notes the full flow can be built on a free account - A Google account connected to Google Sheets - A Google Sheet for the ad data - A competitor profile from the Facebook Ads Library - Static image ads to analyze #### Going Further - Use the flow screenshots or video guide to learn how to pull extra information, like flow run date and competitor name, into your sheet using the “Duplicate” module. - Read the Gumloop documentation on Loop Mode to better understand the concept. - Use the same flow setup to scrape and analyze video ads, Reddit threads, TikTok videos, Tweets, and more. - Add a generate image module to automatically create images based on scraped ads. ### Step 1: Create a new flow in Gumloop Gumloop is generous with free credits, and the source notes you can build this whole flow on a free account. In Gumloop, create a new flow and enter this prompt: ```text Make a flow that takes a url to Facebook ad library account and scrapes all of its static (not video) ads and puts all of the data in a new row in a google sheet. It sends the static creative image to AI with a prompt telling it to reverse engineer a prompt you could put into an image generator to get the creative generated ``` ![1_putting_the_prompt_into_gumloop](https://tru-images.b-cdn.net/guide-assets/69/69330b331b42907f91d1e368/1-putting-the-prompt-into-gumloop.png) > Pro tip: You can use Gumloop’s agent to build this out for you, or you can build it manually by starting with trigger or integration options. You can also ask the agent for help while you’re in a flow. ### Step 2: Set up your Google Sheet Create a new Google Sheet that matches the structure shown below. ![4_google_sheets_empty](https://tru-images.b-cdn.net/guide-assets/69/693309e8bf48afef49f0ab26/4-google-sheets-empty.png) If you have access to the duplicate sheet template from the original guide, you can use that instead of creating the sheet manually. > Pro tip: The Gumloop agent should tell you exactly what headers to put in your sheet. Even if you use a template, once you link it in the next step, the agent will adjust to your column headers. ### Step 3: Connect your Google Sheets account After the Google Sheets module is added, click “Authenticate Credentials” and connect your Google account. You won’t need to add more credentials after this step. ![3_authenticate_with_google_sheets](https://tru-images.b-cdn.net/guide-assets/69/69330abcbfd6fb75639aedb3/3-authenticate-with-google-sheets.png) Once your account is connected, paste the link to your Google Sheet into the module and click continue in the agent sidebar. ### Step 4: Test your new flow - Go to the Facebook Ads Library and search for a competitor. - Verify that the competitor has static image ads. - Click the competitor’s name to go to their profile. - Copy the link to their profile. - Paste the link into your first Gumloop module. ![10_input_competitor_url](https://tru-images.b-cdn.net/guide-assets/69/69330baa02bea1e86d3ea559/10-input-competitor-url.png) Make sure your flow matches the setup shown. Don’t worry if you’re missing the extra two inputs; they’re only needed if you’re running this on a schedule and want to record when it was run. Click the big pink “Run” button to test the flow. ### Step 5: Automate your flow Next, set up a weekly automation that looks at a list of your competitors and pulls ads for each of them. Manually create a new flow that triggers the scraping. It should look like this: ![11_trigger_flow](https://tru-images.b-cdn.net/guide-assets/69/69330be43ecc724c60ee080f/11-trigger-flow.png) - Make a new tab in your Google Sheet with columns for “Competitor Name” and “Competitor Profile URL”. ![sheets_competitor_list_tab](https://tru-images.b-cdn.net/guide-assets/69/69330c2802bea1e86d3ea60f/sheets-competitor-list-tab.png) - Go back to Gumloop and choose the option to start with an integration. - Select Google Sheets Reader, add the link to your sheet, and select the new tab. - Click the “+” button, find the “Time Trigger,” and configure it to run when you want. Weekly should be fine. - Click the “+” button again, find your scraper flow under “Subflow,” and connect it to your Google Sheets module. - Add your Google Sheets “Competitor Profile URL” as the input to your subflow. - Save your flow and run a test if you like. ### Resources - [Analyze Top-Performing Ad Creatives in Minutes](https://tru-images.b-cdn.net/guide-assets/69/6936f4757ec37b2d5b2573c7/analyze-top-performing-ad-creatives-in-minutes.pdf) ### Pro tip > You can use Gumloop’s agent to build the flow for you, or build it manually and ask the agent for help while you’re in a flow. ### Guide: Audit Claude's Context of You and Your Work in 15 Minutes > Audit what Claude thinks it knows about you and your work, correct stale assumptions, update memory, and generate a practical report for improving your AI workflow. - Canonical page: https://app.therundown.ai/guides/audit-claudes-context-of-you-and-your-work-in-15-minutes - Machine-readable page: https://app.therundown.ai/guides/audit-claudes-context-of-you-and-your-work-in-15-minutes.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Productivity & Knowledge Work, Data, Research & Analysis, Automation & Agents - Instructors: Billy Howell - Published: 2026-05-21 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to audit Claude on what it thinks it knows about you and your work. Claude will interview you, clean up stale assumptions, update its memory, and create a report with recommendations for improving your AI workflow and work habits. #### Who This Is Useful For - Heavy Claude or ChatGPT users who use AI across work, side projects, research, and experiments. - Founders, operators, creators, and consultants who want AI advice pointed at their current priorities, not old context. - AI power users who want a lightweight quarterly checkup for their role, KPIs, workflows, tools, and bottlenecks. #### What You Will Build ![final updated work context](https://tru-images.b-cdn.net/guide-assets/192-for-jason/01-final-updated-work-context.png) You will build a work-context audit workflow that produces three useful outputs: - a table of Claude's assumptions about your role, priorities, KPIs, tools, and workflows - an updated picture of your current work context - a Markdown report with recommendations for what to automate, what skills to build, and what to cut A major benefit of going through this process is that you will understand your own work process better too. #### What You Need - A Claude account - Enough Claude history or memory for Claude to have assumptions about your work - 10-15 minutes to answer short interview questions This guide uses Claude, but the same audit works in ChatGPT, OpenClaude, Claude Code, Claude Co-Work, or any other AI you use for real work. ### Step 1 Ask Claude To Audit Its Assumptions Start by asking Claude what it thinks it knows before you correct anything. This is important because the audit only works if Claude separates what it knows from what it is guessing. Use this prompt: ```text Audit your context and memory assumptions about me. Put them in a table with: - what you believe - why you believe it - your confidence level - whether each item is confirmed, an assumption, outdated, or unknown Cover: - my role - priorities - KPIs - tools - workflows - anything you may be over-weighting from old chats or side projects Do not interview me yet. ``` Claude should return a table that shows what it believes about your work. Some of it may be correct. Some of it may be stale. Some of it may be random context from a side project that should not affect your real work. In the recording, Claude correctly understood some core responsibilities, but it also over-weighted one-off projects, personal questions, old trainings, and tools that were not actually part of the main job. That is the point. You are not trying to prove Claude is wrong. You are trying to see what context it is using. > Pro tip: Run this audit anywhere you do serious AI work. Different tools may have different assumptions about you. ### Step 2 Turn The Audit Into An Interview Once Claude shows the table, do not try to correct everything manually. Ask it to turn the uncertain items into an interview. ![claude interview multiple choice](https://tru-images.b-cdn.net/guide-assets/192-for-jason/02-claude-interview-multiple-choice.png) Use this prompt: ```text Now interview me about the assumptions, outdated items, and unknowns from that audit. Ask in rounds. Use multiple-choice questions wherever possible. When multiple choice is not enough, ask for one or two sentences. Cover my role, KPIs, tools, priorities, and work processes. After each round, summarize what changed. ``` The multiple-choice part matters. Without it, this can become a long homework assignment. With it, Claude can ask questions that are fast to answer. For example, it may ask whether a project is still active, whether a workflow is part of your real job, or whether something was just a one-off test. You can usually answer with a number, a letter, or one sentence. > Pro tip: The first two prompts are usually enough. Claude can often walk through the interview rounds on its own after that. ### Step 3 Answer The Interview Rounds Claude should walk through several rounds. In the recorded demo, it moved through: 1. role, mission, and main responsibilities 2. current priorities and what is top of mind 3. follow-up questions about those priorities 4. tools, systems, and where the friction really lives 5. success signals and what "good" looks like ![claude interview workflow questions](https://tru-images.b-cdn.net/guide-assets/192-for-jason/03-claude-interview-workflow-questions.png) Do not worry about answering perfectly. The goal is to correct the big picture. Look for places where Claude thinks a side quest is core work. For example, if you built a random website for a demo, Claude might treat that as an active project. If you asked about travel, personal research, or a one-off tool test, Claude might keep that context around longer than you want. This is where you tell Claude what matters now. If Claude asks an open-ended question, keep the answer short. A sentence or two is enough. The audit should make your work context clearer, not create another huge document you have to maintain. ### Step 4 Let Claude Update Its Memory After the interview, Claude may offer to update its memory. If it does, review the proposed updates and approve the ones that are accurate. ![claude updating memory](https://tru-images.b-cdn.net/guide-assets/192-for-jason/04-claude-updating-memory.png) If Claude does not offer, ask directly: ```text Update your memory based on the corrected context from this audit. ``` This is where the workflow becomes more than a report. You are helping Claude clean up stale assumptions and replace them with a more current picture of your role, responsibilities, priorities, tools, workflows, and success signals. Use judgment here. Do not save sensitive data, private numbers, credentials, client details, or anything you would not want available as general assistant context. Save stable work context only. ### Step 5 Create The Markdown Report Once Claude has the updated context, ask it to create a report. This is the durable artifact from the audit. ![work context audit report](https://tru-images.b-cdn.net/guide-assets/192-for-jason/05-work-context-audit-report.png) Use this prompt: ```text Create a Markdown artifact with a report of the job we completed. Include: - what we audited - what changed in your understanding - recommendations for automations and skills I should build - anything I should cut to improve my work process - any context that should be reviewed again later ``` In the recording, Claude created a report with the updated work context, headline findings, bottlenecks, skill ideas, automation ideas, and things to cut. That is more useful than a generic "about me" file. It gives you a snapshot of how you work right now, where your workflow is noisy, and what Claude thinks would make the biggest difference. Save that Markdown file somewhere you can find it later. You can also use it as the source for an about-me.md file, a Claude Project instruction, or a team operating doc. > Pro tip: Ask for things to cut, not just things to build. The best recommendation may be removing a workflow Claude keeps over-weighting. ### Step 6 Use The Report To Improve Your Workflow The report should not just sit there. Read it like an operations review. Look for three categories: - Automations to build: repeated work Claude noticed during the interview. - Skills or prompts to create: workflows that would be easier if they were repeatable. - Things to cut: projects, habits, tools, or context that distract from your real priorities. ![recommended skills and automations](https://tru-images.b-cdn.net/guide-assets/192-for-jason/06-recommended-skills-and-automations.png) This is where the audit becomes useful for the user, not just the AI. Claude may identify bottlenecks you have been working around for weeks. It may also point out that some "projects" are really old experiments that should stop influencing your priorities. The best outcome is not that Claude knows every detail about you. The best outcome is that Claude knows what to optimize for. ### Going Further Ask Claude to turn the audit into a reusable skill: ```text Turn this interview and report-generation process into a reusable skill. The skill should: - audit your current assumptions about my work - interview me with multiple-choice questions where possible - update your context based on my answers - generate a Markdown report with recommendations - suggest automations, skills, and things to cut ``` Then rerun the audit every quarter. Your role, KPIs, tools, and priorities change over time. Your AI context should change with them. ### Pro tip > Run the audit anywhere you do serious AI work, and save only stable work context instead of sensitive details or one-off project noise. ### Guide: Auto-Generate Video Content From a Blog Post With Manus > Learn how to use Manus to script, generate, and assemble a short promotional video from written content you already have. - Canonical page: https://app.therundown.ai/guides/auto-generate-video-content-from-any-written-content-with-manus - Machine-readable page: https://app.therundown.ai/guides/auto-generate-video-content-from-any-written-content-with-manus.md - Difficulty: beginner - Featured tool or company: Meta - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-03-09 - Updated: 2026-07-23 ### Overview #### The Rundown Learn how to use Manus to script, generate, and assemble a short promotional video from written content you already have. You will produce a 15-20 second portrait-format teaser video, with no video editing experience required. Once Manus starts rendering, generation takes about 5-10 minutes. #### Who This Is Useful For - Content marketers repurposing written content into social video without a production budget - Newsletter brands and publishers driving traffic back to full articles with short teasers - Founders and solo creators who want video output without the time or tools to produce it manually #### What You Will Build A reusable Manus workflow that turns a blog post, article, or press release into a short branded teaser video for Instagram Reels, YouTube Shorts, or TikTok. ![image-2106](https://tru-images.b-cdn.net/guide-assets/69/69af5b5e1347daca8a76c02a/image-2106.png) - A 15-20 second portrait-format teaser video - A short script selected from multiple Manus-generated options - A clear CTA that sends viewers to the full article - A branded closing bumper using your logo #### What You Need - A Manus account - A blog post, article, or press release saved as a PDF or Markdown file - Your brand logo, preferably as a PNG - Optional but recommended: a royalty-free background music file downloaded in advance - Optional: a brand style sheet with colors, fonts, and logo usage - Optional: examples of previous social posts to help Manus match your style #### Going Further - Reuse individual keyframes and clips from the project canvas for story posts, ad creatives, or thumbnails. - Upload B-roll footage, brand photography, and pre-cleared music to Google Drive. - Connect Manus to Google Drive through the integrations panel. - Update your project instructions so Manus pulls from those assets before generating new ones. - Once your preferences are dialed in, use “Make this process into a reusable skill” inside the project to lock in the workflow for that project. ### Step 1: Save Your Content as a File Open your article in a browser and save it as a PDF: File > Print > Save as PDF. Upload a file instead of pasting text. It is fewer steps for the AI and produces more reliable results. PDF and Markdown both work. > Pro tip: The richer your source content, the better Manus’s script options will be. A 500-word article with a clear narrative arc will produce stronger hooks than a short press release with only facts and figures. ### Step 2: Set Up Your Manus Project This is a one-time setup. Everything you configure here carries over to every video you generate in this project. - Go to manus.im > Projects > New Project. - Name your project, such as “Social Videos — [Your Brand].” - Under Instructions, paste the project instruction below. ```text You are a video producer for [your brand name]. Your job is to transform written content into branded videos. ``` Open the Files panel and upload your brand logo. Manus will use it as the closing bumper on every video. If you have a brand style sheet with colors, fonts, or logo usage, upload that too. > Pro tip: Upload examples of your previous social posts alongside your logo. Manus can reference them to match the text overlay style and visual pacing you already use. ### Step 3: Write Your Prompt Open a new task inside your project. Attach your PDF, then use this prompt: ```text Create a 15-20 second teaser video based on the attached article. The CTA is to read the full article. Generate it in portrait orientation for Instagram Reels. Give me four script options before you generate the video. Use default mode. Add background music and a voiceover. ``` Asking for four script options first helps you pick the best angle before rendering begins. Video generation costs 499-800 credits per run, while asking for options first costs almost nothing. Use default mode for test runs because it uses roughly five times fewer tokens than quality mode. Switch to quality mode only for your final render. Each default-mode run also trains Manus’s project memory. > Pro tip: Be specific about audio. Always tell Manus whether you want a voiceover, background music, or both. Leaving it open produces unpredictable results. ### Step 4: Pick Your Script Manus returns four script options before generating anything. Read each one and pick the angle most likely to stop someone mid-scroll. ![image-2107](https://tru-images.b-cdn.net/guide-assets/69/69af5ba51347daca8a76c947/image-2107.png) For most content, the most dramatic or emotionally resonant hook outperforms the factual summary. Reply with your chosen option number and any tweaks you want before Manus starts rendering. ### Step 5: Review and Reuse the Generated Video Generation takes 5-10 minutes. When the video is ready, check the final result carefully. - Does the hook land in the first two seconds? - Is the voiceover pacing comfortable? - Does the CTA clearly direct viewers to the full article? - Does the closing bumper appear cleanly at the end? You can ask Manus to revise specific sections without regenerating the entire video. The individual keyframes and clips also appear as standalone assets in your project canvas, which you can reuse for story posts, ad creatives, or thumbnails. A note on voiceover: Manus does not have a precise sense of where each sentence lands within a clip. Keep the script short and tight. If the voiceover sounds off, it is almost always a script length issue. ### Pro tip > Upload examples of your previous social posts alongside your logo. Manus can reference them to match the text overlay style and visual pacing you already use. ### Guide: Automate Any Manual Task With Codex > Learn how to use Codex Computer Use to handle repetitive, click-heavy work on your computer. - Canonical page: https://app.therundown.ai/guides/automate-any-manual-task-with-codex - Machine-readable page: https://app.therundown.ai/guides/automate-any-manual-task-with-codex.md - Difficulty: beginner - Featured tool or company: OpenAI - Categories: Automation & Agents, Coding & App Development, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-04-28 - Updated: 2026-07-23 ### Overview #### The Rundown Learn how to use Codex Computer Use to handle repetitive, click-heavy work on your computer. You will produce a repeatable workflow for delegating a narrow manual UI task, supervising the first item, and then scaling it to a small batch for a practical time win. #### Who This Is Useful For - Operators and admins cleaning up repetitive backlogs in web apps - Creators and production teams moving through repetitive media workflows - Anyone with a click-heavy process who does not want to build a custom automation for it - This works best when the task is repetitive, the interface is stable, and the right action is visually obvious. #### What You Will Build You will build a repeatable workflow for delegating a manual UI task to Codex. In the video guide, this was tested on a batch video backup task in Google Drive. ![image-2185](https://tru-images.b-cdn.net/guide-assets/69/69f128690d23103ebba5560d/image-2185.png) - Computer Use configured with the right permissions - A cleaned-up workspace with only the target app or window visible - An SOP-style prompt with the exact app, page, action, loop, batch size, and uncertainty behavior - A first-item validation routine before scaling to a small batch #### What You Need - A paid ChatGPT plan with Codex access - Permission to grant Screen Recording and Accessibility on macOS - A repetitive task with a stable interface and visually obvious correct action - The target app, page or view, button or menu label, loop instruction, and batch size #### Going Further - After one manual workflow is stable, reuse the tested setup by turning it into a Codex automation that still uses Computer Use. Keep the job deterministic, watch the first runs closely, and use this approach for boring work that eats time rather than work that needs judgment. ### Step 1 Turn on Computer Use Open Codex settings, find Computer Use, and click Install to install the plugin. ![image-2186](https://tru-images.b-cdn.net/guide-assets/69/69f12875e926b2b0c80d3688/image-2186.png) When macOS prompts you, grant the required permissions. - Screen Recording, so Codex can see the target app - Accessibility, so Codex can click, type, and navigate These system permissions are separate from Codex app approvals. System permissions let Codex operate apps at all. App approvals decide which apps Codex is allowed to touch. > Pro tip: Computer Use is great for repetitive tasks that keep showing up but do not have a clean API-based workflow in Codex or ChatGPT. ### Step 2 Start a new task with Full access Create a new task and toggle on Full access. ![image-2187](https://tru-images.b-cdn.net/guide-assets/69/69f128900d23103ebba55eb2/image-2187.png) ### Step 3 Clean up the environment before you start Before Codex touches anything, reduce the surface area. - Close sensitive tabs and unrelated apps - Use a dedicated browser profile if possible - Keep only the target app or window visible - Decide the batch size in advance When the screen is clean and the task is narrow, Codex has less room to misread the situation. ### Step 4 Write the task prompt like an SOP A vague instruction gets vague behavior. A tight operating procedure gets much better results. Your prompt should include: - The exact app - The exact page or view - The exact button or menu label - The loop instruction - The batch size - What Codex should do when it is unsure Here is a good example: ![CleanShot 2026-04-28 at 17.39.22@2x](https://tru-images.b-cdn.net/guide-assets/69/69f12933e926b2b0c80d37c9/cleanshot-2026-04-28-at-17-39-22-2x.png) The key is to tell Codex which apps to use, what URL to go to, and to specify that it should use computer control for the task. ### Step 5 Let Codex do one item first Do not launch a giant batch immediately. ![image-2188](https://tru-images.b-cdn.net/guide-assets/69/69f129003f1004948bf6d661/image-2188.png) Have Codex complete one item first. Watch the screen and confirm three things: - It is on the right page - It chose the right row or target - It clicked the right action Once that first action is correct, let it continue through the rest of the batch. This first-item check is the simplest way to catch a misunderstanding before it becomes a mess. ### Step 6 Scale to a small batch Once the first action is verified, let Codex continue through a small batch like 5, 10, or 20 items. That gives you a practical time win without making the session hard to supervise. It also makes troubleshooting easier because you can stop, adjust the prompt, and rerun the next batch if something drifts. This is a better operating model than telling it to clear an entire backlog with no guardrails. ### Step 7 Turn the tested workflow into an automation Once one manual workflow is stable, you can turn it into an automation that still uses Computer Use. Automate these tasks after you have tested them live once. The workflow from the demo is: - Create a project for the task - Run the task manually once inside that project with Full access enabled - Create an automation from that existing project - Give it a clear title and restate the job in plain English - Keep the automation set to Local - Choose a smaller model if the task is simple and repetitive - Hit Run now and watch the first few runs closely Codex remembers the setup from the project you already tested. That makes it easier to automate a known-good workflow than to start from scratch inside the automation builder. In the video guide, the demo used a recurring backup task: tell Codex to back up videos to Safari using computer control, give it the Google Drive link, and let it repeat the same upload behavior already verified in the live run. Watch the first couple runs, make sure it is not drifting, and only automate jobs that are deterministic enough to trust. Use it for the boring work that eats time, not for work that needs judgment. ### Pro tip > Don't start with a messy, high-risk task. Pick something boring and predictable first. ### Guide: Automate Any Manual Task With This OpenAI Feature > In this guide, you will learn how to automate any repetitive work task that you do on your computer. - Canonical page: https://app.therundown.ai/guides/automate-any-manual-task-with-this-openai-feature - Machine-readable page: https://app.therundown.ai/guides/automate-any-manual-task-with-this-openai-feature.md - Difficulty: beginner - Featured tool or company: Codex - Categories: Automation & Agents, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-06-29 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to automate any repetitive work task that you do on your computer. OpenAI's new "Record & Replay" tool allows you to teach their Codex agent how to do anything simply by recording your screen while you do the task. #### Who This Is Useful For - Marketers, SEO teams, and analysts who pull recurring reports from tools like Google Search Console. - Creators and operators who upload videos, organize files, rename exports, or move assets between a browser, Finder, and another app. - Anyone with a repeatable computer task who wants to teach an agent by showing the workflow once instead of rebuilding brittle automations. #### What You Will Build ![reusable skill created](https://tru-images.b-cdn.net/guide-assets/local-291/automate-any-manual-task-with-this-openai-feature/01-reusable-skill-created.png) You will set up the Record & Replay plugin in Codex, record your screen while you complete one repeatable task, and let Codex turn that recording into a skill. That skill can reopen the same apps, follow the same browser or desktop steps, and automate the work you recorded. In the video, the example workflow exports a Google Search Console report, applies a country filter, downloads a CSV, and moves the file into the right client folder. Your version could be a monthly analytics export, a YouTube upload checklist, a file-packaging workflow, a CRM cleanup task, or any other repeatable process you normally do by hand. #### What You Need To Get Started - The Codex desktop app (https://developers.openai.com/codex/app) - A ChatGPT account with Codex access - A Mac, because Record & Replay is a macOS workflow - Computer Use enabled in Codex - The Record & Replay plugin installed - One safe, repeatable task you can record without exposing private data #### Going Further ![automation schedule](https://tru-images.b-cdn.net/guide-assets/local-291/automate-any-manual-task-with-this-openai-feature/05-automation-schedule.png) Once the skill works, turn it into a recurring automation. You can do this manually from the Codex schedule menu by creating a new automation, choosing the project, and writing an instruction that invokes the skill. The easier path is to stay in the same Codex thread and ask for it directly: ```text Create an automation that runs this skill every [SCHEDULE] for [CLIENT_OR_WORKFLOW]. Use the same project and inputs we just tested. After it runs, summarize what changed and tell me if anything needs review. ``` For a reporting workflow, that might run every Sunday or on the last day of each month. For a creator workflow, it might prepare an upload folder every Friday. For operations, it might export records, clean up files, or prepare a status packet on a recurring schedule. You can also ask Codex to audit or edit the automation from the same thread: ```text Review the automation you just created. Make sure it invokes the correct skill, uses the right schedule, and stops for review if the output is missing or the site changes. ``` Start with one useful workflow. Record it cleanly, supervise the first replay, improve the skill, and only then schedule it. ### Step 1 Download Codex Download and open the Codex desktop app. If you already use Codex, check for updates before recording. Before you start, set up a clean place for the workflow: - Open the folder where the work should happen, like a reporting folder, client folder, video-production folder, or downloads cleanup folder. - Add that folder as a Codex project if you want the skill to work with local files. - Pick a one- or two-minute task for the first recording. - Choose something with a clear success state, like a saved CSV, uploaded video, renamed file, or completed export. This works best when the task is short, repeatable, and easy to verify. If the steps are mostly the same each time, but the client, date range, folder, or output name changes, it is a good candidate. > Pro tip Start with the most boring version of the workflow. Once the simple version works, you can teach Codex the extra edge cases. ### Step 2 Set Up Computer Use Before Record & Replay can watch and replay a desktop workflow, Codex needs Computer Use. Open Codex, go to Plugins, find Computer Use, and install it. When macOS prompts you, approve the permissions Codex needs to see and operate your screen, including Screen Recording and Accessibility. Computer Use is what lets Codex move the mouse, click buttons, type into fields, use Finder, and operate a browser. Record & Replay uses that capability to turn your demonstration into a skill. > Pro tip Close private tabs, email, Slack, password managers, customer data, and anything else you do not want captured before you start recording. ### Step 3 Install Record & Replay ![record replay plugin](https://tru-images.b-cdn.net/guide-assets/local-291/automate-any-manual-task-with-this-openai-feature/02-record-replay-plugin.png) In Codex, open Plugins, search for Record & Replay, and install the plugin. Depending on the app version, you may start from Plugins > + > Record a skill, or you may invoke the installed plugin directly in a thread with /record and replay. Use a prompt like this: ```text Record me doing this task and turn it into a reusable skill. ``` If you already know the task name, add it: ```text Use Record & Replay to turn this [TASK_NAME] workflow into a reusable skill. ``` For the Search Console demo, that task was a recurring reporting workflow. For your own version, replace it with the real job: exporting a report, uploading a video, moving files, updating a dashboard, or packaging a client deliverable. ### Step 4 Record The Workflow Once ![recording workflow](https://tru-images.b-cdn.net/guide-assets/local-291/automate-any-manual-task-with-this-openai-feature/03-recording-workflow.png) When recording starts, do the task exactly as you normally would. Move slowly enough for Codex to see the path. Open apps from the dock if that is how you normally start. Paste the website URL into a fresh browser tab if the site matters. Navigate through the menus, choose the right filters, export the file, rename it, move it, and verify the final result. In the video example, the workflow was: - Open Google Search Console. - Go to the performance report. - Select the reporting window. - Apply a country filter. - Export a CSV. - Move the download into the right reporting folder. - Rename the file with a clear convention. That same shape works for many other tasks. You could record yourself uploading a YouTube video, exporting monthly financial reports, saving invoices into a client folder, pulling CRM records, or moving finished assets into a delivery folder. ![search console export menu](https://tru-images.b-cdn.net/guide-assets/local-291/automate-any-manual-task-with-this-openai-feature/04-search-console-export-menu.png) When the result is complete, hit the red stop button. Codex will automatically send the completion message, so you do not need to type "I'm done recording" yourself. > Pro tip If something goes wrong during the recording, restart. In the video, Safari glitched during the first attempt, so the recording was discarded with the trash can button and started again. Do the same if you click the wrong thing, expose the wrong data, or take an extra path you do not want the skill to learn. ### Step 5 Review The Skill Codex Creates After you stop recording, Codex inspects the actions and drafts a reusable skill. It may identify the key choices you made, like the report path, filter, export type, naming convention, or destination folder. Review the skill before trusting it. Check that it captured: - what the skill is for - which inputs should change each run - which steps should stay fixed - how to name or save the output - how to verify success - what it should avoid doing If the skill is too specific, ask Codex to generalize it. If it is too broad, ask Codex to narrow it. Use this prompt: ```text Improve this skill based on the recording. Make it reusable for the same kind of task with new inputs. Keep the steps narrow and practical. Include success checks. Do not include accidental clicks, glitches, or one-off detours from the recording. ``` ### Step 6 Test It And Babysit The First Run Run the skill on a new input before you turn it loose. In a new thread, type / and select the skill Codex created. Give it the next version of the task: a new client, date range, file, upload, folder, report, or web app. Then watch it work. Codex may move the mouse around your browser or desktop, click the export button, download a file, open Finder, or move files into place. This is the replay part. Babysit the first run closely. In the video, Codex completed the workflow but was not fully sure it had succeeded. When that happens, tell it what you can see: ```text You succeeded. The output is in the correct folder and the file name is right. Use what you learned from this test run to improve the skill and make it less fragile. ``` You can also steer the implementation: ```text Use Computer Use where possible instead of relying on fragile scripts, unless a script is clearly safer for this step. ``` The first test run is where the skill gets better. Watch whether Codex repeats the clicks correctly, then teach it what success looks like. ### Guide: Automate Email Expense Tracking with Claude for Chrome > In this guide, you’ll learn how to track and label expenses from Gmail into Google Sheets in a couple of minutes using Claude for Chrome. - Canonical page: https://app.therundown.ai/guides/automate-email-expense-tracking-with-claude-for-chrome - Machine-readable page: https://app.therundown.ai/guides/automate-email-expense-tracking-with-claude-for-chrome.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Automation & Agents, Business Operations & Leadership, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-01-07 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you’ll learn how to track and label expenses from Gmail into Google Sheets in a couple of minutes using Claude for Chrome. You’ll produce a populated expense-tracking sheet and use Claude as a personal assistant without building complex automation chains. #### Who This Is Useful For - Small business owners - Freelance consultants - Anyone with monthly expense reports #### What You Will Build You’ll build a reusable Claude for Chrome workflow that finds software subscription receipts in Gmail, labels relevant emails, archives non-receipts, and fills a Google Sheet with expense details. - A Google Sheet with expense headers such as Expense Title, Company, Amount, Date, and Renewal Date - Rows populated from software subscription receipts - A total row at the bottom of the sheet - A repeatable prompt you can reuse for future months ![01_claude_chrome_expenses_what_you_will_build](https://tru-images.b-cdn.net/guide-assets/69/695eb90727eb69f02873d9c5/01-claude-chrome-expenses-what-you-will-build.png) #### What You Need - Claude account, pro or above - Claude for Chrome extension - Google Chrome desktop app - Google Sheets - Access to the Gmail inbox you want Claude to search - A month and year to search for software subscription receipts #### Going Further - Save your prompt as a shortcut by typing / and selecting “Create shortcut.” You can keep using the same Google Sheet and tell Claude which month to run each time you use it. - Once you have a shortcut set up, you can have it run daily or weekly on a schedule. ![05_going_further](https://tru-images.b-cdn.net/guide-assets/69/695eb90527eb69f02873d99d/05-going-further.png) ### Step 1: Set up Claude for Chrome Install Claude for Chrome from the Chrome Web Store. ![01_claude_chrome_store](https://tru-images.b-cdn.net/guide-assets/69/695eb90727eb69f02873d9bd/01-claude-chrome-store.png) ### Step 2: Create your Google Sheet Create a Google Sheet with the headers you want Claude to populate. In this example, use: Expense Title, Company, Amount, Date, and Renewal Date. > Pro tip: You can add other headers like “Category” or “Reimbursable?” and Claude for Chrome will fill them out. ![02_sheetheaders](https://tru-images.b-cdn.net/guide-assets/69/695eb90627eb69f02873d9b5/02-sheetheaders.png) Copy the link to your Google Sheet. You’ll use it in the next step. ### Step 3: Prompt Claude to process receipts In a new Chrome tab, click Extensions → Claude. ![03_open_extension](https://tru-images.b-cdn.net/guide-assets/69/695eb90627eb69f02873d9ad/03-open-extension.png) You should see your current tab highlighted. This means Claude can perform work in the tab. Any tabs Claude opens will also be highlighted. ![04_active_tab](https://tru-images.b-cdn.net/guide-assets/69/695eb90527eb69f02873d9a5/04-active-tab.png) Paste the following prompt into the Claude for Chrome chat box. Make sure you add your Google Sheet link. ```text Look through my Gmail inbox and find all software subscription receipts from [Month + Year]. I have a Google Sheet already created at [YOUR GOOGLE SHEET URL HERE] with headers: Expense Title, Company, Amount, Date, Renewal Date. Navigate to it and fill in each row with the extracted information from the receipts. Add a total row at the bottom. For each email you process: add the “receipt” label to any receipts or subscription sign-ups, and archive everything else that isn’t a receipt, including marketing emails and newsletters. ``` Let Claude work for 10–15 minutes. You may have to prompt it to continue if you have a lot of emails. ### Resources - [Automate Email Expense Tracking with Claude for Chrome](https://tru-images.b-cdn.net/guide-assets/69/695ec294305d2ec845b47c45/automate-email-expense-tracking-with-claude-for-chrome.pdf) ### Pro tip > You can add other headers like “Category” or “Reimbursable?” and Claude for Chrome will fill them out. ### Guide: Automate Financial Research with the OpenClaw of Stocks > Set up Dexter as a single-purpose stock research agent for filings, earnings, market data, and recurring watchlist checks. - Canonical page: https://app.therundown.ai/guides/automate-financial-research-with-the-openclaw-of-stocks - Machine-readable page: https://app.therundown.ai/guides/automate-financial-research-with-the-openclaw-of-stocks.md - Difficulty: intermediate - Featured tool or company: Dexter - Categories: Data, Research & Analysis, Automation & Agents - Instructors: Billy Howell - Published: 2026-06-09 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to set up Dexter, an open-source financial research agent that works like OpenClaw for stock research. Dexter can read earnings reports, interpret SEC filings, pull up-to-date market data, and help you turn a messy company question into a source-backed research brief. #### Who This Is Useful For - Investors and analysts who want a faster first pass on earnings, filings, stock prices, and company context. - Founders and operators who need company research before sales calls, diligence, partnerships, or competitor reviews. - Agent builders who want to see how a narrow, single-purpose agent can be easier to trust than a general-purpose assistant. #### What You Will Build You will set up Dexter as an always-on command-line analyst for the stock market. Once it is running, you can ask plain-English questions about a company, ticker, earnings report, filing, competitor, or market move, and Dexter will pull the research it needs before answering. ![Dexter earnings comparison](https://tru-images.b-cdn.net/guide-assets/local-218/automate-financial-research-with-the-openclaw-of-stocks/01-dexter-earnings-comparison.png) Use it to compare stocks, read SEC filings, summarize earnings, pull current market data, and run follow-up research without opening a new browser tab every time. You can also leave it running with /heartbeat, which turns it into a 24/7 analyst for recurring watchlist checks. #### What You Need To Get Started - A VPS or spare computer where Dexter can run without slowing down your main machine - A terminal-based agent such as Codex or Claude Code to help with setup - The Dexter GitHub repo - An API key from OpenRouter, OpenAI, or Anthropic - A Financial Datasets API key; expect to add about $20 to get started - Optional: a Perplexity API key for web search and an xAI API key if you want Twitter/X sentiment #### Going Further Connect Dexter to X so it can add Twitter sentiment to the research brief. A good next prompt would be: ```text Add Twitter/X sentiment to this research workflow for [COMPANY_NAME]. Use it only as one signal. Compare sentiment against filings, earnings, market data, and recent news before drawing conclusions. Return: - the strongest positive sentiment themes - the strongest negative sentiment themes - whether the market data supports or contradicts the sentiment - what I should verify before acting on it ``` You can also adapt the same setup for earnings prep, competitor monitoring, diligence, or sales research. Keep the same pattern: ask a narrow question, make Dexter pull evidence, inspect the sources, then automate only the checks you actually want repeated. ### Step 1 Set Up a VPS for Dexter In the demo, we used Hostinger VPS hosting and picked a basic VPS plan. Once the server is ready, Hostinger gives you an SSH command that looks like this: ```text ssh root@[YOUR_SERVER_IP] ``` > Pro tip: You can run Dexter locally, but a VPS is cleaner for this kind of agent. It keeps the browser, terminal, and long-running tasks off your main laptop. If setting up a VPS is new for you, start with our Launch Your OpenClaw From Scratch course. It walks through the same basic server setup pattern you will use here. Open a new Codex or Claude Code project, open the terminal, and connect to the server with that SSH command. The nice part is that your coding agent can watch the terminal with you and help debug installs, missing packages, or environment variables. ![Hostinger VPS setup](https://tru-images.b-cdn.net/guide-assets/local-218/automate-financial-research-with-the-openclaw-of-stocks/02-hostinger-vps-setup.png) In the demo, AlmaLinux worked, but Debian would be the safer pick next time. ### Step 2 Install Dexter From GitHub Open the Dexter GitHub repo and follow the install instructions there. The setup is mostly a few terminal commands: install Bun, clone the repo, install dependencies, add environment variables, and start Dexter. The general flow looks like this: ```text # install Bun if you do not already have it curl -fsSL https://bun.sh/install | bash # clone Dexter and install dependencies git clone https://github.com/virattt/dexter.git cd dexter bun install # start Dexter bun start ``` When Dexter starts for the first time, it should walk you through the variables it needs. Treat those variables like passwords. Do not paste them into a public recording, shared doc, or screenshot. After setup, the command you will use most often is: ```text cd dexter bun start ``` That reopens the Dexter interface from the folder where it lives on your VPS. ### Step 3 Add Your Model and Financial Data Keys Now give Dexter the two things that make it useful: a model and stock-market data. For the model, use whichever API key you already prefer: OpenRouter, OpenAI, or Anthropic. In the demo, OpenRouter was useful because it can route to OpenAI, Anthropic, and other models from one account. Inside Dexter, open the model picker with: ```text /models ``` Choose your provider, then paste or type the model name carefully. In the demo, straight pasting a model name caused issues, so typing it out was the safer move. ![Dexter provider picker](https://tru-images.b-cdn.net/guide-assets/local-218/automate-financial-research-with-the-openclaw-of-stocks/03-dexter-provider-picker.png) Next, add a Financial Datasets API key. This is the key that lets Dexter pull company data, earnings reports, SEC filings, stock prices, and other financial context instead of relying only on web search. ![Financial Datasets API key](https://tru-images.b-cdn.net/guide-assets/local-218/automate-financial-research-with-the-openclaw-of-stocks/04-financial-datasets-api-key.png) Financial Datasets charges per use, and the demo notes that the minimum credit load is about $20 to get started. If you want web search, add a Perplexity API key and configure search inside Dexter with: ```text /search ``` If you want Twitter/X sentiment later, add an xAI API key too. ### Step 4 Run the First Research Prompt Start with a narrow research question. Dexter works better when you give it a concrete company and decision context instead of asking it to "tell me about this stock." Use this prompt: ```text Create a source-backed research brief on GOOGL. Focus on: - what the company does - recent catalysts - public financial or operating signals - key risks - competitive context - what I should investigate next Use an iterative research process. Validate important claims before finalizing the brief. Save intermediate research so the workflow can continue without losing context. ``` You can also ask faster questions while testing the setup: ```text What's the stock price of GOOGL? ``` or: ```text How did GOOGL do on its last earnings? How did that compare to META's last earnings? ``` In the demo, Dexter pulled market data, read filing and earnings fields, then returned a quick comparison of revenue, EPS, and growth. That is the workflow you want to verify before trusting bigger research prompts: ask a concrete question, watch which data sources Dexter calls, and make sure the answer cites the right kind of evidence. ### Step 5 Set Up a Heartbeat Watchlist Once a manual run works, use Dexter's heartbeat feature for recurring checks. Use a prompt like this: ```text Set up a heartbeat to check [TICKER] every [TIME_WINDOW]. When you check it, report: - current price movement - any major earnings or filing updates - relevant market news - whether anything needs deeper research ``` In the demo, the heartbeat request was a simple six-hour stock price check. Dexter added it to its heartbeat checklist so the agent could keep checking that stock on a schedule. This is where Dexter starts to feel different from a normal chatbot. You are not opening a new thread every time you want a quick check. You are giving a single-purpose finance agent one recurring job. > Pro tip: Prove the manual workflow before you automate it. If the first GOOGL answer is noisy, fix the model, data, or prompt before you ask Dexter to monitor anything. ### Step 6 Skip WhatsApp Until the Core Workflow Works Dexter also has gateway instructions for WhatsApp-style access, but that part did not work cleanly in the demo. The setup required exiting Dexter with Ctrl+C, returning to the normal terminal, and trying: ```text bun gateway login ``` The gateway setup had more friction than the core stock research workflow. Treat that as optional for now. The stronger first use is simple: 1. Run Dexter on a VPS. 2. Add the right model and market data keys. 3. Ask source-backed questions. 4. Check the data sources it used. 5. Add /heartbeat only after the answers are useful. That keeps Dexter focused on what it is already good at: stock research, market data, filings, earnings context, and recurring watchlist checks. ### Guide: Automate Marketing Assets With ChatGPT Image 2.0 > Use Codex Desktop to build a GPT Image 2 app that turns a marketing campaign and design brief into prompts, test images, and a review gallery. - Canonical page: https://app.therundown.ai/guides/automate-marketing-assets-with-chatgpt-image-20 - Machine-readable page: https://app.therundown.ai/guides/automate-marketing-assets-with-chatgpt-image-20.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: Marketing & Growth, Content & Creative, Automation & Agents - Instructors: Billy Howell - Published: 2026-05-14 - Updated: 2026-09-11 ### Overview #### The Rundown Use Codex Desktop to build a GPT Image 2 app that turns a marketing campaign and design brief into prompts, test images, and a review gallery. #### Who This Is Useful For - Content teams that need social, newsletter, course, or launch images in batches - Marketers who want to test multiple image directions from one brief - Founders and operators who want a reusable local workflow instead of scattered downloads - Codex Desktop users who want to build and test a useful browser app without starting from scratch #### What You Will Build You will build a Node app that runs locally in the browser. ![Finished Review Grid Generated Images](https://tru-images.b-cdn.net/guide-assets/local-207/automate-marketing-assets-with-chatgpt-image-20/01-finished-review-grid-generated-images.png) The app should let you describe a campaign, choose image dimensions, generate prompts, create images, and review the results in one place. In the demo, the app saves generated images locally and gives each card simple review controls like Keeper, Reject, Clear, and Notes. #### What You Need to Get Started - Codex Desktop - Node installed on your computer - A new project folder - An OpenAI API key or OpenRouter API key - A small campaign idea to test Pro tip: we like OpenRouter for keys, and you can use the new ChatGPT Image 2.0 model through OpenRouter too. Before generating real images, make sure the account you use has billing or credits enabled. #### Going Further Once the app works, use Codex Desktop Annotation Mode to refine the interface. ![Codex Desktop Annotation Mode](https://tru-images.b-cdn.net/guide-assets/local-207/automate-marketing-assets-with-chatgpt-image-20/07-codex-desktop-annotation-mode.png) Click or drag over anything you want fixed: labels, spacing, review states, controls, or layout. Then ask Codex to apply the fixes and retest the app. This is where Codex Desktop is especially useful. Instead of describing a vague UI problem from memory, you can point at the exact part of the local app that needs work. Good next upgrades: - add a history view for past campaigns - add a regenerate button for rejected images - add saved brand profiles - export review notes as CSV - create contact sheets for faster review - add OpenRouter as a selectable provider Keep the improvements practical. The win is a local image testing workflow you can reuse for future campaigns, not a bigger app than your team actually needs. ### Step 1 Create a Project Folder Start with a clean folder so Codex has a clear place to write the app. ```text mkdir gpt-image-review-app cd gpt-image-review-app git init ``` The git init step gives you a checkpoint before Codex starts creating files. If the first version drifts, you can inspect the changes and recover more easily. Open that folder in Codex Desktop. You can use Codex from the terminal too, but Desktop is useful here because you can run the app, preview it in the browser, and give visual feedback later. ### Step 2 Ask Codex to Build the App Paste this prompt into Codex: ![Codex Local App Build Prompt](https://tru-images.b-cdn.net/guide-assets/local-207/automate-marketing-assets-with-chatgpt-image-20/02-codex-local-app-build-prompt.png) ```text Build a local browser app for bulk image generation with GPT Image 2.0. Use a clean editorial studio theme. Let me describe a marketing campaign and design brief, choose image dimensions, generate editable prompts, create test images through OpenAI, and review results with Keep, Reject, Needs revision, and Notes fields. Create the files, show me how to add my OpenAI API key to .env, and test the flow. ``` This keeps the request focused. You are asking Codex for a local browser app, not a terminal-only script. You are also asking for the API key to live in .env, which keeps secrets out of the frontend code. If you want to use OpenRouter, say that directly in the prompt. You can also paste the relevant model or endpoint reference into Codex. New image models can be easy for coding agents to miss, so giving Codex the exact provider details can save a few rounds of cleanup. ### Step 3 Add Your API Key After the first build, get an OpenAI API key and add it to the app's .env file. ![Codex Api Key Env Setup](https://tru-images.b-cdn.net/guide-assets/local-207/automate-marketing-assets-with-chatgpt-image-20/03-codex-api-key-env-setup.png) Codex should usually create a .env.example file. Copy that file to .env, then paste your key into the matching variable. ```text cp .env.example .env ``` Your app may use a variable like: ```text OPENAI_API_KEY=... ``` If you are using OpenRouter, the app may use a second variable like: ```text OPENROUTER_API_KEY=... ``` Do not paste API keys into browser JavaScript or commit them to git. The local server should read the key from .env and make the image request from the backend. ### Step 4 Run the App Install dependencies and start the local server: ```text npm install npm run dev ``` Open the localhost URL Codex gives you. It will usually be something like: ```text http://localhost:3000 ``` If Codex Desktop can run and test the app for you, let it. The terminal output helps Codex debug broken routes, missing packages, and API errors. ### Step 5 Describe the Campaign In the app, describe the marketing campaign and design brief you want to test. ![Local App Campaign Brief Input](https://tru-images.b-cdn.net/guide-assets/local-207/automate-marketing-assets-with-chatgpt-image-20/04-local-app-campaign-brief-input.png) Keep the brief short, but include enough direction to shape the images: ```text Create a launch campaign for a local, golf-themed coffee shop. The visuals should feel elegant, modern, useful, and credible. Avoid generic robot imagery. Show interior, exterior, and product shots. ``` The brief is the control panel for the workflow. If the prompts feel generic later, come back here and tighten the audience, visual style, required subjects, or things to avoid. ### Step 6 Choose Image Dimensions Choose the dimensions you want before generating prompts. ![Local App Image Dimensions Settings](https://tru-images.b-cdn.net/guide-assets/local-207/automate-marketing-assets-with-chatgpt-image-20/05-local-app-image-dimensions-settings.png) For a real campaign, you might test: - square images for social posts - portrait images for stories or shorts covers - landscape images for headers or thumbnails Start small. Two formats with two images each is enough to prove the workflow before spending more credits. ### Step 7 Generate Prompts Click Generate Prompts before you create images. ![Local App Generated Prompts Review](https://tru-images.b-cdn.net/guide-assets/local-207/automate-marketing-assets-with-chatgpt-image-20/06-local-app-generated-prompts-review.png) This step matters because it lets you catch weak prompt direction before you spend credits on image generation. Read the generated prompts and edit anything that feels too vague, too generic, or too far away from the campaign. Check for three things: - the prompt matches the campaign brief - the selected image dimension makes sense - the prompt avoids things you explicitly did not want Once the prompts look useful, click Generate Images to Test. ### Step 8 Review the Results When the images finish, review them in the gallery. Use the review controls while the reasoning is fresh. Mark the obvious keepers, reject anything that misses the brief, and leave notes on images that are close but need another pass. The transcript called out one practical detail: image generation can take a while, especially through provider endpoints. Do not assume the app is broken just because a real batch takes longer than the prompt preview. If generation fails, check the simple things first: - the API key is in .env - billing or credits are enabled - your account or organization can access the image model - the app is calling the provider you intended to use - the terminal does not show a missing package or route error ### Guide: Automate Tasks With Claude’s “Record a Skill” Feature > In this guide, you will learn how to record a task with Claude's Record a Skill and turn it into an automation. - Canonical page: https://app.therundown.ai/guides/automate-tasks-with-claudes-record-a-skill-feature - Machine-readable page: https://app.therundown.ai/guides/automate-tasks-with-claudes-record-a-skill-feature.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Automation & Agents, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-07-25 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to record a task with Claude's Record a Skill and turn it into an automation. You do not need to know the best way to complete it or perform every step perfectly. #### Who This Is Useful For - Operators and admins who repeat the same file, folder, or browser workflow every week. - Creators and marketing teams who move finished work between local apps, shared folders, and publishing tools. - Anyone with a manual task who knows what the finished result should look like but not whether to use computer use, a connector, or a script. #### What You Will Build You will build a reusable Claude skill from a task you demonstrate on your computer. Claude will study the task, suggest the best way to automate it, and turn the useful parts into instructions you can review, save, and test. ![Claude skill proposal with its upload workflow and safety guardrails visible](https://tru-images.b-cdn.net/guide-assets/local-413/automate-tasks-with-claudes-record-a-skill-feature/01-skill-proposal-and-guardrails.png) #### What You Need To Get Started - A Pro, Max, or Team Claude plan. Record a Skill is not currently available on Free or Enterprise plans. - The latest version of Claude for Mac with access to Cowork. - macOS Screen Recording and Accessibility permissions for Claude. - A task you already perform, preferably with local files or desktop apps. - Optional access to the Claude in Chrome extension (https://support.claude.com/en/articles/12012173-get-started-with-claude-in-chrome) for workflows that use websites. #### Going Further Once the skill works reliably, add it to a daily or weekly recurring task in Claude Cowork. Start with a low-risk workflow and review the first few runs before letting it operate on its own. Use a request like this: ```text Use my [SKILL_NAME] skill every [DAY_OR_FREQUENCY] to process [INPUT]. Save the finished result to [OUTPUT_LOCATION]. Ask before [RISKY_OR_IRREVERSIBLE_ACTION]. ``` Workflows that depend on local files, browser control, or desktop apps may still need the connected computer available when the task runs. Connector-based or remote workflows are usually easier to schedule. ### Step 1 Pick a Task Pick a task you already know how to complete. You do not need to know the best way to automate it yet. The recording gives Claude enough context to help you choose that later. The strongest first tasks have three things: - a clear starting point - a visible series of actions - a clear finished result Start with a local workflow when possible. You might organize a folder, rename and move finished files, turn meeting notes into project tasks, or connect two desktop apps that do not work well together. Use sample data for your first recording. Close private conversations, passwords, customer information, and anything else you do not want captured. Claude records what appears on your screen and anything you say while the microphone is on. Our original test tried to move a large local video into YouTube through Safari. Claude learned the steps, but the saved workflow could not complete the upload because Safari did not support the needed browser actions and the Chrome path could not handle the large local file. That failure clarified the best starting point. Record the task first, then let Claude help choose a method it can actually run. > Pro tip: Install Claude in Chrome before recording a workflow that touches websites. It gives the saved skill a supported path for browser actions. ### Step 2 Start Recording in Claude Cowork Open the Claude desktop app on your Mac and switch to Cowork. Click the + button in the composer, select Record a Skill, and then click Start recording. Anthropic's Record a Skill instructions (https://support.claude.com/en/articles/12512198-how-to-create-custom-skills) cover the current access requirements, recording permissions, and privacy details. The first time you use the feature, macOS will ask Claude for two permissions: - Screen Recording so Claude can see what is happening. - Accessibility so Claude can track your mouse and keyboard actions. macOS may ask you to restart Claude after granting access. Do that before recording so you do not interrupt the task halfway through. Leave the microphone on. Your narration gives Claude the context it cannot get from clicks alone. You can switch microphones or mute the audio from the recording controls if needed. Recordings can run for up to 10 minutes. Choose a task you can demonstrate comfortably inside that window. ![Claude Cowork with the plus menu open and Record a Skill visible](https://tru-images.b-cdn.net/guide-assets/local-413/automate-tasks-with-claudes-record-a-skill-feature/02-record-a-skill-menu.png) ### Step 3 Do the Task and Explain Your Decisions This is the important part. Do the task as you normally would and explain what you are doing out loud. Narrate the decisions behind the clicks. For example: - "I am choosing this file because it is the newest finished version." - "This item needs approval, so I am not moving it yet." - "I always set this upload to private before saving." - "I am skipping this step because the information is missing." Do not be afraid to make a mistake. Claude reviews your screen, clicks, typing, screenshots, and transcript together. It can usually tell the difference between the real workflow and an extra window resize, a wrong click, or a short detour. Think of the recording as a demonstration of the job, the decisions, and the finished result instead of a fixed macro. Click Done when you finish. Claude will send the recording into a Cowork task and begin analyzing it. ![Claude's recording summary showing captured screenshots and actions](https://tru-images.b-cdn.net/guide-assets/local-413/automate-tasks-with-claudes-record-a-skill-feature/03-recording-actions-summary.png) ### Step 4 Let Claude Recommend the Best Automation Method After the recording ends, Claude analyzes the video, your computer actions, and the transcript. It then creates a proposed skill and may recommend a better way to complete the workflow than the method you demonstrated. This was the most useful lesson from our test. We demonstrated a manual YouTube upload with Finder and Safari. Claude recognized that the more reliable long-term method was the YouTube Data API. We chose computer use instead, and the test stalled on browser and file-upload limits. Treat Claude's recommendation as part of the feature. The goal is to capture what you are trying to accomplish and let Claude identify the most reliable tools available. The proposed setup may use: - Computer use, when available on your plan, for local files and supported desktop actions. - Claude in Chrome for supported browser workflows. - A connector or plugin when the service already exposes the right actions. - A script or API when the task is structured and needs a more reliable path. You do not need to understand every option before recording. Review the recommendation, ask Claude to explain any setup or permissions, and choose the route that matches the tools you can access. ![Claude comparing the YouTube Data API, Claude in Chrome, and computer use](https://tru-images.b-cdn.net/guide-assets/local-413/automate-tasks-with-claudes-record-a-skill-feature/04-youtube-api-recommendation.png) ### Step 5 Save and Test the Skill Open Content on the proposal card before you save anything. Check the skill name, when it should run, the steps Claude captured, the tools it expects to use, and the safety rules. The recording can help Claude add useful guardrails. In our upload test, the proposed skill included rules to keep the video private, avoid scheduling a publish time, and ask before moving the completed folder. Those details made the proposal more useful than a simple list of clicks. Click Save to keep the skill. Saved skills appear under Customize > Skills, where you can edit, share, or delete them. Test the skill on a fresh version of the task. Use a new folder, another set of notes, or a second sample input. Check that it: - starts from the right input - uses the intended tool or integration - creates the expected result - follows the safety rules - stops and asks when something is missing If the test fails, do not force Claude through an unsupported app or browser. Edit the skill, choose a different automation method, or record the improved workflow again. ![Claude explaining why the Safari upload path failed and recommending a reliable alternative](https://tru-images.b-cdn.net/guide-assets/local-413/automate-tasks-with-claudes-record-a-skill-feature/05-browser-upload-limit-diagnosis.png) ### Guide: Automate Your Gmail with Google Workspace Studio (Free Tool) > In this guide, you will learn how to use Google Workspace Studio to automate your Gmail inbox. - Canonical page: https://app.therundown.ai/guides/automate-your-gmail-with-google-workspace-studio-free-tool - Machine-readable page: https://app.therundown.ai/guides/automate-your-gmail-with-google-workspace-studio-free-tool.md - Difficulty: beginner - Featured tool or company: Google - Categories: Automation & Agents, Productivity & Knowledge Work, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-03-12 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Google Workspace Studio to automate your Gmail inbox. You will set up agents that triage incoming messages, extract key information from invoices, and draft replies using Gemini, without connecting third-party tools. #### Who This Is Useful For - Anyone who gets a lot of email and wants AI to sort it before they even open Gmail. - Freelancers and small business owners who deal with invoices, client inquiries, and vendor emails that all need different handling. - Ops and finance teams who want invoice details extracted into a spreadsheet automatically. #### What You Will Build You will build a Gmail automation that uses Gemini to label incoming emails by category, such as invoices, newsletters, and client messages, as they arrive. You can also layer on optional steps that extract specific details from labeled emails into a Google Sheet and draft replies. ![image-2115](https://tru-images.b-cdn.net/guide-assets/69/69b327e2346b7acd675a8db2/image-2115.png) - AI-powered Gmail labels for incoming email categories. - Optional invoice detail extraction into a Google Sheet. - Optional Gemini-drafted replies that appear in Gmail drafts for review. #### What You Need - A Google Workspace account with access to Workspace Studio. - A Google Sheet, optional, for logging extracted data. #### Going Further - Build a full triage system with multiple flows and custom AI-powered labels for the categories that matter to you, such as clients, internal messages, invoices, newsletters, or job applications. - Connect to Google Sheets for reporting beyond invoice tracking, including client inquiry details, support request summaries, and vendor quotes. - Pair this with the Google Workspace Studio form flow if you already set it up, so Forms handle inbound requests and Gmail handles inbound email. ### Step 1 Check out the pre-made email boosters Go to studio.workspace.google.com. Before you build anything custom, look at the email boosters on the homepage. These are one-click templates Google has prebuilt: - Daily email recap that summarizes your inbox. It sends via Google Chat by default, but you can change it to email. - VIP notifications that ping you when you get emails from important people. - Auto-label and star for follow-up emails. These are a good way to get familiar with what Workspace Studio can do. Turn a couple on and see how they work before building your own flow. > Pro tip: The VIP notification booster is useful out of the box, but you can take it further. Create a Google Sheet with a list of your VIP contacts, such as clients, teammates, or your boss, and reference it in a custom flow. That way, you update the sheet instead of editing the flow every time. ### Step 2 Build your auto-labeling flow Click the plus button to create a new flow. Select “When I get an email” as your starter and set it to check all emails. Add an Add labels step. Toggle on AI-powered labels. You will see a list of categories like “Receipts & Invoices,” “Newsletters,” “Social,” and more. Select the labels you care about. ![image-2116](https://tru-images.b-cdn.net/guide-assets/69/69b327f423b0e404f0a49020/image-2116.png) - You can edit each AI-powered label to tell Gemini exactly what to look for. Click into a label and describe what should qualify. - Only select labels you actually want to act on. You do not need to label everything, and fewer labels means fewer AI runs against your daily limit. Hit Turn on, name your flow, and send yourself a test email to make sure it fires. Check the Activity tab in Workspace Studio to see exactly what happened at each step. > Pro tip: It can take a few minutes for a flow to fire after a new email arrives. Be patient. In testing, it took about 5 minutes at most. ### Step 3 Add an invoice extraction step optional You can use Gemini to extract information from your emails. This example walks through an agent flow that extracts invoice details. Create a flow like before. This time, create an AI label for invoices and only check the box for that label. Add a Check if step after the labeling step. Set it to run when the “Receipts & Invoices” label, or whatever label you chose, was added. ![image-2119](https://tru-images.b-cdn.net/guide-assets/69/69b3283b23b0e404f0a4a9d8/image-2119.png) Inside that conditional block: - Add an Extract step. This tells Gemini to read the email and pull out specific fields. For invoices, extract the invoice amount, sender, and due date. - Add a Sheets > Add a row step. Select a Google Sheet you have set up for tracking, then map each extracted field to a column using the variables from the Extract step. ![image-2118](https://tru-images.b-cdn.net/guide-assets/69/69b3281e23b0e404f0a4a143/image-2118.png) Now, every time an invoice lands in your inbox, Gemini labels it, extracts the key details, and logs them in your spreadsheet automatically. - If a step runs outside the Check if box, it runs regardless of whether the condition was true or false. Make sure your Extract and Sheets steps are inside the box. - You can add steps in the “If false” branch too. For example, update a row in your sheet to mark non-invoice emails as “Skipped.” ### Step 4 Draft replies with Gemini optional You can also have Gemini draft email replies as part of any flow. Add an Ask Gemini step and use the variables tool to feed it the email content from a previous step. Use a prompt like: ```text Read this email and draft a professional reply acknowledging receipt and confirming we'll follow up within 24 hours. ``` Then add a Draft a reply step and pull in Gemini’s response using variables. The draft will show up in your Gmail drafts folder. Review it and hit send. > Pro tip: Workspace Studio cannot send emails on your behalf to other people. It can only send emails to you, such as notifications. It can draft replies to anyone, but you have to manually send them. > Another pro tip: Workspace Studio limits are unclear at the moment. If you want to stay under the flow limits, use non-AI flows for simple tasks like keyword filters or sender-based labels, and save AI-powered steps for emails that actually need Gemini to read and interpret them. ### Pro tip > Workspace Studio cannot send emails on your behalf to other people. It can draft replies to anyone, but you have to manually send them. ### Guide: Automated Pre-Meeting Research with Perplexity > Generate instant pre-call briefs on any person or company by connecting Perplexity to your Google Calendar. - Canonical page: https://app.therundown.ai/guides/automated-pre-meeting-research-with-perplexity - Machine-readable page: https://app.therundown.ai/guides/automated-pre-meeting-research-with-perplexity.md - Difficulty: beginner - Featured tool or company: Perplexity - Categories: Data, Research & Analysis, Sales, Consulting & Customer Success, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2025-12-29 - Updated: 2026-07-23 ### Overview #### The Rundown Generate instant pre-call briefs on any person or company by connecting Perplexity to your Google Calendar. You’ll create a workflow that finds upcoming meetings and produces a prep memo with recent news, conversation starters, and smart questions, so you do not have to scramble 10 minutes before every meeting. #### Who This Is Useful For - Anyone with lots of intro calls on their calendar - People without a call prep routine - People who want to try out Perplexity #### What You Will Build You will build a reusable pre-meeting research workflow that turns Google Calendar events into concise call prep memos. - A Perplexity prompt for generating a first pre-call memo from your calendar - A call prep format that includes company context, recent news, person-specific background, public icebreakers, and smart questions - A reusable “Call Prep” Space with custom instructions for future calls and PDF-ready memos ![01_what_you_will_build-1](https://tru-images.b-cdn.net/guide-assets/69/6952d87b84ac693a25cb039e/01-what-you-will-build-1.png) #### What You Need - Perplexity account; free is fine - Gmail and Google Calendar #### Going Further - After your first memo, you can reuse the workflow by creating a dedicated “Call Prep” Space in Perplexity. The Space can generate short call prep memos from upcoming calendar events, save them as PDFs, and respond whenever you ask it to prep for your next call. ### Step 1: Set up the Google Calendar connector - Log in to Perplexity.ai. - In the bottom left, click Account → Connectors. - Enable Gmail with Calendar. ![01_google_connector_toggle](https://tru-images.b-cdn.net/guide-assets/69/6952d9116121bebc8e89b1bc/01-google-connector-toggle.png) ### Step 2: Create a dummy event - Create an upcoming call in your Google Calendar. - Put the person’s full name in the title and add their work email as a guest. - For testing, you can pick a random CEO or public figure and click “Don’t send invite” when you create the call. ### Step 3: Run your first research - Go back to Perplexity and paste in this prompt. ```text It is [current time + today’s date]. Look at my Google Calendar and prep a pre-call memo in the following format: 1. What the company does, in one sentence 2. Recent news or funding in the last 6 months 3. Key things to know about this person, including background, interests, and posts 4. Personal icebreakers that are obviously public and wouldn’t make the person feel creeped out if I knew it 5. 2–3 smart questions I could ask them ``` Perplexity should find your meeting and write the report. ![02_example_prompt](https://tru-images.b-cdn.net/guide-assets/69/6952d94e79ceea6a0cf673da/02-example-prompt.png) ### Step 4 Create a reusable Call Prep Space - Click “Spaces” on the left side of your Perplexity dashboard. - Create a new “Call Prep” Space and paste these custom instructions in: ```text You will write short call prep memos for me based on my upcoming calls on my calendar and save them as PDFs. Here's what you need to cover: 1. What the company does, in one sentence 2. Recent news or funding in the last 6 months 3. Key things to know about this person, including background, interests, and posts 4. 2–3 smart questions I could ask them 5. Give me 3 links to scan on the company and the person 6. Be concise and use bullets so it is easily scannable 7. Give 3 bullets on the biggest recent news stories and/or social posts the person and company were mentioned in 8. Include 3–5 bullets of personal but easily accessible info about the person, like sports teams, hobbies, side projects, or businesses — something that won’t creep them out if I know it but will break the ice ``` ![06_going_further_spaces](https://tru-images.b-cdn.net/guide-assets/69/6952d9c61f9798f2c87b4172/06-going-further-spaces.png) - Anytime you have a call coming up, navigate to this Space and say: “Prep for my next call.” ### Pro tip > To refine your starting prompt, ask Perplexity to interview you. Tell it you want it to write a better prompt by asking you 3–5 questions on your role and key outcomes from these calls. ### Guide: Beginner’s Guide to ChatGPT Work (ChatGPT Projects 101) > In this guide, you will learn how to set up a ChatGPT Work project for a recurring workflow. - Canonical page: https://app.therundown.ai/guides/beginners-guide-to-chatgpt-work-chatgpt-projects-101 - Machine-readable page: https://app.therundown.ai/guides/beginners-guide-to-chatgpt-work-chatgpt-projects-101.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: AI - Instructors: Billy Howell - Published: 2026-08-26 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to set up a ChatGPT Work project for a recurring workflow. You will process the work once, then automate the routine. #### Who This Is Useful For - Operators who repeatedly turn notes, research, or project files into useful deliverables. - Team leads who want decisions and action items organized without manually rebuilding the same reports. - Anyone whose ChatGPT projects have become difficult to find, maintain, or reuse. #### What You Will Build You will build a local meeting-notes project that turns new files into an organized source library, a meeting index, and a weekly action plan. The same pattern works for other repeatable assignments: add new material to one input folder, run the workflow, and review the finished outputs. ![Generated meeting notes outputs](https://tru-images.b-cdn.net/guide-assets/local-439/beginners-guide-to-chatgpt-work-chatgpt-projects-101/01-generated-meeting-notes-outputs.png) The project confirms that the input skill processed all three starter notes. #### What You Need To Get Started - The desktop ChatGPT app with Work available in your workspace. - A few non-sensitive meeting notes or other documents you can use as test files. #### Going Further Adapt the same three-folder system to another recurring assignment. For example, an input folder could collect website screenshots, notes could hold the reviewed issue history, and outputs could contain a prioritized fix list. Keep the project structure small, keep the sources separate from generated work, and improve the reusable skill when you notice the same correction twice. ### Step 1 Create a Numbered Source Folder Create the source folder on your computer before opening ChatGPT. Put a two-digit number at the beginning of its name, such as 01-meeting-notes or 12-client-research. The folder gives the project a stable home in your file system. The numeric prefix keeps related projects ordered and easy to scan, while selecting the folder yourself prevents ChatGPT-created work from disappearing into a location you did not choose. Open the desktop ChatGPT app, then go to New chat > Work > Choose project > New project. Name the project and select the numbered folder as its source folder. Connect the folder even if it is empty; this gives the project a known home before ChatGPT creates anything. ![ChatGPT Work project source folder](https://tru-images.b-cdn.net/guide-assets/local-439/beginners-guide-to-chatgpt-work-chatgpt-projects-101/02-chatgpt-work-project-source-folder.png) The Meeting Notes project is connected to its local demo-materials source folder before any processing begins. ### Step 2 Ask ChatGPT to Propose a Simple Structure Give ChatGPT a small set of representative files and ask it to inspect them before it changes the folder: ```text Inspect the starter files in this project and propose a folder structure for processing meeting notes. The workflow should: - accept new files through one input folder - rename and organize successfully processed notes - extract decisions, action items, owners, deadlines, conflicts, and open questions - produce a meeting index and weekly action plan Show me the proposed structure before creating or moving anything. ``` Treat the proposal as a first draft. In the recorded example, ChatGPT suggested separate input, originals, normalized, and archive folders. That created more file juggling than the job needed. Push back when the structure is too elaborate: ```text This feels duplicative. Simplify the project to three working folders: - input for new files - notes for the permanent source library - outputs for finished indexes and action plans Keep in mind that automations will maintain and process this project. ``` This creates a simple mailbox pattern. New work enters through input, trusted source files live in notes, and regenerated deliverables go to outputs. You can reuse the same pattern for screenshots, research, customer feedback, or any assignment with repeatable inputs and outputs. ### Step 3 Process the Work Once, Then Automate It Once you approve the structure, ask ChatGPT to create it and process the starter files one time: ```text Scaffold the approved folders, then process the starter meeting notes once. For this first run: - rename and move valid sources from input to notes - build the meeting index and weekly action plan in outputs - cite the source filename for every confirmed decision and action - do not overwrite source files or invent missing owners, dates, priorities, or decisions Stop after the first run and show me what changed before creating any schedule. ``` Review the result before automating it. Confirm that each processed file moved into notes, the meeting index includes every source, and the weekly action plan does not invent owners or deadlines. Open the generated files so you can compare important decisions and actions with the underlying notes. Once the manual run works, turn the verified process into reusable routines: ```text Turn the verified process into two reusable workflows: 1. Process new files in input, rename and move valid sources to notes, then rebuild every output. 2. Refresh the meeting index and weekly action plan from all files currently in notes. Create scheduled automations for the timing I approve, then turn both workflows into skills I can run manually. Make the scheduled and manual versions use the same rules. ``` The meeting-notes demo uses one intake skill and one refresh skill. The intake skill processes new material, while the refresh skill rereads the permanent notes library and rebuilds the rollups. Keeping the rules in reusable skills means you do not have to explain the process again every time. Use schedules only when the timing is useful. A weekly refresh may be enough for a team action plan; an inbox that changes throughout the day may need a more frequent run. If Cloud scheduling is available in your account, the blank task form lets you choose where and how often a task runs before you add its instructions. ![Cloud daily schedule settings](https://tru-images.b-cdn.net/guide-assets/local-439/beginners-guide-to-chatgpt-work-chatgpt-projects-101/03-cloud-daily-schedule-settings.png) This blank task form shows one Cloud/Daily schedule option; add a title and instructions before saving it. > Pro tip: Let a new automation run once, then point future runs to the existing project chat when that option is available. Creating a fresh chat for every run can quickly fill the project with duplicate threads. ### Step 4 Match the Mode to the Job Use the lightest capable option for each part of the workflow. Routine intake, renaming, and scheduled refreshes usually do not need the most demanding model. A lighter model can make recurring maintenance more efficient. Switch to Codex when the project needs heavier file edits, scripts, or debugging. The recorded workflow uses this distinction to keep basic maintenance simple while reserving the more capable coding workflow for jobs that actually need it. Specific model names and controls can change. Focus on the job instead: use a routine option for predictable maintenance, and move to a deeper coding or reasoning mode when the work becomes complex. > Pro tip: Run automations locally when they need to change files on your computer. Cloud schedules are better for recurring research or tests that do not depend on those desktop files. ### Step 5 Check the Evidence and Keep the Project Clean Before you share a result, check it against the source material. Open the generated files directly and verify the important decisions, actions, owners, and dates against the notes that produced them. If the work includes web research, ask ChatGPT to open its cited sources in the built-in browser: ```text Open the cited sources in the browser so I can review them before sharing the result. ``` ![ChatGPT browser source check](https://tru-images.b-cdn.net/guide-assets/local-439/beginners-guide-to-chatgpt-work-chatgpt-projects-101/04-chatgpt-browser-source-check.png) The cited research source is opened beside the ChatGPT conversation for a quick evidence check. This gives you a fast way to verify that a source exists and supports the claim. It does not replace reading the source or applying your own judgment. When the job is finished, archive completed chats and keep the reusable project, skills, and source folder. The next run should start with new material in input, not with another round of setup. ### Guide: Brand Kit Creation Guide: Zero to One in Under an Hour > This guide teaches you how to create a full brand kit PDF, multiple logo variations, and social media templates using Nano Banana Pro in Gemini. - Canonical page: https://app.therundown.ai/guides/brand-kit-creation-guide-zero-to-one-in-under-an-hour - Machine-readable page: https://app.therundown.ai/guides/brand-kit-creation-guide-zero-to-one-in-under-an-hour.md - Difficulty: beginner - Featured tool or company: Gemini - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2025-12-09 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches you how to create a full brand kit PDF, multiple logo variations, and social media templates using Nano Banana Pro in Gemini. You’ll move through short phases for setup, brand discovery, logo generation, slide creation, social asset generation, and final export. #### Who This Is Useful For - Business owners hiring a graphic designer or social media intern - Vibe coders building out a new app - Teams with disorganized style guides #### What You Will Build You will build a reusable brand system that can be shared with teammates, contractors, designers, interns, or anyone creating branded materials for your business. - A 10-slide brand guidelines presentation exported as a PDF - 5 logo file variations in PNG format - 6 social media assets in PNG format - An organized folder structure for logos, slides, and social assets - A complete brand system that is consistent, cohesive, and ready to use ![14_brandkit_preview](https://tru-images.b-cdn.net/guide-assets/69/6938a71ee6c6205e37c5dbf2/14-brandkit-preview.png) #### What You Need - Google AI Studio access - Playground in Google AI Studio - A paid API key connected if you haven’t already done that - Nano Banana Pro - Your business name - A one-sentence description of what your business does - Your target customer and their pain point - 3 brand personality adjectives - Tagline ideas for the brand - A local folder for your brand kit files - Google Slides for assembling and exporting the final PDF - Optional: Canva for creating a reusable brand kit after you finish #### Going Further - After you finish, use your logos, typography, and color scheme to create a brand kit in Canva so your assets are ready whenever you build an ad or social post. If you don’t use Canva, create a custom slide theme in Google Slides by going to Slide → Edit Theme. ### Step 1 Setup Google AI Studio and Your Folder Structure - Go to aistudio.google.com. - Click "Playground." - Connect a paid API key if you haven’t yet. Click "System instructions" and paste the following: ```text GENERATION PRINCIPLES: When I request brand assets, structure your prompts using: 1. SUBJECT & PURPOSE: Clearly describe what we're creating and who it's for. For example: "A logo for freelance designers who value both aesthetics and professionalism." 2. VISUAL DESCRIPTION: Use natural language and full sentences. Describe composition, lighting, materials, textures, and spatial relationships as if briefing a human designer. 3. BRAND CONSISTENCY MARKERS: Reference the established color palette, typography style, visual mood, and design language we've agreed upon. 4. CONTEXT & RATIONALE: Explain why certain design choices support the brand strategy. WORKFLOW APPROACH: - Start with foundational assets, like logo and color palette, before moving to applications. - Suggest variations that maintain brand integrity while adapting to different contexts. - Proactively identify when an asset needs adjustment for its specific use case, such as social profile vs. banner. QUALITY STANDARDS: - Leverage Nano Banana Pro's strengths: precise typography, data visualization, consistent branding, and multi-image composition. - Avoid keyword-style prompts like "logo, modern, clean, blue." - Instead use: "A contemporary logo mark that combines geometric precision with organic flow, rendered in deep navy with coral accents, designed to appeal to creative professionals who appreciate sophisticated minimalism." ``` Create a local folder on your computer for your brand kit, then add these subfolders: - [Brand Name] Brand Kit/ - Logo Files/ - Social Assets/ - Brand Kit Slides/ ### Step 2 Run Brand Discovery ```text I need to create a complete brand kit for my business. Let me give you the basics: Business name: [YOUR BUSINESS NAME] What we do: [ONE SENTENCE - e.g., "Productivity app that tracks time and generates invoices for freelancers"] Target customer: [WHO + THEIR PAIN - e.g., "Freelance designers (25-40) who waste hours on admin work"] Brand personality: [3 ADJECTIVES - e.g., "Professional, approachable, energetic"] Interview me briefly to understand my brand, then we'll start generating assets. Keep it to 3-5 focused questions. Make sure you interview me about tagline(s) we can use for the brand. ``` - Gemini will ask you 3-5 clarifying questions. - Answer them naturally; don’t overthink it. - This builds context so all assets stay consistent. ### Step 3 Generate Logo Options ```text Based on our conversation, generate 4 distinct logo concepts for [YOUR BRAND NAME]. Show them in a clean 2x2 grid, labeled "Option 1" through "Option 4". Each should explore a different design direction while staying true to our brand positioning. Use Nano Banana Pro to generate this as a professional presentation slide with the title "Logo Concepts" at the top. ``` ![3_logo_options](https://tru-images.b-cdn.net/guide-assets/69/6938a742e6c6205e37c5dc74/3-logo-options.jpeg) ### Step 4 Extract Your Chosen Logo Variations Use these prompts to generate all the logo variations you will need. Download each result into your Logo Files folder using the filenames provided. ```text I love Option [NUMBER]. Create the primary, full-color version of this logo as a standalone, high-quality asset. This is the main version we'll use most frequently. Show it on a clean white background with adequate spacing. Professional execution ready for all brand applications. Save as: "Logo_Primary.png" in the Logo Files folder. ``` ```text Now create a simplified version of this logo optimized for small sizes like social media profile pictures and favicons. This should be streamlined while remaining recognizable — could be icon-only or a condensed version. Must work well at 48x48 pixels and in circular crops. Save as: "Logo_Secondary.png" in the Logo Files folder. ``` ```text Create a monochrome black version of our logo for situations where color printing isn't available — photocopies, faxes, newspaper ads, embossing. Solid black only, no gradients or gray tones. Same design, just converted to pure black on white. Save as: "Logo_Monochrome_Black.png" in the Logo Files folder. ``` ```text Create a monochrome reverse version — white logo on a dark background. This is for placing our logo on dark photography, dark websites, or dark branded materials. Show it on a dark background so I can see the contrast clearly. Save as: "Logo_Monochrome_White.png" in the Logo Files folder. ``` ```text Create an icon-only version, just the symbol/mark without the wordmark, for extremely small applications like app icons, favicons, and social media profiles. Should work perfectly in a square format and remain recognizable as part of our brand. Save as: "Logo_Icon.png" in the Logo Files folder. ``` ![5_logo_system](https://tru-images.b-cdn.net/guide-assets/69/6938a7607214a4ea5f68c72b/5-logo-system.jpeg) ```text Make sure you switch your system settings aspect ratio to 16:9 for slides. Now create a professional "Logo System" slide showing all 5 variations we just created in a clean grid: 1. Primary (full color) 2. Secondary/Simplified 3. Monochrome Black 4. Monochrome White (on dark background) 5. Icon Only Label each clearly and add minimum size specs: "Minimum width: 120px for primary, 48px for icon." Make this look like a professional brand guidelines page. Save as: "Slide_03_LogoSystem.png" in the Brand Kit Slides folder. ``` ### Step 5 Build Your Brand Kit Slides Now that you’ve got your logo slide set, generate the rest of your brand kit slides. Send these messages one at a time. ![1_brandkit_deck](https://tru-images.b-cdn.net/guide-assets/69/6938a7667214a4ea5f68c760/1-brandkit-deck.png) ```text Create a professional brand kit cover slide for [YOUR BRAND NAME]. Include: - Brand name prominently displayed - Subtitle: "Brand Guidelines" - My tagline: "[YOUR TAGLINE]" - Date: December 2024 Design this using our brand's visual language based on the logo we just created. Clean, modern, professional. This is the first impression of our brand kit. Save as: "Slide_01_Cover.png" ``` ![2_brand_positioning](https://tru-images.b-cdn.net/guide-assets/69/6938a7667214a4ea5f68c768/2-brand-positioning.jpeg) ```text Create a "Brand Overview" slide with these sections: Brand Positioning: [Write 2-3 sentences about what you do, who it's for, and why it matters] Target Audience: [YOUR TARGET CUSTOMER description] Brand Personality: [YOUR 3 ADJECTIVES] Layout: Clean, professional design with good visual hierarchy. Use our brand's visual style. Save as: "Slide_02_BrandOverview.png" ``` ![5_logo_system](https://tru-images.b-cdn.net/guide-assets/69/6938a7607214a4ea5f68c72b/5-logo-system.jpeg) You already generated the Logo System slide in the previous phase. You should have: "Slide_03_LogoSystem.png" ![6_logo_usage](https://tru-images.b-cdn.net/guide-assets/69/6938a7627214a4ea5f68c743/6-logo-usage.jpeg) ```text Create a "Logo Usage Guidelines" slide showing DOs and DON'Ts. Split into two columns: LEFT — DO, with green checkmarks: - Logo on clean white background - Logo on solid brand color - Proper spacing/padding - Appropriate sizing RIGHT — DON'T, with red X marks: - Stretched or distorted logo - Logo on busy background - Logo too small - Altered colors Use visual examples with our actual logo, not just text descriptions. Save as: "Slide_04_LogoGuidelines.png" ``` ![7_color_scheme](https://tru-images.b-cdn.net/guide-assets/69/6938a7617214a4ea5f68c73b/7-color-scheme.jpeg) ```text Create a "Color Palette" slide showing our brand colors. Based on the logo you created, extract and display: - Primary color, large swatch with HEX and RGB codes - Accent color, large swatch with HEX and RGB codes - 2-3 neutral/supporting colors, swatches with codes Include usage notes for each color, such as "Primary: Headlines, key elements" or "Accent: CTAs, energy, highlights." Professional layout with large, clear swatches. Save as: "Slide_05_ColorPalette.png" ``` ![8_typography](https://tru-images.b-cdn.net/guide-assets/69/6938a7627214a4ea5f68c74b/8-typography.jpeg) ```text Create a "Typography System" slide based on the fonts used in our logo. Show: Primary Font: [The font style from the logo] - Large heading sample: "The Quick Brown Fox" (48pt) - Subheading sample: "The Quick Brown Fox" (32pt) - Usage: Headlines, key messaging Secondary Font: [Complementary font] - Body text sample: Full paragraph at 16pt - Usage: Body copy, descriptions Display font specimens with different sizes and weights. Professional typography presentation. Save as: "Slide_06_Typography.png" ``` ![9_brand_voice](https://tru-images.b-cdn.net/guide-assets/69/6938a7617214a4ea5f68c733/9-brand-voice.jpeg) ```text Create a "Brand Voice & Messaging" slide with: Brand Voice: how we communicate. Write 3-4 bullet points describing your tone, such as "Professional but approachable," "Direct without being blunt," or "Energetic without being juvenile." Core Messages: what we say. Write 3-5 key messages about your brand, such as "Time you'll never bill is money you'll never see." Tagline: displayed prominently as "[YOUR TAGLINE]" Clean layout with good hierarchy. Professional brand guidelines aesthetic. Save as: "Slide_07_BrandVoice.png" ``` ![10_social_assets](https://tru-images.b-cdn.net/guide-assets/69/6938a7637214a4ea5f68c758/10-social-assets.jpeg) ```text Create a "Social Media Assets" slide showing a grid with 6 labeled sections for different platforms: 1. Profile Picture (1080x1080px) — square 2. Twitter/X Header (1500x500px) — wide 3. LinkedIn Cover (1584x396px) — wide 4. Instagram Story Template (1080x1920px) — vertical 5. Facebook Cover (820x312px) — wide 6. YouTube Thumbnail (1280x720px) — landscape Show the dimensions and aspect ratios clearly with placeholder frames. We'll generate the actual assets next. Professional grid layout with platform icons. Save as: "Slide_08_SocialGrid.png" ``` ![13_brand_applications](https://tru-images.b-cdn.net/guide-assets/69/6938a75f7214a4ea5f68c713/13-brand-applications.jpeg) ```text Create a "Brand Applications" slide showing 4-6 mockups of how our brand could be used in the real world: - Business card design - Website header - Email signature - Product/app interface - Marketing material - Social media post These should be visual mockups showing our logo, colors, and typography in context. Professional presentation showing brand versatility. Save as: "Slide_09_Applications.png" ``` ![11_quick_reference](https://tru-images.b-cdn.net/guide-assets/69/6938a80ac20dc65f40a17861/11-quick-reference.jpeg) ```text Create a "Quick Reference" one-page cheat sheet with: - Small primary logo at the top - Color palette, compact swatches with HEX codes - Font names, Primary & Secondary - Tagline - One-sentence brand positioning This is a scannable, grab-and-go reference. Everything on one slide. Compact but clear. Could be printed as a reference card. Save as: "Slide_10_QuickReference.png" ``` ### Step 6 Generate Social Media Assets Use these prompts to create everything you’ll need to set up social media accounts for your brand. ```text Create a professional social media profile picture for [YOUR BRAND NAME] (1080x1080px square). Use our simplified logo or icon optimized for circular crops. Works at small sizes (48x48px). Clean, bold design using our brand colors. Should feel [YOUR 3 ADJECTIVES] and be instantly recognizable as our brand. Save as: "Social_ProfilePicture.png" in the Social Assets folder. ``` ```text Create a Twitter/X header banner (1500x500px) for [YOUR BRAND NAME]. Include: - Our logo or logo elements - Our tagline: "[YOUR TAGLINE]" - Our brand colors - Leave safe space for circular profile picture on left side Clean, professional composition that communicates [KEY BRAND MESSAGE]. Appeals to [TARGET CUSTOMER]. Save as: "Social_TwitterHeader.png" in the Social Assets folder. ``` ```text Create a LinkedIn cover banner (1584x396px) for [YOUR BRAND NAME]. More professional/corporate than Twitter. Include our logo, tagline, and brand colors. Leave space for profile picture on left. Communicates credibility and expertise to potential clients/partners. Professional business aesthetic. Save as: "Social_LinkedInCover.png" in the Social Assets folder. ``` ```text Create an Instagram Story template (1080x1920px) for [YOUR BRAND NAME]. Reusable branded template with: - Branded design elements, such as borders, shapes, or patterns - Small logo in corner - Clear content area in middle for photos/text - Brand color accents Professional but Instagram-appropriate. Frames content while maintaining brand presence. Save as: "Social_InstagramStory.png" in the Social Assets folder. ``` ```text Create a Facebook cover (820x312px) for [YOUR BRAND NAME]. Include logo on the right side, tagline, and brand colors. Leave space for circular profile picture bottom left. Community-focused, welcoming, professional. Works with Facebook's interface. Save as: "Social_FacebookCover.png" in the Social Assets folder. ``` ```text Create a YouTube thumbnail template (1280x720px) for [YOUR BRAND NAME]. Reusable template with: - Strong branded design elements - Small logo in corner - Large text area for video titles, readable at small sizes - High contrast for visibility - Placeholder text: "VIDEO TITLE HERE" Stands out in YouTube feed while maintaining brand consistency. Save as: "Social_YouTubeThumbnail.png" in the Social Assets folder. ``` ### Step 7 Assemble and Export Your Brand Kit - Open Google Slides. - Create a new blank presentation. - Delete the default slides. - Go to Insert → Image → Upload from computer. - Add all 10 slides in order, one image per slide, full-size. Use this slide order: - Cover - Brand Overview - Logo System - Logo Usage Guidelines - Color Palette - Typography - Brand Voice & Messaging - Social Media Assets Grid - Brand Applications - Quick Reference Export the presentation as a PDF: - File → Download → PDF Document - Save as "[Brand Name] Brand Kit.pdf" Your final folder structure should include the brand kit PDF, the Logo Files folder, the Social Assets folder, and the Brand Kit Slides folder. - [Brand Name] Brand Kit.pdf - Logo_Primary.png - Logo_Secondary.png - Logo_Monochrome_Black.png - Logo_Monochrome_White.png - Logo_Icon.png - Social_ProfilePicture.png - Social_TwitterHeader.png - Social_LinkedInCover.png - Social_InstagramStory.png - Social_FacebookCover.png - Social_YouTubeThumbnail.png - All 10 individual slide images When you’re done, you should have a shareable 10-slide brand guidelines PDF, 5 logo file variations, 6 social media assets, an organized folder structure, and a cohesive brand system that is ready to use. ### Resources - [Nano Banana Brand Kit Generation PDF](https://tru-images.b-cdn.net/guide-assets/69/6938a4cf4aabcb82e6695a73/53d18a74-89a9-4b2f-81f9-56086d5890d5-nano-banana-brandkit-generation.pdf) ### Pro tip > The more time you spend going back and forth with Gemini during brand discovery, the better your end results will be. ### Guide: Build a “Brand Twin” That Writes Every Post in Your Exact Voice > Learn how to create a custom GPT “Brand Twin” that writes LinkedIn posts, Instagram captions, and other copy in your specific brand voice. - Canonical page: https://app.therundown.ai/guides/build-a-brand-twin-that-writes-every-post-in-your-exact-voice - Machine-readable page: https://app.therundown.ai/guides/build-a-brand-twin-that-writes-every-post-in-your-exact-voice.md - Difficulty: beginner - Featured tool or company: OpenAI - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-02-03 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to create a custom GPT “Brand Twin” that writes LinkedIn posts, Instagram captions, and other copy in your specific brand voice. You will produce a reusable GPT trained on your brand kit, writing samples, and a forbidden-word list, helping you spend less time editing generic AI output. #### Who This Is Useful For - Marketing managers who are tired of rewriting most AI-generated copy - Content leads who want to scale a specific brand voice across a team - Entrepreneurs who do not know where to start with their brand voice #### What You Will Build A custom GPT “Brand Twin” trained on your specific brand kit, best writing samples, and a forbidden list of corporate clichés and AI-isms. - Custom GPT system instructions based on your brand DNA - Uploaded writing samples and optional brand kit inside the GPT - A pinned GPT that your team can reuse for brand-aligned copy #### What You Need - ChatGPT Plus ($20/month) - 3–5 samples of your best writing, such as PDFs, markdown files, or text files - Optional: a brand kit slide deck #### Going Further - Test your new GPT by using the same prompt in your Brand Twin and in a temporary ChatGPT thread, then use the comparison to refine your system instructions. - If a teammate writes copy for your brand, have them add the GPT and ask it to grade how well their copy fits your brand voice on a 1–10 scale. ### Step 1: Analyze your brand voice Open a new ChatGPT thread in the phone or desktop app. Upload 3–5 examples of your best writing, plus your brand kit if you have one. Tell ChatGPT: ```text Analyze these uploads to extract my brand's “DNA” for a Custom GPT System Prompt. Identify the specific voice posture, sentence cadence, and vocabulary preferences, while creating a strict “Forbidden List” of corporate cliches and AI-isms to avoid. Specifically, define syntactic constraints—such as banning em-dashes or “not just A, but also B” structures—to ensure the output never reverts to generic AI slop. ``` ### Step 2: Create your custom GPT - Navigate to https://chatgpt.com/gpts in your web browser and click “Create” in the top right. - Name your GPT and paste in the system instructions you generated in Step 1. - Upload your writing samples and brand kit. - Keep everything else as the default settings. ![liquid_death_brand_twin](https://tru-images.b-cdn.net/guide-assets/69/69829d064a47d8edb96b9ac3/liquid-death-brand-twin.png) ### Step 3: Share and pin your Brand Twin Click Share > GPT Store. Then open the ChatGPT desktop app and click GPTs > Explore GPTs > Search Public GPTs. Find your GPT and click “Start Chat.” Click the name at the top, then click “Pin GPT.” ![pin_gpt](https://tru-images.b-cdn.net/guide-assets/69/69829d3c4a47d8edb96b9bac/pin-gpt.png) ### Pro tip > Use a brand with a unique, recognizable voice as a reference when building your own. The source example used Liquid Death and an old version of their brand kit as a template. ### Guide: Build a 15 Page Travel Itinerary in 20 Minutes With Perplexity Computer > This guide shows you how to use Perplexity Computer to plan a full trip itinerary with flights, a day-by-day schedule, hotel options, transfers, and sources in one run. - Canonical page: https://app.therundown.ai/guides/build-a-15-page-travel-itinerary-in-20-minutes-with-perplexity-computer - Machine-readable page: https://app.therundown.ai/guides/build-a-15-page-travel-itinerary-in-20-minutes-with-perplexity-computer.md - Difficulty: beginner - Featured tool or company: Perplexity - Categories: Productivity & Knowledge Work, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-03-30 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to use Perplexity Computer to plan a full trip itinerary with flights, a day-by-day schedule, hotel options, transfers, and sources in one run. In testing, the itinerary took about 15–20 minutes to finish. #### Who This Is Useful For - Anyone planning personal travel who is tired of bouncing between flights, hotels, maps, and random blog posts - Operators, assistants, and founders who need a fast first draft for a work trip without doing two hours of manual research - Perplexity Pro users who want to see whether Computer is worth spending credits on #### What You Will Build You will create a travel-agent-style itinerary PDF that you can use as a booking-ready first draft. - Flight options with pricing - A day-by-day itinerary - Hotel or stay options - Transfers between cities - Links to sources ![image-2155](https://tru-images.b-cdn.net/guide-assets/69/69cae7e2ef21bb17f6756f49/image-2155.png) #### What You Need - A Perplexity Pro account with access to Perplexity Computer - Optional: check your Perplexity Computer credits before you start - If you have Perplexity Pro, check your credits before starting. Perplexity gave Pro users 4,000 bonus credits to test Computer, expiring April 13, 2026. The source notes that this should work if you sign up for Pro for the first time too. #### Going Further - After generating the itinerary, download the PDF, ask Perplexity for a one-page “at a glance” version, or reuse the plan by asking Perplexity Computer to create an interactive calendar website you can use to plan and tweak your trip. ### Step 1 Open Perplexity Computer Open Perplexity in the app or in your browser and look for the Computer toggle. ![image-2156](https://tru-images.b-cdn.net/guide-assets/69/69cae7eeef21bb17f67577e7/image-2156.png) ### Step 2 Prompt It With Your Travel Brief Be clear up front with your trip parameters, like you are briefing a travel agent instead of chatting with a generic chatbot. ```text Plan a trip itinerary for [DESTINATION] for [DATES / LENGTH]. Constraints: - Departing from: [AIRPORT] - Budget: [budget or range] - Style: [relaxed / packed / foodie / outdoors / etc.] - Must-haves: [2–4 must-haves] Output: 1. Best flight options with pricing 2. A day-by-day itinerary in a table 3. 2–3 hotel or stay options with why they fit 4. Sources/links at the bottom Treat this like you are a travel agent building me a real plan. ``` > Pro tip: Add constraints like “direct flights preferred,” “I prefer flying United,” or “keep hotel changes to a minimum.” The more you treat this like a real travel brief, the better the itinerary gets. ### Step 3 Let the Itinerary Run Let Perplexity Computer run. In the tests for this guide, it took about 15–20 minutes to finish the itinerary and used about 650 credits to plan a 10-day international trip. If you want to reduce the cost on future runs, click the + button, open Select Orchestrator, and pick a cheaper option if one is available. In the recording, Sonnet was called out as a lower-cost option than Opus for first passes. ### Step 4 Compare the Result Against a Standard Search While you wait, you can try the same prompt in regular Perplexity search or any other LLM so you can compare the performance difference. In testing, the sub-agents made a noticeable difference in creating a good itinerary. ### Step 5 Download the Finished PDF When it’s done, download the PDF. Sometimes Perplexity glitches and the “artifacts” do not load right away when you try to access them. > Pro tip: Ask Perplexity to make you a one-pager with the “at a glance” version of the itinerary. ### Pro tip > Treat the prompt like a real travel brief: add constraints such as preferred airlines, direct flights, or minimizing hotel changes. ### Guide: Build a Competitor Database With Claude Cowork > This guide shows you how to set up a competitor intelligence database powered by Claude Cowork. - Canonical page: https://app.therundown.ai/guides/build-a-competitor-database-with-claude-cowork - Machine-readable page: https://app.therundown.ai/guides/build-a-competitor-database-with-claude-cowork.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Data, Research & Analysis, Marketing & Growth - Instructors: Billy Howell - Published: 2026-01-29 - Updated: 2026-07-23 ### Overview #### The Rundown This guide shows you how to set up a competitor intelligence database powered by Claude Cowork. You will download the prompt files, point Claude at the folder, and have Claude structure the system for you. The source notes that the setup can be ready to use in about 5 minutes. #### Who This Is Useful For - Founders and executives who want a “live radar” of market threats - Sales teams that need battlecards for handling objections - Small teams without an established competitor research system #### What You Will Build You will build a reusable competitor teardown workspace that Claude can update as you run analyses. The workspace produces competitor folders, dated killsheets, and an updated Competitive Radar document, with follow-up prompts for sales battlecards and comparison-page mockups. - A competitor_teardowns folder structured from the SOP - A competitor folder inside 01_Competitors - A dated killsheet for each analyzed competitor - An updated Competitive Radar document - Optional follow-up sales battlecards and comparison-page drafts ![comp_teardown_thumb](https://tru-images.b-cdn.net/guide-assets/69/697bdc04f8b025218b1e5365/comp-teardown-thumb.jpeg) #### What You Need - Claude Desktop app, which the source notes only works on Mac - The attached prompt kit - A text editor to view your database #### Going Further - After generating killsheets, you can ask Claude to convert reports into PDFs or update the SOP to do that automatically. You can also drop updated killsheets into NotebookLM to generate quarterly competitor intelligence reports, including slide decks or podcasts. ![daily_download_growth_engine_vs_glass_jaw_infographic](https://tru-images.b-cdn.net/guide-assets/69/697bde7be19fa3a260e514cb/daily-download-growth-engine-vs-glass-jaw-infographic.jpeg) ### Step 1 Set up the competitor teardowns folder Create a new folder named “competitor_teardowns.” Download the prompt kit; it should include six Markdown files. Drop all six files into your new folder. Open the Claude Desktop app, click the Cowork tab, then click the folder button and select your new folder. ![cowork_interface_screenshot](https://tru-images.b-cdn.net/guide-assets/69/697bde51ab193bfde28bf614/cowork-interface-screenshot.png) ```text Read the SOP file and structure our folder as outlined in the SOP. ``` ### Step 2 Run a killsheet analysis Once the system is built, run a competitor analysis. The guide calls this a “killsheet”: a forensic business analysis of your competitors. The source recommends using the latest Opus model for this analysis. Use this when you are doing a quarterly review, reacting to new competitor funding news, or reviewing the loss of a major deal. - Identify the competitor name and URL. - Send Claude the prompt below, replacing the bracketed placeholders with your competitor details. ```text Start an analysis on [competitor] using [competitor domain] [any extra competitor domains] ``` > 💡 Pro Tip: Before running your first analysis, create a small text file called Context.md describing your own company and offer. This helps Claude understand who “we” are when identifying threats. You can even dictate this into Claude to have it write the file for you. ### Step 3 Review your results Open your competitor_teardowns folder in a code editor like Cursor. You should see a new competitor folder inside 01_Competitors, complete with a dated killsheet. Each time you run a killsheet analysis, Claude will update your Competitive Radar document. This becomes your at-a-glance competitor overview. ![competitive_radar_markdown_screenshot](https://tru-images.b-cdn.net/guide-assets/69/697bde6be19fa3a260e513fe/competitive-radar-markdown-screenshot.png) > 💡 Pro Tip: You can tell Claude to convert any report into a PDF, or to update its SOP to do so automatically. ### Step 4 Create follow-up sales and marketing assets After you have your killsheet, use the included prompts to create sales and marketing assets. - For sales, run Prompt_03_Battlecard_Generator.md to create scripts for your sales team based on competitor objections. - For marketing, run Prompt_04_Comparison_Page.md to have Claude mock up a landing page comparing you and a competitor, or competitor versus competitor. ### Resources - [00_Competitive_Radar.md](https://tru-images.b-cdn.net/guide-assets/69/697bdafc7966558d15a63d3d/00-competitive-radar.md) - [00_Workflow_SOP.md](https://tru-images.b-cdn.net/guide-assets/69/697bdb087966558d15a63d66/00-workflow-sop.md) - [Prompt_01_Kill_Sheet.md](https://tru-images.b-cdn.net/guide-assets/69/697bdb117966558d15a63d86/prompt-01-kill-sheet.md) - [Prompt_02_Index_Updater-1.md](https://tru-images.b-cdn.net/guide-assets/69/697bdba18bdd403587b2ce87/prompt-02-index-updater-1.md) - [Prompt_03_Battlecard_Generator.md](https://tru-images.b-cdn.net/guide-assets/69/697bdb1f7966558d15a63dbc/prompt-03-battlecard-generator.md) - [Prompt_04_Comparison_Page.md](https://tru-images.b-cdn.net/guide-assets/69/697bdb187966558d15a63d9c/prompt-04-comparison-page.md) ### Pro tip > Before running your first analysis, create a small Context.md file that describes your own company and offer so Claude understands who “we” are when identifying threats. ### Guide: Build a Custom Blog Writing Agent With No Code (Langflow) > In this guide, you will build a no-code blog-writing subagent in Langflow that writes posts in your website’s style. - Canonical page: https://app.therundown.ai/guides/build-a-custom-blog-writing-agent-with-no-code-langflow - Machine-readable page: https://app.therundown.ai/guides/build-a-custom-blog-writing-agent-with-no-code-langflow.md - Difficulty: intermediate - Featured tool or company: Langflow - Categories: Content & Creative, Automation & Agents, Coding & App Development - Instructors: Billy Howell - Published: 2026-04-29 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will build a no-code blog-writing subagent in Langflow that writes posts in your website’s style. It runs locally, is free besides LLM calls, and can be triggered by Claude, Codex, or another coding agent once exported. The setup is a simple local flow, so you can get started without writing code or setting up a server. #### Who This Is Useful For - Marketers and creators who write repeat blog posts and want the same structure every time - Anyone using Claude or another coding agent across multiple projects who wants to stop wasting context on the same writing task - People curious about AI agents who want a simple local tool to understand how subagents and MCP tools work #### What You Will Build You will build a reusable blog-writing subagent that takes a topic, looks at one of your existing posts for style, and returns a new draft in that voice. ![image-2189](https://tru-images.b-cdn.net/guide-assets/69/69f27e939ace0dda9d0321f8/image-2189.png) - A URL input for the reference post - A text input for the new topic - A prompt template that combines the style reference and topic - A language model node - A chat output #### What You Need - Langflow installed locally - An OpenAI or Anthropic API key - One strong blog post you want the agent to use as a style reference - A test topic for the first draft #### Going Further - Once the flow works, you can export it as an MCP server and connect it to Claude. Claude can then call the flow as a tool whenever you need a blog draft in that style, while saving its context for higher-level work. ### Step 1 Open Langflow and start from the Blog Writer template Download Langflow and open it on your computer. It runs locally, so you can build a custom subagent without setting up a server or writing code just to get started. In the project area, click New Flow, go to the content generation section, and choose the Blog Writer template. ![image-2190](https://tru-images.b-cdn.net/guide-assets/69/69f27eb8e6b36d2c01073b04/image-2190.png) > Pro tip: Templates are the fastest way to learn Langflow. ### Step 2 Add your style reference and topic input Change the reference URL first. Use one of your best blog posts so the flow has something real to imitate. The built-in crawler can also follow more than one page. If you want a broader writing sample, point it at your blog index and increase the crawl depth to 2. Then add a text input called topic. The reference content shapes the style, and the topic input tells the model what to write next. ![image-2191](https://tru-images.b-cdn.net/guide-assets/69/69f27eebe6b36d2c0107441f/image-2191.png) > Pro tip: Pick a reference post that already sounds how you want future posts to sound. If the source post is weak, the draft will be weak too. ### Step 3 Add your model provider and choose a model Open the language model settings and click Manage model providers. Add your OpenAI or Anthropic API key, then pick the model you want to run. ![image-2192](https://tru-images.b-cdn.net/guide-assets/69/69f27f049ace0dda9d032ac3/image-2192.png) In the source test, Anthropic and OpenAI were used. You could also run a model locally with Ollama and use it in Langflow. ### Step 4 Test the flow in Playground Once the reference URL and topic input are wired up, click Playground and test the flow with a topic. A simple topic prompt is enough for the first run: ```text how to use AI for marketing ``` Langflow will pull the reference content, feed the style and topic into the prompt, and return a first draft. Use the first draft to tighten the flow. If the draft is too generic, add clearer instructions into the prompt template. ```text No em dashes. Keep it concise. Sound like our site. ``` > Pro tip: Add voice guardrails early. Short lines like the examples above usually do more than a big abstract style paragraph. If you want better output later, add more sources or drop files into a knowledge base. You can also batch multiple article runs once the basic flow is working. ### Step 5 Export the flow as an MCP server If you want Claude to use this as a subagent, open Share and choose MCP Server. ![image-2193](https://tru-images.b-cdn.net/guide-assets/69/69f27f3de6b36d2c01074d4c/image-2193.png) From there, connect it to Claude. Once it loads, Claude can call the flow as a tool whenever you need a blog draft in that style. The larger workflow is to use Langflow for the repetitive subagent, then let Claude save its context for the higher-level work. ### Pro tip > Pick a reference post that already sounds how you want future posts to sound. If the source post is weak, the draft will be weak too. ### Guide: Build a Daily Command Center With Claude Live Artifacts > Learn how to build a daily command center in Claude Cowork with Live Artifacts. - Canonical page: https://app.therundown.ai/guides/build-a-daily-command-center-with-claude-live-artifacts - Machine-readable page: https://app.therundown.ai/guides/build-a-daily-command-center-with-claude-live-artifacts.md - Difficulty: intermediate - Featured tool or company: Claude - Categories: Productivity & Knowledge Work, Automation & Agents - Instructors: Billy Howell - Published: 2026-04-21 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to build a daily command center in Claude Cowork with Live Artifacts. Instead of opening Slack, email, calendar, tasks, docs, and dashboards one by one, you will create one reusable live view for today’s priorities, this week’s blockers, and this month’s KPIs. The goal is a scan-friendly dashboard you can understand in about 60 seconds. #### Who This Is Useful For - Operators and founders who start every morning by checking too many apps before they know what actually matters - Managers and team leads who need one place for blockers, meetings, follow-ups, and team updates - Creators and consultants who want a lightweight command center for content, clients, pipeline, tasks, and KPIs #### What You Will Build ![CleanShot 2026-04-21 at 16.57.14@2x](https://tru-images.b-cdn.net/guide-assets/69/69e7e4bd7cec2107bf1d1aa8/cleanshot-2026-04-21-at-16-57-14-2x.png) You will build a Claude Live Artifact that works like a daily command center. Version one includes Today, This Week, and This Month views, plus KPI cards, quick stats, charts, and feed panels from your connected apps. Then you will layer on upgrades like priority labels, dashboard skills, refresh buttons, manual override, archive, and click-to-open updates. - A reusable Live Artifact command center - Today, This Week, and This Month dashboard views - KPI cards, quick stats, charts, and app feed panels - Priority labels for updates - Dashboard skills and practical controls for daily use #### What You Need - Claude Desktop app - Any paid Claude plan - Connected apps or safe sample data to test with - A few KPIs or recurring work streams you want to track #### Going Further - Add dark mode, small animations, and cleaner section headers if they make the dashboard easier to scan. - Add a settings panel for update frequency, visible sources, KPI definitions, and priority rules. - Add manual override and archive controls so the dashboard stays useful when a status is wrong or an item is done. - Make updates clickable so you can open the original Slack thread, email, doc, task, or calendar event. - Use the pinned artifact as a morning ritual: refresh the dashboard, read the change report, choose the top 3 actions, and then open individual apps. ### Step 1: Let Claude Interview You First Open Claude Cowork and have Claude interview you before it builds anything. This prevents you from getting a polished dashboard that does not match how your day actually works. ```text Interview me about my connected apps, daily workflow, KPIs, and what counts as urgent. Then propose the modules for a daily command center before creating the artifact. ``` ```text Before you build anything, interview me about my workday. Ask me about: - my connected apps and files - what I check every morning - what decisions I make daily - the KPIs I care about - what counts as urgent - what updates are FYI only - what should show up in Today, This Week, and This Month After the interview, propose the dashboard modules before creating the artifact. ``` Answer with real workflow details. For example, you might say you check Slack mentions, unread Gmail, today's calendar, active Notion projects, CRM pipeline, and weekly content numbers. If Claude asks to use a connector that is not enabled for the current chat, turn it on only if you are comfortable with the access. Claude shows this when a connector exists but has not been enabled for the current chat. ![01-slack-connector-disabled-prompt](https://tru-images.b-cdn.net/guide-assets/69/69e7e51b58b6351fdf17d0aa/01-slack-connector-disabled-prompt.png) > Pro tip: Ask Claude to summarize its understanding before it builds. A quick “Here is what I heard” step catches bad assumptions early. ### Step 2: Create Version One of the Dashboard Tell Claude to create version one of the dashboard. Keep this version simple so you can prove the structure works before adding skills, buttons, animations, or advanced settings. ```text Create a modular Live Artifact command center with Today, This Week, and This Month views. Include KPI cards, quick stats, charts, and app feed panels. ``` ```text Create a modular Live Artifact command center with three views: 1. Today 2. This Week 3. This Month Include: - KPI cards - quick stats - charts for trends or changes - feed panels grouped by app or workflow - blockers and waiting-on items - top 3 recommended actions Keep it scan-friendly. I should understand the day in 60 seconds. Use the interview answers to decide what belongs in each view. ``` Claude may ask how you want the output delivered. Choose Live artifact so the dashboard becomes something you can reopen instead of a one-time chat summary. ![02-live-artifact-delivery-choice](https://tru-images.b-cdn.net/guide-assets/69/69e7e53f58b6351fdf17d0fc/02-live-artifact-delivery-choice.png) - Are the top numbers actually useful? - Are the feeds grouped by app or workflow? - Can you understand the day in under a minute? - Is anything private showing up that should not be there? ![03-created-command-center-artifact](https://tru-images.b-cdn.net/guide-assets/69/69e7e55258b6351fdf17d12b/03-created-command-center-artifact.png) Claude can build the artifact alongside the chat, so you can inspect the dashboard while it works. > Pro tip: Every step after this is an upgrade to add on top of version one. Make sure this version works and is actually useful before proceeding. ### Step 3: Add Priority Tracking on Top Once version one is useful, add priority tracking. The dashboard should rank updates instead of dumping another feed on you. ```text Add priority labels to every update: urgent, needs review, FYI, or blocked/waiting. Rank items by deadline, business impact, customer impact, and whether I am the blocker. ``` ```text Add priority labels to every update: - urgent - needs review - FYI - blocked/waiting Rank items by: - deadline - business impact - customer impact - whether I am the blocker - whether a decision is needed today If you are not sure, mark the item “review manually.” Do not invent status. Cite the source app or source item for each recommendation. ``` This turns the artifact from a nice dashboard into something you can act on. Slack mentions, unread emails, upcoming meetings, overdue tasks, and project updates should not all compete equally. > Pro tip: Keep the labels visible. A fast dashboard should tell you what needs action before you read the details. ### Step 4: Add Dashboard Skills Add dashboard skills so you can start taking action without leaving the dashboard. ```text Add skills to this dashboard: Plan my day, Show blockers, Draft replies, Prep meetings, and Review KPIs. Each skill should be triggered via a button on the dashboard and should return a short next-step list. ``` ```text Add these dashboard skills: 1. Summarize email Read the latest relevant email updates, summarize the important ones, flag anything urgent, and place the results in the Today view. 2. Summarize Slack mentions Read new Slack mentions and DMs, group them by project or urgency, and add anything actionable to the Today view. 3. Prep meetings Look at today's calendar, find related docs or recent messages, and create a short prep brief for each important meeting. 4. Review KPIs Compare today's numbers to this week and this month. Flag anything that changed meaningfully. 5. Draft replies Create short draft replies for urgent messages, but do not send anything without review. ``` The point is not to add every possible button. The point is to turn the command center into the place where the next action starts. > Pro tip: Once the artifact opens, click the pin button to keep it in your Claude sidebar. You do not want to search through old chats every morning. ### Step 5: Add Refresh and Control Buttons Once version one works and the skills are useful, add controls that make the dashboard easier to run every day. ```text Add a Daily refresh button to this command center. When I click it, update the dashboard with current information from connected apps and files. The refresh should show: - what changed since the last version - what got more urgent - what is blocked - what improved - the top 3 actions I should take next Preserve the dashboard structure. Update the state. ``` ```text Add these dashboard controls: - Settings panel for update frequency - Manual override for any incorrect status or priority - Archive button for updates I have handled - Click-to-open links for every source update - Dark mode toggle - Simple animations only where they make status changes easier to notice ``` If Claude cannot fully wire every button to every connector in your setup, keep the control anyway and have it generate the right prompt. The habit still works: open the pinned artifact, refresh the state, review what changed, and take action. ![04-command-center-dashboard-overview](https://tru-images.b-cdn.net/guide-assets/69/69e7e57058b6351fdf17d168/04-command-center-dashboard-overview.png) A finished command center should make the important numbers and connected-source context visible at a glance. > Pro tip: Ask for “since yesterday” and “since last week” summaries. Daily changes tell you what to do now. Weekly changes tell you whether the system is improving or drifting. ### Pro tip > Ask Claude to summarize its understanding before it builds. A quick “Here is what I heard” step catches bad assumptions early. ### Guide: Build a Lindy Agent That Never Drops a Follow-Up > In this guide, you will learn how to use Lindy as an AI teammate for work that falls through the cracks between meetings, Slack, and email. - Canonical page: https://app.therundown.ai/guides/build-a-lindy-agent-that-never-drops-a-follow-up - Machine-readable page: https://app.therundown.ai/guides/build-a-lindy-agent-that-never-drops-a-follow-up.md - Difficulty: beginner - Categories: AI, Productivity - Instructors: Billy Howell - Published: 2026-09-04 - Updated: 2026-09-04 ### Overview #### The Rundown In this guide, you will learn how to use Lindy as an AI teammate for work that falls through the cracks between meetings, Slack, and email. We tested it on real ops work to see where it earns its keep. #### Who Is This Useful For? - Teams that lose follow-ups between meetings, Slack, email, and project trackers - Operators who want an AI teammate for recurring work instead of another chatbot - Anyone who wants useful automation without handing an agent full control on day one #### What You Will Build You will set up Lindy around your real work tools, give it a recurring ops workflow, and use it from Slack. We tested the setup by asking Lindy to analyze our guide library, build a Monday staff meeting deck, and turn a Slack request into a new Notion guide idea. ![The finished Monday staff meeting deck Lindy built as a permanent web link](https://tru-images.b-cdn.net/guide-assets/local-424/build-a-lindy-agent-that-never-drops-a-follow-up/01-finished-monday-staff-meeting-deck.png) #### What You Need to Get Started - A Lindy (https://www.lindy.ai/) account - The work apps you want Lindy to use, such as Slack, Gmail, Google Calendar, Notion, or Google Drive - One annoying recurring task that would be easier if someone gathered the context for you #### Going Further Once your first routine works, add more ways to reach Lindy. You can email the agent, text it, or upload reusable skills and memory in the Files section. Then build out the boring work that keeps coming back: meeting prep, CRM updates, scheduling, follow-up, weekly reports, and internal handoffs. Those routines are where Lindy earns its keep. You can also ask it to build heavier deliverables like presentations or research briefs. Just give it time. Lindy works more like a teammate handling an assignment in the background than a chatbot firing back an instant answer. ### Step 1: Connect Lindy to the Tools You Use Create a Lindy account and start with the integrations. Lindy can connect to Gmail, Google Calendar, Notion, Slack, Google Drive, and a long list of other work apps. Do not connect everything just because you can. Pick the tools Lindy needs for one job, then check the privacy settings for each connection. Lindy gives you more control here than a lot of AI teammate products, so use it. ![Lindy connected to Gmail, Google Calendar, Notion, and Slack](https://tru-images.b-cdn.net/guide-assets/local-424/build-a-lindy-agent-that-never-drops-a-follow-up/02-lindy-integrations.png) Lindy also gives you a few ways to reach your agent. You can use the web app, DM it in Slack, email it, or add your phone number and text it. The phone option is worth setting up if you want to hand off a quick task without opening another app. > Pro Tip: After connecting Slack, open Lindy's settings and turn off public channel auto-join if you do not want the agent appearing in every public channel. ### Step 2: Start With Routines Go to Routines, then open Discover. This is where Lindy becomes useful quickly because you do not have to build every workflow from scratch. You will find prebuilt routines for things like: - meeting prep - email follow-up - scheduling - receipt and invoice filing - Slack reminders and channel summaries - weekly leadership updates - promises and open decisions ![Prebuilt workspace routines inside Lindy's Discover section](https://tru-images.b-cdn.net/guide-assets/local-424/build-a-lindy-agent-that-never-drops-a-follow-up/03-lindy-routines.png) A strong first routine is weekly meeting prep. Tell Lindy to pull the right context from your calendar, Slack, Notion, or Drive and prepare a meeting doc before each recurring meeting. This is the sweet spot for Lindy. Use it for the annoying operational work that has to happen every week, not the creative work you want back immediately. ### Step 3: Give Lindy a Real Task in Slack Once Slack is connected, DM Lindy and give it a real work question. The thread will also show up in the Lindy web app, so you can start in Slack and continue in the browser. We asked Lindy to review the Rundown University guides from this year and summarize where AI products are going, how our coverage has changed, and what kind of content we make. ![Lindy analyzing the Rundown University guide library from a Slack thread](https://tru-images.b-cdn.net/guide-assets/local-424/build-a-lindy-agent-that-never-drops-a-follow-up/04-slack-analysis-result.png) The result was genuinely useful. Lindy found the shift from chatbots to agents that do work, the rise of coding agents, the dominance of OpenAI and Claude, and our move toward outcome-first framing. You can use the same pattern for customer research, sales follow-up, internal handoffs, or any question that requires context from several connected tools. ### Step 4: Let Lindy Build in the Background Lindy runs in the cloud, so it can keep working after you leave the tab. That makes it better for background assignments and recurring routines than instant, on-demand work. We used one of Lindy's suggested prompts to build a Monday staff meeting deck. It took about 10 to 15 minutes, pulled together the right information, anonymized sensitive details when asked, and delivered the finished presentation as a permanent web link. That last part is especially nice. There was no PowerPoint file to download, no Google Slides import, and no hunting around for the final deck. Lindy built and deployed the presentation, then gave us the link. ### Step 5: Use Approvals for Cross-App Handoffs In Slack, we asked Lindy to research Claude 5.1 and turn the result into a new guide idea in Notion. Because it needed to create a page through a connected account, Lindy asked for approval first. After approval, it created the page, filled in the database properties, included the source, and sent the Notion link back in Slack. ![The Notion guide idea Lindy created after approval](https://tru-images.b-cdn.net/guide-assets/local-424/build-a-lindy-agent-that-never-drops-a-follow-up/05-notion-guide-idea.png) This is how we would use Lindy with a team. Let people assign work in Slack, but keep approvals in place whenever the agent needs to create, send, or update something through another person's account. ### Guide: Build a No-Code Voice Agent for Customer Intake With Grok > In this guide, you will learn how to set up a Grok voice agent that answers inbound calls, qualifies leads, and transfers the right callers straight to your phone. - Canonical page: https://app.therundown.ai/guides/build-a-no-code-voice-agent-for-customer-intake-with-grok - Machine-readable page: https://app.therundown.ai/guides/build-a-no-code-voice-agent-for-customer-intake-with-grok.md - Difficulty: beginner - Categories: Sales, Consulting & Customer Success, Automation & Agents - Instructors: Billy Howell - Published: 2026-07-14 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to set up a Grok voice agent that answers inbound calls, qualifies leads, and transfers the right callers straight to your phone. #### Who This Is Useful For - Local service businesses that need to qualify new callers before booking time with the team. - Agencies and sales teams that want a faster first response without handing every call to a person. - Anyone who needs a repeatable inbound-call workflow that captures the right details before a handoff. #### What You Will Build A Grok voice agent with its own phone number that can answer an inbound call, gather lead details, use your business context, and transfer a qualified caller to a real person. You will also have a transcript and recording to review after every test call. ![01-grok-conversation-call-transcript-and-audio.png](https://tru-images.b-cdn.net/guide-assets/local-349/build-a-no-code-voice-agent-for-customer-intake-with-grok/01-grok-conversation-call-transcript-and-audio.png) #### What You Need To Get Started - A fresh xAI account; a Twitter account is optional, and no credit card is needed for the first test. - A credit card when you are ready to use voice tokens. - A business that needs leads qualified and a phone number where you can receive transferred calls. #### Going Further Once the call flow is reliable, use connectors to turn the conversation into work your team can act on. Connect Google Drive or Notion to give the agent more context, or connect HubSpot so a qualified call can create a new contact with the details it collected. You can also use a custom MCP or API tool when the handoff needs to reach an internal system. Start with one useful action: save the caller's details, create a follow-up task, or check a calendar. If you also transfer the caller, have the agent save the note first and test that exact sequence with fictional data. A transfer may end the call, so confirm the record is created before you rely on the workflow. Keep every connector permission as narrow as possible. For example, let an agent check availability or create a new follow-up, rather than giving it broad access to search through your calendar, inbox, or CRM. Once the basics are dependable, you can extend the same agent to update meetings, route different lead types, or add a voice assistant to your website. ### Step 1: Describe Your Business and the Calls You Want Open Grok Voice Agent Builder (https://console.x.ai/team/default/voice/agents?builder=1) and create a new voice agent. Start with one narrow job: answer new-business calls, collect the details your team needs, and transfer the right caller. Describe what your business does, who it serves, and the few questions that decide whether a caller is a fit. Add your website or a short business brief so the agent has accurate context when it answers basic questions. ![02-grok-agent-setup-business-website-calendar-choice.png](https://tru-images.b-cdn.net/guide-assets/local-349/build-a-no-code-voice-agent-for-customer-intake-with-grok/02-grok-agent-setup-business-website-calendar-choice.png) Keep the first version focused. A service business might ask about the caller's need, location, urgency, and callback window. An agency might ask about budget, timeline, and the service they need. The goal is a useful handoff, not a fully automated sales call. > Pro tip: Give the agent one call type before adding every edge case. It is much easier to improve a clear lead-qualification flow after a few test calls. > Pro tip: If the onboarding asks whether the agent should answer every service question or qualify and transfer, start with qualify and transfer. You can always expand the agent's role later. ### Step 2: Connect Your Calendar If availability matters, add the Google Calendar connector and enable only the actions you want the agent to use. This setup can check open times and create an event after a caller confirms the time. ![03-grok-google-calendar-connector-enabled-tools.png](https://tru-images.b-cdn.net/guide-assets/local-349/build-a-no-code-voice-agent-for-customer-intake-with-grok/03-grok-google-calendar-connector-enabled-tools.png) Keep the permission narrow at first. If you enable Create Event, tell the agent to get a clear confirmation before it uses the action. Let a person confirm the appointment until you have tested the full booking flow. > Pro tip: Use a separate scheduling calendar, or turn off calendar search and leave only Availability and the actions you need. That prevents the agent from revealing the details of an existing event. ### Step 3: Set Up the Human Handoff Tell Grok when to transfer a caller to you or someone on your team. A qualified lead, an urgent request, or a caller who explicitly asks for a person are all good handoff rules. In the agent's tool settings, add the transfer action and give it a short rule for when to use it. Also set boundaries: do not quote prices, promise availability, diagnose a problem, or take payment details unless you have intentionally built and tested those workflows. ![05-grok-add-tool-options-transfer-and-end-call.png](https://tru-images.b-cdn.net/guide-assets/local-349/build-a-no-code-voice-agent-for-customer-intake-with-grok/05-grok-add-tool-options-transfer-and-end-call.png) That keeps the agent useful without asking it to make decisions your team should own. ### Step 4: Claim a Number and Test the Call Claim a phone number for the agent, including the area code that makes sense for your callers. Then make a browser test call to hear the flow before you share the number publicly. ![04-grok-provision-new-phone-number.png](https://tru-images.b-cdn.net/guide-assets/local-349/build-a-no-code-voice-agent-for-customer-intake-with-grok/04-grok-provision-new-phone-number.png) Test the moments that matter: a clear lead, a caller who needs a person, and a question the agent cannot answer. Review the call transcript and recording afterward. If a question feels too long or the handoff happens too late, tighten the instruction and run another test. You will need to call the agent from a phone to test a real transfer to your number. > Pro tip: Add the new agent number to your contacts before testing. It is a new number and may otherwise be treated as spam by your phone. ### Guide: Build a Project Room Where Humans and AI Agents Work Together (Buzz Tutorial) > In this guide, you will learn how to set up a team of agents that you and your teammates can collaborate with in a free, open-source chat app called Buzz. - Canonical page: https://app.therundown.ai/guides/build-project-room-humans-ai-agents-buzz-tutorial - Machine-readable page: https://app.therundown.ai/guides/build-project-room-humans-ai-agents-buzz-tutorial.md - Difficulty: beginner - Featured tool or company: Buzz - Categories: Automation & Agents, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-08-03 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to set up a team of agents that you and your teammates can collaborate with in a free, open-source chat app called Buzz. #### Who This Is Useful For - Operators who want to manage several AI agents without losing track of prompts, files, and progress - Consultants and agencies that want a shared project room for internal teammates, clients, and agents - Anyone building repeatable workflows that combine research, content, coding, or other specialist roles #### What You Will Build You will create a Buzz project room with a lead agent, a research agent, and a clearly scoped assignment. The lead agent will delegate the research while beginning a simple webpage in parallel, giving you one place to watch the work, request updates, correct mistakes, and receive finished files. ![Finished AI productivity case study webpage created by the Buzz agent team](https://tru-images.b-cdn.net/guide-assets/local-414/build-project-room-humans-ai-agents-buzz-tutorial/01-ai-productivity-webpage-result.png) This is a case study site agents built via collaborating in Buzz #### What You Need To Get Started - A Macbook computer - A Buzz (https://buzz.xyz) installation - A Codex or Claude Code account with its command-line tool installed and logged in - One small project that can be divided into two clear pieces of work #### Going Further Once the workflow works, ask the lead agent to turn it into a reusable team that you can bring into future project channels. You can also invite human teammates and clients to the relevant rooms, much as you would in Slack, so decisions, agent activity, and deliverables stay together. Buzz's experimental Projects and Pulse views can help you explore project tracking and agent activity across the workspace. Start with one dependable channel workflow first, then add more agents, integrations, or Git-based project management as the work becomes more complex. Next, learn how to turn your ChatGPT subscription into a team of useful agents with Raft. (https://app.therundown.ai/guides/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft) ### Step 1 Download Buzz and connect your coding agent Go to buzz.xyz (https://buzz.xyz), click Get the App, and install Buzz. Open the app and choose the coding harness you already use, such as Codex or Claude Code. Buzz uses that account to power the agents in your workspace, so make sure its command-line tool is installed and authenticated before continuing. During setup, choose a model or accept the recommended defaults. You can adjust agent models and other defaults later from the app's settings. ![Buzz onboarding screen for choosing a default coding harness and model](https://tru-images.b-cdn.net/guide-assets/local-414/build-project-room-humans-ai-agents-buzz-tutorial/02-buzz-default-model-settings.png) Choose the coding harness you already use, then select a default model for new agents. If you are not using the same Mac setup shown in the video, check the official Buzz repository (https://github.com/block/buzz) for the latest packaged releases and platform instructions. ### Step 2 Turn your projects into focused channels Buzz begins with an onboarding conversation. Tell the onboarding agents about your role, the work you manage, and the main projects you want help with. Then ask them to organize those projects into separate chat channels. A useful instruction looks like: ```text I manage several projects that involve research, content, and simple web builds. Create one channel for each active project, and add me to every channel you create. ``` Treat a channel like a project room rather than a general-purpose AI chat. Keeping one outcome per channel makes its messages, files, and agent activity easier to follow. If an agent creates a channel without adding you, ask it to add you before the work begins. ![Buzz workspace with project channels, starter agents, and a shared message composer](https://tru-images.b-cdn.net/guide-assets/local-414/build-project-room-humans-ai-agents-buzz-tutorial/03-buzz-project-workspace.png) Each channel keeps the people, agents, messages, and files for one project in the same room. Channels can also be private to their invited members. That makes it possible to separate internal work, client work, or individual projects inside the same Buzz workspace. ### Step 3 Create a small agent team Open Agents and review the agents created during onboarding. For this workflow, keep the team simple: - A lead agent that receives your instructions and coordinates the project - A research agent that gathers and summarizes the case studies - A maker or coding agent that can turn the research into a webpage You can ask the lead agent to create a dedicated research agent, or create one yourself from the Agents area. Choose a recognizable name and icon, review its instructions and connected tools, and add it to the project channel. An agent cannot participate in a channel it has not joined. Review each agent's permissions before giving it work. Buzz keeps agents in a sandbox by default during the demonstrated workflow, and access outside the workspace should require your approval. Limit who can instruct an agent when a project contains sensitive context or when you want tighter control over token usage. For your first project, avoid an elaborate roster. A lead, a researcher, and one maker are enough to learn how delegation works. You can turn a successful workflow into a more specialized reusable team later. ### Step 4 Give the lead agent one clear project In the project channel, address the lead agent and give it an outcome with two independent parts. Use this prompt from the demonstration: ```text Have the Research Agent find five useful AI productivity case studies. While it researches, draft a webpage that presents each one consistently. ``` This tells the lead agent what should be delegated and what can happen in parallel. The research agent can collect the source material while the lead or maker begins the page structure. ![Buzz project channel showing a lead agent and Research Scout working on the same assignment](https://tru-images.b-cdn.net/guide-assets/local-414/build-project-room-humans-ai-agents-buzz-tutorial/04-buzz-agents-working-in-parallel.png) Buzz shows both agents working beneath the project prompt, making the parallel handoff visible. Make sure the agent mention resolves before you send the message. Type the @ symbol, select the agent from the menu, and confirm its name appears as a mention. If you type an agent's name without tagging it, the agent may never activate. You can also enable the setting that keeps addressed agents active when you expect a longer exchange. If you want a dedicated coding agent to own the build, say so explicitly. For example: ```text Delegate the research to the Research Agent and the webpage build to the coding agent. Keep the page simple and use a familiar stack. ``` Starting with a simple, familiar stack reduces the chance that multiple agents choose incompatible libraries or create unnecessary setup work. ### Step 5 Monitor the work and collect the result Follow the agents inside the channel as you would follow human teammates. Reactions show when an agent has seen a message, and the activity view shows what it is doing. Threads keep follow-up questions attached to the relevant update. Ask for a status update when you need one. If an agent chooses an unsuitable tool, library, or approach, correct it in the channel and restate the constraint. The shared history gives the team the context needed to continue without starting a new chat. When the work is ready, ask the agent to attach the files or screenshots directly in the channel. Review the case studies, confirm that each one follows the same structure, and test the webpage before treating the project as finished. Visible activity is useful, but it does not replace checking the final output. Keep agent permissions narrow as the project grows. Approve access to other folders only when the task requires it, and verify important source claims or code changes before sharing the work outside the project room. ### Guide: Build a Reusable AI Design System With Open Design > In this guide, you will learn to use a free AI design tool to extract your brand's design rules for you. - Canonical page: https://app.therundown.ai/guides/build-a-reusable-ai-design-system-with-open-design - Machine-readable page: https://app.therundown.ai/guides/build-a-reusable-ai-design-system-with-open-design.md - Difficulty: beginner - Categories: AI - Instructors: Billy Howell - Published: 2026-08-25 - Updated: 2026-08-25 ### Overview #### The Rundown In this guide, you will learn to use a free AI design tool to extract your brand's design rules for you. This will let you easily make consistent design assets using AI. #### Who This Is Useful For - Marketers who need social assets, ads, or campaign pages that still look like their brand. - Founders and operators who want a faster starting point for a consistent landing-page prototype. - Anyone using AI for design who wants one set of rules to carry across different visual formats. #### What You Will Build You will turn an approved website or brand reference into a reusable Open Design system, then use it to make a branded motion asset and a landing-page prototype. The goal is not a production-ready site. It is a consistent visual starting point you can review, correct, and reuse. ![A completed branded event countdown prototype made with Open Design](https://tru-images.b-cdn.net/guide-assets/local-326/build-a-reusable-ai-design-system-with-open-design/01-open-design-finished-countdown-prototype.png) #### What You Need To Get Started - Open Design (https://open-design.ai) installed on your computer. - An AI agent you can use from the command line, such as Codex CLI. - An approved website URL, logo, image set, Figma file, code repository, or DESIGN.md to use as brand source material. #### Going Further Use ChatGPT or another planning tool to shape the idea first, then bring the settled brief into Open Design when you are ready to make it visual. Keep one system per brand or campaign, and save useful corrections as reusable rules when the workspace offers that option so the next ad, post, animation, or prototype starts closer to the right answer. ### Step 1 Connect a Local AI Agent Open Design is the visual workspace. You still need to connect the AI agent that will do the design work. During onboarding, choose the local agent option, then select the command-line agent Open Design finds on your computer. In the recording, that was Codex CLI. For the local-agent route shown here, install and authenticate the agent's CLI first; Open Design can then detect it. Once the workspace opens, use the design-system selector near the top of the project. This is where you will create a new system or choose one you have already made. Agent support and intake options change, so check the current Open Design site (https://open-design.ai/) and project README (https://github.com/nexu-io/open-design) before choosing a source or workflow. ### Step 2 Extract and Check Your Brand System Choose Create New Design System. Paste in your website under GitHub or website, click Add, then choose Continue to generation. Open Design uses the source material to assemble a system you can use in later jobs. ![The Open Design screen for adding a website as design-system source material](https://tru-images.b-cdn.net/guide-assets/local-326/build-a-reusable-ai-design-system-with-open-design/02-open-design-website-source-input.png) When the extraction finishes, inspect the result before you generate anything. The website pass in this demo missed the logo, so the presenter reviewed an already-extracted system with the logo, typography, colors, voice and tone, plus light and dark styling and button treatments. ![Recovered brand assets and typography inside the extracted Open Design system](https://tru-images.b-cdn.net/guide-assets/local-326/build-a-reusable-ai-design-system-with-open-design/03-open-design-extracted-brand-assets.png) > Pro tip: Treat the first extraction as a draft. The website pass in the recording did not find the logo. If something important is missing, use the AI optimization option or add the logo, images, or DESIGN.md yourself. Correct the system before asking it to create assets. You can also start from a description, a local code repository, a Figma file, or an existing DESIGN.md. The useful rule is simple: give the tool the clearest approved source material you have, then review what it inferred. ### Step 3 Generate Specific Assets From the System Select the new design system before starting a new generation. Then choose the output type and describe the actual job clearly. The demo used Hyperframes for a short motion asset and created a separate countdown landing-page prototype using the same design system. In the demo, the first prompt was: ```text Build an animation showing the growth that a Rundown University member will have over six months. ``` For your own project, make the output, subject, audience, and constraint explicit: ```text Build a [format] for [audience] that explains [specific outcome]. Use the active [brand] design system. Include [required brand element], and keep the result to [length or scope]. ``` For example, the second demo prompt was a prototype for a countdown until a new Rundown University event. Open Design showed a task list as it worked, including applying brand tokens, composing landing sections, and adding the countdown interaction. Keep the brief specific. A vague request for a Hyperframe in the recording returned something closer to a landing-page demo. Naming the format and desired behavior gives the agent much better direction. ### Step 4 Review, Correct, and Reuse the Rules Open the finished result and check the parts that make it recognizably yours: logo use, typography, colors, layout, motion, and the actual job the asset needs to do. The prototype in the demo included the event countdown, a reserve-a-seat action, and a mobile view to inspect. Use normal feedback when something is off. The first motion asset ran too quickly and did not use the brand logo. The correction was direct: ask it to double the animation length and use the logo from the design system. The revised result addressed those two requested changes. ![A corrected branded motion asset generated with the active Open Design system](https://tru-images.b-cdn.net/guide-assets/local-326/build-a-reusable-ai-design-system-with-open-design/04-open-design-corrected-animated-asset.png) > Pro tip: Feedback is part of the system, not a failure of it. When a result misses the mark, name the visible problem and the exact correction. Where Open Design offers to save a rule, a concrete instruction such as “use the logo from the design system” gives the next request better context. If the result is a prototype, review it as a prototype. You can preview it, inspect the mobile version, and download its code to continue work in an editor such as Cursor. Do not mistake the generated page for a finished production website without your usual content, accessibility, technical, and brand review. ### Guide: Build a Short Form Video Farm With Higgsfield + Claude Code > In this guide, you will learn how to set up a Codex project that lets you test, generate, and improve short-form video with Higgsfield from your terminal. - Canonical page: https://app.therundown.ai/guides/build-a-short-form-video-farm-with-higgsfield-claude-code - Machine-readable page: https://app.therundown.ai/guides/build-a-short-form-video-farm-with-higgsfield-claude-code.md - Difficulty: beginner - Featured tool or company: The Rundown AI - Categories: Marketing & Growth, Content & Creative, Automation & Agents - Instructors: Billy Howell - Published: 2026-06-01 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to set up a Codex project that lets you test, generate, and improve short-form video with Higgsfield from your terminal. Instead of making one-off clips, you will create a video workstation with campaign folders, saved outputs, reusable skills, and a feedback loop. #### Who This Is Useful For - Creators and marketers who want a faster way to test short-form video ideas without rebuilding the same prompts every time - Operators and agencies who need campaign folders, saved outputs, and tracking files instead of scattered generations - Anyone building agent workflows who wants a practical example of CLI tools, skills, feedback, and automation working together #### What You Will Build You will build a Higgsfield video generation workstation inside a normal project folder. The folder gives Claude Code or Codex a place to create campaigns, save generated videos, track feedback, and turn the process into reusable skills. ![01-video-workstation-folder](https://tru-images.b-cdn.net/guide-assets/local-227/build-a-short-form-video-farm-with-higgsfield-claude-code/01-video-workstation-folder.png) By the end, you will have a working setup that can create a campaign, generate two Higgsfield videos from the terminal, save the outputs, improve the prompts from your feedback, and eventually run the same process on a schedule. #### What You Need To Get Started - Claude Code, Codex, or another agent that can work inside a local project folder - A Higgsfield account - The Higgsfield CLI from https://higgsfield.ai/cli - A brand, product, or campaign idea to use for the first test - A few minutes to review prompts before any generation spends credits #### Going Further Use the system manually for about five days or five campaigns before you automate it. That gives Claude enough examples to learn from your real feedback instead of guessing your taste. After that, ask Claude to create a daily automation: ```text Create a daily automation for this Higgsfield video workstation. Every day, it should: 1. Review the previous campaigns and my feedback. 2. Identify patterns in what worked and what failed. 3. Come up with one new campaign idea. 4. Create the campaign folder. 5. Generate one set of videos. 6. Save the outputs and update tracking.md. 7. Suggest one improvement to the generation or iteration skill. Before spending Higgsfield credits, show the planned campaign and commands for approval unless I explicitly turn on automatic generation. ``` Do not automate blind generation on day one. Start by automating the review, idea generation, and skill-improvement loop. Once the campaigns are consistently good, you can decide whether daily generation should run automatically or wait for approval. The reusable pattern is simple: ```text 1. Create the workstation. 2. Generate two videos. 3. Save the outputs. 4. Turn the workflow into a skill. 5. Give feedback. 6. Turn the improvement loop into a skill. 7. Automate after the system has examples to learn from. ``` That is how you move from one-off Higgsfield experiments to a short-form video system you can keep scaling from the terminal. ### Step 1 Create the Project Folder Start with a new folder for the workflow. Do not run this from a random downloads folder. The point is to give the agent one stable workspace where it can create campaign files, skill files, and tracking docs. Use something simple: ```text mkdir higgsfield-video-workstation cd higgsfield-video-workstation ``` Then open Claude Code in that folder: ```text claude ``` You can use Codex instead if that is your preferred agent. The important part is that the agent is operating inside the same folder where the Higgsfield workflow will live. > Pro tip: Use a dedicated folder for this. Skills and campaign tracking only become useful when the agent can find the same files again next time. ### Step 2 Install the Higgsfield CLI Open the Higgsfield CLI setup page and follow the three-part flow: install the CLI, sign in, and add the Higgsfield skills package to your agent. ![02-higgsfield-cli-setup-steps](https://tru-images.b-cdn.net/guide-assets/local-227/build-a-short-form-video-farm-with-higgsfield-claude-code/02-higgsfield-cli-setup-steps.png) Install the CLI: ```text npm install -g @higgsfield/cli ``` Then authenticate: ```text higgsfield auth login ``` The login opens a browser window and asks you to connect Higgsfield. Review the permissions before you approve access. The CLI needs to access generated content, uploaded media, workspaces, subscription plans, and credits so it can run and retrieve generations. After that, add the Higgsfield skills package: ```text npx skills add higgsfield-ai/skills ``` When prompted, choose the Higgsfield generation skill and install it at the project level if this workstation is meant to stay self-contained. ![03-install-higgsfield-skill](https://tru-images.b-cdn.net/guide-assets/local-227/build-a-short-form-video-farm-with-higgsfield-claude-code/03-install-higgsfield-skill.png) Before you generate anything, ask Claude to check the setup: ```text Check whether the Higgsfield CLI is installed and authenticated in this project. Do not generate anything yet. First tell me: - which Higgsfield commands are available - whether the Higgsfield generation skill is installed - what command you would use for a low-risk test generation - anything that might spend credits or require approval ``` > Pro tip: Make the agent show the planned command before it runs a generation. That one approval step saves you from wasting credits on the wrong duration, model, or prompt. ### Step 3 Have Claude Set Up the Workstation Now have Claude interview you about the brand and workflow. This is the setup step that turns a blank folder into a video generation workstation. Use this prompt: ```text Interview me about my brand, audience, video goals, formats, and campaign structure. Then set this folder up as a Higgsfield short-form video workstation. Create: - campaigns/ - skills/ - a brand guidelines file - a campaign tracking template - a README that explains how to run the workflow Inside each campaign, use: - input/ - output/ - drafts/ - finals/ - prompt-generation notes - tracking.md Do not generate videos yet. ``` Answer the interview quickly. Claude should ask about the brand, audience, platform, video format, visual style, and what counts as a good clip. You are not writing a full creative strategy. You are giving the system enough context to make the first campaign coherent. Once Claude creates the files, inspect the folder. If it creates too many layers, ask it to simplify. If it misses tracking, ask it to add a campaign tracking file before moving on. > Pro tip: Keep the first folder boring. A clean campaigns/, skills/, brand file, and tracking file is better than a giant production system you stop using after one day. ### Step 4 Generate the First Two Videos Now create the first campaign. This is the manual run that teaches the system what the reusable skill needs to do later. Use a prompt like this: ```text Create a new campaign for [BRAND_OR_PRODUCT]. Use the brand guidelines and campaign template in this folder. For this campaign: - write a short campaign brief - create two short-form video concepts - turn each concept into a Higgsfield-ready prompt - show me the commands before running them - generate two videos with Higgsfield from the terminal - save the outputs in the campaign output folder - update tracking.md with the prompt, command, output file, status, and notes ``` Claude should prepare the prompts and commands first. Review them before it runs anything. In the demo, Claude caught a practical constraint: Higgsfield's video model had a three-second minimum, so a two-second test needed to become a three-second generation. ![04-claude-runs-higgsfield-generation](https://tru-images.b-cdn.net/guide-assets/local-227/build-a-short-form-video-farm-with-higgsfield-claude-code/04-claude-runs-higgsfield-generation.png) When the generation finishes, check that the outputs are saved in the campaign folder, not just linked in the chat. The folder should include the campaign brief, prompt notes, generated drafts, and tracking file. ![05-campaign-output-folder](https://tru-images.b-cdn.net/guide-assets/local-227/build-a-short-form-video-farm-with-higgsfield-claude-code/05-campaign-output-folder.png) > Pro tip: Do not turn the workflow into a skill until one full campaign has run end to end. The first run shows you what the skill needs to remember and what approval steps should stay manual. ### Step 5 Turn the Generation Flow Into a Skill After the first two videos are saved, ask Claude to package the working process as a skill. Use this prompt: ```text Turn the campaign generation workflow we just completed into a reusable skill. The skill should: 1. Read the brand guidelines. 2. Create a new campaign folder. 3. Ask for missing campaign details. 4. Write two Higgsfield-ready prompts. 5. Show planned commands before generation. 6. Generate two videos. 7. Save outputs in the campaign folder. 8. Update tracking.md. 9. Summarize what worked and what needs another pass. Do not include one-time setup steps like installing the Higgsfield CLI. Assume the project is already configured. ``` That last instruction matters. A skill should capture the repeatable work, not the setup you only do once. If the skill tries to reinstall tools every time, have Claude strip that out. Test the skill on a second campaign or a different product angle. If it asks too many setup questions, point it back to the brand file. If it skips approval before generation, add that as a required step. > Pro tip: Separate setup from execution. The project README can explain installation. The skill should focus on creating campaigns, generating videos, saving outputs, and tracking results. ### Step 6 Turn Feedback Into a Second Skill The first two videos are not the finish line. They are the first data point. The next move is to give Claude feedback and make the prompt-improvement loop reusable too. Use this prompt after reviewing the videos: ```text Review these two generated videos against the campaign brief. Here is my feedback: - Video 1: [WHAT WORKED / WHAT DID NOT] - Video 2: [WHAT WORKED / WHAT DID NOT] Improve the prompts for both videos. Generate the next versions. Save the new outputs next to the originals. Update tracking.md with what changed and why. ``` Once that works, turn the iteration process into its own skill: ```text Turn this prompt-improvement process into a reusable skill. The skill should: 1. Read a campaign folder. 2. Compare the campaign brief, original prompts, outputs, and feedback. 3. Improve the prompts. 4. Generate the next versions. 5. Save the revised outputs. 6. Update tracking.md with the prompt changes and feedback summary. ``` This gives you two clean workflows: one skill creates new campaigns, and another improves existing campaigns. Keeping those separate makes the workstation easier to use. > Pro tip: Feedback should be specific. "Make it better" will not teach the system much. "The first frame needs a clearer product shot, the motion feels too slow, and the hook should happen in the first second" is useful. ### Guide: Build a Website Hands-Free With Claude Voice > In this guide, you will learn how to use Claude Voice to build better websites faster. - Canonical page: https://app.therundown.ai/guides/build-a-website-hands-free-with-claude-voice - Machine-readable page: https://app.therundown.ai/guides/build-a-website-hands-free-with-claude-voice.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Coding & App Development, Content & Creative - Instructors: Billy Howell - Published: 2026-08-05 - Updated: 2026-08-05 ### Overview #### The Rundown In this guide, you will learn how to use Claude Voice to build better websites faster. We'll walk you through a simple technique that you can use every time you want to build a prototype. #### Who This Is Useful For - Creators and freelancers who want to turn a rough portfolio or landing-page idea into a visual prototype. - Founders and marketers who need a fast way to explore a website's message, structure, and brand direction. - Anyone who thinks more clearly out loud and wants to replace a long build prompt with a guided conversation. #### What You Will Build You will build a static website prototype from a voice-led creative brief. In the example, Claude turns a conversation about a wildlife photographer into an HTML and Tailwind artifact with a hero, selected work, an about section, and a contact call to action. ![Finished wildlife portfolio prototype](https://tru-images.b-cdn.net/guide-assets/local-421/build-a-website-hands-free-with-claude-voice/01-finished-wildlife-portfolio-prototype.png) The same workflow works for portfolios, landing pages, blogs, and other simple sites where the first goal is to see and improve the idea. In the recording, the text build phase—from sending Build the prototype to the corrected artifact—took just over eight minutes. The Voice interview and asset preparation happen before that build timer starts. #### What You Need To Get Started - A Claude account. Voice mode is currently a beta feature on Free, Pro, Max, Team, and Enterprise plans across Claude Desktop, the web, iOS, and Android. - The latest Claude desktop app if you want to follow the recorded workflow. - Optional photos, logos, or screenshots of websites whose design principles you like. - A rough website idea or prototype goal. #### Going Further To turn the approved artifact into a real website, give its HTML to Claude Code and ask it to rebuild the direction in a new project. For anything with accounts, a database, payments, or substantial custom behavior, use the Voice interview to create a product requirements document first, then hand that PRD to Claude Code in a separate project folder. You can also create one reusable Claude project dedicated to these interviews. Start a new thread for each idea, let Voice turn the conversation into a concise brief or PRD, and keep the build work in its own project. That gives you the speed of talking through an idea without mixing multiple websites into one long conversation. ### Step 1 Prepare Your Assets and Start Voice If you already have photos or design references, place them in one folder before you begin. This makes them easier to add when Claude is ready to build. You can also start without assets and use placeholders for the first prototype. Use reference sites for principles such as typography, whitespace, image density, and navigation. Do not ask Claude to copy another site's photos, text, logos, or distinctive layout. Open Claude and start a regular chat. Click Voice to begin a live conversation. The Voice control is different from the microphone used to dictate a text message. It also does not appear inside Cowork, so begin in a regular chat and keep that conversation open for the rest of the workflow. Anthropic's Voice mode instructions (https://support.claude.com/en/articles/11101966-use-voice-mode) cover the currently supported Claude surfaces. ### Step 2 Let Claude Interview You Start with the outcome, then ask Claude to interview you. This is the part that replaces the giant build prompt: Claude asks for the missing decisions one at a time, and your answers become the creative brief. ![Claude Voice website interview](https://tru-images.b-cdn.net/guide-assets/local-421/build-a-website-hands-free-with-claude-voice/02-claude-voice-interview.png) Use this reusable opening: ```text I want to build a [SITE TYPE] for [PERSON, BUSINESS, OR PROJECT]. The main goal is [PRIMARY GOAL]. Interview me so we can plan a strong prototype. Ask one question at a time, ask what assets and design references I already have, and do not build anything yet. ``` In the recorded example, the site was a portfolio for a wildlife photographer who works with conservation groups. The goal was to showcase the work and invite project inquiries rather than sell products. Answer the questions that materially change the site. Claude may ask about: - the audience and primary call to action - the sections the page needs - the story or feeling the design should create - existing branding, photos, and reference sites - the visual direction, including color, type, spacing, and image treatment Tell Claude what assets you have even if you have not uploaded them yet. For example, the demo mentioned six wildlife photos and several portfolio references. That gave Claude a concrete image count and a design direction to plan around. When the first version only needs to communicate an idea, keep the build simple. Ask for a static site unless the experience truly needs accounts, live data, payments, or other application logic. > Pro tip: Ask for a static site when you are building a portfolio, landing page, blog, or mockup. It keeps the first prototype simpler for you and Claude. ### Step 3 Turn the Conversation Into a Build Plan Once the audience, page structure, and visual direction feel right, ask Claude to write the full plan into the chat. This gives the text-mode build a clear source of truth. ```text Repeat the full website plan and brand direction in this chat. Include the audience, goal, page sections, call to action, sample copy, visual direction, image treatment, and the static-site approach. Resolve any important gaps before we build. ``` As soon as the complete plan appears as text, click Stop; you do not need to wait for Claude to read it aloud. Review the plan in the chat, make sure it names the intended visitor, the action that visitor should take, and the main sections of the page, and correct anything important. Stay in the same conversation so Claude retains the interview and plan for the build request. ![Voice-generated website plan](https://tru-images.b-cdn.net/guide-assets/local-421/build-a-website-hands-free-with-claude-voice/03-voice-generated-site-plan.png) ### Step 4 Add Your Assets and Build the Prototype Switch to text mode in the same conversation. Upload the photos or other approved assets you want Claude to use. If you prefer to keep the work organized, open the + menu, choose Add to project, and select Start new project. Name the project, keep the existing Voice chat with it, and add your reference files as project context. A project is optional; you can upload the files directly to the conversation instead. ![Wildlife portfolio project context](https://tru-images.b-cdn.net/guide-assets/local-421/build-a-website-hands-free-with-claude-voice/04-jared-portfolio-project-context.png) Then send the simplest build request in the guide: ```text Build the prototype. ``` Claude uses the interview, written plan, and uploaded assets to create the first website artifact automatically. You do not need to restate the entire brief in a second prompt. Open the result at full size and check the parts that matter most: - Does the opening section make the site's purpose clear? - Are the supplied images actually visible? - Does the page follow the agreed structure and visual direction? - Is the primary call to action easy to find? > Pro tip: If Claude skips your uploaded images, say, “Actually use the photos in this artifact.” That one correction fixed the recorded prototype without restarting the build. For any other visible miss, name one change and preserve the rest of the approved direction: ```text Keep the approved structure and copy. Make only this change: [ONE VISIBLE ISSUE]. ``` The finished artifact is a fast prototype, not a production website. Use it to validate the message, hierarchy, visual direction, and content before you invest in a full build. ### Guide: Build a Weekly Marketing Metrics Review That Writes Itself in Claude Cowork > In this guide, you will learn how to turn a messy weekly marketing review into a repeatable Claude Cowork workflow. - Canonical page: https://app.therundown.ai/guides/build-a-weekly-marketing-report-that-runs-itself-in-claude-cowork - Machine-readable page: https://app.therundown.ai/guides/build-a-weekly-marketing-report-that-runs-itself-in-claude-cowork.md - Difficulty: beginner - Featured tool or company: Claude Cowork - Categories: Data, Research & Analysis, Marketing & Growth, Automation & Agents - Instructors: Billy Howell - Published: 2026-05-26 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to turn a messy weekly marketing review into a repeatable Claude Cowork workflow. Claude will prep the data, verify numbers, draft, publish, and package everything. #### Who This Is Useful For - Marketing leads who need a cleaner weekly report without rebuilding the same doc every Monday. - Marketing ops teams who want Claude to pull context from Slack, Gmail, analytics, and meeting notes before drafting. - Founders or team leads who need both the detailed marketing review and the short leadership version from the same source-backed workflow. #### What You Will Build You will build a small reporting workspace that Claude can use every week. It will collect the raw inputs, draft the marketing review, create a leadership brief, write the Slack update, capture action items, and prepare polished files for sharing. ![Main Marketing Review Brief](https://tru-images.b-cdn.net/guide-assets/local-223/build-a-weekly-marketing-report-that-runs-itself-in-claude-cowork/01-main-marketing-review-brief.png) #### What You Need To Get Started - Claude Cowork - Some sort of marketing metrics like sales data, ad spend, etc. - Meeting notes for all recent marketing meetings #### Going Further Once this works once, make the workflow improve itself. After each marketing update, ask Claude: ```text What changed in this review, what did you get wrong, and what should become a durable instruction for next week? ``` Then approve a small update to the reporting skill. That is how the workflow gets better over time without turning into a messy one-off chat. You can also add a weekly prep automation. Have Claude run the input skill before your review meeting, then leave the final writing and publishing steps gated behind your approval. That is the bigger idea here: do not just ask Claude to write a report. Give it a place to work, a repeatable process, and a review loop that makes the next report easier. ### Step 1: Create Your Reporting Workspace Start by creating a folder for the weekly review. Keep the structure simple: marketing-metrics-review/, inputs/, working/, outputs/, and skills/. inputs/ is where Claude will collect the raw material: sales data, revenue numbers, Slack summaries, and meeting note summaries. working/ is where Claude can create drafts, source checks, and intermediate notes. outputs/ is where the final report, leadership brief, Slack update, action items, PDFs, and slides should land. ![Review Folder Structure](https://tru-images.b-cdn.net/guide-assets/local-223/build-a-weekly-marketing-report-that-runs-itself-in-claude-cowork/02-review-folder-structure.png) The point is not to create a fancy file system. The point is to give Claude a predictable place to look every time the workflow runs. Pro tip: keep this in a repo if you can. It makes it much easier to see what changed week to week, recover older reports, and avoid overwriting something important. ### Step 2: Create a Marketing Data Input Skill The first skill should prepare the inputs. Do not make it write the report yet. Its job is to gather and summarize the raw material. Use a prompt like this: ```text Create a skill that prepares our weekly marketing review inputs. It should summarize last week's Slack messages, look at my Gmail for sales updates, check our revenue numbers in Google Analytics, pull in meeting notes from Google Drive, and save the summaries in inputs/. ``` Your inputs will probably look different. You might use HubSpot instead of Gmail, a spreadsheet instead of Google Analytics, or a Notion doc instead of Google Drive. Before you finalize the skill, ask Claude: ```text Look at the connectors I have available and suggest the best weekly marketing review inputs you can gather automatically. ``` That one extra question is worth doing. Claude may have access to useful context you forgot about, and it can help you avoid building the workflow around the wrong sources. ![Review Inputs And Outputs](https://tru-images.b-cdn.net/guide-assets/local-223/build-a-weekly-marketing-report-that-runs-itself-in-claude-cowork/03-review-inputs-and-outputs.png) ### Step 3: Create the Reporting Skill The second skill should turn the inputs into a first draft. Keep this prompt simple: ```text Create a skill that looks at the inputs and creates a first draft of the main marketing report, the leadership brief, the Slack update, and the action items. ``` Once Claude writes the skill, add one important rule: it should separate facts from interpretation. If a number comes from Google Analytics, Slack, Gmail, or a meeting note, Claude should say where it came from. If two sources do not line up, it should flag the mismatch instead of smoothing it over. For example, your sales team might want the report organized by customer segment, while your export still uses old campaign groups. That is exactly the kind of mismatch you want Claude to catch before it writes a polished report. ### Step 4: Run the Workflow Run the input skill first. That should create or update the source summaries in inputs/. Then run the reporting skill. Ask Claude to create the first draft, but do not let it treat the first draft as final. Review the story before it expands the report. Use a direction like this: ```text Use the inputs to draft the weekly marketing review. Focus the story on pipeline quality and Q2 readiness. Keep source-backed facts separate from interpretation, and flag anything that needs human approval before the final report. ``` After you approve the story, Claude can create the final report, leadership brief, Slack update, and action items. ![Leadership Slide Brief](https://tru-images.b-cdn.net/guide-assets/local-223/build-a-weekly-marketing-report-that-runs-itself-in-claude-cowork/04-leadership-slide-brief.png) ![Slack Update](https://tru-images.b-cdn.net/guide-assets/local-223/build-a-weekly-marketing-report-that-runs-itself-in-claude-cowork/05-slack-update.png) ![Action Items](https://tru-images.b-cdn.net/guide-assets/local-223/build-a-weekly-marketing-report-that-runs-itself-in-claude-cowork/06-action-items.png) Pro tip: do not skip the approval step. Claude can find patterns, but you still own the management judgment. The best workflow is not fully automatic writing. It is automatic prep, human-approved story, then automatic packaging. ### Step 5: Add a Publishing Skill The Markdown files are useful, but they are usually not the final format people want. Create one more skill for publishing: ```text Create a publishing skill that turns final Markdown reports into PDFs and the leadership brief into a Google Slide. Save files in outputs/final/. Archive last week's outputs. ``` This keeps the writing workflow separate from the delivery workflow. The reporting skill creates the substance. The publishing skill turns it into the format people will actually read. You can also automate the publishing step after the report is approved. For example, Claude can generate the PDFs, create the slide, archive last week's files, and prepare a Slack-ready summary. ### Guide: Build a Work “Second Brain” That Updates Itself (Town Tutorial) > In this guide, you will learn how to set up Town to handle recurring work from Slack. - Canonical page: https://app.therundown.ai/guides/build-a-work-second-brain-that-updates-itself-town-tutorial - Machine-readable page: https://app.therundown.ai/guides/build-a-work-second-brain-that-updates-itself-town-tutorial.md - Difficulty: beginner - Featured tool or company: Town - Categories: Automation & Agents, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-08-13 - Updated: 2026-08-12 ### Overview #### The Rundown In this guide, you will learn how to set up Town to handle recurring work from Slack. You will connect your apps, see what Town learns about your work, and choose routines that can save you time. #### Who This Is Useful For - People using several AI tools at once who need one place to find and review the reports those tools produce. - Slack-based teams that want an assistant they can direct from a familiar channel instead of managing another complicated workspace. - Busy operators and creators who want a small set of recurring workflows without building or hosting an AI agent themselves. #### What You Will Build ![Town workspace routines grid](https://tru-images.b-cdn.net/guide-assets/local-422/build-a-work-second-brain-that-updates-itself-town-tutorial/01-town-routines-grid.png) This is the finished Town workspace you are working toward: a visible Wiki for reviewing what the assistant knows, a focused set of routines, and a daily review block for checking the work it produces. The system stays useful because you review its context and output instead of turning on every available automation and assuming it will work perfectly. #### What You Need To Get Started - A Slack Workspace. - Optional: Notion, Google, Granola, or Calendly to connect to Town. #### Going Further Repeat the same review loop over the next week: check the Wiki context, inspect one routine's output, flag what is wrong, and keep only what saves time. Once the core system is reliable, you can explore routines for meeting preparation, contact research, content calendars, or inbox cleanup. Keep external actions behind an approval step until you understand exactly what Town reads, changes, or sends. ![Town content routines for newsletter operations](https://tru-images.b-cdn.net/guide-assets/local-422/build-a-work-second-brain-that-updates-itself-town-tutorial/05-town-content-routines-newsletter-ops.png) The durable workflow is simple: make the context visible, choose one routine, review its output, and improve it before adding more. ### Step 1 Create Your Town Account and Choose Starter Routines Go to town.com (https://www.town.com/), click Sign Up, and connect the Slack workspace where you want to use Town. Town will walk you through a set of starter routines. Begin with only the ones tied to work you already do. The recording starts with Auto-inbox, Morning Briefing, and Meeting Briefing. ![Town starter routines during onboarding](https://tru-images.b-cdn.net/guide-assets/local-422/build-a-work-second-brain-that-updates-itself-town-tutorial/02-town-starter-routines-onboarding.png) Review the options inside each routine before moving on. For example, Auto-inbox can include separate actions for labeling mail, assisting with scheduling, or drafting replies. You decide which options are enabled; the routine then produces the recurring output attached to those choices. You can change these settings later, so resist the urge to enable everything during onboarding. ![Town routine library categories](https://tru-images.b-cdn.net/guide-assets/local-422/build-a-work-second-brain-that-updates-itself-town-tutorial/03-town-routine-library-categories.png) ### Step 2 Connect Only the Sources Town Should Use Open Town's connection setup and choose the sources that contain useful work context. That might include Notion pages, a Google account, Granola meeting notes, or Calendly bookings. Treat each connection as a permission decision. Review what Town can access, add only the pages or accounts it needs, and leave private messages or unrelated sources disconnected. If you do not want Town looking at Slack DMs, do not grant that access. Connecting a source gives Town permission to use that information. It does not make every output correct, so you will review the Wiki before relying on any routine. ### Step 3 Review the Wiki and Profile Town Builds Open the browser dashboard and select Wiki. Review the sections Town has created for your work, such as your overview, goals, projects, profile, notes, and people. Start with the facts that are most likely to affect a routine: your current projects, responsibilities, communication preferences, and the people you work with. If anything is missing or inaccurate, treat the Wiki as unfinished context and do not use that section to guide important work. Do not expect one review to create perfect context. The goal is to make Town's understanding visible so you can catch problems before they appear in a briefing, report, or action. ### Step 4 Use Town in Slack and Create a Daily Review Block Open the town-square channel in Slack and tag @Town when you send a request. Start with a low-risk status check: ```text What tasks are you working on? Any blockers? ``` Review the response before asking Town to take an action. A status report is an output you can inspect; creating a calendar event or changing another tool is an action that may require your approval. Next, ask Town to reserve a daily review window and gather the reports you need to check: ```text Find me a recurring time every day and block out 45 minutes for me to read reports generated by AI agents on all research tasks and operational tasks. Set up an automation so that the reports are gathered and linked in that calendar event description every day. ``` Approve only the calendar action you requested, then verify the recurring block, its timing, and the links in the event description. This creates a place to review the work instead of letting reports disappear across different apps and threads. > Pro tip: Town may need your approval the first time it takes an action in Slack, so approve the request before expecting the task to run. ### Step 5 Improve One Routine at a Time Open Routines to see what is already running before you add anything new. ![Town routine library](https://tru-images.b-cdn.net/guide-assets/local-422/build-a-work-second-brain-that-updates-itself-town-tutorial/04-town-routine-library.png) Ask Town to recommend a short list based on the context you just reviewed: ```text Based on what you know about me, recommend the top three automations we can set up to make my job easier. ``` Choose one suggestion, inspect what it will read and what it will do, then test it with a low-risk task. Review the result during your daily 45-minute block and give feedback on whether it was useful. Add the next routine only after the first one is producing something worth keeping. This is the difference between collecting automations and building a dependable system: each routine earns its place through a result you have checked. ### Guide: Build a YouTube Research Bot in 15 Minutes > In this guide, you will learn how to build a Gumloop agent that tracks YouTube channels or search topics, reads transcripts, and turns the useful videos into a ranked research brief. - Canonical page: https://app.therundown.ai/guides/build-a-youtube-research-bot-in-15-minutes - Machine-readable page: https://app.therundown.ai/guides/build-a-youtube-research-bot-in-15-minutes.md - Difficulty: beginner - Featured tool or company: Gumloop - Categories: Data, Research & Analysis, Automation & Agents - Instructors: Billy Howell - Published: 2026-05-12 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to build a Gumloop agent that tracks YouTube channels or search topics, reads transcripts, and turns the useful videos into a ranked research brief. Instead of watching every new video in your niche, you can wake up to a short brief with source links, key takeaways, follow-up ideas, and a usefulness score. #### Who This Is Useful For - Content teams tracking what is changing in a niche - Founders and operators who want a lightweight research scout without building a local agent - Anyone who uses YouTube for research but does not have time to watch every long video manually This works best when you give the agent a tight beat. Channels are best when you already trust the source. Search queries are better when you want broader discovery. #### What You Will Build ![What you will build](https://tru-images.b-cdn.net/guide-assets/local-216/build-a-youtube-research-bot-in-15-minutes/01-what-you-will-build.png) You will build a Gumloop agent called YouTube Scout. The agent will: - keep a list of YouTube channels or search topics - look for recent videos - read the transcript or video content - write a concise research brief - log the output in Google Sheets - score each video by usefulness The finished output is not just a pile of summaries. The useful version tells you what changed, which videos matter, why they matter, and what to do next. #### What You Need - A Gumloop account (Free account works) - A niche, channel list, or search query to test In the video guide, we use Gumloop's agent builder instead of a traditional workflow builder. That made the setup more flexible, and it avoided maintaining a custom workflow. #### Going Further Once the YouTube scout is working, reuse the same pattern on other noisy feeds. The pattern is simple: - Pick an information stream. - Give the agent a tight beat. - Make it collect source links. - Extract only the useful signal. - Rank the findings. - Send the result somewhere you will actually read it. You can use the same structure for podcasts, newsletters, product updates, Reddit threads, competitor research, or internal knowledge feeds. The goal is not to summarize everything. The goal is to stop doing homework and get a short list of what actually matters. ### Step 1 Create the Agent Go to the Gumloop agent builder (https://www.gumloop.com/personal/agents), create a new agent, and name it YouTube Scout. Open the Apps section on the right side and make sure YouTube and Google Sheets are enabled. ![YouTube and Google Sheets tools enabled in Gumloop](https://tru-images.b-cdn.net/guide-assets/local-216/build-a-youtube-research-bot-in-15-minutes/02-youtube-tools-enabled.png) The YouTube tool is the important one. It can search videos, get video details, pull channel videos, get playlist videos, get channel details, and pull comments. For this guide, the core jobs are finding videos and reading enough source material to summarize them. Google Sheets gives the scout a clean place to store the watch list and the final briefs. > Pro tip: If you have not connected Google Sheets before, Gumloop will ask you to approve access through Google. Do that before you build the agent so the first run does not get blocked by a missing connection. ### Step 2 Give the Scout a Tight Job Paste a setup prompt that tells the agent exactly what it owns. ![YouTube Scout setup prompt](https://tru-images.b-cdn.net/guide-assets/local-216/build-a-youtube-research-bot-in-15-minutes/03-youtube-scout-prompt-filled-in.png) Use this prompt as your starting point: ```text Build me a YouTube scout for [NICHE]. Check [CHANNELS_OR_QUERIES], find videos from the last [LOOKBACK_WINDOW], read the transcript or video content, and return a concise research brief. For each important video, include the title, channel, link, 3-5 key takeaways, why it matters, follow-up ideas, usefulness score, and a short "what changed" summary. Track topics and videos in a Google Sheet. Skip repetitive or low-signal videos. ``` For the first test, keep the scope small. Use one niche, a few trusted channels, one or two searches, and a 24-48 hour lookback window. The tighter the scout's beat, the better the brief. You are not asking it to understand all of YouTube. You are giving it one lane to patrol. ### Step 3 Let Gumloop Build the Workflow After you submit the prompt, Gumloop starts building the agent. In the demo, it set up the YouTube and Google Sheets pieces and began running the requested steps in the cloud. ![YouTube Scout agent running the workflow](https://tru-images.b-cdn.net/guide-assets/local-216/build-a-youtube-research-bot-in-15-minutes/04-agent-running-youtube-sheet-workflow.png) This is the part that makes Gumloop useful for lightweight agents. You are not wiring every step by hand in a workflow canvas. You are giving the agent a job, enabling the right tools, and letting it assemble the first working version. That said, do not treat the first build as final. Watch what the agent creates. Make sure it is saving the right source data, pulling recent videos, and putting the output somewhere you can inspect. ### Step 4 Check the Watch List Open the Google Sheet the agent creates. ![YouTube Scout watch list sheet](https://tru-images.b-cdn.net/guide-assets/local-216/build-a-youtube-research-bot-in-15-minutes/05-youtube-scout-watch-list-sheet.png) The Watch List tab should be easy to edit. A useful version includes fields like: - type - value - active status - notes - last scouted Use search when you want the scout to look across YouTube for a phrase like AI agents 2026. Use channel when you want it to track a specific channel URL. Keep an Active column so you can pause a source without deleting it. This matters once the list starts growing. You do not want to rebuild the scout every time one query gets noisy. ### Step 5 Review the Research Brief The Research Briefs tab is the payoff. ![YouTube Scout research briefs sheet](https://tru-images.b-cdn.net/guide-assets/local-216/build-a-youtube-research-bot-in-15-minutes/06-youtube-scout-research-briefs-sheet.png) In the demo output, the agent logged: - run date - rank - title - channel - source URL - key takeaways - why it matters - follow-up ideas - signal score - source query This is what turns the workflow from a generic video summarizer into a research tool. You can scan the ranked list, open the source links, and decide what is worth watching in full. The signal score is the key field to tune. If the agent calls a mediocre video an 8, tell it why that should be a 5. If a video is genuinely useful, explain what made it useful. > Pro tip: Add a User Signal Score column if you want the agent to learn from your review. Over time, that feedback can help it stop assigning high scores to videos that only look useful on the surface. ### Step 6 Make It Repeatable Once the scout works once, make it easy to run again. You can use the agent directly inside Gumloop, or you can add an external channel. ![Gumloop agent email inbox setup](https://tru-images.b-cdn.net/guide-assets/local-216/build-a-youtube-research-bot-in-15-minutes/07-gumloop-agent-email-inbox.png) Gumloop can give the agent its own email inbox. It can also connect agents to Slack. That means the scout can send the brief where you already check updates instead of forcing you to open Gumloop every time. Start weekly for broad research. Use daily only for fast-moving niches where you actually need that much signal. Before you trust it, do one editor pass: - Check that every finding has a source link. - Remove noisy searches or channels. - Confirm the brief says what changed, not just what was posted. - Tighten the prompt if the takeaways are too broad. - Re-run and compare the next brief. The first run is not the final product. Treat it like an editor: check the sources, tighten the beat, and make the brief sharper. ### Guide: Build an AI Case Study Generator > This guide teaches you how to use Claude to turn old project files into information-rich case studies you can use in proposals, on your website, or with prospects. - Canonical page: https://app.therundown.ai/guides/build-an-ai-case-study-generator - Machine-readable page: https://app.therundown.ai/guides/build-an-ai-case-study-generator.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Sales, Consulting & Customer Success - Instructors: Billy Howell - Published: 2026-03-08 - Updated: 2026-07-23 ### Overview #### The Rundown This guide teaches you how to use Claude to turn old project files into information-rich case studies you can use in proposals, on your website, or with prospects. You will set up a reusable Claude Project that stores your instructions, examples, and quality checklist, then generates a publication-ready draft from project documents in about 60 to 90 seconds when extended thinking is on. #### Who This Is Useful For - Agency owners and account managers who have wrap docs and metrics exports that are not being turned into case studies. - Consultants and freelancers who know they should have case studies but do not have time to write them from scratch. - Marketing and sales teams that need to produce client success stories at scale without routing every first draft through a copywriter. #### What You Will Build You will build a Claude Project called “Case Study Generator” that stays loaded with your style guide, example case studies, and structural instructions. Each time a project closes, you can start a new chat, add your project documents, run one prompt, and get a structured case study draft ready to publish or send to a prospect. ![image-2104](https://tru-images.b-cdn.net/guide-assets/69/69adcaf7652d326af04cdc14/image-2104.png) - A reusable Claude Project named “Case Study Generator.” - Project instructions that define the case study structure, tone rules, and missing-data handling. - Reference files that help Claude match your examples and quality standards. - A repeatable workflow for turning wrap memos, metrics, notes, and transcripts into case study drafts. #### What You Need - A Claude account. The free tier works; Pro is recommended for file uploads and longer context. - The template files attached to this guide: project instructions, a structure reference, and a Do’s and Don’ts checklist. - Project documents, such as a wrap memo, metrics export, client email, voice memo transcript, or call notes. - Optional but useful: a short project wrap memo with results and key KPIs. #### Going Further - Review each draft against the Do’s and Don’ts checklist before publishing. - Ask Claude to turn the finished case study into a slide deck version or social media carousel. - Save the final case study as a clearly labeled PDF in a shared folder. - Upload only edited, approved final PDFs back into the Claude Project so future drafts can improve from your best examples. ### Step 1 Start writing project wrap memos When a project ends, write a short wrap memo covering the results and key KPIs. It does not need to be polished; a few bullet points is enough. If you are pressed for time, record a quick voice memo instead. This memo is the raw material the system runs on. The better the wrap memo, the better the case study, but even a messy memo can produce something useful. Use the example wrap memo in the resources as a reference. ### Step 2 Set up your Claude Project Go to Claude, click Projects in the left sidebar, then choose New Project. Name it “Case Study Generator.” Open the project instructions field and paste in the full contents of the “Claude Project Instructions: Case Study Generator” file. This tells Claude how to structure each case study, including the seven-section framework, tone rules, what to do with missing data, and how to handle different types of raw input. Upload the “Case Study Examples and Structure Reference” and “Case Study Do’s and Don’ts” files as project files. These give Claude annotated examples to calibrate against and a quality control checklist to run on every output. You only do this setup once. Every case study you generate from here on should live in a new chat inside the same project. ![CleanShot 2026-03-08 at 15.41.48@2x](https://tru-images.b-cdn.net/guide-assets/69/69add106d7a9d8395f369b6d/cleanshot-2026-03-08-at-15-41-48-2x.png) > Pro tip: Name each chat after the client and project, such as “Vanta Health, Q4 Campaign.” This keeps your case study library organized and searchable inside the project. ### Step 3 Upload your project files and run the prompt Start a new chat inside the project. Upload your project documents directly into the chat. ```text Generate a publication-ready case study from these documents. Lead with the strongest result in the headline. Use the challenge, solution, results framework. If any section is missing data, flag it with [NEEDS INPUT] rather than inventing information. ``` Claude will read what you uploaded, cross-reference the project instructions and examples, and produce a full draft. With extended thinking on, this takes about 60 to 90 seconds. ![image-2105](https://tru-images.b-cdn.net/guide-assets/69/69add0f3d7a9d8395f369b1a/image-2105.png) > Pro tip: If the PDF formatting looks strange, don’t worry. It will look fine when you download it or export it to Google Docs. ### Step 4 Review and export Read through Claude’s output against the checklist in the Do’s and Don’ts file. Verify the main quality markers before you publish or share it. - Does the headline include a specific number or outcome? - Is the “before” state clearly established? - Are there at least three metrics in the results section? - Is there a client quote, or a [NEEDS INPUT] flag where one should go? If you do not have a client quote, ask Claude to draft two or three options based on the results. Send them to your client and ask if they can put their name on one. Once you are happy with the case study, Claude can also generate a slide deck version or a social media carousel from the same case study. ### Step 5 File it and feed it back Save the final case study as a PDF and drop it into a shared folder. Label it clearly with the client name and primary result. Upload the final PDF back into your Claude project as a context file. This is how the system gets smarter over time: each new finished case study becomes an example Claude can reference when generating the next one. Only upload cases you have already edited and are happy with. If you feed Claude unedited drafts, it will optimize for the patterns in those drafts, including the bad ones. ### Resources - [Claude Project Instructions: Case Study Generator](https://tru-images.b-cdn.net/guide-assets/69/69adcf2b146bf650b565a41f/claude-project-instructions-case-study-generator.md) - [Case Study Examples and Structure Reference](https://tru-images.b-cdn.net/guide-assets/69/69adcf35146bf650b565a443/case-study-examples-structure-reference.md) - [Case Study Do’s and Don’ts](https://tru-images.b-cdn.net/guide-assets/69/69adcf3f146bf650b565a4a2/case-study-do-s-and-don-ts.md) - [Example Project Wrap Summary](https://tru-images.b-cdn.net/guide-assets/69/69adcfc62102b3ebd47942f9/02-project-wrap-summary.pdf) ### Pro tip > Name each chat after the client and project, such as “Vanta Health, Q4 Campaign,” so your case study library stays organized and searchable inside Claude. ### Guide: Build an AI Secretary That Finds Open Action Items and Plans Your Day (Works With Slack + Gmail + Calendar) > Build a morning AI secretary that reviews Slack, Gmail, and Calendar, finds open action items, prioritizes your day, and improves from your feedback. - Canonical page: https://app.therundown.ai/guides/build-an-ai-secretary-that-finds-open-action-items-and-plans-your-day-works-with-slack-gmail-calendar - Machine-readable page: https://app.therundown.ai/guides/build-an-ai-secretary-that-finds-open-action-items-and-plans-your-day-works-with-slack-gmail-calendar.md - Difficulty: beginner - Categories: Automation & Agents, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-05-23 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Codex or Claude Code to build an AI taskmaster that checks Slack, Gmail, and your calendar every morning. It turns the mess into a prioritized to-do list that gets better every day you use it. #### Who This Is Useful For - Operators who get tasks from email, Slack, and meetings and need one place to see what actually matters. - Founders and team leads who want a morning list without maintaining another to-do app by hand. - AI power users who want their agent to learn from feedback instead of repeating the same bad suggestions every day. #### What You Will Build You will build a simple one-skill, one-automation task tracking system that works with any coding agent. It automatically updates your to-do list each day in a single Markdown file, then improves based on your feedback in a task-rules.md guidance file. ![Prioritized MonoNote task list](https://tru-images.b-cdn.net/guide-assets/221-for-jason/01-prioritized-mononote-task-list.png) #### What You Need - Codex or Claude Code. - Connected Gmail, Slack, and Calendar access. - A Markdown note app like Obsidian or Cursor. - Automation or scheduled task support. #### Going Further - Create a weekly audit skill that scans your task list. - Find repeated tasks that keep showing up. - Turn repeated tasks into candidates for future AI automations. - Use the same loop for other workflows: create a skill, run it once, automate it, then improve it with feedback. ### Step 1 Ask the Agent to Build the Skill Create a new folder for this workflow, then open Codex or Claude Code inside that folder. You are not manually building the Mono note first. You are asking the agent to create the workflow, the files, and the reusable skill. The starting prompt is simple: ```text Make me a skill that looks at my Slack, Gmail, and calendar every morning, prioritizes tasks high to low, and puts them at the top of MonoNote.md under today's date. Every task should include feedback checkboxes for priority and relevance, plus status checkboxes for done, bump, and cancelled. ``` ![Morning todo skill prompt](https://tru-images.b-cdn.net/guide-assets/221-for-jason/02-morning-todo-skill-prompt.png) The prompt should be copy-pasteable. It needs to include the task source, the destination file, and the feedback checkboxes so the agent builds the right format from the start. ### Step 2 Create the Files and Task Rules Next, tell the agent to create the two Markdown files this system needs: - MonoNote.md for the running task list - task-rules.md for priority and ignore rules MonoNote.md can start blank. The agent will put the first real dated section at the top when you run the skill. Then have it add any priority rules you have per task into task-rules.md. You can dictate this in plain English. Start with priority rules like: ```text High priority means direct requests, same-day deadlines, meetings that need prep, blocked teammates, or anything where silence creates risk. Medium priority means useful follow-ups, active project work, or replies that matter but do not block today. Low priority means optional admin, loose research, maintenance, or things I only need to review when there is time. Ignore newsletters, receipts, automated notifications, FYI-only threads, vendor outreach without a clear ask, and anything already resolved. ``` Keep this file short. It should help the skill make better calls, not become a giant personal operating manual. > Be stricter than you think at first. A short list you trust is better than a giant AI-generated backlog you have to clean up every morning. ### Step 3 Run the Skill Once Run the skill manually before you schedule anything. The first run should create today's dated section at the top of the Mono note and group tasks by priority. Every task should include: - the task itself - a source label or link, such as Gmail, Slack, or Calendar - status checkboxes for done, bump, and cancelled - feedback checkboxes for priority and relevance The source link is the trust check. If the agent says you need to reply to an email or prep for a meeting, you should be able to click back to the message, thread, note, or event that created the task. After the list appears, mark a few items. Use done for finished tasks, cancelled for tasks that should not come back, and bump for anything that should roll into the next business day. For training, mark items that were too high, too low, or not relevant. This is the part that makes tomorrow's list better. If your agent asks before installing the skill globally, approve it only after you have reviewed the files it created. ![Install skill confirmation](https://tru-images.b-cdn.net/guide-assets/221-for-jason/03-install-skill-confirmation.png) ### Step 4 Create the Automation Once the manual run works, ask Codex or Claude Code to create the weekday automation. The automation should not just run the list generator from scratch. It should look at the training you gave yesterday, update the task rules guidance file, then run the full morning loop: - Review yesterday's task section. - Read the prioritization feedback. - Check which tasks were marked cancelled. - Check which tasks were marked Not relevant. - Update task-rules.md from those signals. - Find anything marked bump and move it into today. - Scan Slack, Gmail, and Calendar for new tasks. - Generate today's prioritized list at the top of the Mono note. Use a prompt like: ```text Create a weekday morning automation for this skill. Before generating today's list, review yesterday's Mono note section. Learn from: - prioritization feedback - tasks marked cancelled - tasks marked Not relevant Use that information to update task-rules.md. Then carry forward any task marked bump. Then run the to-do list generation task for today using Slack, Gmail, and Calendar. Put the new list at the top of the Mono note under today's date. Do not send messages, create calendar events, archive email, delete anything, or modify outside systems. ``` In our setup, the agent created the note, the rules file, the skill source, and an active weekday automation. ![Created files and weekday automation](https://tru-images.b-cdn.net/guide-assets/221-for-jason/04-created-files-and-weekday-automation.png) Keep the first version read-only outside of your own Markdown files. The agent can recommend replies, meeting prep, and follow-ups. It should not send or schedule anything until the list is consistently useful. ### Step 5 Use the Feedback Loop The system only gets better if you use the checkboxes. At the end of the day, every task should be in one of three states: - done - bump - cancelled Then use the training checkboxes when the agent guessed wrong. If a task was noise, mark it not relevant. If something was ranked too low, mark the right priority. If you leave the training row blank, the automation can assume the task was already in the correct priority section. That last rule saves time. You should not have to mark the same obvious high-priority task as high priority every day. This fixes the biggest problem with AI task lists. Without feedback, the agent keeps rediscovering the same noisy emails and vague threads. With feedback, it learns what to ignore, what to carry forward, and what actually matters in the morning. ### Pro tip > Be stricter than you think at first. A short list you trust is better than a giant AI-generated backlog you have to clean up every morning. ### Guide: Build an Automated Ad Generator With This New Tool (Eleven Labs Flows) > Learn how to turn a product photo into a finished video ad using ElevenLabs Flows, a node-based canvas that combines image, video, voice, and music models in one place. - Canonical page: https://app.therundown.ai/guides/build-an-automated-ad-generator-with-this-new-tool-eleven-labs-flows - Machine-readable page: https://app.therundown.ai/guides/build-an-automated-ad-generator-with-this-new-tool-eleven-labs-flows.md - Difficulty: beginner - Featured tool or company: ElevenLabs - Categories: Marketing & Growth, Content & Creative, Automation & Agents - Instructors: Billy Howell - Published: 2026-04-08 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to turn a product photo into a finished video ad using ElevenLabs Flows, a node-based canvas that combines image, video, voice, and music models in one place. You will produce a reusable ad-generation Flow that can be duplicated for future products, so each new ad becomes a reference and prompt swap instead of a full rebuild. After the first Flow works, the next ad can be done in minutes. #### Who This Is Useful For - Ecommerce founders who want ads that show their actual product, not a generic AI stand-in - In-house marketers shipping paid social creative who are tired of bouncing between Midjourney, Runway, and Suno in separate tabs - Agency owners and content creators who want one repeatable canvas they can clone across clients and product lines #### What You Will Build ![image-2168](https://tru-images.b-cdn.net/guide-assets/69/69d6e2fed57669b7a4e23ba9/image-2168.png) You will build a reusable Flow that takes 1–3 reference photos of your real product, places it into a generated scene, animates it, and exports a finished video ad with an audio jingle. - A reference-based image generation setup for your product - A connected video generation node that animates the generated product scene - An exportable MP4 video ad - A Flow you can duplicate and adapt for the next product #### What You Need - An ElevenLabs paid plan with Flows enabled - 1–3 clean photos of your real product from different angles, such as front, 3/4, or top-down - A short idea for the scene you want your product placed in #### Going Further - After exporting your MP4, duplicate the Flow and reuse it for the next product by swapping the reference photos and scene prompt. You can also add Text to Speech or Music nodes, connect them into a Mix Audio node, and build a library of reusable ad formats such as 15-second hero ads, 30-second explainers, or UGC talking head canvases. ### Step 1: Open Flows and name the canvas Log into ElevenLabs, click ElevenCreative in the sidebar, then open Flows. Click + New Flow and name it something reusable, such as “[your product line] ad template.” You are building a template, not a one-shot render, so the name matters. > Pro tip: This was tested on the $5 Starter plan and it works. Video generation with Flows is paid only, but you do not need Creator or Pro to run the image and video nodes in this guide. You can test image generation for free. ### Step 2: Add an image node with reference photos Click the plus on the canvas and add an Image Generation node. Click Reference Images and upload photos of your real product. Repeat with different reference photos to help the AI lock in on the product’s design. ![image-2169](https://tru-images.b-cdn.net/guide-assets/69/69d6e30bd57669b7a4e24461/image-2169.png) Now write a prompt that describes the scene, not the product. The references handle the product for you. ```text The product (see references) on a clean seamless studio backdrop, soft diffused lighting from above, subtle shadow underneath, shallow depth of field, photorealistic product shot. ``` Run the node. If the scene is off, tweak the prompt and rerun. Only this node updates, so you are not paying for the rest of the pipeline while you iterate on the look. > Pro tip: Most people try to describe the product in the prompt. Don’t. Let the references do it. That separation is what keeps your real product consistent across every variation. ### Step 3: Connect a video node to the image Add a Video Generation node to the canvas. Drag a line from the image node’s output into the video node’s start frame input. Drop in a short motion prompt and run it. ```text Slow cinematic push-in on the product, subtle camera drift, soft light shifting across the surface, shallow depth of field. ``` ![image-2170](https://tru-images.b-cdn.net/guide-assets/69/69d6e31d63314e8257ab9121/image-2170.png) Because the start frame came from your reference-based image, your real product carries through into the clip without ever being described in the motion prompt. > Pro tip: The most fun part of this workflow is generating different prompts each run. Click the text button on any node to have AI write a different prompt each time. ### Step 4: Swap models, export, and clone Every node in Flows has a model picker in its settings. On the image node, switch between the image models bundled into Flows to see which one handles your product references best. Do the same on the video node for motion, and on the voice node when you add audio later. This is the hidden value of Flows: you can test one model against another without rebuilding the canvas or leaving the app. Remember that clicking Run fires that single node. If you click the Run dropdown, you can click Run till here to re-fire all previous nodes too. ![image-2171](https://tru-images.b-cdn.net/guide-assets/69/69d6e33063314e8257ab99e5/image-2171.png) When the output looks right, click Export on the video node to save the MP4. Then use the canvas menu to Duplicate Flow. Swap the reference photos and the scene prompt for your next product, and the second ad is done in minutes. ### Step 5 Going further: Add audio and build reusable ad formats If you want audio on the clip, add a Text to Speech or Music node and connect it into a Mix Audio node alongside the video. This is how Flows layers voice or a soundtrack onto the clip without leaving the canvas. Once you have one template working, build a small library of Flows for the formats you use most: a 15-second hero ad, a 30-second explainer, or a UGC talking head. Each one is a canvas you only build once, then clone forever. ### Pro tip > Let the reference photos define the product, and use prompts to describe the scene or motion. That separation helps keep the real product consistent across variations. ### Guide: Build and Host a Custom CRM with Google Antigravity > In this guide, you will learn how to use Google Antigravity to turn a plain-English app spec into a hosted full-stack CRM. - Canonical page: https://app.therundown.ai/guides/build-and-host-a-custom-crm-with-google-antigravity - Machine-readable page: https://app.therundown.ai/guides/build-and-host-a-custom-crm-with-google-antigravity.md - Difficulty: beginner - Featured tool or company: Google Antigravity - Categories: Coding & App Development, Sales, Consulting & Customer Success - Instructors: Billy Howell - Published: 2026-06-17 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Google Antigravity to turn a plain-English app spec into a hosted full-stack CRM. The key move is making the agent plan first, then using Firebase Auth, Firestore, and Firebase Hosting so the app has login, saved data, search, and a real URL. #### Who This Is Useful For - Founders and operators who want a quick internal tool without spending the whole day setting up a backend. - Non-engineers who are comfortable prompting AI but get stuck when the app needs auth, a database, and deployment. - Developers who want to test Google's agent-first coding workflow against tools like Cursor, Claude Code, or Codex. #### What You Will Build You will build Northstar CRM, a small internal sales app with login, contacts, companies, deals, notes, search, dashboard cards, and a Firebase Hosting URL. ![Hosted Northstar CRM dashboard with Firebase URL](https://tru-images.b-cdn.net/guide-assets/local-242/build-and-host-a-custom-crm-with-google-antigravity/01-hosted-northstar-crm-dashboard.png) This is not meant to be a finished enterprise CRM. The win is getting a real full-stack app online fast enough that you can test the workflow, then decide what is worth polishing. #### What You Need To Get Started - A Google account. - Google Antigravity (https://antigravity.google/download). - A clean project folder. - Node.js and npm installed. - A Firebase project, or permission to let Antigravity help create one. - A simple app idea. This guide uses a CRM because it needs auth, records, search, and deployment. You can start with a free Google account, but Antigravity usage may hit limits depending on your account and model. In the recording, the app still got built and deployed, but the workflow did hit a usage-limit moment during setup. #### Going Further After the Firebase deploy works, ask Antigravity to add one upgrade at a time. Good next prompts: ```text Add Google sign-in to this Firebase Auth flow. ``` ```text Set up Google Analytics for this Firebase web app and show me what I need to confirm in the Firebase Console. ``` ```text Walk me through connecting a custom domain to this Firebase Hosting app. ``` The bigger lesson is that Antigravity is especially useful when the app touches Google's stack. If you get confused by Firebase, Google Cloud, Analytics, Auth, or Hosting, ask Antigravity for the next exact step and make it explain what to click before you change anything important. ### Step 1 Download Antigravity And Open A Clean Folder Go to antigravity.google/download (https://antigravity.google/download), install the desktop app, and sign in with Google. Then create a clean folder for the project and open it in Antigravity. In the recording, the project folder was called 242-northstar-crm, but any clean folder name works. ![Antigravity create project screen with Northstar CRM folder selected](https://tru-images.b-cdn.net/guide-assets/local-242/build-and-host-a-custom-crm-with-google-antigravity/02-open-project-folder.png) Start clean because the agent will inspect the folder and use whatever it finds as context. If you point it at a messy old project, you are making the first build harder than it needs to be. > Pro tip: Before you start, turn on the Firebase and Google SDK skills in Antigravity. This is what lets the agent help with the Google stack instead of only writing local code. To find them, open Antigravity settings and go to Customizations. Turn on the Google/Firebase skills you plan to use. The most important ones for this build are Firebase, Google SDK guidance, modern web guidance, and Chrome DevTools. ![Antigravity customizations screen with Firebase skill available](https://tru-images.b-cdn.net/guide-assets/local-242/build-and-host-a-custom-crm-with-google-antigravity/03-enable-firebase-skill.png) For the model, use Gemini 3.5 with Medium Thinking. That was a good balance for this build: smart enough to plan and wire the stack, but not so slow that it overthinks every small UI choice. ### Step 2 Give Antigravity A Tight App Prompt Do not start by saying, "Build me a CRM." Make the job concrete and tell Antigravity to plan before it writes code. Use this prompt: ```text Build Northstar CRM with login, contacts, companies, deals, notes, search, and dashboard cards. Use React, Vite, TypeScript, Firebase Auth, Firestore, and Firebase Hosting; plan first. ``` That two-line version is enough for a fast build. If you want more control, add a short out-of-scope list: ```text Keep this as a version-one internal tool. Do not add billing, teams, roles, CSV import, email sending, automations, or SaaS admin features. ``` When Antigravity returns a plan, review it before approving the build. This is where you catch scope creep. If the agent starts adding a multi-tenant SaaS platform, push it back to the smallest working internal app. In the recording, Antigravity asked a few clarifying questions and started a task list before coding. That is what you want. The plan is the control layer. ### Step 3 Let Antigravity Build, Then Test The Local App After you approve the plan, let Antigravity build the first version. It should create the React/Vite app, build the CRM screens, add local or seed data, and run the app locally. If Antigravity does not start the dev server on its own, open the project terminal and run: ```text npm run dev ``` Then test the app like a user, not like a code reviewer. Open the local URL and click through: - Dashboard - Contacts - Companies - Deals - Notes - Create or add record flows - Search and filtering In the recording, the first version already had a functional CRM shape: contacts, companies, deals, notes, dashboard cards, and a dark internal-tool UI. It was not perfect, but it was usable enough to move to backend setup. The important test is not whether the UI looks polished. The important test is whether the app has the right shape before you connect real services. ### Step 4 Connect Firebase Once the app shell works, ask Antigravity how to install and connect Firebase. Use this prompt: ```text How do I install and connect Firebase for this app? Walk me through Firebase Auth, Firestore, and Firebase Hosting for this project. ``` In the recording, the key terminal command was: ```text npx firebase-tools login ``` That command signs you into Firebase from the terminal. You do not need to be a Firebase expert, but you do need to authenticate the CLI so Antigravity can guide the setup. Antigravity may also ask whether it should create a Firebase project for you or whether you want to create one yourself in the Firebase Console. ![Antigravity asking whether to create a new Firebase project](https://tru-images.b-cdn.net/guide-assets/local-242/build-and-host-a-custom-crm-with-google-antigravity/04-firebase-project-choice.png) For a demo, either path is fine. The cleanest version is: - Create or select a Firebase project for this app. - Add Firebase to the project. - Enable Firestore. - Enable Firebase Hosting. - Enable Authentication. - Turn on email/password sign-in. In the recording, Antigravity still needed a manual Firebase Console step for email/password authentication. That is normal. If the agent gets stuck, ask it for the exact click path instead of trying to guess inside the Firebase dashboard. Use this follow-up: ```text I am authenticated in terminal. Tell me the exact Firebase Console steps I need to finish before you deploy this app. ``` For Firestore rules, keep the demo simple. Authenticated users can read and write the demo CRM records. Unauthenticated users should not be able to read or write anything. > Pro tip: That rule is fine for a demo internal tool. It is not a final permission model for a multi-tenant SaaS app. ### Step 5 Deploy To Firebase Hosting Once Firebase Auth, Firestore, and Hosting are connected, ask Antigravity to prepare the deploy. Do not skip the pre-deploy check. Use this prompt: ```text Prepare this app for Firebase Hosting. Verify the production build, Firebase Hosting config, required environment variables, Firestore rules, and exact deploy command before running it. ``` Have Antigravity show you what it plans to run. Then approve the deploy. After deployment, open the live URL and test the app again. In the recording, the app deployed to a web.app URL and the final test was simple: - Sign up or log in. - Create a company. - Add a contact connected to that company. - Add a deal. - Add a note. - Confirm the dashboard updates. That final smoke test matters. A local preview does not prove the app is done. The guide is finished when the hosted app has login, saved data, and a workflow you can repeat from the live URL. ### Guide: Build Any Mobile App Idea In 15 Minutes With Replit > In this guide, you will learn how to turn an app idea into a Replit prototype you can test on your phone. - Canonical page: https://app.therundown.ai/guides/build-any-mobile-app-idea-in-15-minutes-with-replit - Machine-readable page: https://app.therundown.ai/guides/build-any-mobile-app-idea-in-15-minutes-with-replit.md - Difficulty: beginner - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-07-06 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to turn an app idea into a Replit prototype you can test on your phone. The workflow is simple: define one useful user flow, give Replit Agent a focused prompt, preview with Expo Go, then improve the weakest part before adding features. #### Who This Is Useful For - Founders with app ideas sitting in a notes app who want a real prototype before investing in a full build. - Operators and creators who want to test lightweight internal tools, productivity apps, or client-facing mobile workflows. - Anyone who wants to understand the Replit mobile app flow without getting buried in native iOS tooling first. #### What You Will Build You will build a working mobile prototype called Play Day, a playlist-themed Pomodoro app. This is the example from the video guide, but the same structure works for almost any simple app idea: define the user, the job, the core screens, and the first action you want someone to test. In the demo, Play Day takes a rough task dump like research stocks and write a blog post, turns it into Pomodoro blocks, and lets you move through the work like a playlist. ![Play Day running in Expo Go on an iPhone](https://tru-images.b-cdn.net/guide-assets/local-322/build-any-mobile-app-idea-in-15-minutes-with-replit/01-playday-phone-preview.png) The first version is a prototype, not a production App Store release. The point is to get from idea to something testable fast enough that you can decide what to improve. #### What You Need To Get Started - A Replit (https://replit.com/) account. - The Replit desktop app from replit.com/desktop (https://replit.com/desktop), or the Replit web app on desktop. - Expo Go (https://expo.dev/go) installed on your phone. - ChatGPT, Claude, or another chatbot for planning the PRD. - Optional: GitHub connected to Replit so your code is backed up outside the project. #### Going Further Use this workflow for any mobile idea where the first loop matters more than the full business plan. Start with one user, one job, and one believable action path. Then let Replit build the first version and test the result on your phone before adding accounts, payments, onboarding, or admin panels. You can also turn the PRD step into a reusable prompt: ```text Help me write a PRD for a mobile app idea. App idea: [APP_IDEA] Target user: [TARGET_USER] The first useful loop should be: [FIRST_USER_ACTION] Include: - the core user flow - the smallest useful set of screens - design requirements - features to avoid in version one - five simple name ideas Do not choose the tech stack. Focus on product requirements, functionality, and design direction. ``` The bigger pattern is the part worth keeping: clarify the idea outside Replit, build the smallest useful mobile version, preview it on a real phone, then use focused Replit tasks to fix what the phone immediately exposes. ### Step 1: Set Up Replit And Expo Go Start by signing into Replit. You can begin on a free account, but mobile app builds and AI-heavy projects can run into plan limits quickly, so use the account you actually want to build from. Then install the Replit desktop app from replit.com/desktop (https://replit.com/desktop). In the recording, the desktop app made the workflow cleaner because Replit was not fighting for space with a pile of browser tabs. The same core flow also works from the desktop browser. Install Expo Go on your phone before you start building. That is what lets you scan the QR code from Replit and open the generated app on a real device. > Pro tip: Treat Expo Go as the preview path, not the whole publishing process. Replit's mobile app build starts from desktop web or the desktop app, Expo Go lets you test on your phone, and App Store submission comes later through the normal Apple Developer process. ### Step 2: Turn The Idea Into A PRD Before Opening Replit Start in ChatGPT before you open Replit. Talk through the idea until you can describe the smallest useful loop. In the video, the original idea was a broader AI scheduling app, but the stronger demo was simpler: a Pomodoro app where tasks behave like songs in a playlist. Here is the example prompt from the video. You can use it as a fill-in-the-blanks model for your own app: ```text Help me write a PRD (Project Requirements Document) for a mobile app that is a playlist-themed Pomodoro timer. The core concept is that I dictate my tasks, and the AI creates a daily playlist of everything I need to get done as Pomodoro-style sessions. If I finish a task early or get stuck, I can skip to the next item in the playlist. Over time, the app should track my habits and use that data to help the AI: 1. Classify my tasks more accurately 2. Chunk them up effectively 3. Provide a glance at the end of the day showing exactly how long each task took Since this is a native iOS app, we want native system support for: 1. Lock screen widgets and Live Activities 2. Push notifications 3. Dynamic Island support Finally, please give me five ideas for a good, simple, catchy name for this app. ``` ![ChatGPT turning the app idea into a PRD](https://tru-images.b-cdn.net/guide-assets/local-322/build-any-mobile-app-idea-in-15-minutes-with-replit/02-chatgpt-prd-prompt.png) Once ChatGPT gives you the first draft, pick the strongest name and ask it to write the final PRD. In the video, Play Day was the best name, and the visual direction became an old-school iPod Nano style interface. The exact app idea matters less than the shape of the prompt. Give the model a concrete concept, a simple interaction loop, design requirements, and a few naming options so you are not starting from a blank screen inside Replit. Before you copy anything into Replit, add this instruction: ```text Do not choose the tech stack. Focus on the product requirements, functionality, and design direction. Let Replit decide the implementation details. ``` That line matters. Let Replit decide how to build inside its own mobile app flow. Your PRD should explain what the app does, how it should feel, and which screens or behaviors matter most. ### Step 3: Start A Mobile App Build In Replit Open Replit on desktop and start a new build. Choose the Mobile option, then paste the final PRD into the agent chat. Use a short command at the top: ```text Build me this mobile app. ``` Then paste the PRD underneath it. This keeps Replit focused on building the prototype instead of asking it to invent the product from scratch. In the demo, Replit initially picked up the name as Playdate instead of Play Day. Fix small naming mistakes immediately before the build goes too far: ```text The app is called Play Day. ``` ![Replit mobile app build task](https://tru-images.b-cdn.net/guide-assets/local-322/build-any-mobile-app-idea-in-15-minutes-with-replit/03-replit-mobile-build-task.png) While the agent works, you can use the Replit interface to check the preview, database, auth, integrations, and task board. You do not need to click through every panel manually. If the app needs an integration, API key, database, or auth setup, you can usually ask the agent to wire it up. For a project with AI features, expect to connect an AI provider at some point. In the recording, Replit handled the AI side without extra setup, but do not assume every account or project will behave the same way. > Pro tip: Connect GitHub from Replit's integrations if you plan to keep working on the app. It gives you a normal code backup and makes it easier to bring in other agents or review changes later. ### Step 4: Preview The App On Your Phone When Replit finishes the first build, look for the phone preview and QR code. Open Expo Go on your phone, scan the QR code, and give the app a minute to build the JavaScript bundle the first time. This is the point where the workflow becomes useful. A mobile app can look fine in a simulator and still feel wrong in your hand. Test the main loop with real input. In the demo, the test input was: ```text research stocks and write a blog post ``` Play Day turned that into multiple Pomodoro tasks: - Pick stocks to research - Deep dive stock analysis - Outline and draft blog post - Revise and publish blog post That is the proof you are looking for. The app does not need to be finished yet. It needs to show that the core loop works well enough to keep iterating. ### Step 5: Fix The First Mobile Problems With Replit Tasks Once the app is on your phone, do not ask Replit to rebuild the whole thing. Create focused tasks for the problems you can actually see. In the recording, the first issues were practical mobile problems: - The top of the app needed to sit below the Dynamic Island. - The bottom menu covered some playlist controls. - The skip behavior made it too easy to treat an unfinished task as completed. Create separate Replit tasks for fixes like these: ```text Refactor and optimize the main screen to fit everything below the Dynamic Island and make sure it is mobile responsive. ``` ```text The menu at the bottom covers up some of the playlist control buttons. Make sure that does not happen and make the layout responsive. ``` ```text Make sure I can skip back when I am skipping through Pomodoro tasks that are not complete. Skip should mean move to the next task, not mark the task complete. Add a separate completion action if needed. ``` Replit can run those tasks like separate branches. Each task gets its own preview, and if one overlaps with another, Replit may queue it or turn it into a subtask. When a task turns green, review it before applying it to the main version. If the change looks right, click Apply changes to main version. If it is wrong, delete the task or ask Replit to revise it without touching the main app. > Pro tip: Replit's suggested next tasks can be useful, but do not accept them blindly. Open the suggestion, read the scope, and only keep the ones that move the prototype closer to the first real user test. ### Step 6: Publish A Web Build Before App Store Submission After the main prototype works, click Publish in Replit. The first publish step creates a web build on a Replit URL. You can make it public, password protected, or invite only, depending on what you are testing. ![Replit publish screen for the mobile app](https://tru-images.b-cdn.net/guide-assets/local-322/build-any-mobile-app-idea-in-15-minutes-with-replit/04-replit-publish-app.png) Replit may run checks during publishing, including a security scan and build steps for the Expo-based app. Let that finish before you judge the published version. If you want to submit the app to the App Store, continue from the App Store option after the web publish step. You will need an Expo sign-in and an Apple Developer account. That part is normal App Store submission work, not something to rush through in the first 15-minute prototype pass. The clean expectation is: - Replit desktop or web: start and manage the native mobile app build. - Expo Go: preview the app on your phone. - Replit publish: create a shareable web build. - Apple Developer flow: submit to the App Store when the app is actually ready. ### Guide: Build Automated Business Reports With AI (Works With Google Analytics, YouTube, Shopify & More) > Learn how to connect Codex or Claude Code to business tools through Composio, inspect the available data, generate a Markdown report, and turn the process into a reusable automation. - Canonical page: https://app.therundown.ai/guides/build-automated-business-reports-with-ai-works-with-google-analytics-youtube-shopify-and-more - Machine-readable page: https://app.therundown.ai/guides/build-automated-business-reports-with-ai-works-with-google-analytics-youtube-shopify-and-more.md - Difficulty: beginner - Featured tool or company: Composio - Categories: Data, Research & Analysis, Automation & Agents, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-05-19 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to connect Codex (or Claude Code) to Google Analytics with a free tool called Composio. You will use simple prompts to explore your data and build a useful Markdown report that you can schedule to run weekly. #### Who This Is Useful For - Marketing operators who want fast weekly reports from Google Analytics, YouTube, Shopify, or HubSpot without building custom dashboards from scratch. - Founders and small teams who need practical business reporting, but do not want to create Google Cloud projects, manage OAuth, or write API scripts. - AI power users and automation builders who want Codex or Claude Code to produce repeatable reports from real data sources instead of static exports. #### What You Will Build ![terminal google analytics report](https://tru-images.b-cdn.net/guide-assets/209-for-jason/01-terminal-google-analytics-report.png) You will build an agent-powered reporting workflow that connects to Google Analytics through Composio, explores the available account and property, creates a Markdown report, and prepares the process for automation. #### What You Need - A Composio (https://composio.dev/) account - Codex, Claude Code, or another coding agent that can run CLI tools - Access to the business app you want to report from - For this demo: access to the right Google Analytics property - A folder where the agent can save the outputs You can use Google Analytics as the first test, but the workflow works best when you think of it as a general reporting pattern. Start with one connected source, prove the report works, then expand. ### Step 1 Sign Up for Composio and Connect a Toolkit Open Composio and connect the data source you want to use. In this guide, we use the Google Analytics toolkit, but the same pattern works for YouTube, Shopify, HubSpot, Google Sheets, and other business apps. From the Composio dashboard, find the Google Analytics toolkit (https://dashboard.composio.dev/billy_workspace/~/connect/apps/google_analytics) and connect the Google account that has access to the property you want to report on. If you are using a different source, replace Google Analytics with the toolkit you want to report from. The important part is that Composio handles the app login and gives your agent a consistent way to access that data. Before you move on, confirm the toolkit shows as connected. ### Step 2 Install the Composio Script for Codex or Claude Code Next, connect Composio to your agent. ![composio codex cli install](https://tru-images.b-cdn.net/guide-assets/209-for-jason/02-composio-codex-cli-install.png) For Codex, Composio provides a setup prompt that installs the Composio CLI and logs in with your generated key. Copy that setup prompt into a new Codex session, or run the install and login commands yourself in Terminal. In the recording, the CLI path worked well because Codex and Claude Code are already comfortable with command-line tools. MCP can work too, but the CLI setup is often easier for agents to inspect, run, and debug. After installation, restart your Codex or Claude Code session and run a small test. For example: ```text Use the Composio CLI tool to list my three most recent Google Sheets files. ``` The exact test can change based on the connector. The goal is to prove the agent can call Composio before you ask it for a real report. > Pro tip: Composio's Enhanced Control settings are the data protection layer to check before you hand an agent a real account. They let you decide what an agent can do with a connected toolkit. For reporting workflows, start with read-only access when possible. Be careful with write and destructive tools on production accounts. ![enhanced control permissions](https://tru-images.b-cdn.net/guide-assets/209-for-jason/03-enhanced-control-permissions.png) ### Step 3 Ask the Agent to Explore Before It Reports Do not start by asking for the finished report. Start by asking the agent to inspect what it can see. This saves time because agents often need to discover the right account, property, channel, store, dimensions, metrics, and permissions before they can build a useful report. Use this generic prompt first: ```text Use Composio to explore the connected [TOOLKIT] toolkit. Do not create a report yet. Return: - which accounts, properties, channels, stores, workspaces, or data sources you can see - which source looks most relevant for [BUSINESS_OR_REPORTING_GOAL] - which dimensions, metrics, fields, or records would be useful for a [LOOKBACK_DAYS]-day report - any missing access, ambiguity, or risky permissions I should resolve before reporting ``` For Google Analytics, that becomes: ```text Use the Composio CLI tool to inspect our Google Analytics setup. Identify the available accounts and properties, find the property that tracks [SITE_OR_APP], and tell me which dimensions and metrics you would use for a useful traffic report. Do not create the report yet. ``` For a more detailed version, use: ```text Use Composio to explore the connected Google Analytics account. Find the Analytics property that appears to track [SITE_OR_APP]. Do not create a report yet. First return: - which accounts/properties you can see - which property looks correct - which dimensions and metrics you would use for a 30-day page/source report - any missing access or ambiguity I need to resolve ``` ![codex explore data source](https://tru-images.b-cdn.net/guide-assets/209-for-jason/04-codex-explore-data-source.png) The same exploration prompt works for other tools. For YouTube, ask the agent to identify the connected channel, the available channel or video metrics, and the best KPIs for a weekly content report. For Shopify, ask it to inspect products, orders, promotions, and sell-through metrics before it creates anything. ### Step 4 Analyze the Data and Create the Markdown Report Once the agent confirms the right property or data source, give it the actual reporting job. Use this generic report prompt: ```text Use Composio and the confirmed [TOOLKIT] data source to create a [LOOKBACK_DAYS]-day report. Report goal: [REPORT_GOAL] Analyze the data and pull the most important trends, changes, risks, and opportunities. Return the report in Markdown with: - executive summary - supporting tables - most important trends - recommended actions - source notes - access or data quality warnings ``` For Google Analytics, a simple version is: ```text Use Composio + GA to identify the three most important trends over the last 90 days. Return a Markdown report with an executive summary, supporting tables, source notes, and recommended actions. ``` For a more specific page/source report, use this: ```text Use Composio and Google Analytics to create a report for the past 30 days. Scope: - Site or app: [SITE_OR_APP] - Pages: only pages under [PATH_PREFIX] - Questions: 1. Which pages had the most traffic? 2. What were the main traffic sources for those pages? 3. Which pages look under-distributed or worth promoting again? Output: - Executive summary - Table of top pages - Table of traffic sources by top page - 3 recommended actions - Source/tool notes ``` Markdown should be the default first output. Terminal tables are useful while you are working, but Markdown gives you something you can keep, edit, send, and convert. If the report needs to become a PDF, spreadsheet, or website later, you can ask the agent to build from the Markdown. > Pro tip: If you want to build charts, spreadsheets, or deeper visualizations later, ask the agent to save the underlying rows as a CSV too. You do not need that for the first version of the guide, but it is useful when the report becomes a recurring workflow. ### Step 5 Visualize the Report After the Markdown report looks right, ask the agent to turn it into a handoff-ready format. Use a prompt like this: ```text Use the Markdown report to create a [OUTPUT_FORMAT] report for [AUDIENCE]. Focus on: - the top findings - the most important chart or table - the business recommendation - source notes Keep the output clean and readable. ``` For example, the same Markdown report can become a simple PDF traffic snapshot with summary cards and a ranked table. ![final pdf report preview](https://tru-images.b-cdn.net/guide-assets/209-for-jason/06-final-pdf-report-preview.png) You can also ask for a spreadsheet: ```text Create a spreadsheet version of this Markdown report with one tab for summary, one tab for the key table, and one tab for recommended actions. ``` Or ask for an internal website: ```text Build a simple local HTML report from the Markdown report. Include the executive summary, the main table, trend notes, and recommended actions. ``` ### Step 6 Turn the Analysis and Visualization Into Skills Once the workflow works once, do not leave it as a one-off chat. Ask Codex or Claude Code to turn the process into a reusable skill: ```text Create a reusable reporting skill from the analysis and visualization process we just used. Important: - Do not include the initial data-source exploration step in this skill. - Assume I have already confirmed the right toolkit and data source. - Make the skill work with any Composio toolkit, not only Google Analytics. The skill should let me provide: - toolkit name - confirmed data source or account - date range - report goal - desired output format It should: 1. Run the report using the confirmed Composio data source. 2. Analyze and pull the most important trends, changes, risks, and opportunities. 3. Save the main report as Markdown. 4. Optionally create a PDF, spreadsheet, or HTML report from the Markdown. 5. Include source notes, access errors, and privacy warnings. ``` You can keep analysis and visualization in one skill, or split them. Splitting is often better for recurring work: - one skill creates the Markdown analysis report - one skill turns an approved Markdown report into a PDF, spreadsheet, or website That way you can schedule the analysis every week, then only run the visualization skill when the report needs to be shared. If you do split them, use this second prompt: ```text Create a separate visualization skill. Input: - a Markdown business report - desired output format - audience The skill should convert the Markdown report into a clean [PDF_OR_SPREADSHEET_OR_HTML] version without re-running the data analysis. ``` ### Step 7 Schedule the Report as an Automation After the skill works, grab the skill name(s) and create a new automation with this prompt: ```text Use [ANALYSIS_SKILL_NAME] to create a [LOOKBACK_DAYS]-day report. Scope: - Toolkit: [TOOLKIT] - Confirmed data source: [DATA_SOURCE_OR_ACCOUNT] - Business area or report scope: [REPORT_SCOPE] - Report goal: [REPORT_GOAL] Output: - Markdown report first - source notes - access errors - recommended actions If [VISUALIZATION_SKILL_NAME] is provided, run that skill after the Markdown report is complete and create a [OUTPUT_FORMAT] version for [AUDIENCE]. ``` For a Google Analytics weekly report, that might become: ```text Use the business-report-analysis skill to create a 7-day report. Scope: - Toolkit: Google Analytics - Confirmed data source: RU main site - Business area or report scope: app.therundown.ai/guides - Report goal: identify the most important page traffic trends and traffic source changes Output: - Markdown report first - source notes - access errors - recommended actions Then use the business-report-visualization skill to create a PDF version for the marketing team. ``` In Codex, the automation setup should look something like this: ![codex report automation](https://tru-images.b-cdn.net/guide-assets/209-for-jason/07-codex-report-automation.png) Run it weekly for a normal content or marketing report. Run it daily only when someone will actually review it daily. For monthly reporting, change [LOOKBACK_DAYS] to 30. For quarterly reporting, change it to 90. ### Going Further Once this works for Google Analytics, use the same Explore, Analyze, Visualize framework on other business systems. Try prompts like: ```text Use Composio to audit the available toolkits connected to this workspace. Based on what you know about my work, recommend five data sources that could produce useful recurring business reports. For each one, suggest: - what report to build - what data to inspect first - what metrics matter - what output format would be useful - how often the report should run ``` You can use the same pattern for: - YouTube channel and video performance - Shopify product and promotion reports - HubSpot pipeline and lead source reports - Google Sheets operational dashboards - Gmail or support inbox trend reports - Notion project or content pipeline reports ### Pro tip > Start with read-only reporting access, prove the report works once, then turn the analysis and visualization steps into reusable skills before scheduling it. ### Guide: Build Better Slide Decks With AI (Works With Any Tool) > This guide teaches you how to turn raw, unstructured content into a presentation-ready slide deck using Claude and Gamma. - Canonical page: https://app.therundown.ai/guides/build-better-decks-with-ai-works-with-any-tool - Machine-readable page: https://app.therundown.ai/guides/build-better-decks-with-ai-works-with-any-tool.md - Difficulty: beginner - Featured tool or company: Gamma - Categories: Productivity & Knowledge Work, Content & Creative - Instructors: Billy Howell - Published: 2026-02-23 - Updated: 2026-07-23 ### Overview #### The Rundown This guide teaches you how to turn raw, unstructured content into a presentation-ready slide deck using Claude and Gamma. You will start with material like a spreadsheet, meeting transcript, long document, or bullet-point outline, then use Claude to shape it into a deck structure and Gamma to design the slides. The workflow is designed to save time on repeat deck-building tasks, such as monthly performance reports that would otherwise take hours to rebuild manually. #### Who This Is Useful For - Marketing managers who run recurring performance reports and need to update the same deck with new numbers - Agency owners who need to turn a scope of work into a client kickoff deck - Ops leads who need to summarize a planning meeting for executives #### What You Will Build You will build a complete, designed slide deck generated from raw content you already have. The finished product comes out of Gamma with your structure, data, and narrative already in place, so you are not starting from a blank slide or manually dragging text boxes around. - A Claude-generated markdown deck outline - A Gamma-generated designed slide deck - A polished deck that can be presented, exported as PowerPoint, or exported as PDF #### What You Need - Claude, free or Pro, at claude.ai - A Gamma account, with the free tier available, at gamma.app - Raw content such as a CSV export, meeting notes, a document, a transcript, or a rough outline #### Going Further - After you run the workflow once, save the Claude-generated outline as a template. For recurring reports, such as weekly marketing updates, monthly board decks, or client recaps, you can give Claude the same outline and ask it to fill it in with new data and content. - When the deck is ready, you can present directly from Gamma or export it as a PowerPoint or PDF when your audience needs a file. - You can also connect this workflow with other tools, such as using Zapier to auto-export analytics data on a schedule, send it into Claude via the API, and have a draft deck waiting for you every Monday morning. ### Step 1: Gather your raw content Grab the files or notes you want to turn into a deck. This could be a CSV export from Shopify or Google Analytics, a Zoom meeting transcript, a scope of work, or a messy notes document. Your starting point does not need to look presentation-ready. This workflow is most useful when the source material is raw, messy, or too unstructured to drop directly into slides. ### Step 2: Ask Claude to define the deck brief Start a new chat in Claude, attach your documents, and use this prompt: ```text I need you to turn these documents into a slide deck outline. Before you start, ask me five multiple-choice questions about the purpose, audience, tone, and structure so you get it right. ``` > Claude may respond with a question like: “Who is the primary audience for this deck? A) Your direct team (internal update) B) Executive leadership (high-level summary) C) A client or external stakeholder D) Investors or board members” Answering these questions helps Claude tailor the deck before it generates anything. A deck for your internal team should look different from a deck for a client, and Claude needs that context to avoid producing something generic. > Pro tip: You can add your own questions to the prompt. If brand tone matters, ask Claude to include a question about visual style. If there is a specific section you always need, like “next steps” or “budget impact,” tell it to ask about that too. ### Step 3: Generate and refine the markdown outline Answer Claude’s questions and let it generate the outline in markdown. Each top-level heading becomes a slide title, and bullet points become the content on each slide. Claude should pull specific numbers, quotes, and details from your raw content, then organize them around the audience and goal you selected. > Example output: > > # Q2 Planning Summary > > Slide 1: Q1 Results at a Glance > - Revenue: $1.24M (31% above plan) > - Top performer: All-Star drop sold out in 8 hours > - Repeat purchase rate hit 28% > > Slide 2: What Worked > - Scarcity + event timing + strong creative = formula > - Email open rate on ASW drop: 47% > - Bundle pricing increased AOV by 22% > > Slide 3: Q2 Priorities > ... Read through the outline before moving to Gamma. If a slide feels off or you want to add something, ask Claude to revise the markdown first. ```text Add a slide about [your topic] after slide 3. ``` ```text Make slide 5 more visual with a before/after comparison. ``` Iterate in Claude rather than in Gamma. It is much faster to edit markdown than to rearrange designed slides. > Pro tip: If you want Claude to match a format you have used before, paste in a previous deck outline and say, “Follow this same structure but with the new data.” ### Step 4: Create the deck in Gamma Go to gamma.app and click Create New. Select Generate from notes or outline, then paste in your outline from Claude. Click Continue and review the preview of your deck. On the preview page, select the Preserve and Concise options for the best slide deck result. Then click Generate, and Gamma will create a full deck for you. ![image-2075](https://tru-images.b-cdn.net/guide-assets/69/699ce285d61470347f86e0ab/image-2075.png) From there, edit any slide with AI, tweak the design, and export the finished deck as a PowerPoint or PDF when you are ready to share. At this point, you are editing a finished deck rather than building one from scratch. Claude handled the structure, narrative, and data; Gamma handled the design. Your job is final polish. > Pro tip: Gamma lets you present directly from the browser, so you do not need to export to PowerPoint unless your audience specifically needs a .pptx file. ### Pro tip > Iterate on the markdown outline in Claude before moving into Gamma; it is faster than rearranging designed slides later. ### Guide: Build Interactive Lead Magnets With Canva Code > In this guide, you will learn how to create a fun, interactive lead magnet for your company without learning to code, helping you impress prospects and earn more business. - Canonical page: https://app.therundown.ai/guides/build-interactive-lead-magnets-with-canva-code - Machine-readable page: https://app.therundown.ai/guides/build-interactive-lead-magnets-with-canva-code.md - Difficulty: beginner - Featured tool or company: Canva - Categories: Marketing & Growth, Content & Creative, Coding & App Development - Instructors: Billy Howell - Published: 2026-07-28 - Updated: 2026-07-28 ### Overview #### The Rundown In this guide, you will learn how to create a fun, interactive lead magnet for your company without learning to code, helping you impress prospects and earn more business. This is the most beginner-friendly way to dabble in custom-coded marketing websites with ChatGPT and Canva Code. #### Who This Is Useful For - Small business owners and service providers who want a useful way to start conversations with prospective clients. - Marketers and consultants who want to turn a static checklist or PDF into a scored assessment people can complete online. - Anyone new to building websites who wants a beginner-friendly way to create and publish an interactive experience. #### What You Will Build You will build a branded readiness assessment that gives each visitor a score, a personalized action plan, and a clear next step. You will publish it as a shareable Canva website and connect the email and quiz results to a Canva Sheet. ![A published SEO readiness lead magnet with a clear promise and assessment button](https://tru-images.b-cdn.net/guide-assets/local-396/build-interactive-lead-magnets-with-canva-code/01-seo-readiness-lead-magnet-landing-page.png) #### What You Need To Get Started - A Canva account. Canva Code 2.0 is available to Free, Pro, Business, Enterprise, and Education users (https://www.canva.com/newsroom/news/Canva-Code/) - ChatGPT, Claude, Gemini, or another AI assistant you use for planning and writing. - A clear offer and target customer, or a rough checklist you already use with prospects. #### Going Further Reuse this workflow for calculators, product recommenders, client onboarding questionnaires, team intake forms, and other experiences where the next step should change based on a person's answers. Use File > Make a copy before adapting the assessment for another audience. Your teammates can comment and edit inside Canva, so the working version can stay collaborative while the published link remains stable. Start with the free Canva URL if a custom domain slows you down. A useful experience you can test with real people will teach you more than a polished domain attached to an unfinished form. ### Step 1 Define the Lead Magnet Start with the customer problem your lead magnet should help diagnose. Describe your offer, the person you want to reach, and the result they want. In our example, we created an SEO readiness assessment for small business owners. It turned a vague concern, "my website needs better SEO," into a score, priority areas, and a four-week action plan. Open ChatGPT or your preferred AI assistant and use a prompt like this: ```text Create a lead magnet for [TARGET CUSTOMER] who wants to [DESIRED OUTCOME]. Make it a readiness assessment for [TOPIC]. Include practical questions that help the reader understand where they are doing well, where they are falling behind, and what they should do next. Brand it for [COMPANY NAME] and keep the advice useful for someone who may become a client for [OFFER OR SERVICE]. ``` Review the questions before moving on. Remove anything that requires information the visitor will not know, and make sure each answer can lead to a useful recommendation. A simple checklist or PDF is enough for this planning pass. ### Step 2 Turn the Checklist Into a Canva Code Prompt Once the assessment logic looks right, ask the AI to convert it into a build prompt for Canva Code. ```text Turn this lead magnet into a prompt for Canva Code. Build it as a responsive, interactive landing-page assessment. Include: - the assessment questions and answer choices - clear scoring or recommendation rules - distinct result categories - a personalized action plan - a final call to action - [BRAND COLORS], [VOICE], and [VISUAL STYLE] - readable mobile and desktop layouts Keep the logic deterministic so the same answers always produce the same result. ``` Read the generated prompt before you paste it into Canva. Check the questions, result rules, action plan, CTA, and brand direction. This is faster to fix in plain English than after the experience has been generated. ### Step 3 Generate and Customize the Website From the Canva homepage, open Create a design, choose Code from the left sidebar, and start a new project. Depending on your Canva layout, you may also see Canva Code under Canva AI as Code. ![Canva's Create a design screen with Code selected](https://tru-images.b-cdn.net/guide-assets/local-396/build-interactive-lead-magnets-with-canva-code/02-canva-create-design-code-option.png) Paste the build prompt and let Canva generate the full experience. Inspect the complete question and result flow before making cosmetic changes. Click Edit to change the headline, body copy, colors, fonts, images, and other visible elements like a standard Canva design. You can also use Ask Canva to request a specific revision. Keep the scoring and recommendation rules unchanged while you refine the visual design. Open Preview and run the assessment from beginning to end. Check that: - every question can be answered - progress and navigation work - the score matches the answers - the result category makes sense - the action plan is specific - the CTA opens the correct destination - the layout remains readable on mobile ![The finished assessment showing a score, category breakdown, and priority areas](https://tru-images.b-cdn.net/guide-assets/local-396/build-interactive-lead-magnets-with-canva-code/03-seo-readiness-score-results.png) > Pro tip: We hit browser issues with Canva in Safari during testing. If Canva Code behaves strangely, open the project in Chrome before rebuilding anything. ### Step 4 Capture Emails and Save Results Add the capture step only after the assessment itself works. In Ask Canva, use a direct follow-up: ```text Add an email capture at the end of the quiz and save the email with the assessment results. ``` Canva Code can connect the responses to a Canva Sheet. Open the Data panel to inspect the connected sheet and review Canva's access and personal-data notice. ![Canva Code's data-storage notice explaining connected Sheet access and privacy responsibilities](https://tru-images.b-cdn.net/guide-assets/local-396/build-interactive-lead-magnets-with-canva-code/04-canva-code-data-storage-notice.png) Submit the assessment once with your test email. Then open the Data panel and refresh it. Confirm that the email and expected result fields appear in the sheet before you share the link. > Pro tip: Ask Canva to store both the contact information and the quiz result, then run a test submission and refresh the Data panel. A visible email field does not prove the data reached the sheet. Anyone who can access the design may also be able to access its connected data. Use clear consent language, collect only what you need, and follow the privacy rules that apply to your business. ![The finished action plan with review controls and an email-capture section](https://tru-images.b-cdn.net/guide-assets/local-396/build-interactive-lead-magnets-with-canva-code/05-seo-readiness-action-plan-email-capture.png) ### Step 5 Publish and Test the Live Site Click Publish and review the website settings before it goes live. Choose a free Canva URL or an eligible custom-domain option, set the search visibility, add password protection if the page is private, and check the social link preview. Open the published link in a fresh browser window and complete the full assessment again. Test the questions, score, action plan, email capture, connected Sheet, CTA, and phone layout on the live site. Canva Websites also includes website analytics. Open Insights after publishing to review traffic and engagement. Website Insights and some custom-domain options depend on your Canva plan (https://www.canva.com/domains/top-level-domain/), so verify the controls available in your workspace before promising them to a client. ### Guide: Build This AI-Powered Sales Objection Handler > In this guide, you’ll turn the sales call transcripts you are already collecting into a simple weekly objection-handling system. - Canonical page: https://app.therundown.ai/guides/build-this-ai-powered-sales-objection-handler - Machine-readable page: https://app.therundown.ai/guides/build-this-ai-powered-sales-objection-handler.md - Difficulty: beginner - Featured tool or company: OpenAI - Categories: Sales, Consulting & Customer Success - Instructors: Billy Howell - Published: 2026-02-09 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you’ll turn the sales call transcripts you are already collecting into a simple weekly objection-handling system. You’ll use ChatGPT to parse call notes, catalogue objections, and produce a quick reference document you can use during sales calls. #### Who This Is Useful For - Sales reps who need quick answers during “crunch time” on calls. - Founders doing founder-led sales who need to track product feedback patterns. - Managers who want a digestible weekly summary of why deals are stalling. #### What You Will Build You’ll build a reusable system that uses ChatGPT to parse your weekly call notes and catalogue sales objections into a searchable reference. ![CleanShot 2026-02-09 at 16.13.46@2x](https://tru-images.b-cdn.net/guide-assets/69/698a4e16912946607d3e398f/cleanshot-2026-02-09-at-16-13-46-2x.png) - A ChatGPT Project that processes sales transcripts. - A weekly report template for objections, context, prospects, and rebuttals. - A single-page Notion memory bank you can search during calls. #### What You Need - ChatGPT Plus account - ChatGPT desktop app - Notion desktop app - A batch of sales call transcripts - ChatGPT and Notion are recommended, but the same principles can be applied to other AI and note-taking tools. #### Going Further - Connect Notion inside ChatGPT through Settings > Connected Apps so ChatGPT can see the page you are currently working on and more easily reference past weeks. - Reuse the same “Project + Transcript” workflow for all-hands meetings or stand-ups. - Ask for a “fluff-free summary” to send to your boss when a curated, readable summary is more useful than a raw transcript. ![image-2039](https://tru-images.b-cdn.net/guide-assets/69/698a4f05912946607d3e423e/image-2039.png) ### Step 1: Choose a Simple Cheat Sheet Over a Perfect Database Most people over-engineer AI workflows by building beautiful, color-coded databases with too many properties per record. ![image-2040](https://tru-images.b-cdn.net/guide-assets/69/698a4f20912946607d3e4a89/image-2040.png) - Find and open the database - Filter by date - Filter by tag - Click into the page and find the objection rebuttal Instead, use ChatGPT to help update a single-page sales call cheat sheet that is easier to search during a live call. ![image-2041](https://tru-images.b-cdn.net/guide-assets/69/698a4f30912946607d3e52b6/image-2041.png) ### Step 2: Set Up Your ChatGPT Project Use ChatGPT Projects as your processing engine instead of feeding transcripts into a database one by one. - Open ChatGPT. - Click “New Project.” - Name the project. - Create a simple text file listing your product lines, pricing, and core offers. - Upload that text file to the project files so ChatGPT understands what you sell. - Every week, export your sales notes or transcripts. - Make sure each transcript includes a date inside the text, such as “Call with [Client] - Feb 9.” - Drag and drop the weekly transcript files into the Project. ### Step 3: Configure the System Instructions Tell ChatGPT exactly how to format the data so the report is readable at a glance. - Click the three dots next to your Project name. - Select “Edit Instructions.” - Paste instructions similar to the prompt below. ```text Read the attached transcripts. Create a weekly report template. For every objection found: 1. Number the objection. 2. State the objection clearly. 3. Provide 3 descriptive bullets on the context. 4. List which lead or prospect surfaced it. 5. Provide exactly two short, punchy rebuttals. ``` ### Step 4: Run the Weekly Report Do not automate this end-to-end. Fully automated workflows can become output you never review. Manually generating the report forces you to review the week’s performance. - Open your Project. - Prompt ChatGPT to run the weekly report for the week you want to review. - ChatGPT scans the relevant files using the date filtering in your prompt and generates a clean list. ```text Run weekly report for the week starting Feb 9. ``` ### Step 5: Build the Notion Memory Bank Use a blank Notion page as your memory bank. - Open a blank Notion page. - Create a Toggle Heading for the week, such as “Week of Feb 9.” - Copy the ChatGPT output. - Paste it inside the toggle. When you are on a call and hear an objection about a topic like “contracts,” use Cmd+F or Ctrl+F, search “contract,” and quickly find objections and rebuttals related to that topic from the last few months. ### Pro tip > Keep the workflow manual enough that you actually review the weekly output before saving it to Notion. ### Guide: Build Your Own AI SEO Specialist With Gumloop > In this guide, you will learn how to build a team of Gumloop agents that audits one page from your site and organizes the findings in one Google Sheets workbook. - Canonical page: https://app.therundown.ai/guides/build-your-own-ai-seo-specialist-with-gumloop - Machine-readable page: https://app.therundown.ai/guides/build-your-own-ai-seo-specialist-with-gumloop.md - Difficulty: beginner - Categories: Marketing & Growth, Automation & Agents, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-07-23 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to build a team of Gumloop agents that audits one page from your site and organizes the findings in one Google Sheets workbook. The agents will add backlink, content refresh, and technical SEO findings, then combine them into one Issues queue. #### Who This Is Useful For - Content marketers who want a repeatable way to refresh older pages and find useful internal links. - Small website teams that need a practical SEO issue register without adding another paid SEO suite. - Anyone who wants to test cloud agents on one page before scaling them across a larger site. #### What You Will Build You will build a shared SEO audit workbook powered by a Backlink Agent, Content Refresh Agent, and Technical SEO Audit Agent. Each one reviews the same page and adds its findings to the workbook. A final Issues tab turns the recommendations into one prioritized queue for a teammate, contractor, or client. ![A shared SEO workbook with separate tabs for backlinks, content refreshes, and technical SEO.](https://tru-images.b-cdn.net/guide-assets/local-390/build-your-own-ai-seo-specialist-with-gumloop/01-seo-workbook-backlinks-tab.png) #### What You Need To Get Started - A Gumloop account (https://www.gumloop.com/pricing). Gumloop currently lists a Free plan with 5,000 credits per month, but check the live pricing page before estimating a larger crawl. - A Google account that Gumloop can use to create and update a Google Sheets workbook. - A public website to audit. #### Going Further Give your main SEO audit agent your website URL and tell it to coordinate the other agents, audit the first 25 pages it finds, and create the full workbook. ```text Run a full SEO audit for: [WEBSITE URL] Coordinate the Technical SEO Audit Agent, Backlink Agent, and Content Refresh Agent. Audit the first 25 HTML pages you find on the same hostname. Use Firecrawl for public-web research and Google Sheets for the output. Create one Google Sheets workbook with these tabs: - Backlinks - Content Refresh - Technical SEO Audit - Issues Preserve source URLs and coverage limitations. Do not edit the website, publish changes, or send outreach. Return the workbook URL, pages reviewed by each agent, failures, issue totals, and the biggest coverage limitation. ``` Review the workbook and credit use before increasing the crawl limit. If you target older blogs for content refreshes, make sure each page exposes a published date that Firecrawl can read. Keep a human review step before anyone edits a page, publishes content, or sends outreach. ### Step 1: Build the Backlink Agent and Start the Workbook Open Gumloop's Backlink Specialist template (https://www.gumloop.com/templates/backlink-specialist) and select Use this Agent. The template gives you a fast starting point for the internal-link job: inspect one page, map the same site, and suggest relevant pages to connect. ![Start from Gumloop's Backlink Specialist template.](https://tru-images.b-cdn.net/guide-assets/local-390/build-your-own-ai-seo-specialist-with-gumloop/02-backlink-specialist-template.png) Rename the duplicate Backlink Agent. In Connectors, keep Firecrawl and add Google Sheets. You do not need the template's Google Docs connector for this workbook workflow. Use Firecrawl Scrape for the target page and every recommended destination, Map to discover pages on the same site, and Google Sheets to create the workbook and write the findings. Leave email sending, CMS publishing, page editing, and delete tools off. ![Review the template's Firecrawl connector before adding Google Sheets.](https://tru-images.b-cdn.net/guide-assets/local-390/build-your-own-ai-seo-specialist-with-gumloop/03-backlink-agent-connectors.png) Choose one page from your site that you want to audit and copy the URL. > Pro tip: A blog that needs a content refresh or a product page that needs better keywords is a great place to start. Testing one page also lets you review the output and credit use before scaling. Give the Backlink Agent the URL and ask it to verify every destination before recommending a link: ```text Create an internal-link map for this target page: [PAGE URL] Search only this site: [SITE DOMAIN] Use Firecrawl Scrape on the target page, Map on the supplied domain to discover up to eight relevant candidate pages, and Scrape every destination you recommend. Return no more than five recommendations with: - source page - destination page - concise anchor text - exact sentence or paragraph after which to add it - reader benefit - confidence Include rejected candidates and the reason each was excluded. Avoid generic anchors, keyword stuffing, repeated destinations, and recommendations based only on search snippets. End with Verification Notes. ``` Review the suggestions before you save them. Open at least one destination, confirm the anchor reads naturally in the source paragraph, and reject anything that was not fetched successfully. Then create the shared workbook: ```text In Google Sheets, create an SEO audit workbook for: [PAGE URL] Create a tab named Backlinks and add every verified internal-link recommendation. Include these columns: Anchor text | Destination URL | Destination title | Source page | Exact insertion location | Reader benefit | Confidence | Notes Format the header, wrap long text, and keep URLs clickable. Do not publish any link changes. ``` The Backlink Agent creates the workbook for this page. The Content Refresh Agent and Technical SEO Audit Agent will use the same URL and add their findings to the same file. ### Step 2: Add the Content Refresh Agent Create a second agent named Content Refresh Agent. Give it Firecrawl and Google Sheets, then add instructions that keep it focused on one page, one audience, and one editing brief. Run it with this prompt: ```text Create a content refresh brief for this page: [PAGE URL] Audience: [AUDIENCE] Business goal: [BUSINESS GOAL] Use Firecrawl Scrape on the target page first. Then use Firecrawl Search to find and Scrape no more than three current, authoritative comparison sources. Return: 1. What the page already does well and should keep 2. High-, medium-, and low-priority refresh recommendations 3. Missing reader questions 4. Suggested title, meta description, and H2 changes 5. An evidence table with each source URL and the recommendation it supports 6. Verification Notes listing successful fetches, failed fetches, and data you did not have Do not rewrite the full page. Do not claim keyword volume, difficulty, rankings, traffic, domain authority, or Search Console performance. ``` The audience line matters. A refresh brief for first-time buyers should make different recommendations than one for technical practitioners. Put the audience and business goal in the agent instructions if they apply to every page this agent will review. Once the brief is useful, send the existing workbook URL and assign the agent its tab: ```text We have an existing SEO audit workbook for this page: [WORKBOOK URL] Put your findings in a tab named Content Refresh. Preserve every other tab in the workbook. Use these columns for the main recommendation table: Priority | Recommendation | Evidence or source | Type | Notes Add a short summary at the top of the tab and match the workbook's existing style. ``` ![The content-refresh agent compares live sources and prepares a prioritized brief for the workbook.](https://tru-images.b-cdn.net/guide-assets/local-390/build-your-own-ai-seo-specialist-with-gumloop/04-content-refresh-agent-and-output.png) Read the Verification Notes before treating the output as an audit. The agent can compare public evidence and make qualitative recommendations. It cannot honestly provide keyword volume, ranking position, traffic, domain authority, Search Console performance, or conversion data unless you connect those sources. ### Step 3: Add the Technical SEO Audit Agent Create a third agent named Technical SEO Audit Agent. This is the most constrained agent in the system because it can inspect many pages and write a large number of rows. Connect: - Firecrawl Map and Scrape - Google Sheets Get Spreadsheet Info, Get Sheet Names, Add Sheet, Batch Get, Batch Update, Clear Values, and Format Cells Turn AI Discovery off for this agent and disable unrelated abilities. The technical agent already knows the exact connectors it needs, so automatic tool discovery adds noise. Paste this complete instruction block into the agent. It includes the workbook-ownership rules and audit boundaries promised in the walkthrough: ```text You are a single-purpose technical SEO audit agent. Your only job is to inspect a public website with Firecrawl and write an evidence-backed technical SEO issue register to your own tab in a user-supplied Google Sheets workbook. You are one member of a larger SEO agent system. Every agent owns one tab in the same workbook. Your tab is named exactly: Technical SEO Audit Never perform content refresh research, keyword research, backlink research, outreach, or full article rewriting. Never create a new workbook. Never rename, delete, clear, format, reorder, or write to any tab except your own Technical SEO Audit tab. REQUIRED INPUTS Before starting, make sure you have: 1. A public start URL, including the protocol. 2. The exact Google Sheets workbook URL or spreadsheet ID. 3. A crawl limit. If none is supplied, use a maximum of 25 HTML pages. Optional inputs are a sitemap URL, priority page paths, paths to include or exclude, and whether subdomains are in scope. By default, audit only the start URL's hostname and exclude subdomains. If the start URL or workbook is missing, ask for it. If the workbook cannot be opened with edit access, stop and report that clearly. Do not claim that results were saved when a Sheets write fails. WORKBOOK OWNERSHIP RULES 1. Inspect the workbook and list its existing tab names before writing anything. 2. If Technical SEO Audit does not exist, add it to the supplied workbook. 3. Write Agent owner: Technical SEO Audit Agent in cell A2. 4. If the tab already exists and A2 contains that ownership marker, refresh the audit in place by clearing and replacing values only in that tab. 5. If the tab exists but A2 contains a different owner or non-audit content, do not overwrite it. Stop and ask the user to resolve the naming conflict. 6. Never use a whole-workbook clear, replace, sort, or formatting action. 7. On every rerun, preserve all other workbook tabs exactly as they are. AUDIT WORKFLOW 1. Normalize the start URL and record the hostname. 2. Use Firecrawl Map to discover URLs on that hostname, up to the crawl limit. Map returns URL discovery, not page-level proof, so do not treat a mapped URL as audited until it has been scraped. 3. Normalize and deduplicate the discovered URLs. Remove fragments, obvious tracking parameters, mailto or tel links, and non-HTML assets. Keep a note of exclusions. 4. Try to locate and scrape robots.txt and the supplied or discoverable XML sitemap. If either cannot be fetched, record that as a coverage limitation rather than inventing a result. 5. Use Firecrawl Scrape on each in-scope page you audit. Request page metadata, HTML or raw HTML, main content, and links when the tool supports those formats. 6. Keep the run within the crawl limit. If more URLs exist, prioritize the homepage, primary navigation pages, priority paths supplied by the user, and a representative mix of remaining templates. State the coverage limit in the sheet. 7. Compare findings only across the pages actually scraped. Never generalize a sampled issue to the entire site unless the evidence supports it. CHECK ONLY WHAT THE PUBLIC EVIDENCE SUPPORTS For every scraped page, check: - fetch success, final URL, HTTP status when returned, and redirect behavior visible to Firecrawl - title presence and exact duplicate titles across the audited set - meta description presence and exact duplicates across the audited set - canonical presence, whether it is self-referencing or points elsewhere, and obvious conflicts - robots meta directives such as noindex or nofollow - H1 presence and multiple-H1 cases; flag heading-order concerns only when the HTML supports the claim - internal links that resolve to verified 4xx or 5xx pages or fail to fetch - HTTPS issues or mixed-content references visible in the returned HTML - JSON-LD structured-data presence and whether the JSON is parseable; do not claim full schema validity without a validator - image alt-text omissions only when image markup is available in the returned HTML - sitemap inclusion and robots.txt conflicts only when those files were fetched successfully - exact duplicate or near-empty page content within the audited set; label thin-content judgments as qualitative review items Use character counts for titles or descriptions only as review signals, never as universal ranking rules. Do not call a URL an orphan page unless you have enough crawl and sitemap evidence. Use Potential orphan — manual verification required when the evidence is incomplete. DO NOT CLAIM TO HAVE TESTED Do not invent or estimate Core Web Vitals, Lighthouse scores, JavaScript execution problems, server response time, crawl budget, log-file behavior, real Google index status, Search Console coverage, rankings, traffic, keyword volume, domain authority, or backlink metrics. Add these to Not tested / needs another data source when relevant. ISSUE QUALITY RULES - Create one row per verified issue per affected URL. - Every issue must include specific evidence from the fetched page or file. - Do not create filler rows to reach a quota. - Consolidate a truly sitewide pattern into one SITEWIDE row only when multiple verified pages support it. List representative URLs in the evidence. - Use these severities consistently: - Critical: a widespread problem that can prevent important pages from being crawled or served, supported by direct evidence. - High: a verified indexability, canonical, redirect, or broken-internal-link problem on an important page. - Medium: a meaningful metadata, heading, sitemap, structured-data, duplication, or accessibility-related issue. - Low: cleanup or optimization that is useful but unlikely to block discovery or indexing. - If confidence is not High, explain what remains uncertain. SHEET LAYOUT Write this run summary at the top of Technical SEO Audit: - A1: Technical SEO Audit - A2: Agent owner: Technical SEO Audit Agent - A3: Run date/time; B3: the current timestamp - A4: Start URL; B4: the audited start URL - A5: Crawl limit; B5: the applied limit - A6: URLs discovered; B6: the count - A7: URLs successfully scraped; B7: the count - A8: URLs failed or skipped; B8: the count - A9: Coverage notes; B9: a concise scope and sampling note - A10: Not tested / needs another data source; B10: a concise list of unsupported checks Start the issue table on row 12 with these columns in this exact order: Severity | Category | URL | Issue | Evidence | Recommended fix | Confidence | Status | Checked at Use Open as the default Status. Sort issue rows by Critical, High, Medium, then Low. Make the header bold and wrap long text. Freeze the header row, enable a filter, and keep URLs clickable when the Sheets tool supports those actions. Do not paste full page bodies or raw HTML into the workbook. If no verified issues are found, still create the tab and add one informational row explaining that no issues were found within the audited sample. This is not a guarantee that the site has no technical SEO problems. FINAL VERIFICATION After writing: 1. Read back the summary cells and the issue-table header from Technical SEO Audit. 2. Confirm the number of issue rows written. 3. Confirm that no other tab was edited. 4. Reply with the workbook name, tab name, pages discovered, pages scraped, failed or skipped count, issue totals by severity, and the most important coverage limitation. Never edit the website, publish changes, send messages, or modify another agent's tab. All fixes are recommendations for human review. ``` For the one-page test, use this run prompt: ```text Run a technical SEO audit of: [PAGE URL] Write the results to this existing Google Sheets workbook: [WORKBOOK URL] Audit only the supplied page. Create or refresh only the Technical SEO Audit tab and preserve every other tab in the workbook. ``` ![The technical agent records its owner, crawl limit, coverage, and unsupported checks in its own workbook tab.](https://tru-images.b-cdn.net/guide-assets/local-390/build-your-own-ai-seo-specialist-with-gumloop/05-technical-seo-audit-tab.png) Review the summary first. In the demonstrated single-page run, the tab visibly records the crawl limit, successful scrape count, and checks that need another data source. That coverage block is as important as the issue rows because it stops a narrow sample from being presented as a complete site audit. ### Step 4: Build One Issues Queue At this point, the workbook has separate specialist tabs. Give the final consolidation prompt to an agent with Google Sheets access. It reads the completed tabs and writes one consolidated Issues tab, so it does not need Firecrawl. Use this prompt: ```text Review every specialist tab in this SEO audit workbook: [WORKBOOK URL] Create or refresh only a tab named Issues. Preserve every other tab exactly as it is. Collect every actionable recommendation from Backlinks, Content Refresh, and Technical SEO Audit. Deduplicate overlapping items without deleting useful evidence. Use these columns: Status | Priority | Source tab | Category | Page URL | Issue | Evidence | Recommended fix | Owner | Due date | Notes Use New as the default Status. Preserve Critical, High, Medium, and Low priorities from the source tabs. If a source does not provide a priority, assign one and label it as coordinator judgment in Notes. Sort Critical and High items first. Keep source URLs clickable. Add the source tab name to every row so a reviewer can trace the issue back to its full evidence. After writing, read back the header and row count. Confirm that you edited only Issues. Do not edit the website, publish changes, or mark anything complete. ``` This is the handoff tab. A teammate or contractor can filter it by priority, claim an owner, make the page change, and move the status from New to In progress, Ready for review, or Done. The specialist tabs remain the evidence layer. ### Guide: Build Your Own Design System For AI > Learn how to use ShadCN with Codex, or Claude Code, to create a custom design system and use it to build a simple portfolio website. - Canonical page: https://app.therundown.ai/guides/build-your-own-design-system-for-ai - Machine-readable page: https://app.therundown.ai/guides/build-your-own-design-system-for-ai.md - Difficulty: beginner - Featured tool or company: Shadcn - Categories: Coding & App Development, Content & Creative - Instructors: Billy Howell - Published: 2026-01-26 - Updated: 2026-07-23 ### Overview #### The Rundown Learn how to use ShadCN with Codex, or Claude Code, to create a custom design system and use it to build a simple portfolio website. This workflow helps simplify project setup so you can reuse a stronger component and design foundation, with a short PRD planning pass that can take about 5 minutes. #### Who This Is Useful For - Vibe coders who struggle with design systems - Non-designers who want better color palettes and components for their website - Anyone who wants to build a simple portfolio site #### What You Will Build ![shadcn_design](https://tru-images.b-cdn.net/guide-assets/69/6977d4db710ec1d53070abe5/shadcn-design.png) You’ll build a custom ShadCN design system and use it as the foundation for a simple portfolio website. - A custom design system with cards, buttons, and reusable components - A new app scaffold using your selected stack - A portfolio site generated from a PRD and your design system instructions #### What You Need - Cursor - Codex or Claude Code - Context7 agent skill, optional but recommended #### Going Further - Reuse the saved ShadCN preset link and instructions in future projects. - Browse ShadCN components and blocks to find components you like best. - If you have Context7 installed, tell your agent to look for ShadCN components before building new UI. - Add ShadCN charts to track portfolio progress. ### Step 1 Customize your ShadCN design system Navigate to https://ui.shadcn.com/create. Use the dropdowns on the right to create your custom design system. ![shadcn_dropdown](https://tru-images.b-cdn.net/guide-assets/69/6977d5039ad47730e73e47f3/shadcn-dropdown.png) ### Step 2 Create the project command and save it Click “Create Project” in the top right, then copy the command for the package manager and stack you use. If you’re not sure which stack to choose, use Next.js + npm. You’ll get a unique command that looks like this: npx shadcn@latest create --preset "" --template next Before you run the command, save it in your notes. You will need the link from it later. ### Step 3 Scaffold and run your new site Open a new terminal session and run your command. ShadCN will ask you to name your app, then create a new directory and app scaffold with your custom design system and starter components. Use cd [folder name] to navigate to the new directory. Then run npm install and npm run dev to run your new site locally. ### Step 4 Add your design system instructions to your agent Open a Codex or Claude session in your new directory. In your Claude.md or Agent.md instructions file, paste the design system instructions from earlier. To get those instructions, click the link in the command ShadCN gave you and copy them. ![shadcn_customcommand](https://tru-images.b-cdn.net/guide-assets/69/6977d5209ad47730e73e4829/shadcn-customcommand.png) Pro tip: Browse the ShadCN components and blocks to find the ones you like best. ### Step 5 Plan and generate the portfolio with a PRD Before you start coding, plan the website outside of your code editor. Go into Claude, ChatGPT, or Gemini and ask it to create a PRD in Markdown based specifically on ShadCN components. ```text Lay out the page in a PRD in Markdown, based specifically on ShadCN components. ``` Now prompt your AI agent with the PRD: ```text Create a simple portfolio website for a web developer named [your name] based on the attached PRD. Make the portfolio items easily editable with markdown files. Use ShadCN components and the design system outlined in the Agent.md. ``` ![shadcn_portfolio_example](https://tru-images.b-cdn.net/guide-assets/69/6977d53c9ad47730e73e4868/shadcn-portfolio-example.png) The source example shows a logbook built with a custom theme and a PRD made with Claude. The quality improves when you spend 5 minutes making a PRD instead of asking for a generic portfolio site. ![shadcn_billyhq](https://tru-images.b-cdn.net/guide-assets/69/6977d54e9ad47730e73e48aa/shadcn-billyhq.png) Pro tip: Use the markdown method and multiple-choice method to go back and forth with the AI. Ask it specifically: ```text What ShadCN component should I use? Where should I put them? ``` ### Step 6 Go further with Context7 and charts If you have the Context7 skill installed, add this to your Agent.md: ```text Always use context7 to look for ShadCN components before building new UI. ``` You can also tell Codex to add ShadCN charts to your project: ```text Use charts from https://ui.shadcn.com/charts/area to track my portfolio progress. ``` ![shadcn_charts](https://tru-images.b-cdn.net/guide-assets/69/6977d5619ad47730e73e48dc/shadcn-charts.png) ### Pro tip > Browse ShadCN components and blocks to find the ones you like best, then use markdown and multiple-choice prompts to go back and forth with your AI agent. ### Guide: Build Your Own Web Crawler With Manus (Runs in the Cloud) > In this guide, you will learn how to turn Manus Cloud Computer into a private 24/7 web crawler for any site you want to monitor. - Canonical page: https://app.therundown.ai/guides/build-your-own-web-crawler-with-manus-runs-in-the-cloud - Machine-readable page: https://app.therundown.ai/guides/build-your-own-web-crawler-with-manus-runs-in-the-cloud.md - Difficulty: beginner - Featured tool or company: Manus - Categories: Automation & Agents, Coding & App Development, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-05-18 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to turn Manus (https://manus.im) Cloud Computer into a private 24/7 web crawler for any site you want to monitor. Instead of having Manus burn tokens checking the same site each day, your cloud computer does the checking for cheap. #### Who This Is Useful For - AI users who want a simple private automation server without touching AWS. - Researchers and operators who check job boards, RFP pages, public sites, feeds, or changelogs. - Manus users who want to stop spending tokens on “nothing changed” checks. - Founders and consultants who want lightweight monitoring without buying another SaaS tool. #### What You Will Build You will set up a Cloud Computer with a web crawler that checks a site twice a day. The crawler will: - check [website URL] - look for [specific information] - log the results to a CSV - run on a schedule with cron For the demo, we used a Grants.gov search page because it has public data and a CSV export. You can swap in almost any public page, feed, or site you already check manually. #### What You Need - A Manus (https://manus.im) account - The Manus desktop app - A site you want to monitor Do not put API keys or passwords in your prompt. If your script needs credentials later, ask Manus to store them in an .env file on the Cloud Computer. #### Going Further Start with a Manus skill, Claude skill, or any AI workflow you already use. Then ask Manus: ```text Look at this workflow and tell me which repetitive parts can run "for free" on my Cloud Computer before the AI agent gets involved. Separate it into: - cheap checks the cloud computer can run - criteria that should wake Manus - judgment-heavy work the agent should still do - logs or files that prove it ran ``` That is the real pattern. Your cloud computer handles the repetitive work. The agent handles the judgment. ### Step 1 Create the Cloud Computer Open the Manus desktop app, go to Settings > My Computer, and click Create now under Cloud Computer. ![create cloud computer card](https://tru-images.b-cdn.net/guide-assets/local-215/build-your-own-web-crawler-with-manus-runs-in-the-cloud/01-create-cloud-computer-card.png) This gives Manus an always-on server where it can keep scripts, logs, cron jobs, and output files. ### Step 2 Pick the Smallest Plan For web crawlers, API checks, cron jobs, CSVs, and logs, start with the $10 Basic plan. ![cloud computer plan options](https://tru-images.b-cdn.net/guide-assets/local-215/build-your-own-web-crawler-with-manus-runs-in-the-cloud/02-cloud-computer-plan-options.png) Use bigger plans for local models or video. For a crawler that checks a site twice a day, the Basic plan is the right default. > Pro tip: Start small. You can always upgrade later if the job actually needs more compute. ### Step 3 Start a Task on the Cloud Computer Start a new Manus task and select your Cloud Computer in the composer. ![cloud computer task composer](https://tru-images.b-cdn.net/guide-assets/local-215/build-your-own-web-crawler-with-manus-runs-in-the-cloud/06-cloud-computer-task-composer.png) If you skip this, Manus may build in a temporary workspace instead of your new Cloud Computer. > Pro tip: Any time the job needs to keep running after your laptop closes, select the Cloud Computer before you prompt. ### Step 4 Paste the Crawler Prompt Now give Manus the site, the specific information you care about, and what you want it to do with matching results. ```text Set up a web crawler on my Cloud Computer for [website URL]. First inspect the site formatting and tell me the best way to check it reliably. Check for [specific info] twice a day. Log it to a csv. Create a small script, schedule it with cron, log every run, and only send matching results to Manus for [output]. ``` The most important line is the inspection step. Before Manus writes code, it should check the site's formatting and decide the best way to read it reliably. > Pro tip: Do not ask Manus to build first. Ask it to inspect first. That is how you avoid brittle automations. ### Step 5 Run It Once Before Turning It On Have Manus run the script once before turning it on 24/7. Use this: ```text Run the script once before turning it on 24/7. Show me: - what path you used to read the site - what files you created - the CSV you wrote - the latest run log - what Manus would receive if there were a match ``` For the recording, we used Grants.gov as the test site: ![grants search csv export](https://tru-images.b-cdn.net/guide-assets/local-215/build-your-own-web-crawler-with-manus-runs-in-the-cloud/07-grants-search-csv-export.png) That page works well because the results are public and there is a CSV export. But the same pattern works for job posts, pricing pages, competitor updates, policy pages, changelogs, and feeds. > Pro tip: If the test run is messy, tighten the prompt before you schedule it. It is easier to fix the format once than clean up bad logs later. ### Step 6 Turn It On 24/7 Once the test run looks right, have Manus schedule the script with cron. Use this: ```text Schedule this to run twice a day with cron. After you schedule it, show me: - the cron entry - the latest run log - where matching results are saved - where errors are logged ``` Check the log after the first scheduled run. A useful log should show the date, the page checked, what it found, and whether Manus received anything. Once it is running, you can ask Manus to check the collected data anytime from your phone or laptop. The crawler keeps doing the boring part in the background. ### Guide: Build, Test, and Publish an iOS App Without Leaving Codex > In this guide, you will learn how to use Codex to prep your app for the App Store. - Canonical page: https://app.therundown.ai/guides/build-test-publish-ios-app-without-leaving-codex - Machine-readable page: https://app.therundown.ai/guides/build-test-publish-ios-app-without-leaving-codex.md - Difficulty: beginner - Featured tool or company: Codex - Categories: AI - Instructors: Billy Howell - Published: 2026-08-18 - Updated: 2026-08-18 ### Overview #### The Rundown In this guide, you will learn how to use Codex to prep your app for the App Store. You really just need an app, Xcode installed on your Mac, and an Apple Developer account. Codex can handle the rest. #### Who This Is Useful For - A solo builder who has a working iPhone prototype and needs a clear path through Xcode and Apple’s distribution requirements. - A product or operations lead who wants Codex to turn an unfamiliar release checklist into verified actions and handoffs. - Anyone moving a local Swift app from a simulator demo toward a tested, distribution-ready project. #### What You Will Build You will turn a working Swift iPhone app into a tested Xcode project with its identity, icon, signing setup, support pages, and release checks reviewed. We’ll use OurDays, a small couples calendar app, as the example. ![ourdays working result](https://tru-images.b-cdn.net/guide-assets/local-284/build-test-publish-ios-app-without-leaving-codex/01-ourdays-working-result.png) Use the same approach with your own project: have Codex inspect the local files, separate local work from Apple-only actions, and keep the checklist tied to evidence. #### What You Need To Get Started - A working iPhone app or prototype and its local project folder. - ChatGPT with Codex mode available, plus Xcode on a Mac. - An Apple Account and the Apple Developer Program (https://developer.apple.com/programs/) when you need to distribute beyond local development. - The information needed to write a privacy policy and support page for your app. #### Going Further Save the verified checklist with the project so the next build starts from known signing details, release checks, public URLs, and open Apple actions. ### Step 1 Open the Project and Ask Codex to Scan It Switch ChatGPT into Codex mode and open the folder that contains your app. If you only have source files, ask Codex to create the Xcode project and then open the .xcodeproj file in Xcode. Xcode uses that project file for the build and signing work. Give Codex the project path, your experience level, and the finish line. For example: ```text I have a local Swift iPhone app called [APP_NAME] in [PROJECT_FOLDER]. I am a beginner with Apple’s distribution process. I want to move this app toward a tested Xcode project that is ready for Apple’s next distribution step. Scan the project and guide files without changing anything. Check the app target, scheme, display name, bundle identifier, version, build number, icon, signing setup, simulator availability, release build, tests, and any privacy or support requirements you can verify locally. Return: 1. What you verified. 2. What is missing or uncertain. 3. What you can safely do locally. 4. What I must do in Xcode, the Apple Developer app or developer.apple.com, or App Store Connect. 5. The shortest ordered plan to a distribution-ready build. ``` Then use the shorter interview instruction from the video: ```text Scan for App Store readiness, list what’s missing, then interview me one question at a time. Name the right surface: Codex, Xcode, Apple Developer, or App Store Connect. ``` This keeps Codex from guessing which Apple surface owns the next action. It also gives you a checked gap list before any local edit begins. > Pro tip: When Apple or App Store Connect returns an error, paste the exact message into Codex and ask for the smallest related next step. Keep passwords, recovery codes, and two-factor authentication codes out of Codex. Complete those actions yourself in the relevant Apple surface. ### Step 2 Verify Apple Access and the Xcode Project Apple separates this work across three places: - Codex inspects the local project, runs checks, and explains the gap list. - Xcode manages the target, simulator, signing, build, tests, and archive. - Apple Developer manages the membership that unlocks distribution capabilities. Enroll in the Apple Developer Program (https://developer.apple.com/programs/) before you need to upload. Apple currently lists membership at $99 per year and distinguishes free local development from paid distribution access. Approval can take up to 48 hours, so start this step early. In Xcode, open Settings > Accounts and add the Apple Account you will use for the app. Open Settings > Components and install the current iOS simulator runtime if it is missing. If your Mac is low on space, remove simulator runtimes you no longer use. If your Xcode version supports its external-agent workflow, you can connect Codex inside Xcode as an optional convenience. Apple documents this for compatible Xcode versions, but you do not need the in-Xcode connection to follow the core workflow. Codex can inspect the project, and Xcode can build and test it. ![xcode project settings](https://tru-images.b-cdn.net/guide-assets/local-284/build-test-publish-ios-app-without-leaving-codex/02-xcode-project-settings.png) ### Step 3 Use Codex to Audit the Release Gaps With the project open, ask Codex to inspect the app as a release candidate. Have it verify what it can read locally and label anything that requires an Apple account or an external web surface. ```text Review [APP_NAME] for Apple distribution readiness. Check the project target, scheme, bundle identifier, display name, version, build number, icon, signing team, release configuration, export-compliance setting, simulator, tests, and any visible demo or placeholder labels. Do not assume that signing, the App Store record, or the upload is complete. Show evidence for each check, list the remaining gaps, and separate Codex, Xcode, Apple Developer, and App Store Connect actions. ``` Codex can also run the release build and tests. If a command fails because of the environment, let the agent finish diagnosing it before deciding that the app needs a code change. ![codex distribution audit](https://tru-images.b-cdn.net/guide-assets/local-284/build-test-publish-ios-app-without-leaving-codex/03-codex-distribution-audit.png) When Codex finds an issue, answer the current question or complete the named Apple action, then let it update the checklist. Do not hand it a password, recovery code, certificate secret, or payment detail. ### Step 4 Publish the Required Pages and Upload Through App Store Connect App Store distribution needs public policy and support information. You do not need to build a separate marketing website just to provide these URLs. Have Codex draft the pages in Notion, then publish them and test the links while signed out. The recorded example uses OurDays Privacy Policy (https://vaulted-polonium-23c.notion.site/OurDays-Privacy-Policy-3bcb5196e45b81ae87dfc41c354ede13) and OurDays Support (https://vaulted-polonium-23c.notion.site/OurDays-Support-3bcb5196e45b81f8984ec45bb14d903a). For your app, replace them with reviewed public pages and give Codex the URLs so it can place or verify them where your App Store metadata needs them. ![published privacy policy](https://tru-images.b-cdn.net/guide-assets/local-284/build-test-publish-ios-app-without-leaving-codex/04-published-privacy-policy.png) Make the policy match the app you are actually shipping. If you later add accounts, analytics, cloud sync, payments, or another data service, revisit the policy and App Privacy answers before distribution. Make sure you have an app icon before you package the project. Codex can generate one for you if you need it. Run the release build and tests, then review the app icon and the final project identity one more time. Make sure the project is bundled and ready before you move to the Apple-only steps. ![release tests passed](https://tru-images.b-cdn.net/guide-assets/local-284/build-test-publish-ios-app-without-leaving-codex/05-release-tests-passed.png) Apple Developer enrollment can take up to 48 hours, so complete it before you need to upload. Once your membership is approved, create or confirm the app record in App Store Connect using the same bundle identifier. In Xcode, select a generic or real iOS device, choose Product > Archive, and use Distribute App > App Store Connect > Upload. Apple says the build must process before it appears in App Store Connect. After processing, select the build, complete the listing, and submit it for review. See Apple’s App Store Connect workflow (https://developer.apple.com/help/app-store-connect/get-started/app-store-connect-workflow) and upload-builds guidance (https://developer.apple.com/help/app-store-connect/manage-builds/upload-builds) for the current screens. ### Guide: Change One Product Color Without Starting Over > In this guide, you will learn how to use Reve 2. - Canonical page: https://app.therundown.ai/guides/make-precise-image-edits-on-real-photos-with-ai - Machine-readable page: https://app.therundown.ai/guides/make-precise-image-edits-on-real-photos-with-ai.md - Difficulty: beginner - Featured tool or company: Reve - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-07-27 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Reve 2.1 to change the color of one small part of a product photo without regenerating the entire image. You’ll create a new copy with your selected color changes while keeping the rest of the product image intact. #### Who This Is Useful For - Ecommerce teams that need fast color variations for product listings and campaigns. - Designers and social teams that want to update branded details without rebuilding an approved image. - Anyone who needs to test new product colors while keeping the original photo available. #### What You Will Build You’ll create a new color variation of a product photo by changing only the parts you select. In this example, the hat’s brim, rope, and lettering are changed to purple and reused in the finished flat lay. ![Finished product photo with the hat recolored purple](https://tru-images.b-cdn.net/guide-assets/local-405/make-precise-image-edits-on-real-photos-with-ai/01-purple-hat-flat-lay-color-edit.png) #### What You Need To Get Started - A free Reve account (https://reve.com). - A product image with the detail you want to recolor. - A color hex code, if you need an exact brand color. #### Going Further You can make other edits with the same workflow. To change a shadow, material, or background, select that part and edit the text prompt you see instead of changing its color. Describe the result you want, then click Apply to create another version. ### Step 1 Upload Your Product Image After creating your account, open Reve and choose Upload and edit. Select the product image or images you want to change. ![Reve upload and edit start screen](https://tru-images.b-cdn.net/guide-assets/local-405/make-precise-image-edits-on-real-photos-with-ai/02-reve-upload-and-edit-start-screen.png) You can also create an album first if you want to keep several product photos and their color variations together. Once the upload finishes, open the image in the editor. ### Step 2 Find the Exact Part You Want to Recolor Reve automatically scans your image and breaks it into editable parts. In the panel on the right, open the dropdown beside Uploaded image to see the main objects it detected. ![Reve menu showing detected product parts](https://tru-images.b-cdn.net/guide-assets/local-405/make-precise-image-edits-on-real-photos-with-ai/03-reve-detected-product-parts-menu.png) Keep opening the nested dropdown menus to get more detailed. A hat, for example, might contain separate layers for the crown, text, ventilation holes, brim, and shadow. Stop when you find the exact part you want to change. In our test, the clean product photo produced more detailed layers than the busy flat lay. If the part you need is missing, try a simpler source image where the product is the main subject. ### Step 3 Add Your New Color Click the part you want to recolor. Reve will highlight that area and show its editing controls. ![Reve color controls for the selected product part](https://tru-images.b-cdn.net/guide-assets/local-405/make-precise-image-edits-on-real-photos-with-ai/04-reve-open-selected-part-color-controls.png) Click the color palette tool, then click the current hex code. Paste the new hex code you want to test and press Enter. You can also drag around the color picker or use the eyedropper to pull a color from another part of the image. ![Choosing a precise replacement color with a hex code](https://tru-images.b-cdn.net/guide-assets/local-405/make-precise-image-edits-on-real-photos-with-ai/05-reve-choose-new-color-by-hex-code.png) Repeat this for any other detected parts you want to recolor. In our example, the same purple was added to the hat’s lettering and brim before generating the new version. > Pro tip: You can make changes to multiple parts at once before moving to the next step. Copy the hex code so every selected part uses the same color. ### Step 4 Apply the Changes When your color choices are ready, click Apply. Reve will duplicate the original and create a new version with all of your selected color changes. The original stays available, so you can return to it whenever you want to make another color variation. > Pro tip: Start each new color variation from the original image. Repeatedly editing an AI-edited version can add distortion and reduce detail. ### Guide: Claude Cowork Initial Thoughts + Two Surprising Use Cases > This guide introduces Claude Cowork, explains how it differs from regular Claude, and demonstrates two workflows for editing media files from your computer. - Canonical page: https://app.therundown.ai/guides/claude-cowork-initial-thoughts-two-surprising-use-cases - Machine-readable page: https://app.therundown.ai/guides/claude-cowork-initial-thoughts-two-surprising-use-cases.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Productivity & Knowledge Work, Automation & Agents - Instructors: Billy Howell - Published: 2026-01-15 - Updated: 2026-07-30 ### Overview #### The Rundown This guide introduces Claude Cowork, explains how it differs from regular Claude, and demonstrates two workflows for editing media files from your computer. You will use Claude Cowork to work in a folder, compress image copies, speed up a video, and extract audio from another video. #### Who This Is Useful For - AI users who want to level up without opening a code editor - Claude Pro users considering the Max plan - People who want to use Claude to operate on files from their computer #### What You Will Build You will create a reusable Claude Cowork workflow for editing media files in a local folder. - Saved copies of images with file sizes reduced by at least 50% - A copy of a horse video at 2x speed - An MP3 audio file extracted from the small steps video - A basic pattern for queuing multiple file-editing tasks in Claude Cowork #### What You Need - A MacBook - The Claude Desktop app - A Claude Max subscription ($100/month) - Images and videos to use for the demo, either attached files or your own media files - A folder to store the media files #### Going Further - After completing the demo, you can explore Claude Cowork extensions from Settings → Extensions → Browse Extensions. The source recommends Context7 for open source documentation access and suggests starting with extensions developed by Anthropic when expanding what Claude Cowork can automate on your Mac. ### Step 1 Open the Claude Cowork tab Make sure your Claude desktop app is updated, then open it. You should see the new Cowork tab. ![step_1](https://tru-images.b-cdn.net/guide-assets/69/69691e3dc8d33fb39f071baf/step-1.png) ### Step 2 Prepare a media folder Review the suggested workflows. For this demo, you will have Claude edit images and videos. Download the attached images and videos if you have them, or use your own media files, and place them together in a folder. ![step_2](https://tru-images.b-cdn.net/guide-assets/69/69691e4cc8d33fb39f071bcc/step-2.png) ### Step 3 Connect Claude to the folder Click the “Work in a folder” button and select the folder you created. ![step_3_a](https://tru-images.b-cdn.net/guide-assets/69/69691e6cb9cd540b261b57b9/step-3-a.png) ### Step 4 Compress image copies with Claude Ask Claude to save copies of the images and reduce their file sizes. ```text Save copies of the images in this folder then reduce the file size of each by at least 50%. ``` ![step_4](https://tru-images.b-cdn.net/guide-assets/69/69691e76262f8d14b96ae443/step-4.png) You should see Claude editing the images using the terminal. The active task Claude Cowork is completing will show in the top right. ### Step 5 Queue parallel video and audio tasks Ask Claude to make a faster copy of the horse video. ```text Make a copy of the horse video that is 2x speed. ``` Before Claude finishes, add another prompt to the queue. ```text Extract the audio from the small steps video and save it as an mp3. ``` Claude should work through both tasks in parallel. ### Step 6 Explore Claude Cowork extensions Go to Settings → Extensions → Browse Extensions. The source recommends adding the Context7 extension to give Claude Cowork access to open source documentation. You can browse additional extensions to automate more of your Mac with Claude Cowork. A good starting point is extensions that say they were developed by Anthropic. ### Pro tip > When trying extensions, start with ones marked as developed by Anthropic. ### Guide: Close More Deals With Codex Sales Followups > When sales context is spread across meeting notes, Slack signals, CRM exports, and scattered follow-up tasks, it is easy to lose track of what needs attention first. - Canonical page: https://app.therundown.ai/guides/close-more-deals-with-codex-sales-followups - Machine-readable page: https://app.therundown.ai/guides/close-more-deals-with-codex-sales-followups.md - Difficulty: beginner - Featured tool or company: OpenAI - Categories: Sales, Consulting & Customer Success - Instructors: Billy Howell - Published: 2026-06-10 - Updated: 2026-07-30 ### Overview #### The Rundown When sales context is spread across meeting notes, Slack signals, CRM exports, and scattered follow-up tasks, it is easy to lose track of what needs attention first. In this guide, you will use a simple Codex system to send better sales follow-ups, faster. It helps you figure out which accounts need attention, draft the next email, clean up your CRM, and build a close plan so you can save time and keep more deals moving. The workflow stays review-first from the start. Codex reads local files, reasons across them, and creates drafts and review artifacts. It should not send customer emails, post Slack messages, or update a live CRM unless a human explicitly approves that step. #### Who This Is Useful For This is useful if you manage sales follow-up, account prioritization, customer notes, or CRM hygiene and feel like the useful context is always split across too many tools. It is especially helpful for: - reps who want a cleaner post-call follow-up process - founders or operators managing a small pipeline - account teams that need a weekly priority queue - managers who want CRM updates proposed, not silently applied #### What You Will Build You will build a local Codex workspace that acts like a sales follow-up operator. It reads account data, meeting notes, Slack-style signals, CRM field rules, and product notes, then creates the review files a rep needs before sending anything. ![Sales Follow Up Workspace](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/01-01-sales-follow-up-workspace-d15b02c9bfe4.png) The finished workspace gives you a single place to inspect the priority queue, review CRM changes, check the close plan, and open a local Gmail-style draft. In the demo, all of that happens in one local review surface instead of inside live sales tools. By the end, the workspace can produce: - a ranked follow-up priority queue - a follow-up packet for a specific account - a customer email draft - an internal Slack-style update - proposed CRM field updates - a seven-day close plan #### What You Need To Get Started - The Codex desktop app - A paid ChatGPT plan signed into Codex - A clean project folder - The attached starter template files from this guide's resources, or your own exported sales context - Account data from your CRM - Recent meeting notes from your note-taking app - Relevant Slack or team-channel signals - A clear rule that all customer-facing work starts as a draft #### Going Further Once the file-based workflow works, turn it into a reusable Codex skill: ```text Turn this sales follow-up workflow into a project skill. The skill should: - inspect account data, meeting notes, Slack signals, CRM field rules, product notes, and approval rules - rank follow-up opportunities - draft customer follow-up packets - propose CRM updates for review - create seven-day close plans - never send messages or mutate live CRM data without explicit approval Save the skill as skills/sales-followup-operator/SKILL.md. ``` You can also add relevant skills as references. For example, install the public outreach drafting skill: ```text npx skills add https://github.com/anthropics/knowledge-work-plugins --skill draft-outreach ``` Then ask Codex to inspect any local prospecting, Gmail, or Slack skills for patterns. Treat those as references first. Do not give Codex live customer data until the local review workflow is working. The recording also showed an optional dashboard prompt: ```text Make me a simple HTML website that visualizes this data and makes it interactive so I can check boxes and edit things. ``` Codex built a static index.html review surface with editable views for the priority queue, CRM review, close plan, and local Gmail-style draft. That is useful once the Markdown workflow is stable, but it is not required for the core setup. Start with the draft files. Get the review loop right. Then connect live tools and build a nicer interface only after the workflow is reliable. ### Step 1 Set Up Your Sales Follow-Up Workspace Start by downloading the starter folder from this guide's resources. Then create or open a clean blank folder for the workflow in Codex. In the demo, the folder is named codex-sales-followup-demo. The starter folder has this shape: ```text approval-rules.md AGENTS.md data/ accounts.csv crm-fields.md email-style-guide.md product-notes.md slack-signals.md meeting-notes/ acme-robotics-call.md prompts/ outputs/ ``` The data/ folder is where the raw context lives. The outputs/ folder is where Codex saves review artifacts. The approval-rules.md file tells Codex what it can and cannot do. ![Sales Data Folder](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/02-02-sales-data-folder-7c1301819717.png) For the demo, all accounts, companies, deal values, and notes are fake. For a real setup, replace the demo files with exports from your CRM, notes app, and Slack workspace. The attached template files are the starting point. Copy them into the blank project, then replace the sample content with your real exports or let Codex generate fresh versions from scratch if you do not have them yet. Keep the file names stable so Codex can find the expected inputs every time. > Pro tip: Keep approval-rules.md open while recording or running this for the first time. It gives Codex a source to cite when it refuses risky actions like sending email, changing CRM data, or inventing pricing. Your approval rules should say: ```text Codex may inspect files, rank accounts, draft messages, and propose CRM updates. Codex must not: - send customer emails - post Slack messages - update a live CRM - invent pricing, discounts, implementation dates, security claims, or product commitments - use real customer data in a demo - overwrite source data in data/ Every customer-facing message needs human review. Every CRM update must be returned as a proposed update table. ``` ### Step 2 Scaffold the Project in Codex If you are starting from scratch, open your empty project folder in Codex and have it create the workspace. If you already downloaded the attached template files, drop them into the blank folder first. Then open the folder in Codex, or use this prompt to have Codex scaffold the same structure for you. Use this prompt: ```text Create a sales follow-up workspace in this folder. Set up: - AGENTS.md - approval-rules.md - data/ - data/meeting-notes/ - prompts/ - outputs/ Create starter files for: - account data - CRM field rules - product notes - email style rules - Slack or team-channel signals - meeting notes The workflow should help me: - extract account details - rank accounts by follow-up priority - create customer follow-up packets - propose CRM updates for review - create seven-day close plans Important rules: - do not connect to live accounts yet - do not send emails - do not post Slack messages - do not update CRM records automatically - all outputs should be saved as Markdown files in outputs/ ``` Codex should create a folder structure similar to the demo and leave you with prompt files you can reuse. ![Prompt Files](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/03-03-prompt-files-af3f5e498051.png) > Pro tip: Start with exported files, then connect your note-taking app, Slack, and CRM after the local review flow works. That keeps setup simple and prevents Codex from touching live sales data too early. Under Plugins, connect your note-taking app, CRM if you have one, and anywhere you talk to sales prospects only after the local review loop is working. That way Codex can pull account data and propose updates before it touches live records. Once the local version works, connect your real tools carefully: - Connect your note-taking app so Codex can read meeting summaries. - Connect Slack or export account-team signals so Codex can see what changed. - Connect your CRM only after the review workflow is working locally. For the first real run, use exports or read-only access. Do not start by letting Codex write back to the CRM. ### Step 3 Connect Your Sales Context Before Codex can prioritize accounts, it needs a clean picture of the pipeline. Start with a one-time extract. Use this prompt: ```text Perform a one-time extract of account details from the available source files or connected sales tools. Create a normalized account summary with: - account name - owner - deal stage - deal value - forecast category - close date - champion - known blockers - latest interaction - next step - risk level - source evidence Save the result in data/accounts.csv or a Markdown table if CSV is not practical. Do not update the CRM. This is an extraction and cleanup pass only. ``` Then ask Codex to inspect the workspace before it drafts anything: ```text Inspect this sales follow-up workspace. Do not draft messages or propose CRM updates yet. Return: - which files you can read - what each file contributes - what context is missing - the approval rules you will follow - where you will save outputs ``` This step matters because it forces Codex to show its working area before it starts making recommendations. ### Step 4 Extract and Rank Accounts Now have Codex decide where follow-up matters most. In the recording, the prompt was intentionally plain: ```text Look at Slack signals and meeting notes and use them to prioritize our accounts and make a follow-up priority queue. ``` You can make that more reusable: ```text Use the files in data/ to build a follow-up priority queue. Rank each account by: - urgency - clarity of next action - risk if no one follows up this week - evidence quality - closeness to revenue For each account, include: - rank - account - deal stage - reason for priority - source evidence - next-best action - confidence - human review note Save the result as outputs/followup_queue.md. ``` ![Priority Queue Generated](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/04-04-priority-queue-generated-b69aabb16fbc.png) In the demo, Acme Robotics ranked first because it had a recent meeting, a security blocker, and a clear procurement deadline. Nimbus Finance moved up because it had the highest deal value, a commit forecast, and a fresh pricing signal. That is what makes this useful. Codex is not just writing a faster email. It is deciding which follow-up deserves attention and showing the evidence behind that recommendation. ### Step 5 Run the Post-Call Follow-Up Workflow Once you know which account needs attention, run the post-call workflow. In the recording, the live example used Nimbus Finance. The first prompt was: ```text Create a follow-up packet for Nimbus Finance. ``` Codex created outputs/followup_packet_nimbus_finance.md with: - an account summary - a customer email draft - an internal Slack update draft - proposed follow-up assets - claims and assumptions to verify - an approval checklist The next prompt was: ```text Okay, update the CRM entry for this account and then create a close plan. ``` Codex did not edit the source CRM data. It created review files: - outputs/crm_update_review_nimbus_finance.md - outputs/seven_day_close_plan_nimbus_finance.md That is the safer pattern. CRM changes should show current value, proposed value, source evidence, reason for update, risk, and an approve/reject field. The close plan should separate internal actions from customer-facing actions. After approving the CRM review and close plan, the recorded prompt asked Codex to draft the email locally: ```text CRM update approved. 7-day close plan approved. Draft any follow-up emails in my Gmail. Don't really do that in Gmail; this is a demo. ``` Codex created a local file instead: - outputs/gmail_draft_nimbus_finance.md ![Local Gmail Draft Review](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/05-05-local-gmail-draft-review-fec2e2a066f2.png) That file explicitly notes that no Gmail action was performed and no customer-facing send happened. In a live workflow, this is where a rep should review the email, check the assumptions, and decide whether to send. For a reusable post-call prompt, use: ```text I just finished a sales call for [ACCOUNT_NAME]. Read the latest meeting notes, account record, Slack or team-channel signals, CRM field rules, and product notes. Create: - a customer follow-up email draft - an internal account-team update - a list of claims or assumptions to verify - proposed CRM updates in a review table - a seven-day close plan Rules: - do not send email - do not post Slack messages - do not update the CRM - do not invent pricing, dates, discounts, security claims, or product commitments - save all outputs in outputs/ ``` ### Resources - [AGENTS.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/06-agents-a23ecef4a435.md) - [README.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/07-readme-6feab38b7a1f.md) - [approval-rules.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/08-approval-rules-c58d5da084a5.md) - [data-email-style-guide.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/09-data-email-style-guide-78a59fd8ab74.md) - [data-slack-signals.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/10-data-slack-signals-2611ae85884c.md) - [prompt-01-inspect-project.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/11-prompt-01-inspect-project-7573a8ffd6cf.md) - [prompt-02-rank-followups.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/12-prompt-02-rank-followups-705adcaf7cdc.md) - [prompt-03-create-followup-packet.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/13-prompt-03-create-followup-packet-186e68f44823.md) - [prompt-04-propose-crm-updates.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/14-prompt-04-propose-crm-updates-8cbaaef6ec76.md) - [prompt-05-close-plan-and-checklist.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/15-prompt-05-close-plan-and-checklist-4a1f01630ba9.md) - [skill-sales-followup-operator-SKILL.md](https://tru-images.b-cdn.net/guide-assets/close-more-deals-with-codex-sales-followups/16-skill-sales-followup-operator-skill-9a79d72cfba3.md) ### Guide: Code From Your Phone With OpenAI’s Codex > This guide teaches you how to set up OpenAI Codex in Cursor, run Codex in the cloud from your phone, and configure Codex to review your code automatically. - Canonical page: https://app.therundown.ai/guides/code-from-your-phone-with-openais-codex - Machine-readable page: https://app.therundown.ai/guides/code-from-your-phone-with-openais-codex.md - Difficulty: intermediate - Featured tool or company: OpenAI - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-01-16 - Updated: 2026-07-23 ### Overview #### The Rundown This guide teaches you how to set up OpenAI Codex in Cursor, run Codex in the cloud from your phone, and configure Codex to review your code automatically. You’ll produce a Codex-ready project setup and a workflow for local coding, cloud tasks, and pull request reviews. #### Who This Is Useful For - Developers who want AI coding help but find terminal-based tools intimidating. - ChatGPT Plus or Pro subscribers who have not tried Codex yet. - Anyone curious about AI coding agents. #### What You Will Build You’ll install Codex, connect it to a project, and create an AGENTS.md file that tells Codex about your project. You’ll also explore slash commands and set up automated code reviews. - A Codex IDE setup in Cursor or another supported editor. - An AGENTS.md project configuration file. - A cloud Codex workflow connected to GitHub. - Automatic Codex reviews for new pull requests. #### What You Need - A ChatGPT Plus account. - Cursor or another supported code editor, such as VS Code or Windsurf. - A GitHub account, optional for cloud workflows and pull requests. - An existing project, or a PRD or spec sheet if you are starting from scratch. #### Going Further - After setup, you can run Codex tasks from chatgpt.com/codex, create pull requests from Codex’s work, customize project-specific virtual environments, and use cloud-based Codex calls to preserve your 5-hour local limit. ### Step 1 Choose your Codex version - Codex CLI runs in your terminal. - Codex Cloud runs in your browser at chatgpt.com/codex. - Codex IDE Extension runs in your IDE as a chat window. For this guide, use the Cursor extension because it keeps your agent chat separate from your terminal. If you use VS Code or Windsurf, you can follow along as well. Go to https://developers.openai.com/codex/quickstart/ and select your IDE, then follow the brief download instructions. ### Step 2 Set up Codex in your project Open a project in Cursor. Press CMD+Shift+P, search for Codex, and select New Codex Agent. ![02_new_codex_agent](https://tru-images.b-cdn.net/guide-assets/69/696dacdbd8b7fb1874a96af1/02-new-codex-agent.png) You should see a Codex agent window open. ![02_new_agent](https://tru-images.b-cdn.net/guide-assets/69/696dace7d8b7fb1874a96b0e/02-new-agent.png) Ask the agent to create an AGENTS.md file from your project files. ```text Create an AGENTS.md file based on the project files. ``` If you’re starting from scratch, give Codex your PRD or spec sheet instead. After it creates AGENTS.md, ask it to initialize git and write a commit if you have not done that yet. ### Step 3 Explore slash commands Click the slash button to see available commands. ![03_slash_commands](https://tru-images.b-cdn.net/guide-assets/69/696dad0cd8b7fb1874a96b70/03-slash-commands.png) The most important commands are: - /review — Quick code review of your changes. - /status — Shows context limit percentage and your 5-hour usage limit. - /mcp — Shows connected MCP servers. - /context — Toggles auto-context from recent files. Use the official documentation for the full list of slash commands. ### Step 4 Use Codex from anywhere Go to https://chatgpt.com/codex/settings/connectors and connect your GitHub account. ![04_anywhere](https://tru-images.b-cdn.net/guide-assets/69/696dad334090654495857113/04-anywhere.png) This lets Codex access your repositories and create pull requests from its work. Tasks run in isolated cloud sandboxes, so your local machine stays free. To access Codex from anywhere, go to chatgpt.com/codex, select a repository and branch, and give it a prompt. Codex will run the code in a virtual environment. You can edit your project-specific virtual environments so Codex can use your API keys or browse the web. ### Step 5 Set up automatic code reviews Codex can automatically review every pull request in the cloud. - Go to https://chatgpt.com/codex/settings/code-review. - Enable automatic code reviews on new PRs. - Flip Auto code review to “Review my PRs” for your repo. Now whenever you open a pull request, Codex will automatically review it. ![05_code_reviews](https://tru-images.b-cdn.net/guide-assets/69/696dad3f409065449585712d/05-code-reviews.png) ### Step 6 Use the token-saving workflow Cloud-based Codex calls from the web or other connectors do not count against your 5-hour local limit. Use this workflow to save tokens: - Code locally using Codex in Cursor. - When you finish a feature, ask Codex to commit the changes to a new branch and submit a new pull request. - The cloud agent automatically reviews your PR. You can follow along in the GitHub app or website. ```text Commit these changes to a new branch and submit a new pull request. ``` ### Pro tip > Cloud-based Codex calls from the web or other connectors do not count against your 5-hour local limit, so you can code locally in Cursor and use the cloud agent for pull requests and reviews. ### Guide: Comparing ChatGPT Image and Nano Banana Pro Across 5 Use Cases > This guide shows you how to compare Google’s Nano Banana Pro with ChatGPT’s image generation model across a consistent set of use cases. - Canonical page: https://app.therundown.ai/guides/comparing-chatgpt-image-and-nano-banana-pro-across-5-use-cases - Machine-readable page: https://app.therundown.ai/guides/comparing-chatgpt-image-and-nano-banana-pro-across-5-use-cases.md - Difficulty: beginner - Featured tool or company: OpenAI - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2025-12-17 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to compare Google’s Nano Banana Pro with ChatGPT’s image generation model across a consistent set of use cases. You’ll build a reusable comparison matrix, run the same prompts through each model, and score the results across consistency, creativity, utility, and quality. The example test uses five use cases and generates 4 images per model. The source comparison also includes Google’s advanced image generation tool in AI Studio as an optional test. #### Who This Is Useful For - Business owners who want to automate asset creation - Graphic designers using AI for designs - Anyone who wants to generate better AI photos #### What You Will Build You’ll create a reusable model-comparison workflow and scoring matrix that can be used for future image model tests. - A set of image-generation use cases - Structured prompts for each use case - A scoring matrix for consistency, creativity, utility, quality, and overall rating - A repeatable process for comparing ChatGPT Image, Nano Banana Pro, and optionally AI Studio ![1_whatyouwillbuild](https://tru-images.b-cdn.net/guide-assets/69/69432dd51efec033e0ab18b5/1-whatyouwillbuild.png) #### What You Need - Gemini Pro account - Free or paid ChatGPT account - Optional: Google Workspace account with billing enabled for AI Studio - Optional: Notion template for your rating matrix #### Going Further - Try a few of the use cases with your own brand in both ChatGPT Image and Nano Banana Pro. - Use ChatGPT or Gemini to interview you about your brand and turn the answers into 1–3 image prompt variations. - Ask for variations in a 4x4 grid, then generate the best result by itself once you find one you like. ### Step 1: Choose use cases and testing rules Decide which image-generation use cases you want to compare. The example comparison uses five: - Logo for a taco truck - Website graphic for a couch manufacturer - Social post for an energy drink brand - Marketing materials for a travel agency - An arbitrary prompt for something that is easy to identify as fake Outline your testing rules before you generate anything. The example rules are: - Use the same prompt for each model. - Generate 4 images per model. - Grade each model from 1–5. - Score prompt consistency: how well it followed the prompt across variations. - Score creativity: whether the variations are noticeably different, subtly different, or near copies. - Score utility: whether one or more variations could be used for the actual use case. - Score quality: which images look the least like AI-generated “slop.” ### Step 2: Generate structured prompts Feed your use cases into Claude, ChatGPT, or Gemini and ask it to generate structured JSON prompts for each one. ```text Here’s my use cases: [use cases] Write me a JSON prompt for each of the 5 use cases. Invent a different, fun brand for each use case. For the arbitrary image, do something detailed and easy to see if it's fake, like a crowd of sports fans at an NBA game. Include instructions to generate 4 variations per prompt in a 4x4 grid. ``` If you want to test this on your own brand, ask the LLM to interview you about your brand and then write prompts for each use case. > To save time and tokens, ask the LLM to write a prompt that generates 4 variations in a 4x4 grid. ### Step 3: Create your scoring matrix Create a scoring matrix for your comparison. Use this structure: | Use Case | Model | Consistency | Creativity | Utility | Quality | Overall Rating | |---|---|---|---|---|---|---| | Use case 1 | Nano Banana (Gemini) | | | | | | | Use case 2 | ChatGPT Image | | | | | | | Use case 3 | Nano Banana Pro (Vertex) | | | | | | ### Step 4: Open the image tools - Navigate to Gemini. - Click the “Create image” button to add Nano Banana. ![2_nano_banana_input](https://tru-images.b-cdn.net/guide-assets/69/69432b931c08d91d71daa3b9/2-nano-banana-input.png) - Navigate to ChatGPT Images. - Optional: Open Nano Banana Pro in AI Studio. Click the “Playground” link on the left side to open new chats. ### Step 5: Generate and score the outputs Feed each use-case prompt into each tool and generate the images. Open a new chat for each use case so you clear the context between tests. Rate each output against the criteria in your scoring matrix. ### Resources - [Comparing ChatGPT Image and Nano Banana Pro Across FULL GUIDE](https://tru-images.b-cdn.net/guide-assets/69/69432d4e1efec033e0ab17cc/comparing-chatgpt-image-and-nano-banana-pro-across-full-guide.pdf) ### Pro tip > If you don’t have Gemini Pro, you can use Nano Banana Pro for free. Try opening an incognito browser if you have trouble. ### Guide: Conduct Better Stock Research With Perplexity Finance’s Free Features > In this guide, you will learn how to use Perplexity Finance inside Comet as an AI stock research hub. - Canonical page: https://app.therundown.ai/guides/conduct-better-stock-research-with-perplexity-finances-free-features - Machine-readable page: https://app.therundown.ai/guides/conduct-better-stock-research-with-perplexity-finances-free-features.md - Difficulty: beginner - Featured tool or company: Perplexity - Categories: Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-06-16 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Perplexity Finance inside Comet as an AI stock research hub. You'll learn how to quickly research any stock using AI without having to open a bunch of tabs or pay for a bunch of different tools. #### Who This Is For - Retail investors who want a better first pass before opening a dozen finance, news, filing, and analyst tabs. - Operators and researchers who track companies, sectors, competitors, or public market signals as part of their work. - Beginners who want Perplexity to explain what deserves attention without turning the tool into a buy-or-sell machine. #### What You Will Build You will build a repeatable stock research workflow inside Perplexity Finance: - Scan the dashboard for what changed. - Use the screener to find candidates. - Open a stock page and inspect the useful sections. - Ask Perplexity to explain the move, risks, sources, and next questions. - Save stocks, alerts, and follow-up research prompts for later. ![01 Perplexity Finance Dashboard](https://tru-images.b-cdn.net/guide-assets/local-241/conduct-better-stock-research-with-perplexity-finances-free-features/01-perplexity-finance-dashboard.png) The useful part is not that Perplexity replaces your judgment. It replaces the tab chaos at the beginning of research and helps you ask better questions faster. #### What You'll Need - Perplexity's Comet browser. - Access to Perplexity Finance from a Comet new tab. - One ticker, sector, or stock idea to research. - A willingness to verify source dates and important numbers before making any decision. #### Going Further Once the free workflow is working, turn it into a weekly research routine. Use this prompt: ```text Create a repeatable weekly research workflow for tracking [TICKER] and its peers. Every week, I want to review: - major price movement - important news - earnings or guidance changes - analyst or market sentiment shifts - competitor moves - risks that changed - questions to research next Turn this into a checklist I can reuse. ``` If you have access to Perplexity Spaces or automations, save the checklist there and reuse it for the same ticker or sector each week. You can also create separate routines for: - a personal watchlist - competitors in your market - public companies you sell to - defense, energy, or government-exposed stocks - earnings weeks - major macro events The repeatable pattern is simple: ```text 1. Scan Perplexity Finance for what changed. 2. Use the screener to find candidates. 3. Open one stock page. 4. Inspect price movement, stories, financials, earnings, holders, and analysis. 5. Ask Perplexity what deserves attention. 6. Click the sources and verify dates. 7. Save the next research questions before making any decision. ``` Use Perplexity Finance as the starting point for better questions, not the final authority on what to do with your money. ### Step 1 Open Perplexity Finance In Comet Open Comet and start a new tab. Click Finance. That opens the Perplexity Finance dashboard. The recording framed it as the AI-era version of Yahoo Finance: market overview, top movers, stock charts, watchlists, news, earnings, and screeners, with Perplexity sitting beside the data. Start with the dashboard instead of a blank chat. The top navigation gives you the main areas to click through: US Markets, Crypto, Earnings, Predictions, Screener, Politicians, Watchlist, Portfolio, and App Gallery. On the market overview, use the page to answer the first basic questions: - What is moving today? - Which sectors are up or down? - Are there earnings events worth checking? - Are prediction markets or macro signals relevant? - Which stocks deserve a second look? You can also click Earnings to see an earnings calendar. This is useful when you want to know which companies are reporting, what quarter the call belongs to, and whether there are quick bullet summaries available before you open an individual stock page. ![02 Earnings Calendar](https://tru-images.b-cdn.net/guide-assets/local-241/conduct-better-stock-research-with-perplexity-finances-free-features/02-earnings-calendar.png) The dashboard is cleaner than a lot of finance sites because it is less cluttered with ads and SEO-style market noise. Use it as the surface for the research, then use Perplexity to explain what you are seeing. ### Step 2 Use The Screener To Find Candidates Next, click Screener. This is one of the more useful free features because you can describe the kind of stock you are looking for in plain English. For example: ```text Find energy stocks with low PE ratios that recently outperformed earnings expectations. ``` The transcript showed an important caveat here: strict screener prompts can fail or return an error. When that happens, loosen the filter and try again. Use a broader version: ```text Find energy stocks with low PE ratios. ``` Perplexity can return a table with tickers and metrics, then you can click into a company from there. > Pro tip: Screeners are for finding candidates, not choosing winners. Use them to build a shortlist, then inspect each company before trusting the result. Once you have a few names, ask Perplexity to explain why those companies appeared: ```text I used this stock screener and got these companies: [TICKERS] Explain: - what they have in common - why they might have appeared in this screen - which one looks most different from the others - what I should check before doing deeper research Do not make a buy recommendation. ``` That keeps the workflow in research mode. You are asking Perplexity to organize the next questions, not make the decision for you. ### Step 3 Inspect The Stock Page Open one stock page from the screener or search for a ticker directly. ![03 Nvidia Stock Overview](https://tru-images.b-cdn.net/guide-assets/local-241/conduct-better-stock-research-with-perplexity-finances-free-features/03-nvidia-stock-overview.png) The chart is useful for orientation, but the transcript called out a real limitation: the chart can feel difficult to control, especially when zooming or switching display types. Do not over-invest in the chart if it slows you down. Move through the page and inspect the sections that explain the company. On the stock page, click through the page tabs in order: Overview, Financials, Earnings, Holders, Predictions, Historical Data, and Analysis. Start with the at-a-glance numbers: - Market cap - PE ratio - Recent price movement - Notable timeline events - Related stories - Bull case and bear case The timeline is useful because it can connect price moves to earnings, macro context, company statements, or other recent events. The related stories section is also useful because Perplexity tends to screen out some of the thin "stock is up today" articles and surface stories that are more tied to company operations. Then open Financials. Perplexity turns company filings into cleaner tables across recent years or quarters. Use this to inspect revenue, EPS, margins, or other core metrics before you ask for interpretation. Open Earnings next. This can show whether the company beat or missed revenue or EPS, along with highlights from the call transcript. > Pro tip: Be careful with earnings data, especially for smaller companies or non-US tickers. The transcript noted that some values can look strange because data is missing, denominated differently, or incomplete. Verify important figures before using them. ### Step 4 Check The Hidden Tabs Some of the most useful research sections are easy to miss. The Holders tab is one of the strongest examples. Click Holders, then choose Politicians to see politician trading disclosures for that stock. ![04 Holders Politicians Tab](https://tru-images.b-cdn.net/guide-assets/local-241/conduct-better-stock-research-with-perplexity-finances-free-features/04-holders-politicians-tab.png) This view is especially useful if you are researching sectors affected by government spending or regulation, such as defense, energy, aviation, or infrastructure. The same Holders page also lets you switch between Institutional, Insiders, and Politicians. The key is to understand the limitation. Congressional disclosures often report ranges, not exact trade amounts. Perplexity does a good job showing those ranges, but you still need to treat them as directional information. Also inspect Analysis. This can show analyst price targets, changes to targets, and the firms behind those calls. The transcript was careful here: analyst targets are not automatically useful, and some lists can be stale. The value is that Perplexity shows the target, the firm, and the context without turning it into a clickbait headline. For a quick first pass, the highest-signal sections are: - Recent price movement timeline. - Related stories and latest developments. - Bull case versus bear case. - Earnings. - Holders, especially politician trades when relevant. - Analyst price targets. That gives you enough context to ask better follow-up questions. ### Step 5 Ask Better Research Questions Do not start by asking: ```text Should I buy this stock? ``` That is the wrong job for this tool. Use Perplexity to understand what deserves attention, which sources support the explanation, and what you should inspect next. Use this prompt: ```text I am researching [TICKER] as a beginner-friendly first pass. Do not tell me whether to buy or sell it. Help me understand: - what the company does - what has changed recently - the main reasons investors are optimistic - the main risks investors are worried about - which numbers or events I should inspect next Use current sources where possible and separate facts from interpretation. ``` For a current move, use: ```text Explain the stock move today in plain English. What are the most likely drivers, and which sources support each one? ``` Perplexity will usually cite sources at the end of factual claims. Click them. This is not optional for timely stock research. In the recording, Perplexity cited a source from April when the prompt asked about today's stock movement. Some sources were current, but some were stale. That is exactly why you need to inspect the citations before trusting the answer. Use this follow-up when you want a cleaner next-step checklist: ```text What are the three biggest questions I should answer before I spend more time researching this company? For each question, tell me: - why it matters - which section of Perplexity Finance I should inspect - which source type I should verify outside Perplexity ``` The goal is not to get a trade recommendation. The goal is to leave with a focused research plan. ### Step 6 Use The Comet Assistant Without Losing The Dashboard One frustrating behavior from the recording: asking a research question from the stock page can open a new chat thread and pull you away from the finance screen. Comet's assistant is often better for this workflow because it can keep the dashboard open in the browser while answering in the side panel. Use it for questions like: ```text Find me stocks benefiting from macro oil trends. ``` Or: ```text Use my browser to find SEC filings from [COMPANY]. ``` When Comet asks for permission to use the browser, review what it is doing before approving. This is especially useful for SEC.gov and other government websites that are painful to search manually. You can also use the assistant beyond public stocks. The same browser-control pattern works for government contracting research, local business research, company filings, and other public databases. > Pro tip: Keep expensive agentic research for tasks that justify it. The transcript noted that Perplexity Computer can burn credits quickly, so normal stock research is usually better handled in chat, Finance, or the Comet assistant first. ### Step 7 Save Watchlists, Alerts, And Portfolio Context Perplexity Finance also lets you follow a stock. Click Follow on a stock page to add it to a watchlist. The watchlist is useful because it rolls up notable price movement and relevant news across the companies you care about. It also creates comparison graphs, which can make it easier to see how your watched stocks are moving relative to each other. There is a limitation here too. In the recording, asking Perplexity for the best-performing stock on the watchlist did not work cleanly. The dashboard understood the watchlist visually, but the AI answer did not seem to know the same list. Treat the watchlist as a dashboard feature first, then verify what the assistant can actually access. You can also set price targets for a stock. Perplexity can notify you in the app, by email, or by text, depending on your setup. It can also accept follow-up instructions for what to do after a target is triggered. If you want to connect actual portfolio data, Perplexity Finance supports brokerage and bank connections through Plaid. This can bring portfolio analysis into the same dashboard. Only do this if you are comfortable with the permissions. Read the fine print, verify the connection flow, and use view-only access where possible. ### Guide: Create a Custom Codex Pet (No Code Required) > In this guide, you will learn how to make a custom animated desktop pet like Clippy. - Canonical page: https://app.therundown.ai/guides/create-a-custom-codex-pet-no-code-required - Machine-readable page: https://app.therundown.ai/guides/create-a-custom-codex-pet-no-code-required.md - Difficulty: beginner - Featured tool or company: Codex - Categories: Content & Creative, Coding & App Development - Instructors: Billy Howell - Published: 2026-06-23 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to make a custom animated desktop pet like Clippy. It will sit on your screen and give you status updates from your Codex agent as it works. #### Who This Is Useful For - Codex users who want a lightweight way to monitor active tasks without keeping the whole app open. - People who like customizing their workspace, but do not want to hand-build animation files or sprite sheets. - Teams making repeatable Codex workflows who want a small visual reminder that an agent is running, waiting, or ready to review. #### What You Will Build You will create a custom animated Codex pet, add it to the Codex desktop app, and wake it as a floating status companion. ![custom codex pet](https://tru-images.b-cdn.net/guide-assets/local-286/create-a-custom-codex-pet-no-code-required/01-custom-codex-pet.png) Behind the scenes, the pet is more than one static image. The hatch-pet skill turns your mascot idea into the files Codex needs for different pet states and animations. ![pet sprite sheet overview](https://tru-images.b-cdn.net/guide-assets/local-286/create-a-custom-codex-pet-no-code-required/02-pet-sprite-sheet-overview.png) #### What You Need To Get Started - The Codex desktop app (https://developers.openai.com/codex/app) with the latest update - You could also do this in Codex terminal with /pet and @hatch-pet #### Going Further Once you have one pet working, turn the same process into a repeatable customization checklist: update the pet, tune the app theme, set your preferred Codex personality, and document project rules in AGENTS.md. You can also browse the Petdex GitHub repo (https://github.com/crafter-station/petdex) for user-created pet inspiration. Treat outside pet files like any other community-created asset: inspect what you are installing, avoid untrusted commands, and use the gallery for ideas when you are not sure what style you want. ### Step 1: Find The Pet Controls Open the Codex desktop app, check for updates, then go to Settings > Appearance > Pets. This is where you can choose one of the pre-made pets. Select one, and click Wake, then move back into another app so you can see how the pet sits on your desktop while Codex works. ![codex pets settings](https://tru-images.b-cdn.net/guide-assets/local-286/create-a-custom-codex-pet-no-code-required/03-codex-pets-settings.png) You can drag the pet around your desktop, and it will show you progress on open Codex threads and even let you reply to them. > Pro tip: Right-click the pet from anywhere to close it. You can also type /pet in Codex to bring it back without reopening settings. ### Step 2: Prototype The Mascot Before You Build The Pack Do not start by asking Codex to make the full pet pack. Full pet creation can take a long time because Codex generates multiple images, poses, and sprite sheets. Start by asking for a few lightweight concept options: ```text Generate four basic mascot images I could use with the Codex pet feature. Keep them simple, readable at small size, and transparent-background friendly. ``` Pick the mascot you like best. Make sure it's not too cluttered or filled with lots of text. Try to get Codex or ChatGPT to simplify it down; that is what we had to do. > Pro tip: We recommend doing this first because full pet creation takes a long time. It is cheaper to reject a rough concept than to wait through a full animated pack you do not like. ### Step 3: Create The Custom Pet With hatch-pet When you like a concept, tell Codex to run @hatch-pet on the mascot you like. For a clean first pass, use a prompt like this: ```text use @hatch-pet create a small red dragon pet for Codex. Make it friendly, simple, and readable at a small overlay size. Use a transparent background and do not use any logos, text, or real-person likenesses. ``` If the result gets too busy, ask Codex to simplify the mascot before you continue: ```text Simplify the pet into one clear mascot silhouette with a red body, small wings, tiny horns, and one subtle coding-tool detail. Keep the transparent background. Do not add text, branding, scenery, or extra characters. ``` While the skill runs, you should see Codex work through the pet pack. That may include the main look, poses, and sprite-sheet-style assets. ![hatch pet sprite generation](https://tru-images.b-cdn.net/guide-assets/local-286/create-a-custom-codex-pet-no-code-required/04-hatch-pet-sprite-generation.png) We had an instance where the mascot wasn't consistent between sprites, and I think that's just because it was overcomplicated and an abstract concept. If you pick an animal or a creature that's real, it should be more consistent. ### Step 4: Refresh, Select, And Wake The Pet When the pet is done, go back to Settings > Appearance > Pets and click refresh. Your new custom pet should appear in the picker. Select it, click Wake, and then test it while Codex runs a thread. Keep it visible when you want status updates, then close or tuck it away when you need a clean screen. If your new pet does not appear immediately, refresh once more. If it still does not show up, give Codex a moment to finish saving the generated files before you troubleshoot anything else. You can also use /pet from Codex when you want to bring the overlay back quickly. ### Guide: Create a Custom, Royalty-Free Jingle in 30 Seconds > Use Suno AI to create royalty-free brand audio from basic text prompts, including a jingle, background music, and sound effects. - Canonical page: https://app.therundown.ai/guides/create-a-custom-royalty-free-jingle-in-30-seconds - Machine-readable page: https://app.therundown.ai/guides/create-a-custom-royalty-free-jingle-in-30-seconds.md - Difficulty: beginner - Featured tool or company: Suno - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-02-17 - Updated: 2026-07-30 ### Overview #### The Rundown Use Suno AI to create royalty-free brand audio from basic text prompts, including a jingle, background music, and sound effects. Your first generated song takes about 20–30 seconds, then you can refine it into reusable variations. #### Who This Is Useful For - Podcasters who need an intro, outro, and transition sounds that sound like their brand - Content creators who want cohesive audio across YouTube, TikTok, and ads without licensing headaches - Small business owners who need hold music, ad background audio, or a jingle #### What You Will Build You will build a set of brand audio assets from text prompts in Suno. - A main jingle - A couple of variations with different instruments - Short audio stings for transitions and logo reveals #### What You Need - A Suno account; the free plan works, and Pro is $10/month - An idea of your brand’s vibe, such as genre, energy, and mood - CapCut, TikTok, or any video editor if you want to add the audio to video #### Going Further - Download your favorite sounds, then layer them into video projects, logo animations, ads, or podcast transitions. You can also pair them with AI-generated video, such as logo animations in Higgsfield or video ads with Gemini, and bring everything together in CapCut. ### Step 1: Generate Your First Jingle Go to suno.com and click Create. Describe your song in the prompt box. ```text Upbeat indie pop podcast intro with acoustic guitar and light percussion. Energetic but not overwhelming. Think tech podcast. ``` ![CleanShot 2026-02-17 at 13.58.23@2x](https://tru-images.b-cdn.net/guide-assets/69/6994e93c87f491e71039f49d/cleanshot-2026-02-17-at-13-58-23-2x.png) Click Create and Suno will generate your song. It will take 20–30 seconds and show up in the sidebar. Then you will see two variations appear in the feed. If you want to refine the song, the easiest move is to swap out instruments. Try changes like: - Add synth plucks - Switch acoustic guitar to electric guitar - Remove percussion Small changes to the instrument list can give you noticeably different results. ### Step 2: Remix Into Brand Variations Once you have a song or jingle you like, use it as the foundation for a cohesive sonic brand. Click the Remix and Edit button on your song, then select Cover. From there, try different instruments and genres in the prompt. Suno will regenerate the same melodic structure with different instruments, so it sounds different but remains recognizably related to the original. Use this to make variations of one jingle for multiple ads or formats, such as an orchestral version for a TikTok video, a synth-wave variation for a YouTube video, or hold music for background audio in a podcast. ### Step 3: Create Short Stings and Sound Effects Instead of creating a song, click into Sounds. Type a description of the sound effect you want. ```text Punchy, memorable audio sting. Works as a transition sound. Synth pluck. ``` You can choose between One Shot, which creates a single sound effect, and Loop, which creates audio that can repeat seamlessly. For stings and animated logo videos, choose One Shot. ![CleanShot 2026-02-17 at 13.57.44@2x](https://tru-images.b-cdn.net/guide-assets/69/6994e95a87f491e71039f4fd/cleanshot-2026-02-17-at-13-57-44-2x.png) Hit Create and Suno will generate a short audio clip. If you find one you like but it is not long enough, click the three dots, then Remix and Edit, then Extend. Suno will make it longer while keeping the same sound. ### Step 4: Organize and Download Your Sounds Like the sounds you want and add them to a playlist. When you are ready, go to your Library, find your favorites, and click Download. Drop the downloaded audio into CapCut, TikTok, or whatever editor you use. Layer the sounds over logo animations or video ads, or use them as podcast transitions. ### Pro tip > If your first prompt comes out too cheesy, try changing the brand descriptor at the end. “Morning coffee vibes” sounds very different from “tech podcast.” ### Guide: Create Ads and Marketing Videos in Minutes With Remotion > Learn how to programmatically generate ads and marketing videos with code. - Canonical page: https://app.therundown.ai/guides/create-ads-and-marketing-videos-in-minutes-with-remotion - Machine-readable page: https://app.therundown.ai/guides/create-ads-and-marketing-videos-in-minutes-with-remotion.md - Difficulty: advanced - Featured tool or company: Remotion - Categories: Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2026-01-25 - Updated: 2026-09-11 ### Overview #### The Rundown Learn how to programmatically generate ads and marketing videos with code. This workflow helps you conceptualize, template, and generate dozens of videos in minutes using just a few dollars of tokens. #### Who This Is Useful For - Marketers looking to test hooks and CTAs in their ads - Business owners who want to automate content production - Anyone who runs out of Sora tokens quickly #### What You Will Build You will build a reusable Remotion workflow for generating promotional video concepts, turning those concepts into code-based video components, and rendering/exporting finished marketing videos. - Promotional video and ad concepts for your brand - A Remotion project set up from a starter template - Video components generated from your concepts - At least one rendered and exported marketing video #### What You Need - A code editor, such as Cursor - Codex, Claude Code, or Cursor Agent - An AI chat tool, with NotebookLM recommended - Screenshots of your website and products - A link to your website - Optional: YouTube videos in the style of the video you want to create #### Going Further - After generating your first video, you can render additional videos in bulk, export your video as a GIF, or add royalty-free music and sound effects by dropping MP3 files into your project folder and asking Codex to add them with the Remotion skill. Going forward, add music and sound effect cues directly into your video scripts. ### Step 1: Generate ad and promotional video concepts Open your preferred AI chat tool. The source recommends NotebookLM. - Add screenshots of your website and products as sources. - Link to your website as an additional source. - Optionally, link to YouTube videos in the style of the video you want to create. ```text Generate 5 concepts for promotional videos and ads for my brand. Refine the concepts as needed. ``` ### Step 2: Create the Remotion implementation plan ```text Give me a prompt to instruct Codex on how to set up my repo and a plan to create all of those components and make those videos with Remotion. Use Remotion Codex to help with styling. ``` You can replace Codex with Claude Code or Cursor if you prefer. The Remotion Skill works with those too. ### Step 3: Create your Remotion project Open your terminal and run: npx create-video@latest When prompted, make your choices and name your project. This example uses the Hello World template. ![03_terminal](https://tru-images.b-cdn.net/guide-assets/69/69766c44be2cc35a5e26786b/03-terminal.png) At the end, it will prompt you to install the Remotion skill. Hit enter. Otherwise, add the skill manually with: npx skills add remotion-dev/skills Hit enter again to open the project in your code editor. ### Step 4: Install, launch, and generate the components Run these commands to install dependencies and launch the Remotion editor: npm install and npm run dev Drop the markdown instructions you generated into the root of your project. Start a new git repo with git init. This helps Codex with context. ```text Read the PRD and create all the components needed for the video concepts. ``` > Pro tip: If you want to include screenshots or videos of your website or product, drop those into the repo and ask Codex to organize them for you. ### Step 5: Create, preview, render, and export your video Tell Codex to create the video for the first concept. Make sure your server is running with npm run dev. Open http://localhost:3000 in your browser. You should see a video editor and your generated video. ![05_editor](https://tru-images.b-cdn.net/guide-assets/69/69766c520766d7eb32973e08/05-editor.png) Click the rocket ship next to the video title on the left to render and export the video. You can also ask Codex to bulk render videos using the npx remotion render [video name] command. > Pro tip: Click the format dropdown in the render menu to export your video as a GIF. ![05_render](https://tru-images.b-cdn.net/guide-assets/69/69766c5e0766d7eb32973e25/05-render.png) ### Pro tip > You can link to YouTube videos in the style of the video you want to create, or drop product screenshots and videos into your repo and ask Codex to organize them. ### Guide: Create an AI Assistant With Its Own Phone Number (Eleven Labs Walkthrough) > In this guide, you’ll build an AI personal assistant that you can call from your phone. - Canonical page: https://app.therundown.ai/guides/create-an-ai-assistant-with-its-own-phone-number-eleven-labs-walkthrough - Machine-readable page: https://app.therundown.ai/guides/create-an-ai-assistant-with-its-own-phone-number-eleven-labs-walkthrough.md - Difficulty: intermediate - Featured tool or company: ElevenLabs - Categories: Automation & Agents, Productivity & Knowledge Work, Sales, Consulting & Customer Success - Instructors: Billy Howell - Published: 2026-02-24 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you’ll build an AI personal assistant that you can call from your phone. It will know your calendar, your business, and any additional context you add to its knowledge base. The setup takes about 10 minutes and costs $0 to test. #### Who This Is Useful For - Busy founders or operators who want a hands-free way to check their schedule. - Consultants juggling multiple clients who need quick access to project information on the go. - Anyone who wants to experiment with voice AI agents without writing code. #### What You Will Build You’ll create a personal AI phone assistant powered by ElevenLabs that answers calls, uses your calendar and business context, and can later be extended with tools like webhooks and automations. - An ElevenLabs personal assistant agent. - A Twilio phone number connected to the agent. - A knowledge base with your calendar, business page, or other static context. - An optional path for adding live integrations later. #### What You Need - ElevenLabs account. The free plan works. - Twilio account. It is free to sign up and includes $15 in credits. - A calendar export or text file with your schedule. - Optional: a URL to your business or product page. #### Going Further - After the basic assistant is working, you can add tools and integrations for live calendar access, CRM lookups, advanced integrations, or webhooks connected to Make.com, n8n, or Zapier. With webhooks, your assistant can pull live calendar events, schedule meetings, check email, or look up client information from a phone call. ### Step 1 Create your agent in ElevenLabs Sign in to ElevenLabs and go to the Agents section. - Click New Agent. - Select Personal Assistant as the template. - Name it whatever you want. Pick a voice. Browse the voice library, filter by language, gender, and age, preview a few options, and choose one that fits. > Pro tip: If the default voice speed feels slow, bump it up slightly in the voice settings. Review the system prompt. The personal assistant template comes with a solid default. Update your time zone and customize the first message to your liking. Add to the knowledge base. This is what makes your assistant actually useful: - Paste a URL to your website or product page so the agent can scrape it. - Upload a text or markdown file of your calendar. You can ask Claude or ChatGPT to format your Google Calendar as markdown. - Add any other static information, such as client lists, project notes, or your services. Hit Publish when you’re done. ![image-2079](https://tru-images.b-cdn.net/guide-assets/69/699e38ad82f8553134c5b13a/image-2079.png) > Pro tip: The knowledge base is static, so it is best for things that do not change often, such as your offerings, brand info, and current projects. For dynamic data like live calendar events, use the webhook approach in the Going Further section. ### Step 2 Set up Twilio and get a phone number Go to twilio.com and create a new account. Twilio gives you $15 in free credits. A phone number costs about $1 to buy and $1/month to keep. ![image-2080](https://tru-images.b-cdn.net/guide-assets/69/699e38bf82f8553134c5b993/image-2080.png) - Click Phone Numbers → Buy a Number. - Pick one with voice capability. SMS requires extra paperwork and will not be same-day. - Copy down three things from your Twilio dashboard: the phone number you just bought, the Account SID from the main dashboard, and the Auth Token below the SID. ![image-2081](https://tru-images.b-cdn.net/guide-assets/69/699e38df82f8553134c5c253/image-2081.png) > Pro tip: During the free trial, callers will hear a short Twilio message before connecting. It goes away once you upgrade to pay-as-you-go, which is fractions of a cent per minute. ### Step 3 Connect Twilio to ElevenLabs Go back to ElevenLabs. In the sidebar, find Phone Numbers and click Import Number from Twilio. - Paste your Account SID. - Paste your Auth Token. - Paste your Twilio phone number. - Name the number and assign your agent to it. ![image-2082](https://tru-images.b-cdn.net/guide-assets/69/699e38f0428b6c8c05743f86/image-2082.png) That’s it. Your agent now has a real phone number. Call it and start talking. On a free Twilio account, it will make you hit the keypad to connect to your agent. If you upgrade, this will go away. ### Step 4 Extend your assistant with tools and integrations Once the basic assistant is working, you can make it much more powerful by adding tools. Here are the system tools you can install: ![image-2083](https://tru-images.b-cdn.net/guide-assets/69/699e3900428b6c8c057447f3/image-2083.png) You can also add integrations from your agent by going to Add Tool → Integration. Your options include: - Cal.com for live calendar access. - HubSpot for CRM lookups. - Custom MCP server for advanced integrations. - Webhooks that connect to Make.com, n8n, or Zapier. With webhooks, your assistant can pull live calendar events, schedule meetings, check your email, or look up client info, all triggered by a phone call. If you’ve taken any of the Rundown automation workshops, this is a natural next step. ### Pro tip > The knowledge base is static, so use it for information that does not change often. For dynamic data like live calendar events, use webhooks. ### Guide: Create an Animated Intro for Your Brand in 15 Minutes > Learn how to use Manus to generate a full set of branded video bumpers—an intro, transition, and outro—from a logo and a few multiple-choice answers. - Canonical page: https://app.therundown.ai/guides/create-an-intro-animation-for-your-brand-in-15-minutes - Machine-readable page: https://app.therundown.ai/guides/create-an-intro-animation-for-your-brand-in-15-minutes.md - Difficulty: beginner - Featured tool or company: Meta - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-03-15 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use Manus to generate a full set of branded video bumpers—an intro, transition, and outro—from a logo and a few multiple-choice answers. You will use Manus’s built-in video generation workflow to create three consistent video clips in one session, with the generation step taking about 15–20 minutes. #### Who This Is Useful For - Content creators and YouTubers who want polished intros and outros without paying a motion designer. - Marketing teams producing video content at scale who need branded assets that stay visually consistent. - Freelancers and agency owners who want to offer branded video bumpers to clients without the overhead of a full motion design workflow. #### What You Will Build ![image-2121](https://tru-images.b-cdn.net/guide-assets/69/69b70cea98cbbf8ba01a0d1e/image-2121.png) You will build a reusable set of three branded video clips: an intro, a transition, and an outro. The clips will share the same visual style, color treatment, and logo treatment so they look like they were made by the same designer and can be dropped directly into a video editor. - Branded intro bumper - Branded transition bumper - Branded outro bumper #### What You Need - A Manus account ($20/month, or start with the 7-day free trial). - Your brand logos or wordmarks as image files. - Optional: a video editor to assemble the final clips, though the bumpers can also be used as standalone files. #### Going Further - Turn the workflow into a reusable Manus skill with /skill-creator, making sure to clarify that it should work for different brands in the future. - Invite teammates into a Manus chat or share a project link so others can generate new bumpers from the same brand foundation. - Generate variations such as vertical 9:16 versions for Reels and Shorts, seasonal color variations, or versions for sub-brands and campaigns. ### Step 1: Start the task and use the interview method Open a new Manus task. Upload your logo or wordmark, then paste this prompt: ```text I want to create a branded intro bumper, transition bumper, and outro bumper video based on these logos. Interview me and ask me 3-5 multiple choice questions about the style and vibe I want before you start. Use the video generator skill. ``` The interview method helps you avoid writing a detailed creative brief upfront. Manus will ask targeted questions about vibe, animation style, color direction, and whether you want background music. - Ask for all three bumpers—intro, transition, and outro—in the first prompt so Manus can style them as a matched set. - Upload any brand assets you have, such as logos, wordmarks, or brand guidelines. Manus can use them directly or draw inspiration from them. > Pro tip: Don’t try to write out a detailed creative brief yourself. The interview approach gets you better results because Manus asks the right questions, and you just pick from options. ### Step 2: Pick your concept After the interview, Manus will present concept options for your bumpers. Before you commit to one, ask Manus to visualize them: ```text Can you visualize these so I can pick the one I like best? ``` Manus will generate preview images for each concept in the canvas on the right side of the screen. You can click items in the canvas to reference them in the chat, or command-click to select a specific element you want to call out. ![image-2120](https://tru-images.b-cdn.net/guide-assets/69/69b70ccebf1e83ae44968288/image-2120.png) Pick the concept that feels right and tell Manus to go with it. If none of the options work, describe what feels off and ask for new options. ### Step 3: Generate your videos Manus will generate your bumpers. Under the hood, it uses Nana Banana to create keyframes—the start and end frames of each clip—and Google VO3 to animate between them. It can also handle music mixing if you opted for background music. When Manus asks about default mode or quality mode, choose default. Quality mode uses roughly 8x the tokens, and the source did not find a meaningful difference in the output. - This step takes 15–20 minutes, so it is a walk-away-and-come-back task you can let run in the background. - Manus keeps the audio, video, and image tracks separate, which means you can reuse assets or swap things out later instead of being stuck with one burned-in video file. - If Manus gets stuck trying to download royalty-free music, tell it to skip the music. You can add your own track later in your video editor. > Pro tip: Manus sometimes has trouble with logos that are not widely known. If your logo is not rendering correctly, try uploading a cleaner version or telling Manus to use your brand name as text with your brand colors. ### Step 4: Review and download When the videos are ready, play them back in the canvas. If a clip does not look right, you do not need to start over. Tell Manus what to fix: ```text The intro looks great but the transition is too slow. Can you regenerate just the transition clip with faster movement? ``` Once you are happy with the clips, download them and drop them into your video editor as your intro, transition, and outro. ### Pro tip > Use the interview method instead of writing a detailed creative brief yourself. Manus asks targeted multiple-choice questions, and you choose the options that fit your brand. ### Guide: Create Branded Reaction GIFs for Your Company Slack > Learn how to make custom branded reaction GIFs for your company Slack using Higgsfield’s AI image and video generation. - Canonical page: https://app.therundown.ai/guides/create-branded-reaction-gifs-for-your-company-slack - Machine-readable page: https://app.therundown.ai/guides/create-branded-reaction-gifs-for-your-company-slack.md - Difficulty: beginner - Featured tool or company: Higgsfield AI - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-03-25 - Updated: 2026-07-23 ### Overview #### The Rundown Learn how to make custom branded reaction GIFs for your company Slack using Higgsfield’s AI image and video generation. You will create the starting frame first, then animate it into a short branded reaction clip that can be converted into a Slack-ready GIF. #### Who This Is Useful For - Small teams that want internal communication to feel more fun, custom, and on-brand - Founders, operators, or team leads looking for an easy morale boost and a small internal asset pack their team will use in Slack - Marketers and brand teams that want branded GIFs for internal use, with the option to upload them to GIF sites later #### What You Will Build ![image-2145](https://tru-images.b-cdn.net/guide-assets/69/69c4585f5c33dcc5ff1533e9/image-2145.png) You will build a few short branded reaction clips that you convert into GIFs for Slack, using Higgsfield image generation for the look and Higgsfield video for the motion. - A reusable set of branded reaction GIFs - Short MP4 clips generated from still images - Slack-ready GIF files converted from those MP4s #### What You Need - A Higgsfield account - A company logo or another brand reference image - An MP4-to-GIF converter such as CloudConvert, or another converter you already like - Slack, so you can upload and test the GIF at actual Slack size #### Going Further - After you create one good GIF, turn the idea into a small reaction pack, batch-convert MP4s to GIFs if needed, and reuse the best branded reactions in GIF libraries or social replies. - celebration - approval - hype - slow down - this is on fire ### Step 1 Gather a brand reference image If your company is not widely recognizable, gather a logo, wordmark, cover image, or another brand asset that shows your colors, font, and logo before you start generating. This can also help watermark your GIFs with your company’s logo. Here is the reference asset used in the source guide: ![image-2146](https://tru-images.b-cdn.net/guide-assets/69/69c4586a72e418139028a42c/image-2146.png) Pro tip: If you are working with a well-known brand, you can usually skip this because the model may already have a strong visual association. If not, use the logo. ### Step 2 Create the starting frame in Higgsfield Image Go to Higgsfield Image first. At this stage, your job is not to figure out the GIF’s motion. Your job is to figure out the look. If the still image already looks right, the video step becomes much easier. You are giving the model a strong visual starting point instead of asking it to invent the whole thing at once. Pro tip: Do not go straight to video generation. Images are cheaper to iterate on, and they give the video model a much better place to start. ### Step 3 Prompt for the reaction style and text Keep the image prompt simple and direct. Use a recognizable style anchor plus the reaction text you want. ```text ESPN themed reaction gif with words "SLOW DOWN" ``` ```text “LFG” reaction gif in the style of The Rundown ``` ![image-2147](https://tru-images.b-cdn.net/guide-assets/69/69c4587c5c33dcc5ff153c9c/image-2147.png) You do not need to over-describe every detail. The goal is to give Higgsfield a recognizable style anchor and the reaction text you want. Pro tip: Change the dimensions here to match the final video dimensions you want. ### Step 4 Animate the still in Higgsfield Video Generate a few stills and choose the one that feels most usable. Then click the camera Animate button on that image to move it into Higgsfield Video. Set the video length to 3 seconds and turn audio generation off to save tokens. Since you are making a Slack GIF, the audio is not needed. ```text Reaction GIF ``` That is all you need. The still image already contains the brand styling, layout, and mood. Next, the video model will add animation. ![hf_20260325_204401_d06e0fb2-7439-4123-839e-fe894a9e6343-ezgif.com-resize-1](https://tru-images.b-cdn.net/guide-assets/69/69c45a515c33dcc5ff15483b/hf-20260325-204401-d06e0fb2-7439-4123-839e-fe894a9e6343-ezgif-com-resize-1.gif) Pro tip: Use the default video model first. It is usually the best baseline. If you want to optimize for cost or test a different motion style, experiment with other models after you already have one good result. ### Step 5 Convert the MP4 to a GIF and test it in Slack Download the best MP4 and run it through CloudConvert, or use any MP4-to-GIF converter you already like. Then upload it to Slack and check how it looks at actual Slack size. Keep the reaction readable and visually simple. The best Slack GIFs are obvious at a glance. If the first version looks muddy or hard to read in Slack, go back and simplify the still instead of trying to fix everything in the converter. ### Step 6 Expand the GIF into a reusable reaction pack Once you get one good GIF, make a small pack instead of stopping at one. - celebration - approval - hype - slow down - this is on fire If you make a whole batch of MP4s, ask Claude Code to convert them to GIFs in bulk on your desktop so you do not have to use a converter site one file at a time. If you want to push this beyond internal use, upload the best branded reactions to GIF libraries or use them in social replies so the assets can work as lightweight guerrilla marketing too. ### Pro tip > Do not go straight to video generation. Images are cheaper to iterate on, and they give the video model a much better place to start. ### Guide: Create Killer Thumbnails With This Midjourney Technique > Learn a quick prompting and editing system for making thumbnail images that stand out without looking like AI slop. - Canonical page: https://app.therundown.ai/guides/create-killer-thumbnail-images-with-this-midjourney-technique - Machine-readable page: https://app.therundown.ai/guides/create-killer-thumbnail-images-with-this-midjourney-technique.md - Difficulty: beginner - Featured tool or company: Midjourney - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-03-04 - Updated: 2026-07-30 ### Overview #### The Rundown Learn a quick prompting and editing system for making thumbnail images that stand out without looking like AI slop. You will turn a content title into a Midjourney-generated thumbnail background, then finish it with clean text in Canva, Photoshop, or Figma. The source workflow is designed to get you from title to finished thumbnail in about two minutes. #### Who This Is Useful For - YouTube creators and podcasters who need a new thumbnail every week without hiring a designer. - Bloggers and newsletter writers who need hero images and are tired of stock photos. - Marketing teams and social managers producing thumbnails at scale without sacrificing quality. #### What You Will Build ![image-2096](https://tru-images.b-cdn.net/guide-assets/69/69a7a0d6a78dd03b576eb7d5/image-2096.png) You will build a repeatable thumbnail production system: generate thumbnail backgrounds in Midjourney, then edit them with strong, readable text in Canva, Photoshop, or Figma. - Midjourney thumbnail background prompts based on your content title. - A finished 16:9 thumbnail image with negative space for text. - A reusable text-and-depth editing approach for future thumbnails. #### What You Need - A Midjourney account. The source notes that the $10/month starter plan works. - Claude or another AI chat assistant, such as ChatGPT or Gemini, for generating prompt concepts. - Canva, Photoshop, or Figma. Canva Free works, while Canva Pro is needed for the background remover trick. - A content piece that needs a thumbnail, such as a YouTube video, blog post, or social post. #### Going Further - Build a style library in Midjourney. Drag a thumbnail you like into the Style Reference slot and pin it with the lock icon so new generations match that visual style. This is useful for keeping a YouTube series or blog visually consistent. ![image-2098](https://tru-images.b-cdn.net/guide-assets/69/69a7a10aa78dd03b576ec933/image-2098.png) ### Step 1: Get AI to generate prompt concepts Open Claude, ChatGPT, Gemini, or another AI chat assistant and start a new chat. Paste this prompt, replacing the title placeholder with your content title. ```text Use the following formula and article title to create four prompt concepts I can put into Midjourney to generate a thumbnail I can add text to later. Make sure there's negative space for the text. Formula: [Person + Expression] + [Action/Prop] + [Setting] + [Lighting] + [Composition with negative space] --ar 16:9 Title: [paste your title here] ``` You should get four scene concepts based on your title. Each concept should create a real, physical moment rather than an abstract AI-looking background. - If you do not want a person in the thumbnail, remove “Person + Expression” from the formula and ask for concepts without people. - You can also ask for concepts without the formula. This can produce more creative ideas, and sometimes those unstructured ideas work better. - If the concepts are not focusing on the right thing, tell the assistant what the main visual hook should be. For example: “The key concept is translation between languages.” ### Step 2: Generate the thumbnail background in Midjourney Go to the Midjourney web app and paste your first prompt into the imagine bar. - Load all four prompts from Claude so you have options. Some will miss, and some may surprise you. - When you see a generation you like, click it and try Vary Subtle for small tweaks or Vary Strong for bigger changes. - Keep going until you land on an image that feels right, then download it to your desktop. Do not force a weak concept. If none of the four concepts are clicking, go back to Claude and ask for four more. > Pro tip: Click the Draft Mode lightning bolt in the top right of Midjourney to save tokens on generations. ### Step 3: Add simple, readable text in Canva Open Canva and search for a YouTube Thumbnail template, or create a blank design at 1280×720. Drag your Midjourney image into the design and make it full size. Keep the text simple. Add 1 to 5 words of big, attention-grabbing text. This text should be different from your actual title; use an action, an outcome, or a reaction instead. For a video titled “Translate Any Video File Using This Free Local AI Model,” the thumbnail text might be “MAKE GLOBAL CONTENT,” “ANY LANGUAGE,” or “FREE TOOL.” ![image-2097](https://tru-images.b-cdn.net/guide-assets/69/69a7a0f3a78dd03b576ec05d/image-2097.png) - Pick a big, blocky font. - Use white or yellow text. - Add a black outline for contrast. - Bring the opacity down about 5% to blend it with the background. - If readability is still an issue, put a darker color box behind the text and turn the box transparency way down. > Pro tip: If the text does not fit because there is not enough empty space, go back to Claude and ask for a new Midjourney prompt with more negative space where you want the text. This is easier than trying to force the layout afterward. ### Step 4: Add depth with the background remover trick Use this layering move to make the thumbnail look more designed and add depth. - Take your Midjourney background image in Canva and duplicate it. - Put the duplicate directly on top of the original. - Use Background Removal on the top layer. This is a Canva Pro feature. - Arrange the design so you have three layers: the original image on the bottom, text in the middle, and the foreground cutout on top. This makes the text look like it is sitting between the foreground and background, which adds depth and makes the thumbnail feel more intentionally designed. - This works especially well when the image has a person in it. Put the text behind their head. - Do not use this on every image, and do not cover too much of the lettering. - The goal is to make the viewer’s brain work a little harder when it sees the thumbnail. That curiosity helps earn the click. ### Pro tip > Use Midjourney Draft Mode to save tokens on generations. If your text does not fit, ask your AI assistant for a new Midjourney prompt with more negative space where you want the text. ### Guide: Create Winning Ads With Claude in One Command > In this guide, you will learn how to use Claude to scan your website, extract your brand, and generate ad creatives. - Canonical page: https://app.therundown.ai/guides/create-winning-ads-with-claude-in-one-command - Machine-readable page: https://app.therundown.ai/guides/create-winning-ads-with-claude-in-one-command.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2026-06-30 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to use Claude to scan your website, extract your brand, and generate ad creatives. The Goose Ads skill gives Claude the ability to do this. You will learn how to install it, set it up, and use the free version. #### Who This Is Useful For - Founders and marketers who want quick ad creative drafts from an existing website. - Small teams that have a website, product page, or offer page and need faster creative ideas. - Anyone testing paid social who wants Claude to help with brand setup, template selection, credit estimates, and review links. #### What You Will Build You will set up Goose Ads inside Claude Code and use it to create a small batch of ad creative drafts from a website. By the end, you should have a GooseWorks brand project, extracted brand details, selected ad templates, a credit estimate, a brand gallery link, and individual creative links you can approve or reject. ![Generated Goose Ads creative preview with batch candidates](https://tru-images.b-cdn.net/guide-assets/local-290/create-winning-ads-with-claude-in-one-command/06-goose-ads-generated-preview.png) #### What You Need To Get Started - A Claude account with Claude Code available. - A GooseWorks account. - A website for the brand you want to scan. - Logos, colors, and a product or offer page if GooseWorks does not pull everything automatically. - Enough GooseWorks credits for a small test batch. > Pro tip: You can do this workflow in Claude Code inside the Claude desktop app. If you are on a Mac, install Claude from the Claude download page (https://claude.com/download), then open the Code tab. #### Going Further For the most cost-effective setup, use GooseWorks to generate draft ads set to low quality, then refine the best concepts with ChatGPT image tools or Nano Banana. This uses GooseWorks for the part it is best at: scanning your brand, finding templates, and creating fast creative directions. Then you can use another image tool to clean up the final asset, fix text, improve the logo, or resize the ad. Once you find templates that work, save them. You can ask Claude to search for new templates each week, estimate credits, and wait for approval before generating anything. Use this prompt: ```text Find 5 Goose Ads templates that could work for [brand/product]. Return the template links, why each fits, likely offer angle, brand risks, and estimated credits for a small low-quality batch. Do not submit until I approve. ``` The workflow to keep is simple: install GooseWorks, start Claude Code, run /goose-ads, check the brand kit, approve templates and credits, generate one small batch, then refine the best ads before launch. ### Step 1 Install GooseWorks In Terminal Open a new Terminal window. You should see a blank window where you can paste commands. > Pro tip: If you are using the Claude desktop app, open the Code tab and ask Claude Code to run npx gooseworks install --all for you. If you are following the exact recorded setup, use Terminal. Paste this command and press Return: ```text npx gooseworks install --all ``` ![GooseWorks CLI install terminal](https://tru-images.b-cdn.net/guide-assets/local-290/create-winning-ads-with-claude-in-one-command/02-gooseworks-cli-install-terminal.png) Follow any login or approval steps. This installs the Goose Ads skill so Claude Code can use GooseWorks to create brands, search templates, estimate credits, and generate ads. The GooseWorks CLI docs (https://docs.gooseworks.ai/concepts/cli) describe the CLI as the installer for GooseWorks skills and MCP servers. ### Step 2 Start Claude Code And Run Goose Ads After the installer finishes, stay in Terminal. Type: ```text claude ``` Press Return to start a new Claude Code thread. Then run: ```text /goose-ads create ads for my brand [brand name] [link to your website] ``` For the demo, the command was: ```text /goose-ads create ads for my brand The Rundown University https://www.rundown.ai/ai-university ``` ![Goose Ads command entry](https://tru-images.b-cdn.net/guide-assets/local-290/create-winning-ads-with-claude-in-one-command/03-goose-ads-command-entry.png) Claude will create or find the brand in GooseWorks and start scanning the site. The first scan can take a few minutes. If Claude says research is pending, let it keep working. Do not run the same command again. ### Step 3 Review The Brand Project Claude should send you a Goose Ads project link. Open it to see the brand information GooseWorks is extracting from your site. Check the basics: - Brand name - Logo - Colors - Product or offer - Audience - Pricing or trial details - Call to action If GooseWorks misses something, add it before generating ads. In the demo, we added logos and a product page so GooseWorks could extract the Rundown University offer. ![Goose Ads brand kit context fields extracted from the site](https://tru-images.b-cdn.net/guide-assets/local-290/create-winning-ads-with-claude-in-one-command/02-goose-ads-brand-kit-context.png) Before you generate, ask Claude to summarize the setup: ```text Read the brand kit. Summarize assets, product_name, offer, audience, pricing/trial details, CTA, and open questions. Pick 3 templates, estimate credits, and wait for approval. ``` This gives you one checkpoint before credits are used. ### Step 4 Pick Templates And Approve Credits Goose Ads has a library of ad templates. Use that library for direction, then let Claude pick a few templates that fit your brand. ![Goose Ads template library with example ad formats](https://tru-images.b-cdn.net/guide-assets/local-290/create-winning-ads-with-claude-in-one-command/03-goose-ads-template-library.png) Ask for a small batch first. Three templates is enough to test the workflow. Claude should show: - The templates it picked - Why each one fits - The product name it will use - The estimated credits - Whether it is waiting for your approval ![Claude approval gate before submitting a Goose Ads batch](https://tru-images.b-cdn.net/guide-assets/local-290/create-winning-ads-with-claude-in-one-command/04-goose-ads-approval-gate.png) ![Goose Ads credit estimate comparison before generation](https://tru-images.b-cdn.net/guide-assets/local-290/create-winning-ads-with-claude-in-one-command/05-goose-ads-credit-estimate.png) Do not skip the estimate. In the recorded demo, Claude estimated 180 credits for three ads. Treat that as a demo number. The final cost can change based on model, quality, settings, and GooseWorks pricing. > Pro tip: GooseWorks offers free credits, but pricing and credit rules can change. Check GooseWorks pricing (https://gooseworks.ai/#pricing), credits and billing (https://docs.gooseworks.ai/help-center/credits-and-billing), and credit usage (https://docs.gooseworks.ai/help-center/how-credits-are-used) before quoting exact numbers. ### Step 5 Generate And Review The Ads After the templates and estimate look right, approve one small batch. Claude should return: - A brand gallery link - Individual creative links - Batch status - Pending, completed, or failed counts ![Goose Ads brand gallery with example generated ad candidates](https://tru-images.b-cdn.net/guide-assets/local-290/create-winning-ads-with-claude-in-one-command/01-goose-ads-batch-gallery.png) If the batch is queued, wait. Re-running too quickly can reserve credits twice. Once the drafts are ready, ask Claude to review them: ```text Review the generated ads before launch. Return a table with: creative link, template, strongest angle, offer accuracy, brand fit, unsupported claims, edits needed, and whether I should approve, revise, regenerate, or reject it. ``` Then do your own review. Check whether the ad matches the brand, describes the offer accurately, uses the right logo, keeps the text readable, and avoids unsupported claims. Some drafts will need cleanup. That is fine. The goal is to get useful ad directions faster, then polish the strongest ones before launch. ### Guide: Create Your First Agentic Workflow in Google Workspace Studio > In this guide, you will create an agentic workflow in Google Workspace Studio that reads Google Form submissions, tracks them in a spreadsheet, and emails you action items. - Canonical page: https://app.therundown.ai/guides/create-your-first-agentic-workflow-in-google-workspace-studio - Machine-readable page: https://app.therundown.ai/guides/create-your-first-agentic-workflow-in-google-workspace-studio.md - Difficulty: beginner - Featured tool or company: Google - Categories: Automation & Agents, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-03-11 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will create an agentic workflow in Google Workspace Studio that reads Google Form submissions, tracks them in a spreadsheet, and emails you action items. You will use Google’s AI automation tool, Workspace Studio, which comes preconfigured to work with Gmail, Google Drive, and other Google Workspace tools. When you test the finished flow, it can take up to 10 minutes to fire, though it is usually closer to 5. #### Who This Is Useful For - Agency owners and freelancers who get inbound leads through forms and want to auto-qualify them before spending time on a call. - Ops and support teams who handle internal requests, such as IT tickets, onboarding, or troubleshooting, and want Gemini to triage them automatically. - Anyone on Google Workspace who wants to automate a repeatable process without connecting external tools like Zapier or Make. #### What You Will Build You will build a Google Workspace Studio flow that fires when someone submits a Google Form. Gemini reads the response, summarizes it, logs the summary in a Google Sheet, decides whether the submission meets your criteria, and emails you if it does. The whole flow runs on autopilot. - A Google Form trigger for new submissions. - A Gemini summary step for each form response. - An optional Google Sheet log with the submitter details, AI summary, and response link. - A decision step that qualifies or rejects the submission. - An email notification for qualified submissions. ![image-2110](https://tru-images.b-cdn.net/guide-assets/69/69b1f52374030095a92b61ae/image-2110.png) #### What You Need - A Google Workspace account or Gmail. - Access to Workspace Studio. - A Google Form, either existing or new. - A Google Sheet, optional, for tracking responses. #### Going Further - Client onboarding: reuse the same flow structure and add steps that create a new folder in Google Drive, generate a kickoff document, or draft a welcome email to the client. - Internal troubleshooting: replace the lead qualification logic with triage logic so Gemini can categorize support requests by urgency and route them to the right person by email. - Draft responses: instead of only notifying you, have Gemini draft a reply to the form submitter that you can review before sending, or send automatically if you trust the output. ### Step 1 Create your Google Form Create a Google Form for the process you want to automate. This guide uses a web design agency new-client inquiry form as the example, but the same setup works for inbound requests like support tickets, client onboarding, project kickoffs, and internal troubleshooting. ![image-2111](https://tru-images.b-cdn.net/guide-assets/69/69b1f53374030095a92b6a27/image-2111.png) Make sure the form is named and saved. You will connect it to Workspace Studio in the next step. ### Step 2 Create a new flow in Workspace Studio Go to studio.workspace.google.com and click the plus button to start a new flow. - Select “When a form response comes in” as your starter. This is the automation trigger. - Choose the form you just created. - Leave “Also run when a form response is edited” unchecked. ### Step 3 Add the AI summary step Click Add step and select Summarize > Content from previous steps > Variable > Form response. This feeds the entire form submission to Gemini. In the prompt box, give Gemini context about what this form is for. For example: ```text Read this form response and summarize the potential engagement. We are a web design agency that specializes in homepage design and e-commerce design. Our average project budget is about $4,500. ``` The more context you give Gemini here, the better the summary will be. Tell it what your business does, what a good submission looks like, and what matters to you. > Pro tip: If you use an “Ask Gemini” step, you can let Gemini use context from your workspace to answer questions. ### Step 4 Log responses in a Google Sheet Add a new step: Sheets > Add a row. Select a Google Sheet you have set up for tracking. - Pull in the submitter’s name and email using form variables. Click into the field and select the variable from the form node. - Add the AI summary as its own column. - Add a column for the form response link so you can always go back to the raw data. Use the “Link to form response” variable. > Pro tip: Google Forms already has a built-in “Link to Sheets” feature, but this custom sheet is cleaner. You get just the fields you care about plus the AI summary, instead of every raw form field. You can also add the link to the full form response with a variable. ### Step 5 Add the decision step Add a Decide step. This is where Gemini reads the summary and makes a call. Paste a prompt like: ```text Read this inbound lead summary and decide whether this lead is qualified based on the fact that we are a web design agency that specializes in homepage design and e-commerce design. Our average project budget is about $4,500. ``` You can skip the summary step entirely and put the raw form response here instead. But having the summary first means you get a clean version in your sheet and in your email. The Decide step creates two branches: If true and If false. You will set up what happens in each. ![image-2112](https://tru-images.b-cdn.net/guide-assets/69/69b1f54574030095a92b72bb/image-2112.png) ### Step 6 Set up the email notification Inside the If true branch, click Add substep > Notify me by email. Build out the email: - Subject line: something like “New Qualified Lead” with the submitter’s name or company as a variable. - Body: paste in the AI summary variable, plus the submitter’s name and email. - Include a link to the form response so you can verify the raw data. If you set up the tracking sheet, you can also add an Update row step in each branch to mark the submission as “Qualified” or “Rejected.” Match the row by the email column. It can be tricky to add static values here. You may need to put “Qualified” and “Rejected” in two existing rows, then select them from the dropdown. ![image-2113](https://tru-images.b-cdn.net/guide-assets/69/69b1f55774030095a92b7b49/image-2113.png) > Pro tip: Add the same Update row step in the If false branch too, marking those as “Rejected.” That way your sheet always shows the status of every submission without you checking manually. ### Step 7 Turn it on and test Name your flow and hit Turn on. Submit a test response through your form. - It can take up to 10 minutes for the flow to fire, but it is usually closer to 5. - You can also run a test from inside Workspace Studio by clicking Test run and selecting a previous form response. - Check the Activity tab in Workspace Studio to see exactly what happened at each step: what Gemini summarized, what it decided, and whether the email sent. ![image-2114](https://tru-images.b-cdn.net/guide-assets/69/69b1f56838ff76f8308a5948/image-2114.png) If all goes well, you should get an email from yourself. You can click the Activity button back in your flow to see the details of each run. > Pro tip: You can click the “Test run” button to re-trigger any previous form submissions as a test. ### Pro tip > You can click the “Test run” button to re-trigger any previous form submissions as a test. ### Guide: Cut Onboarding Time in Half With Loom and ChatGPT > In this guide, you will learn how to turn a short Loom walkthrough into a clear onboarding SOP with ChatGPT. - Canonical page: https://app.therundown.ai/guides/cut-onboarding-time-in-half-with-loom-and-chatgpt - Machine-readable page: https://app.therundown.ai/guides/cut-onboarding-time-in-half-with-loom-and-chatgpt.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: AI, Productivity - Instructors: Billy Howell - Published: 2026-08-09 - Updated: 2026-08-09 ### Overview #### The Rundown In this guide, you will learn how to turn a short Loom walkthrough into a clear onboarding SOP with ChatGPT. You will end with a Markdown SOP you can paste into Google Docs for the next person to use. #### Who This Is Useful For - Operations leads who need a dependable way to document recurring team processes before a new hire needs them. - Consultants and client-service teams who want to turn a recorded handoff into a clear procedure for a specific client. - Anyone who repeats a task such as submitting timesheets, writing reports, or sending client emails and wants a reusable explanation of it. #### What You Will Build You will turn one short screen recording into an SOP a teammate can follow: a purpose, links to the resources they need, the steps in order, and a checklist that shows when the work is done. The weekly-timesheet example below is the model; use the same process for your own repeatable work. ![Private SOP handoff](https://tru-images.b-cdn.net/guide-assets/local-75/cut-onboarding-time-in-half-with-loom-and-chatgpt/01-private-sop-handoff.png) #### What You Need To Get Started - A Loom walkthrough of one repeatable task. - The Loom transcript, plus links or screenshots for details that only appear on screen. - ChatGPT. - Google Docs for the final SOP. #### Going Further This is really important and helps you stay organized: build a Google Sheet index for finished SOPs by team or process, so teammates can find the right handoff quickly. Add one row per SOP with its title, owner, team, process, last-reviewed date, and a link to the Google Doc or Notion page. Start with the tasks people ask about most often. Then update the recording and the written SOP together whenever the process changes. The goal is not to document everything at once; it is to make the next repeated handoff easier to find and follow. ### Step 1 Record a Task Someone Else Needs to Repeat Choose a task you do often and record yourself doing it in Loom. Good starting points include filling out a timesheet, writing a report, or sending an email to a specific client. Explain the clicks, the decisions you make along the way, any exception that should stop the process, and what a finished result looks like. In the video, the task is logging weekly hours in Google Sheets. The walkthrough shows the hours entered, the weekly total, the READY TO FILE status, and the next action: make a copy and save it in the completed-timesheets folder. Those details turn a screen recording into something another person can actually follow. ![Source timesheet workflow](https://tru-images.b-cdn.net/guide-assets/local-75/cut-onboarding-time-in-half-with-loom-and-chatgpt/02-source-timesheet-workflow.png) ### Step 2 Clean Up the Transcript When you finish recording, use Loom's transcript editor to remove mistakes or long silences. You do not need a perfect script. You do need a transcript that reflects the useful path through the task. > Pro tip: Clean up mistakes and long pauses before you hand the transcript to ChatGPT. A cleaner source produces a cleaner SOP. ### Step 3 Copy the Transcript Into ChatGPT Copy the cleaned Loom transcript into ChatGPT. Add the links, screenshots, or short visual notes a new teammate needs, especially when a value, file name, destination folder, or status is only visible on screen. Then tell it to write the SOP in Markdown. Keep the request focused on the process you recorded instead of asking it to fill in missing details. Here is a compact starting prompt: ```text Use this Loom transcript and the linked resources to write an onboarding SOP in Markdown. Keep the process in the same order, and do not add details that are not in the source. [PASTE TRANSCRIPT] ``` ChatGPT gives you a first written version to work from. Treat it as a draft: read it against the recording and correct anything that no longer matches the task. ### Step 4 Give the SOP a Useful Format If the first version is not organized this way, ask ChatGPT to reshape it into four parts: purpose, links to resources, steps, and then a checklist. That gives the reader context before they start, the files they need while they work, and a simple way to check the result at the end. In the timesheet example, the resource section points to the template and completed-timesheets folder. The steps explain how to log the hours and make the copy. The checklist confirms the date, hours, total, filing status, and copied Sheet all carried over. ![Reviewed SOP purpose](https://tru-images.b-cdn.net/guide-assets/local-75/cut-onboarding-time-in-half-with-loom-and-chatgpt/03-reviewed-sop-purpose.png) Make the exception plain rather than hiding it in a long paragraph. For example, the demo SOP says to stop and message Ops if the weekly total is over 40 hours or the status is not READY TO FILE. Your own SOP should name the point where a teammate should pause and ask for help. ### Step 5 Paste the Final SOP Into a Google Doc Once the SOP reads like the process you recorded, paste the final version into a Google Doc. Use clear headings so the purpose, resource links, steps, and checklist are easy to scan while someone works through the task. > Pro tip: If you ask it to build the SOP in Notion, it can embed the Loom video so it is playable right on the page. The Google Doc is the written reference. The Loom remains useful when someone needs to see the original clicks or hear the explanation behind a decision. ### Guide: Cut Recurring Meeting Times in Half With Claude + Granola > Use Granola and Claude to audit recurring meetings, create a pre-read, and move repeat work out of the live call. - Canonical page: https://app.therundown.ai/guides/cut-recurring-meeting-times-in-half-claude-granola - Machine-readable page: https://app.therundown.ai/guides/cut-recurring-meeting-times-in-half-claude-granola.md - Difficulty: beginner - Featured tool or company: Claude + Granola - Categories: Productivity & Knowledge Work, Automation & Agents - Instructors: Billy Howell - Published: 2026-06-08 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Granola and Claude to make recurring meetings shorter. You will run an audit on one recurring meeting, get a templated pre-read, and find the parts Claude can automate. #### Who This Is Useful For - Team leads and operators who run the same meeting every week and want less live status reporting. - Founders and managers who want recurring calls to focus on decisions instead of catch-up. - Granola and Claude users who want their meeting notes to improve the next meeting, not just summarize the last one. #### What You Will Build ![Audit Findings Memo](https://tru-images.b-cdn.net/guide-assets/local-228/cut-recurring-meeting-times-in-half-claude-granola/01-audit-findings-memo.png) You will walk away with three practical takeaways: - Save time on any meeting by getting a report that tells you what to automate and use AI for before the meeting starts. - Get a template for a one-pager pre-read that AI can generate. - Receive suggested pre-work for all attendees of the meeting. #### What You Need To Get Started - A Granola account - A Claude account - At least a few notes from one recurring meeting #### Going Further Once the pre-read works manually, ask Claude to set up automations that create and populate the document before each recurring meeting. Use this: ```text Help me automate the pre-read workflow for [RECURRING_MEETING_NAME]. Every week before the meeting: - review the latest Granola notes - create a fresh one-page pre-read - include open threads, decisions, and required attendee inputs - suggest AI tasks that can happen before the call - save or send the pre-read to [GOOGLE_CALENDAR / NOTION / SLACK / EMAIL] Before scheduling anything, show me: - when the automation will run - which Granola notes it will inspect - where the pre-read will be saved - what I should review before it is shared ``` That is the larger pattern: Granola captures the meeting history, Claude audits the pattern, and the recurring meeting gets a little sharper before people join the next call. ### Step 1 Set Up Granola for Recurring Meetings Start by getting the recurring meeting into Granola. You can use Granola for free. It does not join the call, and if the meeting is already on your calendar, Granola can detect it and start a new note when the meeting begins. If the meeting is impromptu, start a quick note manually. The important part is consistency. You want Granola to become the memory bank for this meeting over time. One note is useful. Five notes from the same recurring meeting are much more useful because Claude can start seeing patterns. If you use spaces in Granola, create one for the recurring meeting you want to improve. For example: ```text Weekly content slate meetings Client check-ins Leadership syncs Project standups ``` That extra organization makes the audit cleaner because the meeting history is easier to isolate. > Pro tip: Use one Granola space for each recurring meeting you want to audit. Claude will have a much easier time finding the right notes later. ### Step 2 Connect Granola to Claude Open Claude and go to: ```text Customize > Connectors > Browse connectors ``` ![Browse Connectors Menu](https://tru-images.b-cdn.net/guide-assets/local-228/cut-recurring-meeting-times-in-half-claude-granola/cleanshot-2026-06-04-at-13-15-59-2x.png) Search for Granola, connect your account, and approve the connection. ![Granola Connector Details](https://tru-images.b-cdn.net/guide-assets/local-228/cut-recurring-meeting-times-in-half-claude-granola/02-granola-connector.png) Once connected, Claude can search your notes, list meetings, and pull meeting transcripts. That is what makes this different from copying one meeting summary into a chat. Claude can look across the recurring meeting history and find the work that keeps repeating. Before running the full audit, confirm Claude can find the right meeting: ```text Use my connected Granola notes to find the most recent meetings for [RECURRING_MEETING_NAME]. Return: - the meeting dates you found - a one-line summary of each meeting - whether there is enough context to audit this recurring meeting - anything you cannot access or are unsure about Do not recommend changes yet. ``` If Claude finds the wrong meetings, tighten the meeting name, date range, attendee names, or Granola space. ### Step 3 Audit the Last Five Meetings Now ask Claude to audit the recurring meeting. The goal is not to summarize the meetings. The goal is to find what should move out of the live call. Use this prompt: ```text Look at my five most recent [RECURRING_MEETING_NAME] meetings in Granola. Audit them so we can make this recurring meeting more efficient and shorter. Look for: - repeated status updates - delayed decisions - unresolved topics - repetitive questions - work that keeps happening during the meeting - friction points that slow the meeting down - tasks AI could automate or prepare before the meeting Give me a report on: 1. What we can improve 2. What should happen before the meeting 3. What still needs live discussion 4. What AI could help automate or do faster before the meeting 5. Any missing context or assumptions I should verify Use evidence from the Granola notes. Do not invent owners, dates, or decisions. ``` ![Recurring Meeting Audit Prompt](https://tru-images.b-cdn.net/guide-assets/local-228/cut-recurring-meeting-times-in-half-claude-granola/03-recurring-meeting-audit-prompt.png) Review the output before asking Claude to create anything. In the recording, the audit was useful because it found work that was happening live but should have happened before the meeting. It also separated status updates from decision-making. That is the shift you are looking for: less live catch-up, more live decisions. ### Step 4 Generate the Pre-Read and Automation Ideas Once the audit looks right, ask Claude to turn it into concrete meeting assets. Use this prompt: ```text Based on the recurring-meeting audit, create three outputs. Output 1: A one-page pre-read document for attendees. It should include: - meeting purpose - what changed since the last meeting - required attendee inputs - open threads from previous meetings - decisions we need to make live - a proposed shorter agenda - a "come prepared with" checklist Output 2: AI tasks or automations we can run before the meeting. For each one, include: - what the task does - what source it uses - where the output should go - whether it needs human review Output 3: A simple weekly workflow. Show how the pre-read could be created and shared in [GOOGLE_CALENDAR / NOTION / SLACK / EMAIL] before each meeting. Keep the pre-read short enough for attendees to complete in five minutes. Do not include sensitive notes that attendees should not see. Phrase uncertain owners, dates, or decisions as confirmation questions. ``` The best output should give you two things at once: a memo you can send before the next meeting and a list of work that no longer needs to happen live. Do not accept the first draft blindly. Remove anything too vague, too sensitive, or too time-consuming for attendees. A good pre-read should make the meeting lighter, not create a second meeting before the meeting. ### Step 5 Create a Better Granola Template The final step is to improve the notes you will create in future meetings. Ask Claude to create a Granola template for this recurring meeting: ```text Create a Granola note template for [RECURRING_MEETING_NAME]. The template should make future notes easier to audit and reuse. Include sections for: - final decisions - action items - unresolved topics - repeated questions - prep work that should happen before next time - things to improve about the meeting Keep the section titles simple enough to use inside Granola. ``` Then open Granola and create a custom template for that recurring meeting. Add the sections Claude recommends and apply it to future notes. This is the part that makes the system improve over time. Better notes make the next Claude audit more useful, and the next audit makes the next meeting easier to prepare. ### Guide: Cut Typing Time In Half With AI Voice Commands (Typeless Tutorial) > In this guide, you will learn how to use Typeless to research an idea, summarize a source, open a draft, and dictate a clean message without typing every sentence by hand. - Canonical page: https://app.therundown.ai/guides/cut-typing-time-in-half-with-ai-voice-commands-typeless-tutorial - Machine-readable page: https://app.therundown.ai/guides/cut-typing-time-in-half-with-ai-voice-commands-typeless-tutorial.md - Difficulty: beginner - Featured tool or company: Typeless - Categories: Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-06-22 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Typeless to research an idea, summarize a source, open a draft, and dictate a clean message without typing every sentence by hand. #### Who This Is Useful For - Knowledge workers who write emails, prompts, posts, notes, or project updates all day. - AI power users who think better out loud and do not want to clean up every false start. - Anyone who wants a voice-first workflow for research, navigation, drafting, formatting, and quick translation. #### What You Will Build You will build a voice workflow that moves from research to draft: - Ask Typeless for a useful angle. - Summarize source material. - Open the draft destination. - Dictate naturally. - Review before sending or publishing. The example uses an X post, but the same flow works for email, docs, Slack messages, meeting follow-ups, notes, and AI prompts. ![typeless thread draft modal](https://tru-images.b-cdn.net/guide-assets/local-280/cut-typing-time-in-half-with-ai-voice-commands-typeless-tutorial/01-typeless-thread-draft-modal.png) #### What You Need To Get Started - A Typeless account. - Typeless installed on your computer. - Microphone access enabled. - Accessibility or input permissions enabled if Typeless asks for them. - A safe draft destination, like Google Docs, Notes, Gmail, or an unpublished X post. #### Going Further Once the core flow works, test the secondary features. For translation: ```text Translate this into Italian. ``` ![typeless language translation settings](https://tru-images.b-cdn.net/guide-assets/local-280/cut-typing-time-in-half-with-ai-voice-commands-typeless-tutorial/04-typeless-language-translation-settings.png) For app navigation: ```text Open a new Notion page for me. ``` Use this pattern anywhere you write: ```text Ask, navigate, dictate, format, verify. ``` Good next use cases: - Turn a rough meeting follow-up into a clean email. - Dictate a long AI prompt without fixing every false start manually. - Summarize selected article text before drafting a post. - Open a work app with Ask Anything, then dictate the update. - Translate a short message while traveling or working with someone in another language. Start in draft mode until you trust the shortcuts, cursor placement, and cleanup behavior. ### Step 1 Install Typeless And Set Up The Shortcuts Go to Typeless (https://www.typeless.com/) and download the desktop app. Complete the tutorial and set up the two shortcuts you will use most: - Ask Anything, for research, search, and opening pages. - Dictation, for writing into the current text field. ![typeless keyboard shortcuts](https://tru-images.b-cdn.net/guide-assets/local-280/cut-typing-time-in-half-with-ai-voice-commands-typeless-tutorial/02-typeless-keyboard-shortcuts.png) > Pro tip: Make sure the keyboard shortcuts do not conflict with other apps. Open Notes or Google Docs and test one sentence: ```text We need a 30-day report on our social media accounts by the end of the week. ``` If the sentence lands in the right place, your microphone, cursor, and shortcut are working. Before using Typeless heavily, check the plan fit. As of June 22, 2026, the Typeless pricing page (https://www.typeless.com/pricing) lists: - Free: 8,000 words per week, standard accuracy, standard high-demand access, and core features like cleanup, Translate, Ask Anything, personalized writing style, app-specific tones, and Whisper mode. - Pro: $12 per member per month when billed yearly, or $30 monthly, with unlimited words, enhanced accuracy, priority access, team management, prioritized feature requests, and early access. Check the live pricing page before subscribing, because plan details can change. ### Step 2 Use Ask Anything To Research An Angle Use Ask Anything to research the topic you want to write about. Try: ```text What are some recent trends in consumer AI use cases that I could use to draft an interesting tweet? ``` If the answer is too broad, tighten the search: ```text Search Google for the hottest [TOPIC] trends in [MONTH YEAR]. ``` Pick one clear angle before you draft. In our test, the best angle came from an article about how people are actually using AI: - research - emotional support - work help - personal productivity - bottom-up habits inside companies ### Step 3 Summarize Source Material Before You Write Open a useful article and select the relevant text. Then ask Typeless: ```text Summarize this text. ``` To turn the summary into a draft format, use: ```text Rewrite the selected text as an X thread. ``` You can swap the format: - Rewrite the selected text as a short email. - Rewrite the selected text as three bullets for a project update. - Rewrite the selected text as a prompt I can paste into ChatGPT. > Pro tip: Select the text first so Typeless knows exactly what to summarize. Keep commands small: - Summarize the selected text. - Move the summary where you want to work. - Rewrite it in the format you need. - Open the final destination separately. ### Step 4 Open Your Draft Destination Use Ask Anything to open the place where the final message should go. For a social post: ```text Open x.com and open a new tweet. ``` For email: ```text Open Gmail and start a new email. ``` For a document: ```text Open a new Google Doc. ``` You may still need to click once or confirm the right field. Before dictating, make sure your cursor is in the draft. If your cursor is in the browser address bar, Typeless can put the text there instead. ### Step 5 Dictate Naturally And Let Typeless Clean It Up Put your cursor in the draft and talk through the message like you would explain it to a coworker. For example: ```text Here are the three most interesting things Harvard Business Review found for 2026. First, AI slop and cognitive decline are becoming a concern as people over-rely on AI. Second, emotional support, therapy, companionship, and relationship advice have become top AI use cases. Third, AI at work is growing from the bottom up, with individuals using tools before companies formally roll them out. ``` Typeless can clean up rough speech and corrections. Try this: ```text We need a 90-day report, no actually a 30-day report, on our social media accounts by the end of the week. ``` Typeless should write the final ask as a 30-day report. You can also speak a list naturally: ```text There are three teams working on the account: Mike's team, April's team, and Sean's team. ``` Typeless can format that as a list instead of one long sentence. ![x draft from dictation](https://tru-images.b-cdn.net/guide-assets/local-280/cut-typing-time-in-half-with-ai-voice-commands-typeless-tutorial/03-x-draft-from-dictation.png) > Pro tip: If you make a mistake while dictating, correct yourself naturally. Typeless's AI will understand the correction and only write your final idea. Review the draft before sending or publishing. Check names, numbers, links, and any claims that came from research. ### Guide: Delegate Team Tasks to Claude Inside Slack > In this guide, you will learn how to set up and install Claude Tag in Slack. - Canonical page: https://app.therundown.ai/guides/delegate-team-tasks-to-claude-inside-slack - Machine-readable page: https://app.therundown.ai/guides/delegate-team-tasks-to-claude-inside-slack.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Business Operations & Leadership, Automation & Agents, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-07-02 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to set up and install Claude Tag in Slack. This lets you and your teammates tag @Claude like a real teammate, then have Claude work in the cloud and report back when it is done. #### Who This Is Useful For - Team leads who want one shared Claude inside Slack instead of each teammate using a separate personal Claude account. - Operators and founders who want Claude to work from a Slack thread, use approved tools, and report back in the same place. - Anyone evaluating Claude Tag who wants the setup path, token-spend warnings, and first-test workflow before rolling it out. #### What You Will Build You will set up Claude Tag in a Slack test channel, connect the tools Claude is allowed to use, and run one cloud task from a Slack thread. In our test, Claude picked up the request, showed a checklist, opened a cloud session, worked through connected Google Drive files, and reported back in Slack. ![Claude working inside a Slack thread](https://tru-images.b-cdn.net/guide-assets/local-289/delegate-team-tasks-to-claude-inside-slack/01-claude-working-in-slack-thread.png) #### What You Need To Get Started - A Claude Team or Enterprise plan (https://claude.com/pricing). - Slack admin permissions for the workspace where you want to install Claude. - Access to Claude Tag admin settings (https://claude.ai/admin-settings/claude-tag). - One dedicated Slack test channel. - Optional: tool accounts you want Claude to use, such as Google Drive, Notion, Gmail, or GitHub. #### Going Further After the first run, check the admin panel for token spend, plugins used, and memory files created. You can edit or delete persistent Claude Tag memory files there. Also review: - Claude Tag memory docs (https://claude.com/docs/claude-tag/users/memory) to understand what persists. - Claude Tag customization docs (https://claude.com/docs/claude-tag/admins/customize) to manage connections, plugins, instructions, and model choices. - Claude Tag spend-limit docs (https://claude.com/docs/claude-tag/admins/restrict-access#set-spend-limits) before expanding beyond the test channel. Once the first channel is clean, repeat the same pattern slowly: - Add one channel. - Connect only the tools that channel needs. - Run one narrow test task. - Check spend, plugins, and memory files. - Expand only after the team understands what Claude can access. ### Step 1 Verify Your Account Setup Claude Tag is for Claude Team and Enterprise accounts. Anthropic's Claude Tag announcement (https://www.anthropic.com/news/introducing-claude-tag) describes it as a team workflow for Slack, where @Claude can work in a channel and respond back in the thread. You also need Slack admin permissions so you can install and pair the Slack app. > Pro tip: If you do not have a business plan for Claude, you can still follow along with the regular Claude in Slack (https://claude.com/docs/connectors/slack) experience. That path uses your own Claude account instead of a shared team Claude. But Anthropic says this legacy Claude in Slack experience is being deprecated in August 2026. ### Step 2 Install The Claude Slack App In Slack, open Admin > Apps and workflows, search for Claude, and install the app. You can also open the Claude app in the Slack Marketplace (https://slack.com/marketplace/A08SF47R6P4-claude) directly. ![Claude app in the Slack Marketplace](https://tru-images.b-cdn.net/guide-assets/local-289/delegate-team-tasks-to-claude-inside-slack/02-claude-slack-app-marketplace.png) Slack will send you into the browser for most admin actions. That is normal. The Slack desktop app is mostly the starting point; install, permissions, and pairing happen in browser windows. ### Step 3 Connect Your Claude Account Back In Slack Back in the Slack app, you should see Claude in your bottom-left sidebar under Apps. Open it, click the Home tab, and connect your Claude account. ![Claude app Home tab connected in Slack](https://tru-images.b-cdn.net/guide-assets/local-289/delegate-team-tasks-to-claude-inside-slack/03-claude-slack-app-home-connected.png) Once connected, the app Home tab is also where you will see account details, routing options, and the regular Claude in Slack setup. Claude Tag setup is separate, and happens in the browser. ### Step 4 Set Up Claude Tag And Connect Tools In your browser, open Claude Tag admin settings (https://claude.ai/admin-settings/claude-tag). You may also see a Claude Tag setup prompt from Claude settings. ![Set up Claude Tag card in Claude settings](https://tru-images.b-cdn.net/guide-assets/local-289/delegate-team-tasks-to-claude-inside-slack/04-claude-tag-setup-card.png) Follow the setup flow from Anthropic's setup instructions (https://claude.com/docs/claude-tag/admins/setup-overview): - Pair Claude Tag with your Slack workspace. - Connect the tools you want Claude to use. - Add usage credits and set a spending limit. - Launch Claude Tag. Connect only the tools you are comfortable exposing to the channel. In the recording, we tested Google Drive and Notion. Gmail or shared inbox access is more sensitive because anyone in the allowed Slack scope can tag Claude and ask it to use that connection. > Pro tip: Only enable Claude Tag in one test channel first. That gives you a small group of testers, a clear access boundary, and an easier way to watch token spend. ### Step 5 Pair Claude Tag From Slack In the Slack channel you want to test, run: ```text @Claude connect ``` Claude will return a pairing code. Copy that code into the Claude setup dialog in your browser. During pairing, choose the narrowest scope that works for your test. For a first run, pick one channel instead of the whole workspace. This keeps Claude away from unrelated channels while you confirm the setup, permissions, and spend behavior. If your Team plan needs usage credits, Claude will prompt you to add them before launch. Channel work draws from the organization usage balance, not an individual user's normal Claude chat limits. ### Step 6 Lower The Model And Run A Test Task Before testing, open Advanced in Claude Tag admin settings (https://claude.ai/admin-settings/claude-tag) and switch from the expensive default model to Sonnet. ![Claude model and routing settings in Slack](https://tru-images.b-cdn.net/guide-assets/local-289/delegate-team-tasks-to-claude-inside-slack/05-claude-model-routing-settings.png) In our test, the default model burned through usage credits much faster than expected. Sonnet was the safer starting point for a first test. Then run your first task inside the dedicated Slack thread: ```text @Claude analyze our brand kit and drafted social posts in Google Drive. Suggest social strategy improvements. ``` Claude should reply in the thread, show a short checklist, and keep working in the cloud. You can open the linked Claude session to see more of what it is doing. The result was not perfect in our demo because the test Drive did not have enough useful files. But the workflow worked: Claude picked up the Slack request, looked through the connected tool, and reported back like a teammate. ### Guide: Deploy a Mini-SaaS in 10 Minutes With ChatGPT Sites > In this guide, you will learn how to turn one public API into a small hosted web app with ChatGPT Sites. - Canonical page: https://app.therundown.ai/guides/deploy-a-mini-saas-in-10-minutes-with-chatgpt-sites - Machine-readable page: https://app.therundown.ai/guides/deploy-a-mini-saas-in-10-minutes-with-chatgpt-sites.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-07-18 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to turn one public API into a small hosted web app with ChatGPT Sites. You will create a product plan, build an MVP, and deploy it as a site people can actually use. #### Who This Is Useful For - Anyone who wants a practical first app idea instead of starting with a blank canvas. - Operators who want to turn a repetitive lookup or decision into a simple tool. - Non-developers who want to build and share a small web app without designing a full product first. #### What You Will Build You will build a mini-SaaS with one clear job. Our example is WeatherLedger: a tool that uses weather data to show contractors the best hours to work at a jobsite. ![The finished app: one decision, powered by one API.](https://tru-images.b-cdn.net/guide-assets/local-389/deploy-a-mini-saas-in-10-minutes-with-chatgpt-sites/01-weatherledger-finished-app.png) #### What You Need To Get Started - Access to ChatGPT and Codex, plus a workspace where ChatGPT Sites is available (https://help.openai.com/en/articles/20001339-creating-and-managing-chatgpt-sites). Sites availability can vary by plan, region, and workspace settings. - An API from the Public APIs list (https://github.com/public-apis/public-apis), plus the API provider's own documentation. - One small use case. Version one should answer a single question or complete a single task. #### Going Further In the Sites menu, you can see and track web analytics for your deployed apps all in one place. This is also where you can add custom domain names. Once the first version is working, expand intentionally. Add only the next feature that helps the same user get the same outcome faster or more reliably. A small app that does one job well is a much stronger starting point than a large product plan. ### Step 1: Pick One Public API Open the Public APIs list (https://github.com/public-apis/public-apis). It is a huge list of mini-SaaSes waiting to be built. Pick one API to build your mini-SaaS around. ![Start with the Public APIs list and pick one data source.](https://tru-images.b-cdn.net/guide-assets/local-389/deploy-a-mini-saas-in-10-minutes-with-chatgpt-sites/02-public-apis-list.png) We used weather data to create WeatherLedger. It takes a jobsite and turns the forecast into a recommendation for when contractors can work. If you do not know which API to choose, pick a couple that look interesting, give them to ChatGPT, and ask it for a good mini-SaaS idea. That is how we came up with WeatherLedger. Before you commit, open the provider's documentation and terms. The Public APIs list is a discovery tool, not the source of truth. Confirm the required inputs, free-tier limits, attribution rules, and whether the API fits the use you have in mind. ### Step 2: Turn the Idea Into a PRD Tell ChatGPT to turn the API and your idea into a PRD. Ask it to plan only, not code. You want a clear product promise, the smallest possible user flow, the API inputs and data you need, a few acceptance checks, and a list of what to leave out. Use a prompt like this: ```text Plan, but do not build yet, a mini-SaaS that uses [API NAME] for [AUDIENCE]. The one job it must do is: [CORE JOB]. Return: - a one-sentence product promise - the smallest user flow - required API data and inputs - a distinct brand and UI direction - three acceptance checks - a short list of features to keep out of version one ``` For WeatherLedger, the plan was simple: a contractor selects a jobsite and sees the best work window based on weather conditions. We left out accounts, billing, a database, and every other feature that did not help answer that one question. > Pro tip: Approve the plan before you build. A good-looking landing page is not the same thing as an interactive app that completes the job. ### Step 3: Build the MVP in Codex Switch to Codex and give it the approved plan. Ask it to build the MVP, track the project with Git, and include the Sites plugin from the start. ![Build the MVP in Codex.](https://tru-images.b-cdn.net/guide-assets/local-389/deploy-a-mini-saas-in-10-minutes-with-chatgpt-sites/03-build-in-codex.png) You can keep the request direct: ```text Build the MVP from this approved plan: [PASTE PLAN]. Keep the first version focused on the core user flow. Use Git to track the project and include the Sites plugin from the start. Use live data from [API NAME], add a useful error state, run the available checks, and open the preview when it is ready. ``` Review the preview before you publish. Test the main user flow with live data and make sure the app gives a useful answer, not just a polished screen. In our example, that means WeatherLedger needs to return a readable work-window recommendation from real weather data. ### Step 4: Deploy With Sites Once you are happy with the preview, tell Codex to deploy the site with Sites. ![Deploy the site privately for a team or publicly for anyone to use.](https://tru-images.b-cdn.net/guide-assets/local-389/deploy-a-mini-saas-in-10-minutes-with-chatgpt-sites/04-deploy-with-sites.png) You can deploy it privately so only people on your team can use it, or deploy it publicly so anyone can visit it. In the Sites sidebar, you can see and manage these deployments. Use the narrowest access setting that fits the test. Then open the live URL and repeat the core workflow one more time. A deployment URL is a real hosted version of the app, so it should show the behavior you expect before you share it. ### Guide: Design Better Websites With This New Cursor Feature > Learn how to quickly set up and use Cursor’s visual design editor to refine your frontend design without switching back and forth with a design tool like Figma. - Canonical page: https://app.therundown.ai/guides/design-better-websites-with-this-new-cursor-feature - Machine-readable page: https://app.therundown.ai/guides/design-better-websites-with-this-new-cursor-feature.md - Difficulty: beginner - Featured tool or company: Cursor - Categories: Coding & App Development, Content & Creative - Instructors: Billy Howell - Published: 2025-12-15 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to quickly set up and use Cursor’s visual design editor to refine your frontend design without switching back and forth with a design tool like Figma. You will use a simple HTML and CSS project, preview it in Cursor’s browser, and make design edits through the Design pane. #### Who This Is Useful For - People tired of switching between design and coding software - Cursor users - Solo developers #### What You Will Build You will build a refined frontend design using Cursor’s visual editor and agent workflow. The final output is an updated HTML and CSS website with design changes applied and kept in your project. ![1_what_you_will_build](https://tru-images.b-cdn.net/guide-assets/69/694090b3f80c9808d91f754e/1-what-you-will-build.png) #### What You Need - Cursor app - A basic HTML and CSS project - Optional: ask the Cursor agent to build a simple index.html and styles.css website, or use a template - Live Server extension: settings → extensions → search “live server” #### Going Further - Use a prompt to extract your main colors and use them in the color picker dropdown. - Read more about the design sidebar. ### Step 1: Open your project in Cursor Open a new project in Cursor with an HTML file and a CSS file. Make sure you are running the latest Cursor update. ### Step 2: Start a Live Server preview Make sure you have the Live Server extension installed. ![3_live_server_extension](https://tru-images.b-cdn.net/guide-assets/69/694090b3f80c9808d91f755e/3-live-server-extension.png) Hit CMD+Shift+P and search for “Open with Live Server.” ![4_live_server_extension2](https://tru-images.b-cdn.net/guide-assets/69/694090b3f80c9808d91f7566/4-live-server-extension2.png) ### Step 3: Open the preview in Cursor’s browser - A live preview of your site should open in your default browser. - Grab the URL and go back to Cursor. - Hit CMD+Shift+B to open the Cursor browser. - Paste in the URL and hit Enter. ### Step 4: Select and edit elements in the Design pane Make sure you have toggled on the element selector option. ![6_selector_toggle](https://tru-images.b-cdn.net/guide-assets/69/694090b4f80c9808d91f7576/6-selector-toggle.png) Now you can click any element on your site and edit its properties in the Design pane. ![cursor_design_pane_preview](https://tru-images.b-cdn.net/guide-assets/69/69409de734c5ff3db093ff51/cursor-design-pane-preview.png) You can also click an element and it will be mentioned in the active agent chat. If you do not know how to make the changes you want in the Design pane, tell the agent what you want instead. ### Step 5: Apply and keep your changes Hit the Apply button and a Cursor agent will make the changes for you. Click “Keep” or “Keep All” in the agent chat to save those changes. ![7_keep_all](https://tru-images.b-cdn.net/guide-assets/69/694090b4f80c9808d91f757e/7-keep-all.png) The agent is smart enough to understand classes. If you have a style class like the dino cards in the example and make style changes to one, the agent will update the class so the changes are made to the other divs as well. ### Step 6 Use color tokens in the picker Use this prompt to extract your main colors and use them in the color picker dropdown: ```text Can you extract and put the color tokens from my page's theme in the root of @styles.css so that I can use them as color tokens? ``` ![2_color_tokens](https://tru-images.b-cdn.net/guide-assets/69/694090b3f80c9808d91f7556/2-color-tokens.png) ### Resources - [Design Better Websites With This New Cursor Feature](https://tru-images.b-cdn.net/guide-assets/69/69409d81fed5fda318384cbf/design-better-websites-with-this-new-cursor-feature.pdf) ### Pro tip > Make sure you are saving your progress with Git as you go. Git will make it easier to roll back any unwanted style changes. You can even ask the agent to write Git commits for you. ### Guide: Edit Videos Faster With AI (For Non-Editors) > In this guide, you'll learn how to edit videos using AI like we do on the rundown, and then teach ChatGPT to edit for you hands-free. - Canonical page: https://app.therundown.ai/guides/edit-videos-faster-with-ai-for-non-editors - Machine-readable page: https://app.therundown.ai/guides/edit-videos-faster-with-ai-for-non-editors.md - Difficulty: beginner - Categories: AI, Video - Instructors: Billy Howell - Published: 2026-07-23 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you'll learn how to edit videos using AI like we do on the rundown, and then teach ChatGPT to edit for you hands-free. #### Who This Is Useful For - Creators who record talking-head or screen-share videos but do not want to learn a traditional timeline editor. - Podcasters and interviewers who need to remove retakes, filler words, and long pauses without outsourcing every first pass. - Teams that want a repeatable, source-safe editing process they can run from Descript, ChatGPT, or Codex. #### What You Will Build You will build a five-part editing workflow: Protect → Clean → Tighten → Review → Delegate. Descript handles the first pass, and a reusable skill lets ChatGPT or Codex repeat it. ![The completed AI clean edit, ready for human review](https://tru-images.b-cdn.net/guide-assets/local-364/edit-videos-faster-with-ai-for-non-editors/01-ai-clean-edit-result.png) You still make the final calls on meaning, pacing, screen action, layouts, captions, and export. #### What You Need To Get Started - The Descript desktop app (https://www.descript.com/download). - ChatGPT Plus or higher (optional if you already have a source file to start with). #### Going Further Ask the skill to find three standalone 30–60 second moments without changing the clean composition. Review each opening and ending, then finish the strongest clip. Let the skill handle the mechanical 90%. You handle scenes, layouts, B-roll, effects, captions, and final export. Get resources, prompts, and full step-by-step video guide here. (https://app.therundown.ai/guides/edit-videos-faster-with-ai-for-non-editors) ### Step 1 Start With the Most Accurate Transcript Record in the Descript desktop app or import your source video. Descript edits through the transcript, so accuracy matters. Open Settings with Command/Control + ,, choose AI models, and select the most accurate transcription model available. If you know the language, select it instead of relying on automatic detection. Add unusual product names or technical terms to the transcription glossary before processing. Model names vary by plan, so check Descript’s current automatic transcription (https://help.descript.com/hc/en-us/articles/10249424286477-Automatic-transcription) and app settings (https://help.descript.com/hc/en-us/articles/10606771145869-App-settings) documentation. > Pro tip: When you make a mistake while recording, say “cut that” or a unique marker word such as “strawberry.” You—and the AI—can search for that signal later instead of hunting through the entire video. ### Step 2 Protect the Source Before the AI Edits Open your project, confirm the source composition name and duration, then duplicate the composition. Rename the copy [GUIDE] - Clean Edit and make every cleanup change there. ![The untouched source beside the duplicate used for AI cleanup](https://tru-images.b-cdn.net/guide-assets/local-364/edit-videos-faster-with-ai-for-non-editors/02-protected-source-duplicate.png) Use the duplicate as your safety copy. For early cuts, prefer Ignore because struck-through text can be restored. See Descript’s guides to composition duplication (https://help.descript.com/hc/en-us/articles/10612359826061-Duplicate-a-project-composition-selection-or-layer) and Ignore versus Delete (https://help.descript.com/hc/en-us/articles/10164872017933-Deleting-vs-ignoring-script-text). If you want Underlord—Descript’s AI co-editor—to perform the first pass, paste this prompt: ```text Create a duplicate of the current source composition named "[GUIDE] - Clean Edit" and edit only that duplicate. Leave the source unchanged. On the duplicate, perform a conservative first pass: - trim dead air at the beginning and end; - remove clear repeated takes and sentence restarts; - remove spoken editor notes such as "cut that"; - remove only nonessential filler words. Preserve meaning, screen actions, loading time, and useful comprehension pauses. Prefer reversible ignored or struck-through transcript edits. Do not shorten all word gaps yet. Do not add captions, publish, export, or modify any other composition. Report what changed and flag uncertain cuts for human review. ``` Before continuing, verify that the source still exists at its original duration and that only the named duplicate changed. ### Step 3 Remove Filler Words and Tighten Long Pauses On the clean duplicate, open AI Tools → Remove filler words. Preview a few results before using bulk removal, and enable Avoid harsh cuts to prevent clipped speech. ![Filler-word candidates being reviewed before a bulk removal](https://tru-images.b-cdn.net/guide-assets/local-364/edit-videos-faster-with-ai-for-non-editors/03-filler-word-review.png) Not every filler should disappear; some preserve emphasis or cover a visual action. Use the Remove filler words (https://help.descript.com/hc/en-us/articles/10164806394509-Remove-filler-words) panel to preview each candidate. After the pass, watch the edited composition at up to 2× speed and listen across every cut boundary. If a word sounds clipped or the delivery becomes rushed, restore that edit before moving on. ### Shorten Long Pauses Open AI Tools → Shorten word gaps and start by finding gaps longer than 1.0 second. Review individual candidates before using Shorten all. ![A long word gap being checked before it is shortened](https://tru-images.b-cdn.net/guide-assets/local-364/edit-videos-faster-with-ai-for-non-editors/04-word-gap-review.png) Our recommended starting points are 0.25–0.5 seconds inside a sentence and 0.5–0.8 seconds between ideas. These are not Descript defaults—just what we recommend. Keep longer pauses when the viewer needs time to read the screen, watch a click, wait for loading, or absorb a point. For a safer delegated pass, use: ```text On "[GUIDE] - Clean Edit" only, review word gaps longer than 1.0 second. Shorten unintentional hesitation gaps conservatively, but preserve pauses needed for emphasis, screen actions, loading, or a clear break between ideas. As a starting point, use 0.25-0.5 seconds within a sentence and 0.5-0.8 seconds between ideas. Keep changes reversible. Do not alter the source, publish, or export. Report the gaps changed and any pauses left for review. ``` Descript’s Shorten word gaps (https://help.descript.com/hc/en-us/articles/10164807277453-Shorten-word-gaps) tool supports individual and bulk actions. Review individual gaps before automating them. ### Step 4 Review the AI Edit Like an Editor Run Edit for Clarity at Low. Preview each suggested removal, then accept or reject it. Removed phrases appear as ignored text so they can be reviewed or restored. See Descript’s Edit for Clarity (https://help.descript.com/hc/en-us/articles/36841959272717-Edit-for-Clarity) guide. Then watch the entire clean composition and check: - Meaning: Did any cut remove context or change the claim? - Retakes: Is the intended take present and the false start gone? - Word boundaries: Did a cut clip a consonant, breath, or word ending? - Pacing: Does the speaker still sound human rather than rushed? - Visual timing: Is there enough time to read, click, or watch the screen action? - Source safety: Does the untouched composition still have its original duration? Studio Sound is optional and affects the source file, not just the composition. Adjust it carefully; Descript’s Studio Sound (https://help.descript.com/hc/en-us/articles/10327603613837-Studio-Sound) guide explains the scope. For a final no-change verification pass, use: ```text On "[GUIDE] - Clean Edit" only, identify any remaining hesitation gaps of at least 1.0 second. Do not shorten them. Report each safe candidate with its timestamp, current duration, suggested duration, and reason. Separately list gaps that must remain for screen action, loading, emphasis, ignored content, or a clear break between ideas. Then verify that the source composition still exists at its original duration, only the named duplicate changed, and no caption, publish, or export action occurred. Do not make further edits after reporting. ``` ### Step 5 Turn the Workflow Into a Reusable Skill Run this process manually once or twice before automating it so you know which edits the skill should make. Connect Descript from the ChatGPT app marketplace and authorize the correct Drive. ChatGPT requires Plus or higher, and workspace admins may need to enable third-party connectors. Follow Descript’s current connection instructions (https://help.descript.com/hc/en-us/articles/46215649218445-Connect-Descript-to-ChatGPT). Then invoke Skill Creator in ChatGPT or Codex and paste: ```text Use Skill Creator to build a reusable video-editing skill that works through the connected Descript app/MCP. The skill must duplicate the named source composition and edit only the duplicate. It should remove dead air, clear retakes, spoken edit notes such as "cut that," and nonessential filler words; enable conservative filler-word handling; shorten only unhelpful word gaps; preserve meaning, screen actions, loading time, emphasis, and useful pauses; and keep early transcript edits reversible. Require verification after every pass. The source must remain unchanged, uncertain cuts must be reported for human review, and the skill must stop before captions, publishing, or export. If the available tool can only shorten all remaining gaps globally, report the candidates and stop instead of applying that action. Name the skill "descript-clean-edit" and include a clear example showing how to run it with either a Descript project link or an exact project and composition name. ``` Keep your script, outline, and editing notes in the same ChatGPT or Codex project folder so the assistant can reference them. To run the finished skill, paste: ```text Use $descript-clean-edit on this Descript project: [PROJECT LINK OR EXACT PROJECT NAME]. Source composition: "[SOURCE NAME]" Create: "[GUIDE] - Clean Edit" Leave the source untouched. Run the staged conservative cleanup: obvious retakes, editor notes, and filler words first; long gaps second. For the final hesitation-gap pass, identify and report individual candidates only—do not apply a global shorten-word-gaps action. Verify the project after every pass, preserve screen actions and useful pauses, and stop before captions, publishing, or export. Report the source duration, clean-edit duration, and everything that still needs human review. ``` ![The reusable Descript editing skill running from ChatGPT or Codex](https://tru-images.b-cdn.net/guide-assets/local-364/edit-videos-faster-with-ai-for-non-editors/05-descript-editing-skill.png) The job may take 10–15 minutes, and progress can be limited. Older projects can fail, so inspect the completed duplicate and verify the source. ### Guide: Exploring How Comet Works (Perplexity’s AI Browser) > This guide introduces the basics of Comet, Perplexity’s AI browser. - Canonical page: https://app.therundown.ai/guides/exploring-how-comet-works-perplexitys-ai-browser - Machine-readable page: https://app.therundown.ai/guides/exploring-how-comet-works-perplexitys-ai-browser.md - Difficulty: beginner - Featured tool or company: Perplexity - Categories: AI Foundations & Prompting, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-01-12 - Updated: 2026-07-30 ### Overview #### The Rundown This guide introduces the basics of Comet, Perplexity’s AI browser. You’ll learn where the main tools live, what you can try with a free or paid Perplexity subscription, and preview several useful workflows including templates, page summaries, voice mode, and Assistant-controlled browsing. #### Who This Is Useful For - Researchers who want relevant citations. - Shoppers who want to find the best deal without manual effort. - Anyone who needs a virtual assistant that’s easy to use. #### What You Will Build You will build a practical starting workflow for using Comet as an AI browser and assistant. - A basic understanding of Comet’s sidebar and navbar tools. - A tested Space template workflow. - A first Assistant workflow that can interact with a shopping page. - A plan for enabling integrations such as Assistant, Connectors, and Purchases. #### What You Need - Perplexity’s Comet browser. - A free or paid Perplexity subscription. #### Going Further - After the walkthrough, set up Assistant, Connectors, and Purchases from your profile icon. Purchases can let Perplexity use your credit card to make purchases on your behalf, track packages in the Purchases menu, and use a similar menu for travel reservations. ### Step 1: Set up Comet - Install the Comet desktop app. - Follow the onboarding instructions. - Turn your volume up for the onboarding music. > Ignore the bit about making Comet your default browser to block ads. It blocks ads even if it’s not your default browser. ### Step 2: Take a quick tour of Comet The sidebar and the top right of the navbar are where your tools are. ![comet_navbar](https://tru-images.b-cdn.net/guide-assets/69/6965505b8a28efc3014da7b8/comet-navbar.png) Sidebar features: - Library: a list of your chat history. - Discover: similar to the front page of an AI newspaper. You can customize which trends you are interested in. - Spaces: where you can build Perplexity agents with custom context. Context can include documents, MCP tools, links, and instructions. - Finance: a mix of Yahoo stocks and Bloomberg Terminal, but with AI. - More: additional tools you can add to your sidebar. ![comet sidebar](https://tru-images.b-cdn.net/guide-assets/69/696550748a28efc3014da7ee/comet-sidebar.png) Navbar features: ![comet_navbar_features](https://tru-images.b-cdn.net/guide-assets/69/6965508e8a28efc3014da82a/comet-navbar-features.png) - Summarize this webpage: press the button with three lines or press ⌥ + S to open a new assistant that will summarize whatever page you’re on. - Launch Voice Mode: press the sound-wave button or ⌥ + Shift + V to open an assistant in voice mode that can interact with the page and do tasks for you. - Assistant: press the Assistant button to open the assistant on the right side. The assistant can take control of your browser tab and do tasks for you. ### Step 3: Test your first template - Click Spaces → Templates. - Choose a task to try. The source recommends Trip Planner Pro. ![comet_templates](https://tru-images.b-cdn.net/guide-assets/69/696550bc174b08d0102c951a/comet-templates.png) ```text Plan me a trip from [starting city] to [destination city] on [dates]. ``` Templates are like custom GPTs, but a little more predefined. For example, the trip planner will sometimes, but not always, give you a form to select choices like budget and activities before it starts research. Try a couple prompts to see what you get. > There’s a “Get Started with Perplexity” template that can help you learn how to use it better. ### Step 4: Try your first workflow - Go to a retailer like Amazon or Walmart. - Open Assistant with ⌥ + A or by clicking in the top right corner. - Try the prompt below. ```text Find a computer mouse with 5k+ reviews averaging above 4.5 stars for under $60 and add it to my cart. ``` - You should see a blue overlay appear on your browser window. This means Perplexity is controlling your tab. ![comet_shopping](https://tru-images.b-cdn.net/guide-assets/69/696550ef174b08d0102c9554/comet-shopping.png) ### Step 5 Going further: Set up integrations and services To get the most out of Comet, set up a few integrations and services. Click your profile icon and set up Assistant, Connectors, and Purchases. ![comet_going_further](https://tru-images.b-cdn.net/guide-assets/69/696550fbc9011fcab6c20382/comet-going-further.png) - Purchases lets Perplexity use your credit card to make purchases on your behalf. - You can track packages in the Purchases menu. - Comet has a similar menu for travel reservations too. ### Pro tip > Comet blocks ads even if it is not set as your default browser. ### Guide: Find 5 Extra Prospects a Day With This Agentic Framework > In this guide, you will learn how to build a Codex prospecting system that finds five qualified prospects a day, ranks them, and drafts the next outreach step. - Canonical page: https://app.therundown.ai/guides/find-5-extra-prospects-a-day-with-this-agentic-framework - Machine-readable page: https://app.therundown.ai/guides/find-5-extra-prospects-a-day-with-this-agentic-framework.md - Difficulty: beginner - Featured tool or company: The Rundown AI - Categories: Sales, Consulting & Customer Success, Automation & Agents - Instructors: Billy Howell - Published: 2026-06-05 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to build a Codex prospecting system that finds five qualified prospects a day, ranks them, and drafts the next outreach step. Instead of a giant spreadsheet you never use, you get a daily report with fit reasons, source links, and a running prospect log. #### Who This Is For - B2B consultants and agencies who want a few qualified conversations every day instead of a giant cold list. - Founders and operators who need a lightweight outbound system before buying another prospecting tool. - Codex users who want to turn research, scoring, and outreach prep into reusable skills and automations. #### What You Will Build ![Ranked Prospect Report](https://tru-images.b-cdn.net/guide-assets/local-231/find-5-extra-prospects-a-day-with-this-agentic-framework/01-ranked-prospect-report.png) You will build a small prospecting system inside Codex. The system will: - define your ICP from an interview - find five qualified prospects from public sources - write a dated prospecting report - keep a running prospects.csv - track stale or overused sources in burned_sources.csv - rank the final five and draft short outreach for the best leads #### What You'll Need - The Codex desktop app - A paid OpenAI account #### Going Further Once the reports are useful, ask Codex to package them into a PDF or email them to you each morning. If you want to connect this to your sales workflow, keep a human approval step in the middle. Codex can draft outreach or prepare CRM notes, but you should review the lead, the source, and the personalization before anything gets sent. ### Step 1 Set Up the Project Folder Create an empty folder and open it in Codex. Ask Codex to create the files it will maintain every day: ```text Set up this prospecting project folder: - reports/ - prospects.csv - burned_sources.csv - source_queue.md Do not prospect yet. ``` ![Prospecting Project Files](https://tru-images.b-cdn.net/guide-assets/local-231/find-5-extra-prospects-a-day-with-this-agentic-framework/02-prospecting-project-files.png) Use reports/ for the daily Markdown reports, prospects.csv for accepted prospects, burned_sources.csv for sources that are getting stale, and source_queue.md for fresh places to search next. > Pro tip: Codex will not keep a prospecting project organized automatically unless you tell it exactly which files to maintain. ### Step 2 Build the ICP Brief Before Codex researches anything, have it interview you. Use this prompt: ```text Here is my company, offer, and URL: - company: [COMPANY] - offer: [OFFER] - URL: [URL] Interview me about my ICP and sales process before finding prospects. When the interview is done, create a reusable prospecting brief with: - ideal company profile - ideal buyer roles - buying signals - pain points - hard disqualifiers - region or market focus ``` ![Prospecting Brief](https://tru-images.b-cdn.net/guide-assets/local-231/find-5-extra-prospects-a-day-with-this-agentic-framework/03-prospecting-brief.png) This is where the rest of the system gets sharper. If the ICP is vague, the daily reports will be vague too. > Pro tip: Add your company URL if Codex does not already know your business. It gives the interview better context and saves you from re-explaining the offer. ### Step 3 Run a Dry Prospecting Pass Now run the prospecting workflow once before turning it into a skill. Use this prompt: ```text Use my ICP brief to find 5 qualified prospects today. Inspect sources first. Build a longlist. Reject weak matches. Return the best 5. Create a dated report in reports/. Append accepted leads to prospects.csv. Update burned_sources.csv. ``` The best output should explain why each company fits, what public source supports the fit, and what buying signal made the lead worth keeping. In the recording, Codex used public job posts and company pages to look for RevOps pain. That worked because the offer was a Revenue Operations Sprint, and job posts often reveal CRM, reporting, routing, and handoff problems. > Pro tip: Do the dry run before creating the skill. There is no point automating a list that is noisy or missing the real buying signal. ### Step 4 Turn the Finder Into a Skill Once the dry run looks useful, ask Codex to turn the process into a reusable skill. Use this: ```text Create a Daily Prospect Finder skill from the dry run. The skill should: - use the ICP brief - inspect sources before recommending leads - return exactly 5 qualified prospects - reject weak matches - write a dated report in reports/ - append accepted leads to prospects.csv - update burned_sources.csv ``` Then schedule it to run every weekday. Keep this automation focused. It should find and qualify prospects. It should not draft outreach yet. ### Step 5 Rank the Final Five and Draft Outreach Create a second skill for the follow-up work. Use this: ```text Use today's final 5 prospects to prepare the next outreach step. Return: - ranked priority order - why each lead is ranked there - one personalization angle per lead - short outreach for the top 1 to 2 leads - anything that needs human verification Do not invent private facts, recent events, or metrics. ``` ![Outreach Draft](https://tru-images.b-cdn.net/guide-assets/local-231/find-5-extra-prospects-a-day-with-this-agentic-framework/04-outreach-draft.png) This second skill turns the report into a next action. You are not just collecting names. You are deciding who to contact first and what to say. Schedule this after the finder runs. In the recording, the finder ran first, then the ranking and outreach prep ran after it. ### Step 6 Add a Source Refresh Automation Prospecting systems get stale when they keep searching the same public lists. Add a weekly automation that refreshes the source queue: ```text Once a week, review source_queue.md, burned_sources.csv, and the current ICP. Find fresh public sources or source categories to test next week. Update source_queue.md. Do not run prospecting in this automation. ``` This is what keeps the system alive. The daily finder searches. The weekly refresher keeps the search surface from getting exhausted. ### Guide: Find Dozens of Free AI Tools With Google Labs > Learn how to use free AI tools in Google Labs to build marketing content for your business. - Canonical page: https://app.therundown.ai/guides/find-dozens-of-free-ai-tools-with-google-labs - Machine-readable page: https://app.therundown.ai/guides/find-dozens-of-free-ai-tools-with-google-labs.md - Difficulty: beginner - Featured tool or company: Google - Categories: AI Foundations & Prompting - Instructors: Billy Howell - Published: 2026-01-09 - Updated: 2026-07-23 ### Overview #### The Rundown Learn how to use free AI tools in Google Labs to build marketing content for your business. You’ll try Pomelli, Google Flow, and ImageFX to create ad creatives, videos, and visual frames, with the option to add custom background music later. Some tools, such as Flow, use daily credits and may prompt you to upgrade after you run out; the Pomelli business analyzer takes about five minutes to run. #### Who This Is Useful For - Marketers who need ad ideas - Businesses that need social posts - Anyone interested in testing AI audio or video generation for free #### What You Will Build You’ll create reusable marketing assets with Google’s AI tools, including ad creatives, video scenes, visual frames, and the foundation for videos with custom background music. - A brand analysis in Pomelli - Suggested marketing campaigns and downloadable ad creatives - AI-generated video scenes in Google Flow - Styled image frames from ImageFX - Optional custom audio using MusicFX ![002_biz_analyzer](https://tru-images.b-cdn.net/guide-assets/69/6961445edcad2de47fcdcd5b/002-biz-analyzer.png) #### What You Need - A free Google account - A business or brand website URL for Pomelli - Optional: enough daily credits for tools like Flow #### Going Further - Use MusicFX from the ImageFX dropdown to generate music with a similar interface. - Combine Flow and MusicFX to create videos with custom audio. - Generate audio for ads you created earlier with Pomelli. ### Step 1: Open Google Labs Navigate to labs.google/experiments. ### Step 2: Analyze your business with Pomelli Copy your business or brand website URL. Go to the Pomelli tool in Google Labs. Paste your URL into the analyzer and let it run for about five minutes. You should get a brand analysis similar to this: ![002_biz_analyzer](https://tru-images.b-cdn.net/guide-assets/69/6961445edcad2de47fcdcd5b/002-biz-analyzer.png) Correct anything that does not match your brand. Click Campaigns in the top-left hamburger menu to see suggested marketing campaigns for your brand. Open a campaign to download ad creatives or generate new ones in the same style. ![002_pom_output](https://tru-images.b-cdn.net/guide-assets/69/6961448bd566f256e3084a77/002-pom-output.png) > Pro tip: Click Animate on a creative you like to generate a video ad or story post. Note: You can only have one business analyzed in Pomelli at a time. ### Step 3: Create a video scene with Google Flow - Navigate to labs.google/flow. - Select “Create Image” in the prompt dropdown. ![003_google_flow](https://tru-images.b-cdn.net/guide-assets/69/696144dbd566f256e3084b29/003-google-flow.png) Describe the starting frame of a scene. Try this example: ```text Beef Jerky brand called Adventure Strips on a picnic table ``` Go to “Frames to Video” in the dropdown. Use the first plus button to select the generated picture as the starting frame. Flow will create a scene based on that starting frame. You can also select an end frame. > Pro tip: Flow is a great tool for creating video transitions between two clips. ### Step 4: Experiment with styles in ImageFX Use ImageFX to practice Nano Banana prompting and explore suggested styles. Every style you add becomes a dropdown with related options. ![004_imagefx](https://tru-images.b-cdn.net/guide-assets/69/696144fbd566f256e3084b73/004-imagefx.png) Lock the seed so your outputs stay as consistent as possible while you experiment with different styles. > Pro tip: Use ImageFX to quickly iterate on frames you can drop into Flow. ### Resources - [Intro to Google Labs PDF Guide](https://tru-images.b-cdn.net/guide-assets/69/696145bfd566f256e3084c6d/99-intro-google-labs-pdf-guide.pdf) ### Pro tip > Use ImageFX to quickly iterate on frames you can drop into Flow. ### Guide: Find Your Winning Ad Angles With Claude and Meta Data > In this guide, you will learn how to use Claude Cowork to audit the angles in your Meta ads. - Canonical page: https://app.therundown.ai/guides/find-your-winning-ad-angles-with-claude-and-meta-data - Machine-readable page: https://app.therundown.ai/guides/find-your-winning-ad-angles-with-claude-and-meta-data.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Marketing & Growth, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-07-13 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Claude Cowork to audit the angles in your Meta ads. It will analyze your ad data alongside your live Ad Library, then create a report you can use to decide what to test next. #### Who This Is Useful For - Performance marketers who need to turn a Meta export into a focused creative brief. - Creative strategists who want evidence for which hooks, promises, and visual proof to develop next. - Anyone running ecommerce Meta ads who wants a repeatable review instead of another spreadsheet read-through. #### What You Will Build You will end with a self-contained HTML angle report: the strongest creative patterns, the ads and placements that support them, the high-click/low-purchase traps to avoid, and three next tests with clear decision rules. ![Claude Cowork with the finished Meta ad angle report](https://tru-images.b-cdn.net/guide-assets/local-340/find-your-winning-ad-angles-with-claude-and-meta-data/01-claude-cowork-angle-report.png) Use the report to turn a comparable slice of campaign data into a smaller set of evidence-backed creative hypotheses. #### What You Need To Get Started - A Claude account with access to Cowork. - A CSV export from Meta at the Ads level. - Your brand’s public Meta Ad Library (https://www.facebook.com/ads/library/) link. - One completed, comparable reporting period: the same objective, offer, attribution setting, and currency. #### Going Further Make this a recurring creative-intelligence workflow. Save the project instruction and report prompt, then add each new export to a dated folder. Once you trust the review, schedule Cowork to prepare the report after the CSV arrives—but keep a person responsible for validating the evidence and deciding on any campaign changes. Over time, compare the same angle by placement, audience, and offer. The useful pattern is not “this ad won once.” It is a creative structure that continues to earn purchase efficiency across enough comparable tests to deserve more production. If you want to explore broader ad-platform audit workflows, check out AgriciDaniel/claude-ads (https://github.com/AgriciDaniel/claude-ads). It is an open-source collection of audit skills covering Meta alongside Google, YouTube, LinkedIn, TikTok, Microsoft, Apple, Amazon, and other paid-ad workflows. ### Step 1: Export One Comparable Meta Ads CSV Start with one slice of your account that you can fairly compare. Export at the Ads level, not Campaign or Ad Set level, so each row points to one creative. For a purchase-focused ecommerce review, include the fields that let Claude identify the ad, understand where it ran, and judge the business outcome: - Identity: campaign name, ad set name, ad name, and Ad ID. - Delivery: amount spent, impressions, reach, frequency, CPM, link clicks, link CTR, and CPC. - Funnel outcomes: landing-page views, adds to cart, initiate checkout, purchases, cost per purchase, purchase conversion value, and website purchase ROAS. If you want a Facebook-versus-Instagram read, add the Publisher platform breakdown before you export. If you want a more granular placement read, include Platform position too. Do not export every available column. The useful export is focused enough to audit. More importantly, do not combine campaigns with different objectives, offers, attribution settings, currencies, or date windows in the same review. > Pro tip: Meta’s purchase labels can vary by account setup. Export the purchase event you optimize for, plus its cost, value, and ROAS fields. The exact label matters less than keeping it consistent across the rows you compare. ![Meta Ads performance export with ad-level and placement fields](https://tru-images.b-cdn.net/guide-assets/local-340/find-your-winning-ad-angles-with-claude-and-meta-data/02-meta-ads-csv-export.png) ### Step 2: Create a Cowork Project and Give Claude the Creative Reference Create a new Cowork folder called meta-angle-review. Add the CSV, then paste in your brand’s public Meta Ad Library link. The export tells Claude what happened. The Ad Library gives it a creative reference for active ads: copy, offers, format, and the visual execution it can inspect. Ask Claude to be strict about this match. An old exported row may no longer be active, and a similar-looking ad is not necessarily the same one. Use this project instruction before you begin the review. It keeps the analysis grounded instead of letting Claude fill gaps with plausible-sounding creative interpretations. ```text You are analyzing a Meta Ads creative review for [BRAND]. Treat the performance CSV as the source of truth for delivery and purchase outcomes. Treat the public Meta Ad Library as the source of truth for active-ad copy and visual context. Match an exported ad only when the ad name, offer, or copy makes the match clear. Label every export row as matched, ambiguous, or unmatched. Do not describe creative for ambiguous or unmatched rows. Use purchases, cost per purchase, purchase conversion value, and website purchase ROAS as the primary evidence. Use CTR only as a diagnostic. Keep source files unchanged. Do not claim that an angle caused performance or will win in the future. ``` ![Dew Drop mock Ad Library used as the creative reference](https://tru-images.b-cdn.net/guide-assets/local-340/find-your-winning-ad-angles-with-claude-and-meta-data/03-dew-drop-ad-library-reference.png) ### Step 3: Ask Claude for the Angle Report Now give Claude one clear analysis job. This prompt asks for the report you will use with your creative team, rather than a loose summary of top ads. ```text Analyze the Meta performance CSV in this project and open the provided Meta Ad Library link. Keep the source CSV unchanged. First, flag non-comparable rows or thin evidence. For every export row, label the creative match as matched, ambiguous, or unmatched. Do not describe a creative for ambiguous or unmatched rows. Find repeated ad angles only from matched rows when the hook, promise, or visual treatment is genuinely similar. Use purchases, cost per purchase, purchase conversion value, and ROAS as the primary evidence; use CTR only as a diagnostic. Create a self-contained visual report called meta-ad-angle-report.html. Include: - an executive summary; - angle clusters with supporting ads and placements; - performance ranges, confidence, and likely confounders; - misleading high-CTR/low-purchase examples, if present; and - three distinct next concepts, each with a hook, visual treatment, hypothesis, primary metric, and pause-or-remix rule. Clearly separate observed evidence from hypotheses. Do not claim an angle caused performance or will win. ``` When the report opens, review the evidence behind one conclusion before you act on it. A useful angle is more specific than “video” or “UGC.” It should name the repeatable combination that the creative team can actually remake: for example, a problem-first hook plus a Day 1-to-Day 7 proof structure, or a customer result tied to one recognizable routine moment. > Pro tip: A high CTR is not proof that an angle is working. If the ad attracts clicks but has weak purchases, CPA, or ROAS, treat it as a diagnostic: the opener may be strong while the promise, product fit, or landing-page expectation is off. ### Step 4: Turn the Findings Into Three Testable Creative Briefs Keep the final output narrow. Ask Claude to package only three next tests, each based on a different observed pattern. Each test should tell a copywriter or designer what to make and how you will decide whether it earned another iteration. Use this follow-up: ```text Using only the matched evidence in the report, create next-creative-tests.md with three test cards. For each card, include: - the angle to test; - a distinct hook; - the visual treatment and proof moment; - the hypothesis; - the primary purchase metric; and - a pause-or-remix rule. Do not repeat the same concept three times. Keep the recommendations proportional to the evidence and flag assumptions that still need validation. ``` Then run the same review on a CSV filtered to only Instagram or only Facebook. That placement-specific pass helps you isolate where an angle is worth adapting, instead of assuming the same execution belongs everywhere. ### Guide: Generate 100 Pain-Point Ad Hooks From Customer Feedback With AI > In this guide, you will learn how to turn customer feedback into fresh ad angles in bulk with ChatGPT Work. - Canonical page: https://app.therundown.ai/guides/generate-100-pain-point-ad-hooks-from-customer-feedback-with-ai - Machine-readable page: https://app.therundown.ai/guides/generate-100-pain-point-ad-hooks-from-customer-feedback-with-ai.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2026-08-12 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to turn customer feedback into fresh ad angles in bulk with ChatGPT Work. You will create two CSVs, generate visual ads, and turn your best process into a simple reusable app. #### Who This Is Useful For - Marketers who want ad ideas grounded in the language customers already use - Founders and agencies that need to turn reviews, support messages, or social comments into repeatable creative work - Anyone who wants to test an AI workflow manually before turning it into an automation #### What You Will Build You will build a local ad factory that turns customer comments into a structured pain-point library, ad copy, image prompts, and visual mockups. The same workflow can process a small test batch or scale across dozens or hundreds of comments. ![A local Source to Signal app that turns customer comments into evidence-backed ad concepts](https://tru-images.b-cdn.net/guide-assets/local-385/generate-100-pain-point-ad-hooks-from-customer-feedback-with-ai/01-source-to-signal-ad-factory.png) This local ad factory turns customer language into reviewable concepts and queues the strongest ideas for image generation. #### What You Need To Get Started - An eligible paid ChatGPT plan and workspace with ChatGPT Work (https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex) available - The ChatGPT desktop app for the local-folder workflow shown here - Optional: An OpenRouter API key if you want to run the reusable app #### Going Further Keep the CSVs as the durable layer of the system. You can add new feedback, reuse the image prompts in another tool, or switch models without rebuilding the process from scratch. Because the files and images stay in the local project folder, you can always inspect and reuse them. If the app works well for you, publish it for your team with ChatGPT Sites (https://app.therundown.ai/guides/deploy-a-mini-saas-in-10-minutes-with-chatgpt-sites). This example keeps the workflow in CSVs, so you can publish a first internal version without adding a database. ### Step 1 Create a local ChatGPT Work project Create a new folder on your computer before you open ChatGPT. Give it a clear name such as customer-feedback-ad-factory. This will hold your source CSVs, generated images, app files, and exports. Open the ChatGPT desktop app, select ChatGPT, then switch from Chat to Work. Open the folder you just created as a new project and start a Work chat inside it. ChatGPT Work is designed for longer tasks and finished deliverables, while the project keeps the files and conversation together. If Work is missing, confirm that your paid plan is eligible and ask your workspace admin to check the role's Work permission. The feature is still rolling out across eligible accounts, and local desktop access can appear later than cloud access. > Pro tip: Create the folder yourself before starting the project. You can then use the project's three-dot menu and Reveal in Finder to find every CSV, image, and code file without hunting through ChatGPT's default folders. ### Step 2 Turn customer comments into a pain-point CSV Gather an initial batch of 10–20 comments from places where customers already explain what is frustrating them. Useful sources include: - Product reviews on Google, Trustpilot, or another public review site - Instagram or Facebook comments - Customer support messages - Sales-call notes - An internal Slack channel that contains customer feedback Use only information you are authorized to analyze. Remove unnecessary personal details before uploading private messages. You can paste or upload the feedback yourself. Work can also gather it from an approved source when you are logged in and grant the required permission. Add one example CSV to the project if you already have a preferred structure. Then use this prompt: ```text Create a CSV that gathers pain points about [BRAND OR PRODUCT]. Use the attached example as the structure when one is available. Include these columns: - source - customer or persona - pain point - frustrating moment - exact phrase - current workaround - desired outcome - primary objection Preserve the customer's wording in the exact phrase field. Do not invent details that are not present in the feedback. Start with [10 OR 20] comments from [APPROVED SOURCES]. ``` Review the resulting CSV before moving on. The pain point needs enough detail to support a useful ad idea, and the exact phrase should still sound like the customer rather than polished marketing copy. ![A CSV organizing customer feedback by frustrating moment, exact phrase, desired outcome, and objection](https://tru-images.b-cdn.net/guide-assets/local-385/generate-100-pain-point-ad-hooks-from-customer-feedback-with-ai/02-customer-feedback-csv.png) The first CSV keeps the original customer language beside the pain, desired outcome, and objection. ### Step 3 Create the ad-concept CSV and scaffold the factory Ask Work to create a second CSV instead of modifying the first one. Keeping the evidence and the ad concepts separate makes both files easier to inspect. ```text Create a second CSV that turns each pain-point row into an ad concept. For every row, include: - the source row or ID - pain point - ad headline - CTA - an idea for a square feed image for Facebook or Instagram Keep every concept traceable to its source row. Save this as a separate CSV. ``` Once the concept CSV exists, type /side to open a temporary side chat in the same Work project. Use it to build the reusable folder structure while the main chat keeps working with the customer feedback. ```text Scaffold this folder as a factory for creating ad variations from the pain-point CSV and the ad-concept CSV. Create clear folders for inputs, generated images, and exports. Keep the output tied to the source row so each file can be traced back to the original customer feedback. ``` Work may create the script and folder structure automatically after this instruction. Review the plan and files it proposes. You do not need another prompt for work that is already underway. ### Step 4 Generate a small batch and correct the creative direction Test the workflow on a small batch before asking for 100 outputs. In the main Work chat, use this prompt: ```text Generate the first five ad concepts and create two visual variations of each. Save them in the correct export folder. Give every file a descriptive name that includes its source spreadsheet row or ID so I can match the image to the concept. ``` Work will generate the images one at a time. Open them in the canvas view to compare the batch, comment on an individual image, or select several images for the same revision. ![Several generated ad mockups open together for comparison in ChatGPT](https://tru-images.b-cdn.net/guide-assets/local-385/generate-100-pain-point-ad-hooks-from-customer-feedback-with-ai/03-chatgpt-ad-mockup-review.png) A small batch makes it easy to compare the creative direction and correct weak concepts before scaling it. Look for whether the visual actually communicates the pain point. If it does not, give a direct correction such as: ```text Create five more variations with more visual emphasis on the pain point. Keep the source concept and customer language unchanged. ``` > Pro tip: Automating 100 things at once is usually the worst way to get good output. Do the task manually, fiddle with it until you get what you like, then turn that into a repeatable automation. Repeat the CSV and image batches until you reach the number of usable ideas you need. Treat the larger sheet as a bank of hypotheses, not proof that every concept is ready to run as an ad. ### Step 5 Turn the tested process into a reusable app After you have corrected the manual workflow, Work has enough context to turn it into a simple local app. Ask for the app in the same project so it can reuse the CSV structure, filenames, and creative decisions you already approved. ```text Build a simple local web app that accepts my customer-feedback CSV and an OpenRouter API key. Use the CSV to create pain points, ad angles, hooks, ad copy, and image prompts. Let me review and select concepts before generating image mockups. Save the enriched CSV and generated images into organized local output folders. Make the app easy to use. Let me enter the OpenRouter key in the interface, but do not hard-code it or save it in the CSV, source files, or output folders. ``` Open the app, import the CSV, enter the key in the interface, and process the feedback. The app should automatically create the pain, angle, hook, copy, and image-prompt fields. Select only the concepts you want to visualize, then generate the images. The app reduces the tested workflow to a few clicks. You still review the source language and creative direction before scaling the output. ![A finished food-ad image generated from a customer pain-point concept](https://tru-images.b-cdn.net/guide-assets/local-385/generate-100-pain-point-ad-hooks-from-customer-feedback-with-ai/04-generated-food-ad-concept.png) This is one visual ad concept generated after the customer-feedback workflow was reviewed and refined. ### Guide: Generate an Actionable SEO Audit Using This Strategy (Works with Any LLM) > Learn how to run a full SEO audit on any website using AI, then turn the results into three usable deliverables: a PDF report, a client-ready slide deck, and a prioritized task workbook. - Canonical page: https://app.therundown.ai/guides/generate-an-actionable-seo-audit-using-this-strategy-works-with-any-llm - Machine-readable page: https://app.therundown.ai/guides/generate-an-actionable-seo-audit-using-this-strategy-works-with-any-llm.md - Difficulty: intermediate - Featured tool or company: OpenAI - Categories: Marketing & Growth, Data, Research & Analysis, Sales, Consulting & Customer Success - Instructors: Billy Howell - Published: 2026-03-18 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to run a full SEO audit on any website using AI, then turn the results into three usable deliverables: a PDF report, a client-ready slide deck, and a prioritized task workbook. The workflow was tested in Manus, Claude, ChatGPT, Gemini, and NotebookLM. #### Who This Is Useful For - Small business owners and freelancers who know their website has SEO problems but do not want to pay for dedicated SEO tools to find them - Marketing managers and consultants who need to present SEO findings to a client or stakeholder - Anyone managing a website who wants a quick way to find and prioritize high-impact SEO fixes without learning dedicated SEO tools #### What You Will Build You will create a reusable SEO audit package from one website audit. - A full PDF report covering technical SEO, on-page optimization, content quality, and backlinks - A client-ready slide deck - A prioritized Excel workbook with every issue, how to fix it, and which page it affects ![image-2129](https://tru-images.b-cdn.net/guide-assets/69/69bb253f5e050ff0ab3feeae/image-2129.png) #### What You Need - An account on at least one of: Manus, Claude, ChatGPT, or NotebookLM - A website URL to audit, such as your site, a client site, or a competitor site #### Going Further - Break the workbook into separate sections or CSVs: Overview and Scores, Technical SEO Tasks, On-Page Optimization, Content Refresh, and Backlink Opportunities - Save the process as a reusable skill or custom GPT so you can rerun the workflow with a new website URL - If you have dedicated SEO tools like SE Ranking or Ahrefs, connect them so the AI can add analysis and presentation on top of those reports ### Step 1: Run the audit Open whichever AI tool you have access to and paste in the audit prompt with your website URL. ```text Perform a comprehensive SEO audit of [your website URL]. Analyze technical SEO, on-page optimization, content quality, and backlink profile. Give me the top 10 highest-impact fixes ranked by effort vs. results. Cite your data sources for each section. Export the full audit as a PDF report with sections for each category. ``` Make sure web browsing is enabled so the LLM can look up data like keywords and domain ratings. ![image-2130](https://tru-images.b-cdn.net/guide-assets/69/69bb254d5e050ff0ab3ff738/image-2130.png) ### Step 2: Turn it into a slide deck Review the PDF report, then ask the AI tool to turn it into a presentation. ```text Turn this audit into a slide deck I can present to a client or stakeholder. Keep it under 10 slides. Include an executive summary, top issues by category, and a recommended action plan with priorities. ``` If you are using Claude or ChatGPT, make sure you have the PowerPoint skill installed. In NotebookLM, you can use the slide deck option when it is done. ![image-2131](https://tru-images.b-cdn.net/guide-assets/69/69bb255e5e050ff0ab3fffc2/image-2131.png) ### Step 3: Build the SEO workbook The deck is for presenting. This spreadsheet is for doing the work. ```text Create an SEO workbook as a CSV with every issue from the audit. Columns: Issue, Category, Priority (High / Medium / Low), How to Fix, Affected Page URL. Sort by priority. ``` Open the CSV in Excel or Google Sheets. You will have a working task list with every issue, how to fix it, which page it affects, and how urgent it is. You can import this into any project management tool, hand it to a developer, or work through it yourself. ![image-2132](https://tru-images.b-cdn.net/guide-assets/69/69bb256c5e050ff0ab400847/image-2132.png) ### Step 4: Create a reusable skill Once you are done, tell the tool you are using to create a skill or custom GPT based on the process you just ran. That way, next time you can drop in a website URL and have it run the full audit, slide deck, and workbook flow automatically. ### Pro tip > After the audit finishes, ask the tool: “List every tool, connection, and data source you used to complete this audit.” This helps you catch hallucinations and learn what tools you might need to install to improve the report. ### Guide: Generate an Agent-Native CLI From Any API or Website > In this guide, you will learn how to use Printing Press (https://printingpress. - Canonical page: https://app.therundown.ai/guides/generate-an-agent-native-cli-from-any-api-or-website - Machine-readable page: https://app.therundown.ai/guides/generate-an-agent-native-cli-from-any-api-or-website.md - Difficulty: beginner - Featured tool or company: Printing Press - Categories: Coding & App Development, Automation & Agents - Instructors: Billy Howell - Published: 2026-05-21 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to use Printing Press (https://printingpress.dev/) to download agent-friendly tools for hard to use APIs like Google Flights or ESPN. You will also learn to build your own CLI tools to access any site with an API. We use Codex but this works great with OpenClaw and Claude Code too. #### Who This Is For - Developers and AI operators who want agents to use websites and APIs without long browser sessions. - Codex, OpenClaw, Hermes, or Claude Code users who want better local tools for repeated workflows. - Anyone who wants to turn a data-rich website or API into reusable read-only commands. #### What You Will Build You will install the Printing Press starter kit, have your agent test a working tool like ESPN, then install the Printing Press binary so your agent can generate its own CLI commands. ![ESPN CLI returning live NBA scores](https://tru-images.b-cdn.net/guide-assets/local-217/generate-an-agent-native-cli-from-any-api-or-website/01-espn-cli-working.png) #### What You Need - Node and npx for the starter kit install. - Go for the Printing Press binary install. - An AI coding agent such as Codex, OpenClaw, Hermes, or Claude Code. - A simple website, API docs page, or OpenAPI spec to test after setup. #### Going Further If your agent struggles to use a specific MCP or integration, you can tell it to build its own commands with Printing Press. You just need a site with useful data, API documentation, or an API key. The bigger idea is to stop making agents rediscover the same surface area every time. If your agent needs the same website or API every week, turn that system into a small local tool. ### Step 1: Start With Printing Press Open the Printing Press website (https://printingpress.dev/) and skim the examples before you install anything. The point is not just that Printing Press creates CLIs. The useful part is that it creates agent-friendly tools: compact commands, help output, and repeatable workflows your agent can call without browsing around every time. You can also read the Printing Press GitHub repo (https://github.com/mvanhorn/cli-printing-press) for install notes and implementation details. Keep that link handy. In our test, giving the agent the relevant GitHub page helped it recover when it did not know how a starter skill was supposed to be configured. ### Step 2: Install the Starter Kit Install the starter kit first. This gives your agent working examples before you ask it to generate anything new. Run: ```text npx -y @mvanhorn/printing-press install starter-pack ``` ![Printing Press starter kit install command](https://tru-images.b-cdn.net/guide-assets/local-217/generate-an-agent-native-cli-from-any-api-or-website/02-starter-kit-install-command.png) The starter kit includes example tools such as ESPN and flight-goat. Those are useful because they let you prove the setup works before you start designing your own commands. > Pro tip: do not debug installation and tool generation at the same time. First prove the starter kit works. Then move on to generating your own tools. ### Step 3: Have Your Agent Test a Starter Tool Now ask your agent to inspect the starter kit and try one safe read-only command. Use this prompt: ```text Inspect the starter kit. List the tools, read their help output, and try one safe read-only command from ESPN or flight-goat. Tell me which tool worked and which command you ran. ``` For ESPN, the agent should be able to inspect the CLI help and run a command that pulls live or recent sports data. In our Codex run, the ESPN CLI returned completed NBA playoff games from ESPN data. ![ESPN CLI scoreboard commands](https://tru-images.b-cdn.net/guide-assets/local-217/generate-an-agent-native-cli-from-any-api-or-website/03-espn-cli-scoreboard-commands.png) If the agent gets stuck, point it back to the Printing Press website or GitHub repo and ask it to inspect the specific skill. Many generated skills include a doctor or verification command. That is the first thing to run when a starter tool does not work. > Pro tip: make sure your agent has access to GitHub. Printing Press tools often need the repo context, install instructions, or skill files to finish setup cleanly. ### Step 4: Install the Printing Press Binary Once the starter kit works, install the Printing Press binary so your agent can generate its own tools: ```text go install github.com/mvanhorn/cli-printing-press/v4/cmd/printing-press@latest ``` Then ask your agent to confirm the binary is available: ```text Confirm printing-press is installed. ``` Do not skip this check. If the binary is not available, your agent should fix that before it tries to generate a new CLI. ### Step 5: Generate Commands for Your Own Website Start with a simple site that has a lot of public data. A docs site, public catalog, sports site, recipe site, or API-backed directory is easier than a complex app with private endpoints. Use this prompt: ```text Use Printing Press to generate read-only commands for [WEBSITE]. Propose the useful commands first, then generate and test them. ``` Ask for read-only commands first. Good starter commands are usually search, lookup, compare, export, or summarize. Save write actions for later, after you know the generated tool is safe and accurate. If the site has an API key, docs page, or OpenAPI spec, give that to your agent. If it does not, ask the agent to look for public API endpoints and avoid private endpoints unless you explicitly provide credentials. ### Guide: Generate Editable Infographics in 15 Minutes With AI > This guide teaches you how to turn raw research into an editable infographic using Perplexity, Gemini, and Canva Magic Layers. - Canonical page: https://app.therundown.ai/guides/generate-editable-infographics-in-15-minutes-with-ai - Machine-readable page: https://app.therundown.ai/guides/generate-editable-infographics-in-15-minutes-with-ai.md - Difficulty: intermediate - Featured tool or company: Canva - Categories: Content & Creative, Marketing & Growth, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-04-12 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches you how to turn raw research into an editable infographic using Perplexity, Gemini, and Canva Magic Layers. You will move from a data-heavy topic to a strong first-draft graphic you can clean up and ship instead of rebuilding manually. #### Who This Is Useful For - Marketers and content creators who want to turn reports, case studies, or campaign recaps into shareable graphics - Operators and analysts who need to turn messy data into something people will actually read - Teams that already use AI for research or image generation and want a faster way to move from insight to a finished visual #### What You Will Build You will build a repeatable workflow that turns a topic, memo, report, or dataset into an infographic you can actually edit in Canva. The final output is not a perfect AI image; it is a strong first draft you can clean up fast. - Infographic concepts based on the strongest storylines in your data - AI image prompts for those concepts - A generated PNG first draft - An editable Canva version separated into layers with Magic Layers #### What You Need - Perplexity or another AI tool for research - Gemini or another AI image generator - Canva Pro for Magic Layers - A topic, report, memo, or dataset you want to turn into an infographic #### Going Further - You do not always need to start with fresh research. If you already have a memo, case study, or report, feed it into Perplexity or Gemini and ask for an infographic prompt based on the strongest data. That lets you skip the research step and go straight from existing work to a graphic you can edit. ### Step 1: Research the story, not just the facts Start in your favorite AI research tool. The source workflow used Perplexity's deep research. Tell the tool that the end goal is an infographic instead of asking for a generic summary. Ask for outliers, comparisons, and narrative angles. ```text Do deep research on [your topic]. Look for outliers and narrative angles. Our end goal is four infographic concepts recapping the most interesting data trends. ``` ![CleanShot 2026-04-12 at 13.28.52@2x](https://tru-images.b-cdn.net/guide-assets/69/69dbd68443c43aefee672c23/cleanshot-2026-04-12-at-13-28-52-2x.png) In the source example, the topic was Daytona 500 data. The goal was not to collect every stat; it was to find the strongest storylines that could anchor a graphic. ### Step 2: Turn the research into image prompts Once the research is done, ask the tool to turn each concept into an image prompt. ```text Give me a prompt for each concept that I can paste into an AI image generator. Make the data easy to copy and paste. ``` ![CleanShot 2026-04-12 at 13.30.25@2x](https://tru-images.b-cdn.net/guide-assets/69/69dbd6ba43c43aefee672c8d/cleanshot-2026-04-12-at-13-30-25-2x.png) This gives you a prompt set with the story and the data already packed in. Some prompts will be usable right away. Others will be too long or too cramped. Treat them as a starting point. ### Step 3: Generate the first visual in Gemini Paste one of the prompts into Gemini and generate a few infographic options. The source workflow also tested Midjourney, but Gemini handled the data-heavy prompts better. Download the strongest result as a PNG. If you use another image tool, the same rule applies: you only need a clean first pass you can edit later. ![CleanShot 2026-04-12 at 13.31.49@2x](https://tru-images.b-cdn.net/guide-assets/69/69dbd712be5df7e60be0475f/cleanshot-2026-04-12-at-13-31-49-2x.png) > Pro tip: Text-based infographics tend to work best. Once the image has lots of tiny charts, thin lines, or crowded labels, the next step gets harder. ### Step 4: Upload the image to Canva and run Magic Layers Open a new Canva project, upload the PNG, select the image, then go to: - Edit - Magic Studio - Magic Layers Canva will scan the image and separate it into editable parts. In the source test, each image took about 30 to 60 seconds, and Canva could process more than one at once. ### Step 5: Clean up the layers Now fix the parts AI got wrong. This is where the workflow becomes useful. The source workflow found three common cleanup jobs: - Correct misspelled names or labels - Replace cramped text that was too dense - Swap in real product photos or logos when the generator guessed wrong Text-heavy infographics came out closest to ready. Mixed text-and-image layouts usually needed a few replacements. Dense charts were the easiest way to break Magic Layers, so keep the concept simple. ### Pro tip > Tell the research tool what the final format is going to be. “Infographic concepts” is much better than “data summary.” ### Guide: Get a Deal On World Cup Tickets With AI Price Tracking (ChatGPT) > In this guide, you will learn how to connect ticketing and travel apps in ChatGPT, run a one-off World Cup ticket search, turn that search into a recurring task, and use Agent Mode when you want a more advanced automation. - Canonical page: https://app.therundown.ai/guides/get-a-deal-on-world-cup-tickets-with-ai-price-tracking-chatgpt - Machine-readable page: https://app.therundown.ai/guides/get-a-deal-on-world-cup-tickets-with-ai-price-tracking-chatgpt.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: Automation & Agents, Productivity & Knowledge Work, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-06-24 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to connect ticketing and travel apps in ChatGPT, run a one-off World Cup ticket search, turn that search into a recurring task, and use Agent Mode when you want a more advanced automation. #### Who This Is Useful For - Fans trying to find World Cup tickets without refreshing five marketplaces every day. - Travelers who care about the all-in price, not just the cheapest ticket listing. - Anyone who wants a repeatable ChatGPT workflow for ticketed events, concerts, conferences, or sports. - ChatGPT users who want to understand when to use apps, scheduled tasks, and Agent Mode. #### What You Will Build You will build a recurring ChatGPT ticket scout that checks World Cup tickets every morning and reports the best options based on your budget, travel constraints, and preferred sources. The finished result should look like a ranked table: match, date, venue, ticket price, travel cost, all-in total, source links, and the next human action. ![ranked ticket scout results](https://tru-images.b-cdn.net/guide-assets/local-283/get-a-deal-on-world-cup-tickets-with-ai-price-tracking-chatgpt/01-ranked-ticket-scout-results.png) The basic flow is: ```text Connect the apps -> run a one-off search -> automate the search -> use Agent Mode for more advanced runs ``` #### What You Need To Get Started - A ChatGPT account. - Any ChatGPT account works for the basic setup. - A Plus account or higher for Agent Mode and advanced automation. - Any version of the ChatGPT app. In this guide, we use the ChatGPT desktop app on Mac. #### Going Further This same workflow works for more than World Cup tickets. Replace the venues, dates, sources, and thresholds, and you can use it for concerts, flights, conferences, or other events where prices change over time. For heavier scouting runs, try the Codex desktop app. If you have a ChatGPT plan, you should have access to Codex. Codex gives the agent more room to work longer, check more sources, and keep a reusable prompt sheet or tracking file on your computer. ### Step 1: Connect The Ticketing And Travel Apps Start by connecting the apps you want ChatGPT to use for one-off searches. Open Settings > Apps > Browse apps, then look for ticketing and travel apps such as: - Ticketmaster - SeatGeek - Vivid Seats - Expedia - Booking.com Some apps may require an account. Others may show a Connect without account option. ![optional ticket apps](https://tru-images.b-cdn.net/guide-assets/local-283/get-a-deal-on-world-cup-tickets-with-ai-price-tracking-chatgpt/02-optional-ticket-apps.png) You do not need every possible ticket marketplace. Start with the sources you actually trust, then add more later if the results are too thin. > Pro tip: If you want ChatGPT to use a connected app inside a chat, type the app name with @, like @SeatGeek, then select it from the menu. ### Step 2: Run A One-Off Ticket Search Before you automate anything, ask ChatGPT to run the search once. This lets you see whether the connected apps return useful results and whether your prompt has enough detail. Use a simple version first: ```text Check SeatGeek, Ticketmaster, and Vivid Seats for the best World Cup ticket deals for 2 people near Philadelphia or New York/New Jersey. I am starting from Washington, DC. Include train or flight context when it changes the total cost. Return a ranked table with: Match | Date | Venue | Ticket source | Ticket price for 2 | Travel cost | All-in total | Link | Next action ``` ![seat geek one off search](https://tru-images.b-cdn.net/guide-assets/local-283/get-a-deal-on-world-cup-tickets-with-ai-price-tracking-chatgpt/03-seat-geek-one-off-search.png) If the first answer is too broad, tighten the brief. Give ChatGPT the information a real assistant would need: - Where you are based. - How many tickets you need. - Which venues or cities you will travel to. - Which dates or match windows you care about. - Which ticket sources to check. - Your max budget. - Your alert thresholds. Here is a stronger version: ```text You are my World Cup 2026 ticket scout. Goal: Find realistic World Cup ticket deals for 2 people departing from Washington, DC. Include: - Philadelphia or New York/New Jersey matches that are realistic day trips by car or train - group stage matches first - other June matches only if the deal is exceptional Budget: - hard cap: $1,500 total all-in for 2 people - worth watching: under $1,100 all-in - check now: under $750 all-in Sources: - Ticketmaster - SeatGeek - Vivid Seats - TickPick - SeatPick - Expedia only if travel or lodging context is needed Return a ranked table: Match | Date | Venue | Ticket source | Ticket price for 2 | Travel cost | All-in total | Status | Link | Next human action Rules: - multiply ticket price by 2 - label fees as included, excluded, or unclear - include travel and transit caveats - cite direct links - do not log in, join queues, bypass verification, cart, reserve, buy, transfer, or resell tickets - if a page requires login, queue access, verification, carting, or checkout, mark it manual review ``` Click the links ChatGPT returns. You are checking that the sources open to the right match, the prices look real, and the all-in math makes sense. ### Step 3: Turn The Search Into A Recurring Task Once the one-off search gives you a result you would actually use, ask ChatGPT to automate it. Use the simple version: ```text Turn this into a recurring task every morning at 8:00 a.m. ``` Or use the more explicit version: ```text Turn the World Cup ticket search we just ran into a recurring task. Schedule: Every morning at 8:00 a.m. What to do: - repeat the approved ticket scout workflow - check Ticketmaster, SeatGeek, Vivid Seats, TickPick, and SeatPick - include travel and transit context when needed - return the same ranked table format - compare results against my thresholds Notify me only if: - a deal is under $1,100 all-in for 2 people - a deal is under $750 all-in for 2 people - a tracked match drops by more than $100 - fees, availability, or transit uncertainty makes a row manual review Safety: Do not log in, join queues, bypass verification, cart, reserve, buy, transfer, or resell tickets. ``` Once the task exists, open it and confirm the schedule, prompt, and notification settings. ![daily scheduled task](https://tru-images.b-cdn.net/guide-assets/local-283/get-a-deal-on-world-cup-tickets-with-ai-price-tracking-chatgpt/04-daily-scheduled-task.png) This is the main product wrinkle: you can create and edit tasks directly in chat, but the visual Scheduled Tasks list is easiest to find in the ChatGPT web or mobile app. If you are using the desktop app and cannot find the full task settings, switch to web or mobile. You can also tell ChatGPT to edit the task later: ```text Update my recurring World Cup ticket scout task. Keep it running every morning at 8:00 a.m., but only notify me for deals under $900 all-in for 2 people. Keep the direct source links and manual-review safety rules. ``` ### Step 4: Use Agent Mode For More Advanced Automations The app-based version is good for a simple recurring search. Use Agent Mode when you want ChatGPT to do a deeper comparison across sources, open more pages, check travel options, and come back with a more complete ranking. Open a new thread in ChatGPT, turn on Agent Mode from the tools menu, or type /agent in the composer if that shortcut is available in your account. Then run the same scout prompt from Step 2. Agent Mode is better suited for tasks like: - Comparing multiple ticket marketplaces in one run. - Checking flights, trains, buses, and driving time against the ticket price. - Explaining why one option is a better value than another. - Returning a fuller table with caveats, source links, and manual-review notes. After Agent Mode finishes, verify the result by clicking the source links. If the output is better than your original app-based task, toggle Agent Mode off and say: ```text Turn the task you just did into a recurring task every morning at 8:00 a.m. ``` The pattern is the same. Run the search once, make sure the results are good, then automate the version that worked. ### Guide: Get a Personal Newspaper Written by Claude Every Morning > This guide shows you how to turn daily updates from Claude-connected work tools into a personalized Morning Edition: one ranked brief instead of bouncing between Slack, Gmail, Notion, and your calendar. - Canonical page: https://app.therundown.ai/guides/get-a-personal-newspaper-written-by-claude-every-morning - Machine-readable page: https://app.therundown.ai/guides/get-a-personal-newspaper-written-by-claude-every-morning.md - Difficulty: intermediate - Featured tool or company: Notion - Categories: Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-04-23 - Updated: 2026-07-23 ### Overview #### The Rundown This guide shows you how to turn daily updates from Claude-connected work tools into a personalized Morning Edition: one ranked brief instead of bouncing between Slack, Gmail, Notion, and your calendar. You will produce a static newspaper-style daily brief, refine it, turn it into a reusable skill, and schedule it once it works. Plan for a few minutes to refine the first draft before you automate it. #### Who This Is Useful For - Operators and founders who start the day in five different tabs and want one clean brief - Managers and team leads who need action items, meeting prep, and internal updates in one place - Anyone already running agents or reports who wants Claude to assemble the final edition instead of doing all the research itself #### What You Will Build You will build a static newspaper-style daily brief that pulls together your most important updates, then becomes a reusable skill and scheduled task. - A front page with inbox, Notion, calendar, and Slack summaries - An inbox section that turns Gmail updates into a readable dispatch with clear priorities - A research section that folds in AI reports or task lists from Notion - An automation layer that schedules the Morning Edition to run every morning in Claude CoWork ![image-2181](https://tru-images.b-cdn.net/guide-assets/69/69ea82c65bd3a97ffd7d2da2/image-2181.png) ![image-2182](https://tru-images.b-cdn.net/guide-assets/69/69ea82d85bd3a97ffd7d3659/image-2182.png) ![image-2183](https://tru-images.b-cdn.net/guide-assets/69/69ea82ec02b183ca13a441f5/image-2183.png) #### What You Need - Claude or Claude CoWork - Slack, Notion, Gmail, and Calendar connected if you want the full version - Optional: a centralized task list or daily AI reports database in Notion - A few minutes to refine the first draft before you automate it #### Going Further - The best version of this workflow is not one model doing everything from scratch. Use other agents and automations to gather outside news, AI reports, customer notes, or workflow updates first, then drop those into a Notion database so Claude can read the reports and assemble the final paper. - Let other systems do the research - Let Claude do the editing - Separate analysis from assembly - That setup is less fragile, cheaper to run, and easier to improve over time. ### Step 1 Connect your sources Open Claude or Claude CoWork and make sure your integrations are connected. - Slack - Notion - Gmail - Calendar If you already keep a central task list or a daily AI reports database in Notion, connect that too. That is where this gets better: Claude stops acting like a researcher and starts acting like an editor. ### Step 2 Create the first Morning Edition Prompt Claude to make the paper. Use a prompt close to this: ```text Take a stab at making me a quick morning newspaper called Morning Edition. It should update me on: - my Slack updates - my Notion updates - my Gmail updates - my calendar updates Use the last 24 hours. Make it a static newspaper, not a live artifact. Format it like a newspaper with top stories, action items, and schedule prep. ``` The “last 24 hours” instruction keeps the edition focused. The other important instruction is “static newspaper, not a live artifact.” You want something you can keep, review, and compare over time. > Pro tip: If Claude gives everything equal weight, ask it to make stronger editorial choices. Action items and calendar prep should beat random reading. ### Step 3 Refine it and turn it into a skill Once Claude gives you a first version, refine it before saving the workflow. - Review the layout - Cut weak sections - Fix the order - Make the important items more prominent Then tell Claude: ```text Turn this process into a skill. ``` Claude will package the workflow so you can run it again without rebuilding the prompt every time. > Pro tip: Do not waste Opus on this if your reports are already gathered upstream. In the demo, the better move was Sonnet or Haiku, because the hard work had already been done. ### Step 4 Schedule it in Claude CoWork After the skill is saved, go to Claude CoWork and create a recurring task. Your instruction can be simple: ```text Run /morning-edition every morning. ``` If you want delivery built in, use: ```text Run /morning-edition and then email it to me at [your email]. ``` ![image-2184](https://tru-images.b-cdn.net/guide-assets/69/69ea830702b183ca13a44abc/image-2184.png) If Slack or email is connected, you can have the edition delivered there. Otherwise, it will show up in a new Claude CoWork thread. Once the skill is stable, the morning brief becomes a product, not another prompt you have to remember. ### Pro tip > Start with connected work tools you already trust. If you are testing with sensitive data, use sample content first. ### Guide: Get Better AI Outputs With This Multiple Choice Technique > Learn a universal multiple-choice prompting strategy for getting better outputs from AI models while reducing unhelpful context and token usage. - Canonical page: https://app.therundown.ai/guides/get-better-ai-outputs-with-this-multiple-choice-technique - Machine-readable page: https://app.therundown.ai/guides/get-better-ai-outputs-with-this-multiple-choice-technique.md - Difficulty: beginner - Featured tool or company: Google - Categories: AI Foundations & Prompting - Instructors: Billy Howell - Published: 2026-01-20 - Updated: 2026-07-23 ### Overview #### The Rundown Learn a universal multiple-choice prompting strategy for getting better outputs from AI models while reducing unhelpful context and token usage. You’ll also learn the basics of tokens and context windows, then use the technique to produce clearer AI-generated concepts. #### Who This Is Useful For - Marketers and designers who generate lots of images - People who want to prompt better - Anyone frustrated with lengthy AI chats that seem to go in circles #### What You Will Build You’ll build a reusable multiple-choice interview prompt structure that helps an AI gather useful context before producing outputs. In the example workflow, you’ll use it to generate four distinct logo concepts. - A goal, task, and next-steps prompt structure - A multiple-choice AI interview for gathering context - Four distinct concept directions generated from the interview answers - A token-saving grid of logo concept options #### What You Need - Access to any chat-based AI or AI image generator - For this tutorial specifically: Nano Banana Pro in AI Studio and Google’s NotebookLM #### Going Further - For more complex workflows like email copy ideas or blog post outlines, copy and paste each concept into a fresh chat and refine it there. - If the interview method gives you strong results, export and reuse the context by asking the AI to extract the context as system instructions in markdown. ### Step 1 Understand tokens and context When you interact with an AI, you use text. As a general rule, every four characters read or generated by an AI equals one token. Tokens are the chunks of text the model processes. If you send a prompt with 40 characters, you will be billed for about 10 tokens, plus however many tokens the AI uses in its response. Input tokens are used when the AI parses what you send. Output tokens are based on the words, images, or videos generated by AI. Output is often pricier depending on the model. Context is the relevant tokens stored in the AI’s “brain” while you’re talking to it. It is text summarizing what you have talked about so far that gets passed to the AI on each new request. The more context you have, the more tokens the AI has to process. The goal of this technique is to minimize unhelpful context so you can improve outputs and reduce token usage. ### Step 2 Break your request into goal, task, and next steps Think of a task you want AI to accomplish. The example in this guide uses AI Studio to create logo concepts for a brand. To minimize unhelpful messages stored in context, break your request into a goal, a task, and next steps. - Goal: Generate a logo for my student-run moving company, “Uni Movers.” - Task: Interview me with 5–10 multiple-choice questions about my brand and vision in order to build context for the task. - Next steps: Create 4 distinct logo concepts. ### Step 3 Send the multiple-choice interview prompt Put your prompt together. The source example recommends writing the goal, task, and next steps conversationally. ```text Your job is to generate a logo for my student-run moving company, “Uni Movers.” Interview me with 5–10 multiple-choice questions about my brand and vision in order to build context for the task. Our goal is to create 4 distinct logo concepts. ``` Send your prompt to your AI. ### Step 4 Answer the AI’s multiple-choice questions You should get back a list of questions with multiple-choice answers. Respond with each choice separated by commas, such as “A, B, A, C…” ![04_multiple_choice](https://tru-images.b-cdn.net/guide-assets/69/696facfb4b339c13c4eacaf6/04-multiple-choice.png) > Why this works: the multiple-choice constraint removes potential for miscommunication. It ensures your response uses terminology that AI will understand. ### Step 5 Generate and refine the concepts The AI should respond with four logo concepts. To save on tokens, you can ask it to generate a 4x4 grid with each logo concept in a separate grid cell. Then, ask it to generate the concepts you like as standalone images. ```text Generate a 4x4 grid with each logo concept in a separate grid cell. ``` ![05_4_logo_options](https://tru-images.b-cdn.net/guide-assets/69/696fad1214e40e844cb21743/05-4-logo-options.jpeg) ### Guide: Get to Inbox Zero With This Claude Prompt > In this guide, you will build a daily email triage workflow using one Claude Cowork scheduled task. - Canonical page: https://app.therundown.ai/guides/get-to-inbox-zero-with-this-claude-prompt - Machine-readable page: https://app.therundown.ai/guides/get-to-inbox-zero-with-this-claude-prompt.md - Difficulty: intermediate - Featured tool or company: Claude - Categories: Productivity & Knowledge Work, Automation & Agents - Instructors: Billy Howell - Published: 2026-04-07 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will build a daily email triage workflow using one Claude Cowork scheduled task. The workflow gives you a sorted inbox each morning, with draft replies waiting for the emails that actually need you. #### Who This Is Useful For - Founders and operators who repeat the same morning inbox cleanup every day - People with lots of newsletter, product, and marketing subscriptions mixed in with real work email - Claude Cowork users who want a practical scheduled task that saves time on day one #### What You Will Build ![image-2167](https://tru-images.b-cdn.net/guide-assets/69/69d56d8e03d6ed46c75f5804/image-2167.png) You will create a recurring Claude Cowork task that scans your unread inbox each morning, sorts every message into one of four buckets, drafts replies for messages that need them, and delivers an interactive HTML report you can check off in seconds. - A clean inbox - A queue of drafts ready to review and send - A clear list of anything that needs a human decision - An interactive HTML triage report with approvals #### What You Need - Claude Desktop app and Pro Plan ($20/month) - The Gmail connector enabled in Claude - A local folder for Cowork to save reports and memory files - One real morning’s worth of unread email to triage #### Going Further - After the core workflow is running, wire Gmail filters to your Claude/* labels. You can have Claude/Archive leave the inbox automatically, Claude/Needs-Response and Claude/Needs-Attention get starred, and Claude/Unsub stay visible as a weekend cleanup queue. ### Step 1 Run one real triage session in regular Claude Before you schedule anything, open regular Claude with the Gmail connector on and run one triage pass while you watch. This helps turn a generic prompt into one that understands your inbox. ```text Generate an interactive email triage report for the last 24 hours of unread email in my inbox. Sort each email into exactly one of these buckets: 1. Needs response 2. Needs attention 3. Archive 4. Archive and unsubscribe For each email, include the sender, subject, a one-line reason for the category, a direct link to the email, and the item number. ``` Review Claude’s triage and give it feedback. Tell it which emails it got wrong, name the senders that always matter, flag newsletters that should always be archived, and call out the product updates you actually want to keep. That feedback teaches Claude the difference between useful system updates, junk promos, teammate messages, and edge cases that only show up in your inbox. > Pro tip: If you want the lazy version, skip the test run entirely and paste the prompt in the next step as-is. ### Step 2 Have Claude write your custom recurring prompt Once Claude has completed one real pass and you have corrected it, ask it to turn the workflow into a reusable prompt. ```text Turn this workflow into a recurring task prompt for my inbox. Include my common senders, archive rules, unsubscribe rules, and which messages should be drafted or flagged. ``` Below is the structure from this process. Replace the bracketed placeholders with your own email address, team domain, tools, and senders that show up most in your inbox. ```text You are running the daily email triage for [your@email.com]. Follow these steps: ## Step 1: Scan Inbox Use `gmail_search_messages` with query `is:unread in:inbox` and maxResults 50. Paginate through all results. ## Step 2: Read Important Threads For emails from @[yourcompany].com teammates, or any email that looks like it needs a personal response, use `gmail_read_thread` to pull full context. ## Step 3: Categorize Every Email **Needs Response** - direct messages from teammates - calendar invites that need accept or decline - real customer or subscriber replies with questions or complaints - security alerts - product updates from tools I actively use **Needs Attention** - anything that needs a personal follow-up but not a drafted reply - billing or account notices that require an action **Archive** - my own newsletter copies - routine recap emails - calendar RSVP responses - auto-replies and out-of-office messages - bounce notifications - receipts and purchase confirmations - self-sent emails **Archive and Unsubscribe** - unsolicited product promos - repeat marketing senders - non-essential promotional senders ## Step 4: Deliver an Interactive HTML Report Do not label, draft, or archive anything yet. Instead, save an HTML file to my Cowork workspace named `triage_YYYY-MM-DD.html` using today's date. The report should include: - a dark theme - a stats bar with counts per category - numbered rows with a checkbox next to each email - Gmail deep links in the format `https://mail.google.com/mail/u/0/#all/{threadId}` - sender, subject, date, and snippet for each email - a suggested action for each Needs Response item - a "Select All" button per category - a "Download Approvals" button that exports a JSON file of the checked items Then stop and wait. Do not take any further actions in Gmail until I come back with the approvals file. ## Step 5: Act On the Approvals File When I upload the approvals JSON back to you, read the checked items and act only on those. For each approved email: - Needs response: create a draft reply with `gmail_create_draft` using the thread ID, then apply the label `Claude/Needs-Response` - Needs attention: apply the label `Claude/Needs-Attention` - Archive: apply the label `Claude/Archive` - Archive and unsubscribe: apply the label `Claude/Unsub` Ignore any row I did not check. Only pause to ask me if a thread looks like phishing or involves money or legal decisions. When you're done, report back with the counts: drafts created, threads labeled, and anything you skipped. ``` ### Step 3 Save it as a Claude Cowork scheduled task Now move into Claude Cowork and make the workflow recurring. Create a local folder called something like “Email Triage.” This becomes the workspace where Cowork saves the daily HTML reports and memory files. In Cowork, click Scheduled > Create New Task, then paste in your custom prompt from the previous step as the instructions. Click More options > Folder to work in and select your new folder. > Pro tip: You can also ask Claude to turn the prompt into a skill and reference it in the instructions of the scheduled task. ### Step 4 Approve the report and let Claude execute When the task fires, Claude will not touch your inbox yet. It creates an interactive report with every unread email sorted into buckets, a checkbox next to each row, and a direct Gmail link so you can audit any call quickly. Open the report in a real browser window, not inside Cowork. If you only open the file inside Claude, the browser sandbox will not let you download the completed checklist after you mark it up. In Chrome or Safari, scan the list and check the boxes for every email you want Claude to act on. Uncheck or ignore anything it got wrong. When you are done, click Download Approvals. The page saves a JSON file of exactly what you approved. Upload that JSON back into the same Claude conversation and tell it to get to work. Claude reads the checked rows, drafts replies for the Needs Response items, applies the right Claude/* label to each approved thread, and reports back with the counts. Open the Claude/Needs-Response label in Gmail, send the drafts that look right, tweak the ones that need a human touch, and you are done for the morning. ### Pro tip > Open the generated HTML report in a real browser window, not inside Cowork, so you can download the completed approvals checklist. ### Guide: Give an AI Agent a Credit Card (Safely) > In this guide, you will learn how to give an AI agent a real payment method without handing over your main credit card. - Canonical page: https://app.therundown.ai/guides/give-an-ai-agent-a-credit-card-safely - Machine-readable page: https://app.therundown.ai/guides/give-an-ai-agent-a-credit-card-safely.md - Difficulty: intermediate - Featured tool or company: AgentCard - Categories: Automation & Agents, Business Operations & Leadership, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-06-26 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to give an AI agent a real payment method without handing over your main credit card. The workflow uses AgentCard to create prepaid or single-use virtual cards, then lets an agent such as Codex attempt a supervised checkout. #### Who This Is Useful For - People who want an AI agent to buy a low-risk item online while they supervise the final checkout. - Builders testing agent commerce, browser automation, and checkout workflows. - Teams that need a safer alternative to pasting a real company card into an agent chat. - ChatGPT, Codex, Claude Code, Cursor, or other MCP users who want payment controls before experimenting. #### What You Will Build You will build a controlled agent purchasing workflow: install AgentCard, sign in, attach a low-risk funding source, issue a small card, and give the agent a checkout task with strict spend and approval rules. The safe version is not fully autonomous. The agent can help navigate checkout, but you should keep the spending limit small, watch the browser, and require a human approval before the order is placed. #### What You Need - Node.js 18 or newer and npm. - A terminal you are comfortable using, preferably inside a throwaway playground folder. - An AgentCard account and access to email for the magic link login. - A low-limit payment method or temporary card. Do not start with your primary checking debit card or main company card. - A small test purchase and a merchant site you are comfortable using. - Codex, Claude Code, ChatGPT with an MCP server, or another agent that can use a browser or terminal tools. #### Going Further Agent commerce is still early. Some stores have anti-bot flows, confusing checkout forms, 3DS checks, or payment widgets that agents handle poorly. Start on a simple, low-stakes purchase before trying a merchant with queues, limited inventory, or complex shipping rules. AgentCard also publishes an MCP server and experimental agentic checkout commands. Those are useful when you want the agent to request card details through tools instead of relying on copied text, but you should still keep approval and spend limits in place. ### Step 1: Set The Safety Rules First Before you install anything, decide what the agent is allowed to do. The card details are real payment credentials, so treat them like secrets. - Use a card or bank product where you can set a low limit, lock the card, or let it expire quickly. - Pick a tiny first purchase. In the video, Billy used a $50 test card. - Keep your billing card, phone number, ID, and home address off-screen if you are recording or sharing the workflow. - Tell the agent to stop before placing the order unless you explicitly approve the final screen. - Do not save card numbers, CVVs, or personal addresses in project files, prompts, screenshots, or logs. > Use a prepaid or tightly limited card for testing. The point is to make the agent useful without giving it access to your real financial blast radius. ### Step 2: Install AgentCard Open a terminal in a scratch folder. Keeping this in a playground directory makes it easier to clean up if you are just experimenting. ```text mkdir agent-card-playground cd agent-card-playground npm install -g agentcard ``` Confirm the CLI is available: ```text agentcard --help ``` The current CLI uses the package name agentcard and exposes commands such as signup, setup, request, details, balance, and agentic. ### Step 3: Sign In And Set Up Your Payment Method Sign in with your email address. AgentCard sends a magic link, so keep your email open. ```text agentcard signup --email you@example.com ``` After you click the magic link, run setup. This is where you add your profile, phone number, payment method, and any required verification. ```text agentcard setup ``` If AgentCard asks for identity verification, complete it through the provider flow. The video showed Stripe-hosted payment and identity screens, but you should still do your own review before submitting personal information. ### Step 4: Create A Small Single-Use Card Once setup is finished, request a small card amount for your test purchase. The current CLI uses request new for single-use virtual cards. ```text agentcard request new --amount 50 ``` If you need to re-fetch the card details later, use: ```text agentcard request get ``` Only reveal card details in the moment you need them. If you paste them into an agent chat, assume they are now part of that conversation history. ### Step 5: Give The Agent A Narrow Checkout Task Give the agent a specific, low-risk task. Separate billing information from shipping information, and make the approval rule explicit. ```text Try buying this item with AgentCard: Budget: $50 maximum, including shipping and taxes. Payment: use the AgentCard virtual card only. Shipping: use the shipping address I provide in this chat. Rules: - do not buy substitutions - do not save payment details to the store account - stop at the final review screen - ask me before clicking Place Order - if a page asks for login, a queue, 3DS, or identity verification, stop and ask me ``` Watch the browser while the agent works. In the video, Codex eventually completed an NBA Store checkout, but it struggled with form controls and payment-method dropdowns before the order went through. ### Step 6: Expect Checkout Friction Agents can still get stuck on checkout pages. Anti-bot protection, limited inventory, payment widgets, and browser handoffs can all confuse them. - If the agent cannot add an item to cart, add it yourself and let the agent continue from checkout. - If the agent keeps clearing a payment form, pause and manually inspect the selected payment method. - If the site uses 3DS, queues, login, or identity checks, keep those steps manual. - For a first test, use a simpler merchant before trying a site with limited drops or heavy anti-bot rules. After the order, check AgentCard and the merchant confirmation to make sure the charge amount, merchant, and shipping details are correct. ### Step 7: Add MCP Or Agentic Checkout Later Once the basic CLI flow works, you can expose AgentCard through MCP so compatible agents can request setup instructions or create cards through tools. ```text { "mcpServers": { "agentcard": { "command": "npx", "args": ["-y", "agentcard-mcp"] } } } ``` AgentCard also exposes experimental agentic checkout commands. Use these only after you understand the basic card request flow and have a clear approval process. ```text agentcard agentic connect agentcard agentic buy --merchant "Example" --url https://example.com --item "Test item" --amount 10 --checkout ``` ### Pro tip > Start with a prepaid, single-use, low-limit card and keep a human approval step before the agent places the order. ### Guide: Give Your Coding Agent Access to Better AI Models with OpenRouter > In this guide, you will learn how to give your AI agent more models to work with using OpenRouter. - Canonical page: https://app.therundown.ai/guides/openrouter-101-add-new-ai-models-to-codex-and-claude-code - Machine-readable page: https://app.therundown.ai/guides/openrouter-101-add-new-ai-models-to-codex-and-claude-code.md - Difficulty: intermediate - Categories: AI, openrouter, codex - Instructors: Billy Howell - Published: 2026-09-15 - Updated: 2026-09-15 ### Overview #### The Rundown In this guide, you will learn how to give your AI agent more models to work with using OpenRouter. We'll set up an API key, compare models, and test cheaper options for making ad mockups. #### Who This Is Useful For - Marketers who want their agent to make better product images and ad mockups. - Developers who want to try cheaper models for routine coding tasks. - Anyone who likes their current AI agent but wants it to use another model for a particular job. #### What You Will Build ![OpenRouter model directory](https://tru-images.b-cdn.net/guide-assets/local-507/openrouter-101-add-new-ai-models-to-codex-and-claude-code/02-openrouter-model-directory.png) You'll set up an OpenRouter key and use it with your agent to try another model. We tested this with our GrokBot, making ad mockups for a dad hat. The goal was to find cheaper ways to try ideas before paying for a finished image. #### What You Need To Get Started - An OpenRouter account - An AI agent that can make API requests, plus a private place to store its API key - A small task to test; for product mockups, have a photo of your product ready #### Going Further In Part 2, turn this into a skill your agent can reuse: make cheap mockups, let you choose a favorite, then use a better image model for the final version. You can try several directions before spending more on the one you want to keep. ### Step 1: Browse the Model Catalog OpenRouter lets your agent use models from different providers through one API key. Start with a job you want it to do better: making product images, reviewing code, or handling a routine task more cheaply. Browse the model catalog and filter for that task. Sort by price from low to high, then open a model's page to check its capabilities, price, and usage limits. Copy the model name when you find one you want to try. > **Pro tip:** Check that an image model can generate images. We accidentally selected a vision model, which reads images instead. That sent us back to the catalog. ### Step 2: Create an API Key and Set a Budget Go to **API Keys** and choose **New key**. Name it after the project or agent you'll use it with so you can recognize its charges later. Set a spending limit and an expiration date before creating the key. We set ours to expire after one day because it was just for the test. Choose a budget you're comfortable spending and check whether it resets daily, weekly, or monthly. This is worth doing before you start experimenting. When we checked our earlier image usage, we'd spent more than we realized. A spending limit gives you a place to stop. ### Step 3: Give Your Agent a Small Test Ask your agent to help you set up OpenRouter and tell you where to store the key privately. Follow its setup instructions; keep the key out of shared chats and project files that might be published. Once it's connected, give the agent the model name and a small task. We asked our GrokBot to make more ad mockups with a cheaper image model. Here's a version you can adapt: ```text Use [model name] through OpenRouter to make three ad mockups for [your product]. Use the attached product photo as a reference. Keep the product recognizable and try a different layout for each ad. ``` Our demo uses GrokBot. If you're using Codex or Claude Code, ask it to check its OpenRouter setup before running the same test. ### Step 4: Compare the Images and Check the Cost Compare the images with your original product photo. Does the product still look right? Is the text readable? Is there a layout you'd want to develop? Our cheaper mockup gave us an idea to work with, but the hat didn't look as close to the real product as it did in the GPT Image 2.5 version. It was useful for choosing a direction, not ready to publish as a product ad. Check the key's usage in OpenRouter after the test. Ours showed $0 at that point. Check the price and limits of the model you choose rather than assuming your run will also be free. ### Guide: Hand Off Real Work Across Your Apps With Grok Bot > In this guide, you will learn how to turn Grok Bot into a focused team that works across your apps. - Canonical page: https://app.therundown.ai/guides/hand-off-real-work-across-your-apps-with-grok-bot - Machine-readable page: https://app.therundown.ai/guides/hand-off-real-work-across-your-apps-with-grok-bot.md - Difficulty: beginner - Categories: AI, Productivity - Instructors: Billy Howell - Published: 2026-08-15 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to turn Grok Bot into a focused team that works across your apps. Here’s the three things you’re going to learn: how to set it up, which agents to create, and how to connect them to apps and routines so they can work while you sleep. #### Who This Is Useful For - Busy operators who want one AI team to keep projects moving across email, calendars, files, and meeting notes. - Founders and creators who want specialized agents for research, writing, sales, and inbox work without managing a pile of separate chats. - Anyone juggling a big project who wants AI agents to finish reports and recurring work while they focus on everything else. #### What You Will Build You will build a focused team of Grok Bots led by a Chief of Staff. The team can work across your connected apps, create specialized agents, use a shared cloud computer, and report back on what it gets done. ![Focused Grok Bot agent team map](https://tru-images.b-cdn.net/guide-assets/local-430/hand-off-real-work-across-your-apps-with-grok-bot/01-focused-agent-team-map.png) Your Chief of Staff sits at the center of the team and delegates work to specialized bots. #### What You Need To Get Started - The Grok Bot desktop app (https://docs.x.ai/grok-bot/get-started). - A Cursor account to sign in and start the seven-day free trial. - The work apps you want Grok Bot to connect to. #### Going Further Tell your Chief of Staff to set up a daily debrief routine for each of your Grok Bots so you can see what they worked on. Ask for one short report that lists what each agent finished, what is still in progress, and what needs your decision. Grok Bot can save that report in Notion or Drive, or email it to you at the end of the day. This turns the setup into an AI team you can actually manage. The agents keep working in their own roles, the Chief of Staff keeps everything moving, and you get one clear update instead of checking every thread yourself. ### Step 1 Download Grok Bot and Start the Free Trial Download Grok Bot, open the app, and choose Sign in with Cursor. Follow the prompts to start the seven-day free trial. Once you are inside, you will see a simple workspace with your bots in the left sidebar and one conversation for each bot. The whole app feels more like a team chat than a giant AI dashboard, which makes it much easier to stay focused. ### Step 2 Create Your First Bot Grok Bot will ask you to create your first bot. Choose the premade Chief of Staff. This bot can review your work, recommend the rest of your team, and assign jobs to the right agents. Keep this Grok Bot workspace focused on one project, one business, or one job. The agents share the same cloud computer and work best when they are all moving toward the same goal. ![New Grok Bot settings](https://tru-images.b-cdn.net/guide-assets/local-430/hand-off-real-work-across-your-apps-with-grok-bot/02-new-bot-settings.png) Give each bot a clear name, title, and job description in its settings. > Pro tip: Grok Bot gives every agent one thread by design, so use a separate agent for each role. ### Step 3 Add the Apps You Want Grok Bot to Use Next, add the apps that you want Grok Bot to use. You can connect Gmail, Drive, Calendar, Notion, etc. Open Plugins in the bottom-left corner of the app, find the services you want, and click Add. Grok Bot can also connect to tools such as Granola for meeting notes. If you do not see an app in the marketplace, you can ask your Chief of Staff whether there is another plugin or connector it can install. ![Grok Bot plugins marketplace](https://tru-images.b-cdn.net/guide-assets/local-430/hand-off-real-work-across-your-apps-with-grok-bot/03-plugins-marketplace.png) Add the work apps you want Grok Bot to use from the Plugins marketplace. Some apps will wait to ask you to sign in until a bot needs them. If a website needs a password, passkey, or anti-bot check, Grok Bot can hand you its cloud computer. Sign in yourself, then let the bot continue. ### Step 4 Tell Your Chief of Staff What Team to Build Now tell your Chief of Staff to review your connections, learn how you work, and recommend the agents that will save you the most time. Use this prompt: ```text Review my connections, build a profile of my workstreams and projects, then suggest the three most useful agents, automations, and any plugins I should add. ``` The Chief of Staff will look across the apps you connected and recommend a small team. In our setup, that included agents for social, inbox management, writing, sales, and newsletter work. Review its recommendations, then tell the Chief of Staff to create the agents you want. It will name them, write their instructions, and keep track of what each one is working on. You can open the agent messages at any time to see how the Chief of Staff is delegating the work. ### Step 5 Build the First Deliverable or Report Now tell the Chief of Staff to build its first deliverable or report for you. Start with something you already need. Ask it to research a question, summarize what your team finished, create a project report, or pull together a draft from the apps you connected. The Chief of Staff can assign parts of the job to several agents, let them work in parallel, and combine the results for you. Grok Bot also gives the team a shared cloud computer. The bots can browse the web, store files, open connected services, and keep working even when you switch computers. Reports can live in Notion, Google Drive, or Markdown files on the bot’s computer. Choose the place where you will review the work. > Pro tip: Have the bots save finished work or sync documents once at the end of the day. Frequent connector updates can use more tokens and slow them down. ### Guide: Here's How to Get This $200 ChatGPT Feature On Your $20 Plan > This guide shows you how to recreate much of ChatGPT Pulse’s proactive briefing experience on a ChatGPT Plus plan by using ChatGPT Tasks. - Canonical page: https://app.therundown.ai/guides/heres-how-to-get-this-dollar200-chatgpt-research-feature-on-your-dollar20-plan - Machine-readable page: https://app.therundown.ai/guides/heres-how-to-get-this-dollar200-chatgpt-research-feature-on-your-dollar20-plan.md - Difficulty: beginner - Featured tool or company: OpenAI - Categories: Productivity & Knowledge Work, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-03-10 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to recreate much of ChatGPT Pulse’s proactive briefing experience on a ChatGPT Plus plan by using ChatGPT Tasks. You will set up recurring prompts that run automatically on a schedule and deliver research briefings to your ChatGPT inbox. #### Who This Is Useful For - Anyone paying for ChatGPT Plus who does not hit their daily token limit - Founders and marketers who want daily competitive intel without manually searching - Anyone who checks the same news, markets, or topics every morning and wants it waiting for them #### What You Will Build You will build a set of recurring ChatGPT tasks that automatically research and brief you on topics you care about, delivered to your ChatGPT inbox on the schedule you choose. Once it is running, no typing is required. ![image-2109](https://tru-images.b-cdn.net/guide-assets/69/69b08b50fbf78844cdbe235d/image-2109.png) - Daily briefings on topics like markets, AI news, crypto, or industry updates - Weekly competitor reports for specific companies - Scheduled briefings that appear directly in ChatGPT conversation history #### What You Need - ChatGPT Plus ($20/month) or higher - The ChatGPT web app, desktop app, or mobile app - Optional: linked Gmail or Google Calendar if you want tasks that summarize your inbox or prepare you for meetings #### Going Further - You can have up to 10 active tasks at once. If you do not use your full ChatGPT token limit each day, Tasks can help put those unused tokens to work. - Create a morning news stack with 2–3 daily briefings on different topics, such as AI news, your industry, and market moves. - Schedule weekly competitor reports, with one task per competitor on Monday mornings, covering news, product updates, and press mentions from the past week. - Use connected services, if you have linked Gmail or Google Calendar, to summarize your inbox or prepare you for tomorrow’s meetings. - Create one-off reminders for a specific date, such as following up on a proposal next Thursday. ### Step 1 Understand how Tasks can replicate Pulse ChatGPT Pulse, available on the $200 Pro plan, proactively sends you daily news and suggestions based on your chat history. You do not have to ask for it; it shows up automatically. ChatGPT Tasks is available on Plus. Tasks let you write a prompt that runs automatically on a schedule, even when you are not logged in. The output appears in your ChatGPT conversation history like a normal message. By setting up a few daily tasks, you can get much of the same utility as Pulse on the Plus plan. ### Step 2 Create your first recurring task Open ChatGPT on the web, desktop, or mobile app, then prompt it directly. ```text Create a daily, recurring task that briefs me on the daily stock market moves at 5 PM. ``` ChatGPT will confirm the task and set the schedule. The next day at 5 PM, you will get a briefing in that chat without doing anything else. You can swap the topic for whatever you care about. Try prompts like these: ```text Create a daily task at 7 AM that summarizes the top 3 AI news stories from the past 24 hours. ``` ```text Create a weekly task every Monday at 9 AM that searches for any news about [competitor name] from the past week. ``` ```text Create a daily task at 8 AM that gives me a 2-sentence summary of overnight moves in crypto markets. ``` ### Step 3 Manage tasks and turn on notifications To edit or delete tasks, go to Settings > Notifications > Tasks > Manage Tasks. From there, you can pause a task, edit the schedule, or delete it. ![image-2108](https://tru-images.b-cdn.net/guide-assets/69/69b08b40fbf78844cdbe1acd/image-2108.png) Turn on push notifications for Tasks and for the ChatGPT app on your phone. In the source example, task notifications did not fire until both were enabled. ### Step 4 Check your scheduled briefings Your briefings will start arriving in the chat where you set up the task. Open ChatGPT in the morning and they will be waiting for you: no typing, no searching, and no prompting required. ![CleanShot 2026-03-10 at 17.21.34@2x](https://tru-images.b-cdn.net/guide-assets/69/69b08b72289d44a4e6782581/cleanshot-2026-03-10-at-17-21-34-2x.png) ### Pro tip > The Manage Tasks menu might not show up in the mobile app. The source example had to use ChatGPT.com in a browser to find it. ### Guide: How this Hidden “Insights” Feature Improves Claude Code > In this guide, you’ll learn how to use Claude Code’s hidden /insights feature to improve your coding habits. - Canonical page: https://app.therundown.ai/guides/how-this-hidden-insights-feature-improves-claude-code - Machine-readable page: https://app.therundown.ai/guides/how-this-hidden-insights-feature-improves-claude-code.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-02-10 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you’ll learn how to use Claude Code’s hidden /insights feature to improve your coding habits. You’ll generate a built-in report from your Claude Code history, review Claude’s feedback on your process, and turn the report’s recommendations into reusable CLAUDE.md instructions, custom skills, or hooks. #### Who This Is Useful For - Vibe coders who aren’t confident in their coding habits - Anyone curious what Claude Code thinks of their process - Data nerds interested in what tools and skills their Claude Code uses the most #### What You Will Build You’ll produce a local Claude Code insights report and reusable improvements for future coding sessions. - A report.html insights report generated by Claude Code - Copy-paste-ready CLAUDE.md instructions based on your patterns - Optional custom skills and hooks suggested by the report ![image-2042](https://tru-images.b-cdn.net/guide-assets/69/698ba08995eabb62d18949ed/image-2042.png) #### What You Need - Claude Pro or higher subscription - A computer with the Claude CLI installed - At least a few weeks of Claude Code usage; more history means a better report - A code editor such as Cursor, VS Code, or just your terminal #### Going Further - Run /insights once a month. As your habits improve, Claude Code can surface new skills and instructions you can reuse. Feedback can also be project specific, so if you’ve been struggling with a project all month, this is a useful way to get feedback. ### Step 1: Generate your insights report Create an empty folder and open it in your IDE. Open a new terminal session and run: claude /insights. > Pro tip: If this command fails, close Claude Code with ctrl + c and reopen it. This should update Claude Code if you needed an update to use /insights. ### Step 2: Open the report Claude should begin working on your insights report. When it’s done, it will give you a link to a file named report.html. Copy it into your current folder. Start the webpage with cmd + shift + p and find the “Open Live Server” tool. > Pro tip: You can also give the HTML file to Claude or ChatGPT and have them run it for you. ### Step 3: Review your usage patterns Scroll through the report sections. Here’s some of what you’ll find: - What You Work On shows cards for each project you’ve used Claude Code on, with session counts and a summary of what happened. It also breaks down what you asked for most, which tools Claude used, what languages you work in, and your session types. - Impressive Things You Did highlights what went well across your sessions. - Where Things Go Wrong helps you learn where the disconnects are between you and Claude Code. It flags patterns like Claude ignoring your instructions, producing buggy code, or taking wrong approaches. ![image-2043](https://tru-images.b-cdn.net/guide-assets/69/698ba09a95eabb62d1895217/image-2043.png) ![image-2044](https://tru-images.b-cdn.net/guide-assets/69/698ba0a495eabb62d1895a3c/image-2044.png) ![image-2045](https://tru-images.b-cdn.net/guide-assets/69/698ba0af95eabb62d1896270/image-2045.png) ### Step 4: Add suggested CLAUDE.md instructions Scroll down to “Existing CC Features to Try.” This section has copy-paste-ready lines for your CLAUDE.md. Each suggestion comes with a “Copy” button and an explanation of why it matters for you specifically. For example, if Claude keeps ignoring your theme preferences, you might see a CLAUDE.md line like this: ```text This project uses TypeScript with a custom shadcn/ui dark theme. ALWAYS use the initialized theme tokens instead of arbitrary Tailwind color classes. ``` Open your CLAUDE.md in any project, paste in the suggestions you want, and Claude will follow them in future sessions. ### Step 5: Set up custom skills and hooks The report also suggests Custom Skills and Hooks based on your patterns. ![image-2047](https://tru-images.b-cdn.net/guide-assets/69/698ba0c695eabb62d18972cf/image-2047.png) Custom skills are reusable prompts you trigger with a single command. If you keep doing the same task over and over, such as generating CLAUDE.md files for new repos, the report will suggest turning that into a skill so it runs with one command instead of a full session. To set one up, paste the skill code from the report into Claude Code and say: ```text Set up this skill ``` Claude creates it for you. You can then use it anytime with a custom slash command in Claude Code. > Pro tip: To use newly installed skills, exit and restart Claude Code. ### Pro tip > If `/insights` fails, close Claude Code with `ctrl + c` and reopen it to update Claude Code before trying again. ### Guide: How to 3D Model Anything With Claude + Blender > Connect Blender to Claude Code with MCP, then use plain English to create, edit, light, and arrange a 3D scene. - Canonical page: https://app.therundown.ai/guides/how-to-3d-model-anything-with-claude-blender - Machine-readable page: https://app.therundown.ai/guides/how-to-3d-model-anything-with-claude-blender.md - Difficulty: beginner - Featured tool or company: Blender - Categories: Content & Creative - Instructors: Billy Howell - Published: 2026-05-18 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to connect Blender to Claude Code with MCP, then use plain English to create and edit a 3D scene. #### Who This Is Useful For - People who are curious about Blender but feel intimidated by the interface. - People who already use Blender and want a faster way to test scene ideas, lighting changes, camera angles, or quick 3D mockups. - People who want to start creating their own 3D animations without learning every Blender control first. #### What You Will Build You will connect Blender to Claude Code, confirm the MCP server is running, and test the connection by asking Claude to create a simple 3D name scene with disco balls and studio lighting. ![What You Will Build 3d Name Scene](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/01-what-you-will-build-3d-name-scene.png) #### What You Need You need: - Blender: https://www.blender.org/ - Claude Code - Terminal access - the Blender MCP extension from Blender's MCP Server page: https://www.blender.org/lab/mcp-server/ You can also use the GitHub repo as a setup reference: ```text https://github.com/ahujasid/blender-mcp ``` #### Going Further Once the connection works, you can use Blender MCP for more than a quick name scene. Try using it to create thumbnail backgrounds, arrange product mockups, test animation ideas, or prepare 3D print concepts. The best workflow is simple: - Open or download a Blender scene. - Confirm the MCP server is running. - Ask Claude to inspect the scene before editing. - Give a specific creative direction. - Review the result in Blender. - Ask for targeted revisions. Blender is still the source of truth. Claude is the control layer. Use it to move faster, explore ideas, and make Blender less intimidating. ### Download Blender Start by downloading Blender from blender.org: https://www.blender.org/. Blender is free and open source. Install it like any other Mac app, then open it once before you start the MCP setup. The connection works best when Blender is already open and the MCP extension is running inside the app. ### Install the Blender MCP Extension Open Blender's MCP Server page: https://www.blender.org/lab/mcp-server/. The page gives you two options: drag the extension into Blender or download it and install from disk. ![Drag And Drop Blender Mcp Extension](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/02-drag-and-drop-blender-mcp-extension.png) The easiest route is drag and drop. Drag the MCP installer into Blender. If Blender asks you to add the Blender Lab repository first, allow it. ![Add Blender Lab Extension Repository](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/03-add-blender-lab-extension-repository.png) > Pro tip: The drag-and-drop install can take two passes. The first pass adds the Blender Lab repository. The second pass installs the MCP add-on itself. If nothing appears after the first try, drag it in again or restart Blender and check Extensions. Now open Edit -> Preferences. Search for MCP. You should see the MCP extension from Blender Lab. ![Enable Mcp Extension In Blender](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/04-enable-mcp-extension-in-blender.png) Make sure the checkbox is enabled. ### Turn on the Blender-side MCP Server Expand the MCP extension details inside Blender Preferences. This is where the server settings live. You should see settings like: - Host: usually localhost - Port: usually 9876 - Auto Start - Stop MCP Server or Start MCP Server - Server is running ![Enable Online Access In Blender](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/05-enable-online-access-in-blender.png) If the server does not start, check Blender's network setting. In our setup, Blender needed online access enabled before the MCP server would behave normally. ![Confirm Mcp Server Running](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/06-confirm-mcp-server-running.png) > Pro tip: If Claude cannot connect, this screen is the first thing to check. It shows the exact host, port, file path, version, and whether the server is running. If you are stuck, screenshot this panel and paste it into Claude Code so it can reason from the real settings. ### Connect Claude Code from Terminal Open Terminal and start Claude Code in the project folder you want to use. Then run: ```text claude mcp add blender -- uvx blender-mcp-server claude mcp get blender ``` You want Claude Code to show the blender MCP server as connected. ![Confirm Blender Connected In Claude Code](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/07-confirm-blender-connected-in-claude-code.png) If you already added a Blender MCP server and it points to the wrong command, remove and re-add it: ```text claude mcp remove blender -s local claude mcp add -s local blender -- uvx blender-mcp-server claude mcp get blender ``` > Pro tip: The key is that Claude Code needs a working MCP config for the project you are in. If your mcp.json or Claude MCP config is wrong, Blender can be running perfectly and Claude still will not connect. ### Let Claude Help With the Setup If the setup feels annoying, use Claude Code as the setup assistant. Paste the GitHub repo and ask it to verify the config. ```text Use this GitHub repo as a reference: https://github.com/ahujasid/blender-mcp Connect Claude Code to Blender and make sure my mcp.json or Claude MCP config is set up correctly. Check that Blender is running the MCP server, confirm the host and port, register `blender` if needed, and verify the connection before editing. ``` This is useful because Claude can inspect your local MCP config, check whether the command is registered, and tell you what to fix. You still need to do the Blender GUI steps yourself: install the extension, enable it, and confirm the server is running. There is also another setup option in the Claude desktop app. Open the directory, search for Blender under connectors, and add it there if it is available in your workspace. ![Find Blender Connector In Claude Directory](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/08-find-blender-connector-in-claude-directory.png) This route can be easier if you prefer a UI, but the terminal route is still the cleanest way to verify the exact MCP command. ### Test the Connection With a Simple Scene Once Claude Code shows Blender as connected, open Blender and try a small creative prompt. ```text Use Blender MCP to model my name in 3D. Add disco balls, dramatic lighting, reflective materials, and a camera angle that makes it look like an event poster. ``` In the recording, the test prompt created large 3D letters for R and U. Earlier tests had Claude adding spheres and pyramids. The exact output will vary, but the important signal is that Blender changes in response to the prompt. ![Claude Created 3d Letters In Blender](https://tru-images.b-cdn.net/guide-assets/local-packages/208-for-jason/09-claude-created-3d-letters-in-blender.png) If the scene changes, the connection works. ### Use Existing Models When You Want Better Results Claude can create basic geometry from scratch, but it will often be better if you start with existing Blender assets. For example, you can download models from: - BlenderKit: https://www.blenderkit.com/ - Free3D - other Blender model libraries Then ask Claude to arrange and improve the scene instead of inventing every object from scratch. Use a prompt like: ```text Before creating anything from scratch, look for existing objects in this Blender scene and use those first. Arrange the models into a clean hero composition. Improve the lighting, set a camera angle, adjust materials if needed, and prepare the scene for a still render. ``` > Pro tip: For polished results, treat Claude like a scene director. Give it real assets, then ask for lighting, camera, material, and composition changes. That usually works better than asking it to model every object from zero. ### Resources #### Setup Links - Blender: https://www.blender.org/ - Blender MCP Server: https://www.blender.org/lab/mcp-server/ - blender-mcp GitHub repo: https://github.com/ahujasid/blender-mcp - BlenderKit: https://www.blenderkit.com/ - [Blender](https://www.blender.org/) - [Blender MCP Server](https://www.blender.org/lab/mcp-server/) - [blender-mcp GitHub Repository](https://github.com/ahujasid/blender-mcp) - [BlenderKit](https://www.blenderkit.com/) ### Pro tip > For polished results, treat Claude like a scene director. Give it real assets, then ask for lighting, camera, material, and composition changes. ### Guide: How To Audit Your Business With Notion's Built-In Claude Agents > Learn how to use Notion's prebuilt AI agents, powered by Claude, to audit a Notion workspace for efficiency. - Canonical page: https://app.therundown.ai/guides/how-to-audit-your-business-with-notions-built-in-claude-agents - Machine-readable page: https://app.therundown.ai/guides/how-to-audit-your-business-with-notions-built-in-claude-agents.md - Difficulty: beginner - Featured tool or company: Notion - Categories: Business Operations & Leadership, Automation & Agents, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-04-15 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use Notion's prebuilt AI agents, powered by Claude, to audit a Notion workspace for efficiency. You will produce a scored audit report, a prioritized fix list, and, if permissions allow, apply selected fixes directly in Notion. The source describes the full clean-up loop as taking under 10 minutes. #### Who This Is Useful For - Operators and founders whose Notion workspace has become a graveyard of half-built databases. - Team leads who want a second opinion on a process or SOP without hiring a consultant. - Anyone who over-engineers in Notion and needs critical feedback on what to clean up. #### What You Will Build ![CleanShot 2026-04-15 at 17.08.24@2x](https://tru-images.b-cdn.net/guide-assets/69/69dffe59322b920e774b48e6/cleanshot-2026-04-15-at-17-08-24-2x.png) You will create a real audit of one Notion database, teamspace, or SOP, with scored findings, prioritized fixes, and the option to let the agent apply those fixes itself. - A scored audit report. - A prioritized fix list. - Recommended database or SOP improvements. - Optional applied fixes inside Notion. #### What You Need - A Notion workspace with Notion AI enabled. - A Notion Business or Enterprise plan. - A database, teamspace, or SOP page you want audited. #### Going Further - After the first audit, chain agents together: run the Workspace Auditor, feed its findings into the Business Process Audit, then let the Task Triager turn the fixes into assigned work. If you outgrow the presets, open a preset agent, tweak its system prompt, and save your own version. ### Step 1: Find the preset agents Notion has a full library of prebuilt agents, and most people are scrolling right past them. - In the app: open the sidebar, scroll to Agents, click the + button, and choose Browse templates. - In your browser: go to notion.com/custom-agent-templates and filter by category. > Pro tip: The browser view is easier to filter. Sort by popularity. Some of these agents have tens of thousands of downloads, so you do not need to reinvent the wheel. ### Step 2: Install the Business Workspace Auditor Start with the Business Workspace Auditor. It scans a page or database and returns a scored report with the biggest scalability risks at the top. ![CleanShot 2026-04-15 at 17.13.22@2x](https://tru-images.b-cdn.net/guide-assets/69/69dfff82c1a96302b3806726/cleanshot-2026-04-15-at-17-13-22-2x.png) - Open Business Workspace Auditor. - Click Get template. - Pick the workspace you want it to live in. - Under Permissions, add any private pages you want it to see. By default, it can only see shared pages. - Hit Save. > Pro tip: Managing permissions is the one annoying part, but do not skip it. Scoping the bot to the pages you actually want audited keeps it focused and keeps it from touching things you do not want it touching. ### Step 3: Run the audit Open the agent and @ mention the page or database you want reviewed. You can also drop the link directly in the chat. The prompt can be simple because the agent is already preconfigured with a long system prompt written by Notion experts: ```text Audit this database. ``` > Pro tip: You can also @ mention the agent in a comment on any page to trigger it, the same way you would tag a teammate. This is useful for async audits. ### Step 4: Read the report and let it fix things The agent returns a scored report with: - Schema issues, such as duplicate status fields or orphaned properties. - Missing relations between databases. - Naming inconsistencies. - Severity on each finding. - A recommended fix. If you gave the agent edit permissions, you can ask it to apply selected fixes directly: ```text Go ahead and apply fixes 1, 2, and 4. ``` The agent will make the changes in the database itself, without copy-paste or manual cleanup. > Pro tip: These agents are more critical than many AI tools. Notion's preset agents can call out structural issues directly, so lean into that feedback. ### Step 5: Watch your token usage These agents use credits, so it is worth knowing where to check usage. - Pricing is $10 per 1,000 credits. - You can check usage per agent in the agent settings. > Pro tip: Check the per-agent usage panel before and after a big run. That is the fastest way to figure out which agents are worth keeping on and which are burning credits on tasks a cheaper tool could do. ### Step 6: Keep going Once the Workspace Auditor has cleaned up one database, the same install flow works for the rest of the marketplace. The ones worth installing next: - Business Process Audit: paste in an SOP or describe a process, and it maps the steps, flags bottlenecks, and suggests fixes. - Task Triager: paste a messy brain dump, and it routes each item into your Tasks database with project, priority, and due date filled in. - Weekly Planning agents: useful out of the box for roll-ups and weekly reviews. - Email help agents: morning inbox triage without the setup work. You can also pair these with free Notion templates from the marketplace. Filter to Notion and Free, grab a starter kit like a Tasks or Reports database, and point a preset agent at it. ### Pro tip > Use the browser view to browse Notion agent templates, then sort by popularity. It is easier to filter, and popular agents can save you from reinventing the wheel. ### Guide: How To Automate Your Business With Custom Notion Agents > This guide teaches you how to build Notion Custom Agents that run recurring company work on a schedule using a simple dual-database system. - Canonical page: https://app.therundown.ai/guides/how-to-automate-your-business-with-custom-notion-agents - Machine-readable page: https://app.therundown.ai/guides/how-to-automate-your-business-with-custom-notion-agents.md - Difficulty: beginner - Featured tool or company: Notion - Categories: Automation & Agents, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-04-09 - Updated: 2026-07-23 ### Overview #### The Rundown This guide teaches you how to build Notion Custom Agents that run recurring company work on a schedule using a simple dual-database system. You will create shared Tasks and Reports databases, then use them with a Weekly Planning agent that turns inbox items into action items and logs its work. #### Who This Is Useful For - Founders and operators juggling five to ten AI tools and losing track of what each one is doing - Consultants and agencies running recurring client work who want one place to see every agent’s output - Teams collaborating around AI who need a cloud source of truth their teammates can actually see #### What You Will Build ![image-2173](https://tru-images.b-cdn.net/guide-assets/69/69d830d4b8520a96a0fd2c18/image-2173.png) You will build a reusable operating pattern for AI agents: a shared Tasks database for inputs, a shared Reports database for outputs, and a Weekly Planning agent that reads your Gmail inbox every Monday, drafts action items into Tasks, and logs a summary in Reports. - Shared Tasks database - Shared Reports database - Weekly Planning agent - Reusable pattern you can clone for other recurring business jobs #### What You Need - A Notion workspace on any plan - A Notion Business or Enterprise plan if you want to use Notion Custom Agents specifically. Free Notion AI tokens to try them are included through May 3, 2026 - Or a connected AI that can write to Notion, such as Claude, ChatGPT, Perplexity, Manus, or another tool using the Notion connector - The Gmail connector if you want the example Weekly Planning agent to read your inbox #### Going Further - Once the two-database pattern is set, you can clone it for other recurring business workflows. You can also connect external AI tools with Notion connectors and instruct them to write summaries into the same Reports database, giving you one dashboard for what every agent in your stack is doing. ### Step 1: Build the Tasks Database Open a new page, add a database, and choose Build with AI. Prompt Notion with: ```text I need a task database for tracking recurring tasks that agents complete, and for agents to assign one-off tasks to me. Include Name, Type (recurring or one-off), Assignee, Priority, and Status. ``` Notion will create the database and property types for you. The key properties are Type, so agents and humans can filter recurring versus one-off work, and Assignee, so you can assign a task to a specific agent or to yourself. ![image-2172](https://tru-images.b-cdn.net/guide-assets/69/69d83095b8520a96a0fd22dc/image-2172.png) > Pro tip: Add a Kanban view on the Tasks database grouped by Assignee. You will be able to see every agent’s column at a glance, what each one is working on, and what is waiting on you. ![image-2174](https://tru-images.b-cdn.net/guide-assets/69/69d830eeb8520a96a0fd3509/image-2174.png) ### Step 2: Build the Reports Database Add a second database and prompt Notion to build it: ```text I need a reports database that teammates and AI agents can add reports to, with a field that marks who (or which agent) created each report. Add views for one-off, daily, weekly, and monthly reports. ``` Reports is the audit trail. Every time a scheduled agent runs, it should add a row here summarizing what it did. This keeps you from digging through Claude project threads or Codex logs to figure out what an agent did yesterday. ![image-2175](https://tru-images.b-cdn.net/guide-assets/69/69d83100b8520a96a0fd3e67/image-2175.png) Every AI task has inputs and outputs. Tasks are the inputs: what needs to happen. Reports are the outputs: what the agent actually did. Hold every agent to that standard to reduce complexity. ### Step 3: Create Your First Custom Agent Open Notion AI and click + Create custom agent. Instead of filling fields by hand, prompt it to build the agent for you: ```text Create a weekly planning agent that looks at my inbox every Monday in Gmail, creates tasks for me in the Tasks database, and then writes a report in the Reports database. ``` ![image-2176](https://tru-images.b-cdn.net/guide-assets/69/69d8310cb8520a96a0fd4746/image-2176.png) Notion drafts the agent with a name, trigger, instructions, and data sources based on your prompt. Review it, connect the Gmail connector if it is not already wired in, confirm the schedule, and save. ![image-2177](https://tru-images.b-cdn.net/guide-assets/69/69d83119b8520a96a0fd4ff8/image-2177.png) > Pro tip: Make sure the agent has edit access to the Tasks and Reports databases. ### Step 4: Assign Agents Like Teammates Anytime you @mention a custom agent in Notion, it triggers the agent to start working. Open a row in your Tasks database, add the agent as the Assignee, @mention it in a comment, and it goes to work on that task. > Pro tip: You can turn this trigger off in the agent settings on a case-by-case basis. Stop thinking about AI as a bunch of apps you have to babysit. Start thinking about AI like a business with employees: tasks get assigned, reports get logged, and work gets reviewed on Monday morning. ### Step 5 Optional: Connect Other AI Tools to Reports You do not need Notion Custom Agents for this system to help simplify and automate your business. Any AI that has a Notion connector can plug into the same two databases, including Claude, ChatGPT, Perplexity, Manus, or anything else with scheduled tasks. Tell the external agent about your Reports database in its system instructions. For example, when you set up a scheduled task in Claude, paste this at the end of the instructions: ```text When you are done, go to the Reports database in Notion at [paste your Reports database URL] and write a report on what you did today, with Agent = [your agent name] and a one-paragraph summary. ``` Now Claude logs its work in the same place your Notion Custom Agents do. Do the same for every AI tool you use and you will have one dashboard showing what every agent in your stack is doing. ### Pro tip > Add a Kanban view on the Tasks database grouped by Assignee so you can see every agent’s work, what is waiting, and what is waiting on you. ### Guide: How To Automate Your Chrome Browser With Gemini > Learn how to turn on and use Gemini in Chrome to compare open tabs, then optionally use Google AI Pro to enable Chrome automation. - Canonical page: https://app.therundown.ai/guides/how-to-automate-your-chrome-browser-with-gemini - Machine-readable page: https://app.therundown.ai/guides/how-to-automate-your-chrome-browser-with-gemini.md - Difficulty: beginner - Featured tool or company: Google - Categories: Automation & Agents - Instructors: Billy Howell - Published: 2026-04-14 - Updated: 2026-07-23 ### Overview #### The Rundown Learn how to turn on and use Gemini in Chrome to compare open tabs, then optionally use Google AI Pro to enable Chrome automation. You will create a simple browser workflow for faster product research while keeping the higher-risk parts under your control. #### Who This Is Useful For - Shoppers and researchers who keep too many tabs open and want Gemini to compare them quickly - Operators and assistants who want a faster way to summarize options without copying links into a separate chat - Anyone curious about browser agents who wants to start with safer Gemini in Chrome features before turning on full automation #### What You Will Build You will build a reusable Chrome workflow for comparing open tabs inside Gemini, with an optional upgrade into a browser automation workflow if you have Google AI Pro. - A tab-comparison workflow using Gemini in Chrome - A prompt for comparing options across selected tabs - An optional automation prompt for finding better or cheaper alternatives #### What You Need - Chrome desktop app - A personal Gmail account - Optional: Google AI Pro ($20/month) #### Going Further - After using tab sharing for low-risk comparison tasks, you can use browser automation for repetitive parts with clear stop instructions. If Google AI Pro can access saved Chrome passwords to sign in to accounts, review Gemini’s plan carefully before approving anything involving money, forms, scheduling, or messages. ### Step 1 Turn on the Gemini in Chrome settings In Chrome, click the three dots in the top right, then go to Settings > AI Innovations > Gemini in Chrome. Turn on these two settings first: - Show Gemini at the top of the browser - Share current tab with Gemini If you have Google AI Pro, also turn on “Let Gemini browse for you.” That toggle unlocks browser automation. The first two settings are the ones most people should start with. ![CleanShot 2026-04-14 at 18.28.26@2x](https://tru-images.b-cdn.net/guide-assets/69/69dec0f582717e244fa607b8/cleanshot-2026-04-14-at-18-28-26-2x.png) ### Step 2 Open the tabs you want Gemini to compare Open a few tabs on the same topic. The source example uses three standing desks, each on its own product page. This works best when the tabs are clearly comparable. Products, pricing pages, hotel options, or software plans are good examples. The less cleanup Gemini has to do, the better the answer will be. ### Step 3 Attach the tabs Click Gemini at the top of Chrome to open the side panel. In the message box, type @ and attach each tab you want Gemini to use. ```text Compare these desks and tell me which one has the best features for the best price. ``` This is the easy version of browser automation. Gemini is not roaming the web on its own. You are giving it the exact tabs to work from. ![CleanShot 2026-04-14 at 18.29.24@2x](https://tru-images.b-cdn.net/guide-assets/69/69dec11382717e244fa60827/cleanshot-2026-04-14-at-18-29-24-2x.png) The source example also tested Gemini extracting and comparing several recipe tabs, then using Chrome automation to find more recipe recommendations. ### Step 4 Let Gemini pick the best option Gemini will compare the tabs and give you a recommendation. This works even without Google AI Pro because you are sharing the tabs directly. If the answer is too vague, ask a follow-up: ```text Now rank them from best to worst and explain the tradeoffs. ``` ### Step 5 Turn on Chrome automation if you have Google AI Pro If you have Google AI Pro and the toggle is available, you can go one step further. Ask Gemini to browse for a better option for you. ```text Find me 5 cheaper standing desks with these same features. Use browser automation. ``` Gemini will send you a plan first. Read that plan carefully before you approve it. ![CleanShot 2026-04-14 at 18.36.52@2x](https://tru-images.b-cdn.net/guide-assets/69/69dec194e2bd3b934fe8de49/cleanshot-2026-04-14-at-18-36-52-2x.png) ### Pro tip > Start with tab sharing for low-risk comparison tasks, then use browser automation only for repetitive parts with clear stop instructions. ### Guide: How To Build a Lead Magnet Tool for Your Website With Replit > This guide shows you how to use Replit to build a custom calculator lead form for your website. - Canonical page: https://app.therundown.ai/guides/how-to-build-a-lead-magnet-tool-for-your-website-with-replit - Machine-readable page: https://app.therundown.ai/guides/how-to-build-a-lead-magnet-tool-for-your-website-with-replit.md - Difficulty: intermediate - Featured tool or company: Replit - Categories: Marketing & Growth, Coding & App Development - Instructors: Billy Howell - Published: 2026-02-08 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to use Replit to build a custom calculator lead form for your website. You will produce a simple internal lead-generation calculator builder that captures email leads, stores responses, and can generate embeddable calculators. If you can copy and paste, you can follow the workflow. #### Who This Is Useful For - Agencies that want to build custom lead magnets for clients without hiring a dev team. - Marketing teams tired of paying per-seat pricing for simple calculator tools. - Non-technical founders who want to capture leads with interactive tools. #### What You Will Build You will build a reusable calculator lead form app for your website. The app lets you create custom calculators, generate an iframe embed, and store lead responses and emails in a Replit database. - A custom calculator lead form. - Editable calculator colors, text fields, and formulas. - An iframe code snippet you can embed on your website. - A place in Replit to view captured lead data. ![leadcalcbuilt](https://tru-images.b-cdn.net/guide-assets/69/6988e219218ea48053178b5a/leadcalcbuilt.png) #### What You Need - Replit Core membership ($25/month). - A website where you want to embed the calculator. - Calculator details such as the color scheme, text fields, and formulas you want to use. #### Going Further - Embed the generated iframe into your website. - Send lead data somewhere else by using an automation tool like n8n or Make.com to receive the data via webhook and store it in a Google Sheet. - Build automations to deliver lead magnets to the user. ### Step 1 Create the lead calculator app in Replit Go to Replit.com and toggle on Plan mode. ![replit_plan_mode](https://tru-images.b-cdn.net/guide-assets/69/6988e26d218ea48053178c5a/replit-plan-mode.png) Paste in a prompt like this: ```text Build a "CalcFactory" app using Python (Flask) and Replit SQL. It is a LeadGen Calculator Builder where users create embeddable cost-savings calculators that capture email leads and track analytics. Make it so the user can create new calculators with custom color scheme, text fields and calculator formulas editable. Store the lead's response and email in a Replit database. ``` Answer any relevant questions the agent has, then hit “Build.” > 💡 Pro tip: Keep it simple. This is an app only you and your team will use. You are not turning this into a billion-dollar SaaS app. ### Step 2 Build and embed a custom calculator Try building a custom form with your new tool. The example form looked like this: ![custom_calc_1](https://tru-images.b-cdn.net/guide-assets/69/6988e294218ea48053178cdb/custom-calc-1.png) Make sure your app generates a code snippet called an iframe. ![calc_embed](https://tru-images.b-cdn.net/guide-assets/69/6988e29f218ea48053178d0b/calc-embed.png) Copy that iframe and embed it into your website. If that feels too technical, you can hand the iframe code to Replit, Codex, or another AI tool and it will know what to do. > 💡 Pro tip: Deploy your lead form app with the “Publish” button in the top right. If you embed forms from your Preview build, they will disappear when you leave Replit. ### Step 3 Test the form and find your lead data Try filling out the form. The example had an email field at the bottom. ![calc_closeup](https://tru-images.b-cdn.net/guide-assets/69/6988e2be218ea48053178d50/calc-closeup.png) To find your lead data, go back into Replit, then open Database > Production Database. ![replit_database](https://tru-images.b-cdn.net/guide-assets/69/6988e2d8218ea48053178da1/replit-database.png) ### Pro tip > Keep it simple: this is an app only you and your team will use, not a billion-dollar SaaS app. Also, publish your app before embedding it so your forms do not disappear when you leave Replit. ### Guide: How To Build a Productivity Tool With Replit + Interview Method > Learn how to build a simple productivity tracker that counts what you actually did each day and turns it into a clean weekly report. - Canonical page: https://app.therundown.ai/guides/how-to-build-a-productivity-tool-with-replit-interview-method - Machine-readable page: https://app.therundown.ai/guides/how-to-build-a-productivity-tool-with-replit-interview-method.md - Difficulty: beginner - Featured tool or company: Replit - Categories: Productivity & Knowledge Work, Coding & App Development - Instructors: Billy Howell - Published: 2026-04-01 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to build a simple productivity tracker that counts what you actually did each day and turns it into a clean weekly report. The method starts with Claude or ChatGPT interviewing you about your work, then uses that conversation to create a Replit build prompt. In the guide test, the first working version came together in less than ten minutes. #### Who This Is Useful For - Managers, operators, and client-facing workers who get asked what they accomplished each week and do not want to answer from memory - People juggling many small outputs, such as calls, proposals, meetings, follow-ups, deliverables, or published work - Anyone who dislikes time tracking and wants a lighter system that counts work instead of logging every minute #### What You Will Build ![084_finished_dashboard_tally_4_with_heatmap_nav-1](https://tru-images.b-cdn.net/guide-assets/69/69cd9c09ea506755deee694a/084-finished-dashboard-tally-4-with-heatmap-nav-1.png) You will build a lightweight Replit tracker you can update in a few seconds each day and reuse for manager updates, client recaps, or team check-ins. - Daily task-count inputs - Optional notes - A calendar heatmap - A report generator for any date range - Weekly or monthly report outputs with totals, averages, most active day, and a short written summary #### What You Need - A Claude or ChatGPT account for the interview step - A Replit account - A rough understanding of the kind of work you do each week - Replit Starter is enough to build and publish one app; Replit Core gives you more room if you keep building tools like this #### Going Further - After the first version works, you can make the tracker more useful by adding a client or project tag to each entry for filtered reports by account. You can also add simple authentication if you plan to keep using the app over time, but the first goal is a useful personal accountability tool you will actually maintain. ### Step 1: Use AI to Figure Out What You Should Track Start with Claude or ChatGPT instead of opening Replit first. The goal is to let AI help you figure out what the app should track before you build anything. ```text Interview me about my daily tasks so that I can prompt Replit to build me a daily task counting app. ``` Claude or ChatGPT should ask you a few simple questions and turn your answers into a build prompt for Replit. - Sales calls - Deliverables sent - Client replies - Internal meetings - Content published - Proposals sent ![084_replit_agent_qa_task_types_and_report_preferences](https://tru-images.b-cdn.net/guide-assets/69/69cd9c2cee4ae4b414652e85/084-replit-agent-qa-task-types-and-report-preferences.png) This part matters because the app is only useful if the categories are useful. In the demo, Claude surfaced categories it already knew from previous work, so the interview felt more like a quick sorting pass than a formal questionnaire. > Pro tip: Keep the labels tight. “Admin” is too vague. “Client response” or “Proposal sent” is much easier to count and much more useful in a report later. ### Step 2: Build the Tracker in Replit Once Claude or ChatGPT gives you the build prompt, paste it into Replit and let it build the app. - Daily number inputs for each task - An optional note field - A calendar heatmap - A report generator for any date range In the test for this guide, the generated prompt was pasted directly into Replit and the first working version came together in less than ten minutes. The important mindset is not “build me a perfect app.” It is “build me something I will actually open every day.” ### Step 3: Log a Few Fake Days First Before you trust the app with real data, add a few fake days so you can see the system working. Once the calendar starts filling in, the tracker stops feeling theoretical because you can see your work instead of trying to remember it at the end of the week. - Enter the count - Add a note if needed - Save the day If logging takes more than 30 seconds, people usually stop doing it. One tweak worth making early is your time zone. AI-built apps often get dates slightly wrong if you do not specify it. Either tell Replit your time zone up front or ask it to let you edit dates manually. > Pro tip: This works because it is task counting, not time tracking. You are not trying to account for every minute. You are just capturing the outputs that matter. ### Step 4: Generate the Weekly Report Once you have a few days of data in the app, click Generate Report and choose a weekly or monthly range. - Total tasks completed - Daily averages - Your most active day - A short written summary This is the part that makes the tracker useful. The data is already formatted for a manager update, client recap, or team check-in. - “23 client replies” - “8 meetings” - “5 proposals” - “2 launches” Now you are not guessing what you did. You have something clean you can send. ### Step 5: Turn It Into a Daily Habit The app only works if you use it consistently, so the habit matters just as much as the build. The easiest version is to update it at the end of your workday. Open it, log the numbers, add one quick note if something unusual happened, and move on. The point is not to become obsessed with measuring yourself. The point is to have a clean record of what you did so you are not reconstructing your week from Slack, email, and memory. ### Pro tip > Keep task labels tight. “Client response” or “Proposal sent” is easier to count and more useful in reports than a vague category like “Admin.” ### Guide: How To Connect ChatGPT to iMessage (and What It Can Do) > In this guide, you will learn how to connect iMessage to ChatGPT Work. - Canonical page: https://app.therundown.ai/guides/how-to-connect-chatgpt-to-imessage-and-what-it-can-do - Machine-readable page: https://app.therundown.ai/guides/how-to-connect-chatgpt-to-imessage-and-what-it-can-do.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: AI - Instructors: Billy Howell - Published: 2026-08-31 - Updated: 2026-08-31 ### Overview #### The Rundown In this guide, you will learn how to connect iMessage to ChatGPT Work. You will learn the connection's capabilities, limitations and some quick ways to make it work better. #### Who This Is Useful For - People with tons of unread messages. Especially those who fall behind on follow-ups. - Consultants who want a quick list of client action items before the workday starts. - Anyone doing business on an iPhone. #### What You Will Build You will connect ChatGPT to Messages, ask it to check your unread conversations, and text the recap to yourself. You will also learn what it can read, where it works, and how to keep each conversation behind an approval. ![ChatGPT found two unread iMessages and texted the recap to the user's self-chat](https://from-ava.b-cdn.net/guide-assets/local-442/how-to-connect-chatgpt-to-imessage-and-what-it-can-do/01-unread-imessage-recap-texted-to-self.png) #### What You Need To Get Started - A current ChatGPT desktop app on an Apple-silicon Mac - Apple Messages signed in on that Mac - Your own phone number saved in Contacts - Automation, Contacts, and Full Disk Access enabled for ChatGPT #### Going Further For now, the iMessage plugin only works in the desktop app. To use it from your phone, tap Remote in the ChatGPT iOS app and pair it with your Mac. Your Mac still does the work, but you can control the task from your phone. You can also ask ChatGPT to create a noon routine that checks project progress and important Slack @mentions, then texts the recap to your self-chat. In the video, we set up that routine as one daily alert so the most important work lands in a channel that is hard to miss. ![ChatGPT set up a daily noon project and Slack recap routine](https://from-ava.b-cdn.net/guide-assets/local-442/how-to-connect-chatgpt-to-imessage-and-what-it-can-do/05-daily-project-slack-recap-automation.png) Use a dedicated self-chat for automated recaps. Keep every other recipient behind per-send approval, and run the routine once yourself before relying on it. ### Step 1 Install the Messages Plugin Open the ChatGPT desktop app on your Mac. Click Plugins, search for Messages, install the official plugin, then allow Automation, Contacts, and Full Disk Access. Automation lets ChatGPT send an approved message through Messages. Contacts lets it find people by name. Full Disk Access lets it search the Messages stored on your Mac. ![The Apple Messages setup screen asks for Automation, Contacts, and Full Disk Access](https://from-ava.b-cdn.net/guide-assets/local-442/how-to-connect-chatgpt-to-imessage-and-what-it-can-do/02-chatgpt-messages-permissions-setup.png) The plugin only runs in the ChatGPT desktop app on an Apple-silicon Mac. You can use it in ChatGPT Work or Codex, but not in a regular chat. The official Apple Messages documentation (https://learn.chatgpt.com/docs/plugins?surface=app#app-use-apple-messages-from-codex) has the full compatibility list. > Pro tip: There are additional safety filters built in after this step. You must opt each conversation in before ChatGPT can read it. We tested the plugin from a separate macOS work profile and work Apple Account. If you do not want to connect your personal Messages history, this gives ChatGPT a smaller set of conversations and contacts to work with. ### Step 2 Ask ChatGPT to Check Your Unread Messages Start a chat in ChatGPT Work or Codex. Make sure your own phone number is saved in Contacts, then try: ```text Check my unread iMessages and let me know what I need to catch up on. Then text myself the recap. ``` Click the plus button beside the prompt, attach the Messages plugin, and run the task. The slash shortcut did not show the plugin in our test, but the plus button worked. ![Attach the Messages plugin from the plus menu before running the task](https://from-ava.b-cdn.net/guide-assets/local-442/how-to-connect-chatgpt-to-imessage-and-what-it-can-do/03-attach-messages-plugin-to-the-task.png) ### Step 3 Review the Access and Send Approvals ChatGPT will ask for read access to each unread conversation. Use for this query to approve a one-off read. You can also approve a conversation for the rest of the current ChatGPT session or allow it globally. Keep the task on Ask for approval or Approve for me so ChatGPT can show the send confirmation. Full access may stop that panel from appearing. ![Choose whether ChatGPT can read the Messages chat for one query, one conversation, or always](https://from-ava.b-cdn.net/guide-assets/local-442/how-to-connect-chatgpt-to-imessage-and-what-it-can-do/04-choose-message-reading-scope.png) Use the smallest scope that works. Full Disk Access exposes the message history stored on that Mac, so avoid global access for personal or sensitive conversations. ChatGPT will search the conversations you approve and write the recap. Check the recipient and message before you approve the send. ### Step 4 Confirm the Recap Arrived You should receive a text from ChatGPT summarizing your inbox. Open Messages and check your self-chat. Make sure the names, requests, and deadlines match the original conversations. In our test, ChatGPT found two unread chats, pulled out the call and deadline we needed to catch up on, and texted the recap to Myself (Billy). You can also narrow the request to one person or topic. For example, ask it to find the restaurant someone mentioned in an older conversation. ### Guide: How To Design a High-Converting Landing Page in Claude Design > In this guide, you will design a high-converting landing page in Claude Design without opening Figma. - Canonical page: https://app.therundown.ai/guides/how-to-design-a-high-converting-landing-page-in-claude-design - Machine-readable page: https://app.therundown.ai/guides/how-to-design-a-high-converting-landing-page-in-claude-design.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2026-04-20 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will design a high-converting landing page in Claude Design without opening Figma. You will go from a rough brief to a polished, on-brand page that can be deployed from the browser, all in one sitting. #### Who This Is Useful For - Marketers and growth teams that need to spin up new landing pages regularly without a designer on standby - Solo founders and indie makers who want a page that converts instead of a generic template - Consultants and coaches building lead-capture pages for specific services, offers, or audience segments #### What You Will Build You will build a complete landing page mockup with four variations in Claude Design, resulting in an interactive mockup that you can share with others or keep refining with Claude. ![CleanShot 2026-04-20 at 16.40.39](https://tru-images.b-cdn.net/guide-assets/69/69e699670c4619ac4c7d0569/cleanshot-2026-04-20-at-16-40-39.png) #### What You Need - A Claude Pro, Max, Team, or Enterprise account, since Claude Design is included - Access to claude.ai/design - A reference landing page in your category - Screenshots of reference pages or conversion patterns you want Claude Design to borrow from - A brief that defines the audience, offer, and CTA #### Going Further - Export the page as standalone HTML, Canva, PDF, or PPTX depending on how you plan to deploy, review, or present it. - Generate three hero variants and A/B test them while keeping the design system consistent. - Build separate landing pages for different audience segments using the same brand and a different brief for each segment. - Hand off to Claude Code when the page needs backend logic such as bookings, payments, signups, live forms, or calculators. - Claude Design’s UI may change, but the durable workflow stays the same: specific brief, brand upload, wireframe, refine, export. ### Step 1 Find a reference page and write a specific brief The single biggest lever on output quality is the brief. Vague briefs get pretty, generic pages. Specific briefs get pages that convert. Before you open Claude Design, find one landing page in your category that you know works. For e-commerce, find a Shopify template or an Amazon product page. For consulting, find a coach or agency page you respect. Search for “best [your niche] landing pages” if you need examples. Screenshot your reference. Then write a brief that names three things: - Audience: Who is this page for, and what do they already know? - Offer: What specifically are you selling, at what price, and on what terms? - CTA: What is the one action a visitor should take? > Pro tip: Go one step further. Screenshot a conversion pattern from a site people already trust, such as Amazon’s add-to-cart flow or YouTube’s subscribe button, and tell Claude Design to mimic it for your CTA. People pattern-match with what they already know, so borrow the pattern. ### Step 2 Start with Create Wireframe, not a full build Go to claude.ai/design and start a new project. Click Create Wireframe to start, not the full site build. There are two reasons to start with a wireframe: it uses fewer tokens, and seeing the structure first makes it easier to iterate before Claude commits to colors, imagery, and typography. Drag your reference page screenshot and any pattern screenshots into the input. Paste your brief. Add the URL of a site whose visual style you want to match, and Claude Design’s web-capture tool will pull colors, fonts, and components automatically. Here is the example prompt used in the guide: ```text ProductivitybyBilly.com make four variations of a landing page to get people to sign up for a one hour free call where I diagnose what's wrong with their current productivity system and give them a report and then eventually upsell into a four-hour coaching package. Make at least one of the heroes have the elements of the attach example site and then I want you to make the check out pain as similar to Amazon as Possible to take advantage of pattern matching. Give me four variations of the mock-up. ``` ![CleanShot 2026-04-20 at 17.26.09@2x](https://tru-images.b-cdn.net/guide-assets/69/69e69a000c4619ac4c7d072d/cleanshot-2026-04-20-at-17-26-09-2x.png) > Pro tip: Claude Design does not accept HEIC files, which are often the iPhone default when you AirDrop from your phone. Convert to PNG or JPEG first or the upload will fail. ### Step 3 Generate variations and scroll before you edit Claude Design briefly interviews you after the first prompt. It asks things like price, number of clients, review counts, and what specific elements a section should borrow. Answer concisely. Select how many wireframe variations you want. Three to five is the sweet spot. Hit generate. The first pass takes 60 to 90 seconds. Scroll every variation from top to bottom before you edit anything. The structural pieces—layout, hierarchy, and section order—are usually right on the first pass. What usually needs fixing is copy specificity and CTA wording. ### Step 4 Refine with Tweaks first, comments second Claude Design gives you two refinement paths. Use them in order. Start with the bottom-right Tweaks menu to test different color schemes and variations without changing anything. Then use inline comments for local polish. Click any element and leave a note. You can either save the comment for later or hit Send to Claude to make the change right now. - “Rewrite this CTA to be outcome-specific, not generic.” - “Add a testimonial placeholder here.” - “Tighten this paragraph; it is three sentences of fluff.” ![CleanShot 2026-04-20 at 17.30.55@2x](https://tru-images.b-cdn.net/guide-assets/69/69e69b290c4619ac4c7d0b01/cleanshot-2026-04-20-at-17-30-55-2x.png) > Pro tip: There is a Grab Web Element bookmarklet in the tools menu that most people miss. Drag it to your bookmarks bar once, then click it on any webpage to pull any element—nav, pricing card, or hero block—directly into Claude Design. It is useful for grabbing a pattern you saw somewhere and wiring it into your page. ### Step 5 Export and ship, or hand off to Claude Code When the page is ready, you have four export paths: - Standalone HTML to deploy to any host - Canva for team review or extra visual polish - PDF for stakeholder sign-off - PPTX if the page doubles as a sales asset When the page needs real logic, the better move is the Claude Code handoff. Claude Design and Claude Code share a project format, so you can copy a single command into a local Claude Code session or send the whole project to Claude Code on the web without spinning up a local instance. Use the handoff when your page needs live forms, calculators, signup flows, or booking integrations. Claude Design is best for the visual shell. Claude Code makes it functional. ### Pro tip > Start with a specific brief and a trusted conversion pattern. Vague briefs produce generic pages; specific briefs, reference screenshots, and familiar CTA patterns give Claude Design much better material to work from. ### Guide: How To Do a Brand Refresh in Five Minutes With Claude Design > This guide shows you how to use Claude Design to create a refreshed brand design system from an existing website and brand brief. - Canonical page: https://app.therundown.ai/guides/how-to-do-a-brand-refresh-in-five-minutes-with-claude-design - Machine-readable page: https://app.therundown.ai/guides/how-to-do-a-brand-refresh-in-five-minutes-with-claude-design.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-04-26 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to use Claude Design to create a refreshed brand design system from an existing website and brand brief. You will produce typography direction, color tokens, reusable web components, website direction, slide templates, and product UI concepts. The source says Claude’s generation can take about five minutes, but you should budget closer to 10–15 minutes if the tool is busy. #### Who This Is Useful For - Founders and consultants who already have a real brand but need a sharper visual system quickly - Marketers and operators who want a better starting point before handing work to a designer or developer - Anyone building decks, landing pages, or product mockups who wants one reusable system across everything #### What You Will Build You will build a reusable brand system, not just a prettier homepage. The output is a set of assets and directions you can keep building on. - Typography direction and design tokens - A cleaner color system with primary, accent, neutral, surface, and status colors - Reusable web components like buttons, cards, inputs, and badges - A refreshed website direction - Slide templates and product UI concepts built from the same system #### What You Need - A paid Claude plan with access to Claude Design - A website you can screenshot - Another AI tool for the brand refresh step; the source example used ChatGPT, but you can use whatever you prefer - Optional logo files, wordmarks, and brand fonts if you already have them #### Going Further - After Claude finishes, review the marketing page, web app page, and slide decks it created. Then move the strongest parts into a real project and keep building from the system. - Click Share and export the design files to Claude Code - Turn the marketing page into a real web project - Use the slide templates as the starting point for future decks - Turn the system into a PDF brand kit for clients or teammates ### Step 1: Create the refreshed brand description Screenshot your current site and drop it into your preferred AI tool. Then paste this prompt: ```text Analyze this website and create a refreshed brand description. Keep what works, but sharpen the positioning, visual direction, typography, and color palette. Give me a concise brand summary I can use in Claude Design. ``` In the source test, Claude did not want to redesign an existing brand directly, so ChatGPT was used for this part. The key output is a clean refresh brief you can reuse. If the response is too long, ask the tool to tighten the brand description before you move on. ### Step 2: Generate a logo and wordmark Take the refreshed brand description and ask your AI tool to generate a logo and a wordmark logo. The source example used ChatGPT for its image model, but you can use whichever tool you prefer. After generating the logo and wordmark, ask for the full brand refresh in one clean block of text so you can copy and paste it into Claude Design without cleaning it up by hand. Download the logo and wordmark files. Even rough first-pass assets can help Claude Design produce a better result. ### Step 3: Build the design system in Claude Design Go to claude.ai/design and click the Design systems tab. Create a new design system and follow this setup: - Paste the name of your site in the first box. - In the last box, paste the refreshed brand description from Step 1. - Upload the logo, wordmark, and any other photos or assets you want Claude to use. Claude Design may also offer GitHub and Figma inputs. Those are useful if you have them, but you do not need them for a strong first pass. Start the generation when everything is in place. Claude says this takes about five minutes. In the source test, it often took longer, so budget closer to 10 or 15 minutes if the tool is busy. ### Step 4: Review the system and export the good parts Once the system finishes, do not stop at the first polished screen. Open the foundations and review what Claude made. In the source run, Claude generated: - Teal as the primary color family - Lime as the accent family - Graphite neutrals, surfaces, and status colors - Spacing rules, corner radii, and elevation - Component groups for buttons, status pills, inputs, cards, list rows, and charts The neutral system was especially strong, including Rundown Black #111112, Academic Black #0F0F12, Graphite #232326, Graphite Raise #2D2D31, Soft White #F7F5F0, Chalk White #F5F2EA, Muted Stone #A8A29E, and Stone Deep #7A7670. The same system also produced a refreshed learning dashboard and could be stretched across other outputs like slide templates and app UI. ### Pro tip > Ask for a refresh, not a reinvention. You want the brand to feel better, not unrecognizable. ### Guide: How To Fix Bugs and Ship Features From Slack Using Claude Code > Learn how to add Claude Code to Slack and use it to fix bugs or implement new features without opening a code editor. - Canonical page: https://app.therundown.ai/guides/how-to-fix-bugs-and-ship-features-from-slack-using-claude-code - Machine-readable page: https://app.therundown.ai/guides/how-to-fix-bugs-and-ship-features-from-slack-using-claude-code.md - Difficulty: intermediate - Featured tool or company: Claude - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2025-12-10 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to add Claude Code to Slack and use it to fix bugs or implement new features without opening a code editor. By the end, you’ll have a Slack-based workflow for assigning coding work to Claude Code from a channel or thread. #### Who This Is Useful For - Small teams working in Slack - Claude Code users who need to work on the go - Operators who want to minimize context switching #### What You Will Build You’ll set up Claude Code inside Slack so you can assign coding tasks from Slack, including from the mobile app. - A connected Claude Code and GitHub setup - The Claude app added to your Slack workspace - A Slack channel where Claude can receive coding tasks - A workflow for asking Claude to fix bugs or ship features from a thread ![05_mobile_demo-ezgif.com-resize](https://tru-images.b-cdn.net/guide-assets/69/6939f671832269e20994f818/05-mobile-demo-ezgif-com-resize.gif) #### What You Need - Claude Pro account - GitHub account - At least one repository to test this on; a repository with a single HTML file will do - Paid Slack workspace #### Going Further - Ask Claude to create a single-page HTML file, like the simple galaxy animation from the source example. - Set up GitHub Pages with GitHub Actions so approved and merged Claude Code pull requests can automatically deploy your updated site. ### Step 1: Connect Claude Code to GitHub Connect Claude Code to your GitHub account. ![02_connect_claude_code_github](https://tru-images.b-cdn.net/guide-assets/69/6939dd84d5ccaa22378e9cc4/02-connect-claude-code-github.png) ### Step 2: Add Claude to Slack and connect your account Add the Claude app to your Slack workspace. After it appears in Slack, click “Connect Account” and give Claude access to Slack. ![01_claude_in_slack](https://tru-images.b-cdn.net/guide-assets/69/6939dd85d5ccaa22378e9ccc/01-claude-in-slack.png) > By default, the app will be set to “code only” mode. You can switch to “Code + Chat” if you still want to use Claude like an AI assistant in your workspace. You might do this if you have Claude connected to other services like Notion or Google Drive. ![03_claude_slack_options](https://tru-images.b-cdn.net/guide-assets/69/6939dd83d5ccaa22378e9cb7/03-claude-slack-options.png) ### Step 3: Add Claude to a Slack channel Claude Code will not work if you mention it in a direct message. It must be used in a Slack channel. Add Claude to an existing channel, or create a dedicated channel for coding tasks. - Open your Slack channel and type “@claude”. - Hit enter. - When Slack asks you to add Claude to the channel, approve it. ### Step 4: Give Claude coding tasks from Slack You can now @Claude and give it coding tasks. Claude will build context based on recent messages, and it works best if you start a thread on a particular issue. ![04_claude_response](https://tru-images.b-cdn.net/guide-assets/69/6939dd83d5ccaa22378e9caf/04-claude-response.png) As a bonus, you can now access Claude Code from your phone through the Slack app. ### Step 5 Optional: Set up a simple site and automatic deployment As a next project, ask Claude to write a single-page HTML file. In the source example, Claude created a simple galaxy animation. ![06_galaxy_screenshot](https://tru-images.b-cdn.net/guide-assets/69/6939de49193e1e46713377b6/06-galaxy-screenshot.png) - Open your GitHub repository and go to Settings > Pages. - Follow the instructions to set up static deployment through GitHub Actions. - After that, when you approve and merge a pull request from Claude Code, GitHub will automatically deploy your updated site. ### Resources - [Claude Code + Slack Guide](https://tru-images.b-cdn.net/guide-assets/69/6939dcc8ed9c7de4141df76c/claude-code-plus-slack-guide.pdf) ### Pro tip > Add the URL or name of the relevant GitHub repository to the channel details so Claude always picks the correct project. ### Guide: How To Generate a Cohesive E-Commerce Product Shoot Using Grok AI (for Free) > This guide shows you how to generate a cohesive ecommerce product photo set using Grok’s Aurora model. - Canonical page: https://app.therundown.ai/guides/how-to-generate-a-cohesive-e-commerce-product-shoot-using-grok-ai-for-free- - Machine-readable page: https://app.therundown.ai/guides/how-to-generate-a-cohesive-e-commerce-product-shoot-using-grok-ai-for-free-.md - Difficulty: intermediate - Featured tool or company: xAI - Categories: Marketing & Growth - Instructors: Billy Howell - Published: 2026-03-17 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to generate a cohesive ecommerce product photo set using Grok’s Aurora model. You will start with one phone photo, create an anchor studio image, then build additional images from that shot so the full set looks like it came from the same session. #### Who This Is Useful For - Anyone selling a physical product online who cannot justify a $500+ photo shoot every time they launch or refresh a listing - Ecommerce marketers who need consistent product imagery across dozens of SKUs without a photographer for every batch - Side hustlers testing a product idea who need professional-looking images before investing in a full production run #### What You Will Build You will create a reusable set of 5–6 product images covering the kinds of visuals high-converting listings use. - Clean hero shot - Styled studio shot - Alternate angle - Detail close-up - Lifestyle image - Flat lay ![image-2127](https://tru-images.b-cdn.net/guide-assets/69/69b9d1d2eb42983dce7b9a28/image-2127.png) #### What You Need - A photo of your product; a phone photo works - A free account at grok.com - Optional: X Premium ($8/month) or SuperGrok ($30/month) for higher image limits - Access note: Free accounts get roughly 3–10 image generations per day. X Premium gets around 40 prompts per 2-hour window, and SuperGrok is effectively unlimited. xAI does not publish official limits, so these numbers come from user reports and may shift. #### Going Further - Create seasonal refreshes by swapping the surface and lighting for holiday campaigns or seasonal drops. - Batch your catalog by running the same prompt framework across every SKU for consistent brand imagery across your store. - Layer on infographics by using the clean studio shots as base images, then adding text overlays, dimensions, or feature callouts in Canva for secondary listing images. ### Step 1: Generate your anchor shot Go to grok.com, upload your product photo, and generate your first styled studio shot. This will become the anchor image that the rest of the shoot builds from. Aurora responds well to photography terminology, so write your prompt like a creative brief you would hand to a photographer. ```text Product photo of [your product] on a matte black surface, soft box lighting from the upper left, subtle shadow, dark gray gradient background, shot on 85mm lens, f/2.8, shallow depth of field, commercial product photography, minimalist aesthetic ``` > Pro tip: Switch up the surface and background gradient colors to better match your brand. Once this image is generated, you have your anchor shot. ### Step 2: Build the rest of the studio set To keep the studio environment consistent, select your anchor shot before running the next prompts. If you cannot find it, drag the image in or upload it again. ![image-2125](https://tru-images.b-cdn.net/guide-assets/69/69b9d1b70dc60da3a2ad85b2/image-2125.png) Generate a three-quarter angle shot: ```text Same product, same studio setup. Three-quarter angle from the right, soft box lighting from upper left, matte black surface, dark gray background, 50mm lens, sharp focus with soft background blur ``` Generate a detail close-up: ![image-2126](https://tru-images.b-cdn.net/guide-assets/69/69b9d1c4eb42983dce7b9185/image-2126.png) Swap out “logo and fabric texture” for details that are relevant to your product. ```text Extreme close-up of [your product] focusing on the logo and fabric texture, macro photography, soft directional light from the left, 100mm macro lens, f/4, same studio as previous shots ``` Lay these images next to your anchor shot. They should feel cohesive: the same studio energy, with different angles and details. ### Step 3: Add lifestyle and context shots Create stylized images that can work on your product page, in ads, or on social media. Swap in your specific product and the way people use it. Generate a lifestyle shot: ```text A person wearing the hat, in the same studio, shot on an 85mm lens, editorial sports photography. ``` Swap out “editorial sports photography” for a photography style that fits your brand. Generate a flat lay: ```text Flat lay of [your product] on a desk in the same studio alongside sunglasses, a game ticket, and a coffee cup, overhead shot, natural window light, warm tones, magazine editorial style ``` Replace the sunglasses, game ticket, and coffee cup with items relevant to your product. You can also reuse the same warm or cool color grading language across your lifestyle shots to make them feel like a set. > Pro tip: When writing lifestyle prompts, name a real camera and lens. “Shot on Canon EOS R5 with 85mm lens” gets noticeably better results than just saying “photograph.” ![image-2128](https://tru-images.b-cdn.net/guide-assets/69/69b9d238eb42983dce7ba431/image-2128.png) Lay out all your shots in Canva or another workspace and adjust them as needed. ### Pro tip > Once your prompts are dialed in, reuse the same framework across products, seasons, and listing formats to keep your brand imagery consistent. ### Guide: How To Generate a TV Commercial With AI > In this guide, you’ll learn how to create a professional-looking TV commercial using AI. - Canonical page: https://app.therundown.ai/guides/how-to-generate-a-tv-commercial-with-ai - Machine-readable page: https://app.therundown.ai/guides/how-to-generate-a-tv-commercial-with-ai.md - Difficulty: beginner - Featured tool or company: Higgsfield AI - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-02-12 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you’ll learn how to create a professional-looking TV commercial using AI. The same workflow can also be used for YouTube ads, Facebook ads, or other short video ads. The source workflow uses $9 software and takes around 20 minutes. #### Who This Is Useful For - Small business owners who want to create polished video ads without hiring a production team - Content creators and marketers looking to prototype commercial concepts quickly and cheaply - AI enthusiasts who want to explore the latest video generation tools in a hands-on, practical workflow #### What You Will Build You’ll build a short AI-generated commercial from a two-scene script, four generated key frames, and two AI-generated video clips that are spliced into a final ad. ![image-2054](https://tru-images.b-cdn.net/guide-assets/69/698e609a420e08c0eb8ee7e1/image-2054.png) - A 20-second, two-scene commercial script - Start and end frame prompts for each scene - Four generated images to use as scene frames - Two generated video clips - A spliced commercial edited in CapCut or with ffmpeg #### What You Need - Higgsfield.ai basic plan or higher ($9+) - CapCut desktop app (free) - A TV commercial idea - Gemini for scripting and frame prompt generation - Optional: custom voiceover, sound effects, or music ideas #### Going Further - After generating your clips, download them from Higgsfield, splice them together, and improve the audio. For more polish, mute the generated clips and add sound effects, music, or voiceover using CapCut, Suno, or ElevenLabs. ### Step 1 Write a two-scene commercial script in Gemini Go to Gemini.google.com and describe your TV commercial idea. Ask Gemini for a 20-second script with exactly two scenes. Review the script and revise it as needed. If you already have a custom voiceover, sound effects, or music in mind, include those details in the script request. ```text I have a TV commercial idea: [describe your idea]. Write a 20-second script with exactly two scenes. Include voiceover, sound effects, or music notes if relevant. ``` Next, ask Gemini to generate prompts for the start frame and end frame of each scene. Tell it to use photography and film terms to keep a consistent aesthetic. ```text Generate prompts for the start frame and end frame of each scene. Use photography and film terms to keep a consistent aesthetic. ``` > 💡 Pro tip: Use photography terms to describe the camera and the angles of your frames. Try phrases like: anamorphic lens, macro lens, 14mm wide angle, hero shot, or Dutch angle. ### Step 2 Set up your image generation in Higgsfield Sign in to Higgsfield.ai. Click Image > Create Image > Select Nano Banana Pro. Set the quality to 4K and variations to 4. ![image-2055](https://tru-images.b-cdn.net/guide-assets/69/698e60b7420e08c0eb8ef01b/image-2055.png) Set the aspect ratio to 21:9. This is the cinematic aspect ratio Google uses when training Nano Banana on films. ### Step 3 Create and download the four key frames Paste in the prompt for your scene 1 start frame. When it is finished, choose the version you like best and click the heart button so you can find it later. Repeat the process with the scene 1 end frame prompt and your scene 2 frame prompts. By the end, you should have 4 images total. Download the 4 images to your desktop. > 💡 Pro tip: Drag the start frame into your prompt box along with your end frame prompt. This helps keep characters and settings consistent within each scene. ### Step 4 Generate the video clips In Higgsfield, navigate to Video > Create Video and select the latest Kling model. The source example uses Kling 3.0. Upload your start frame and end frame images. ![image-2056](https://tru-images.b-cdn.net/guide-assets/69/698e60dba01da8fcd4149372/image-2056.png) Copy this scene’s section from your Gemini script and put it in the prompt box. Adjust the clip length as desired, then click generate. Repeat the same process for scene 2. ### Step 5 Download, splice, and polish the commercial When both scenes are generated, download them from your asset library in Higgsfield. Splice the scenes together in a free desktop app like CapCut. Alternatively, you can ask Claude Code, Codex, or Claude Cowork to splice them together using ffmpeg in the terminal. For next-level polish, focus on audio. Try muting your generated clips, then use CapCut’s audio library to add sound effects and music. You can make even better audio using tools like Suno for music and ElevenLabs for voiceovers. ![image-2057](https://tru-images.b-cdn.net/guide-assets/69/698e60f3a01da8fcd4149bcc/image-2057.png) ### Pro tip > Use photography terms to describe the camera and angles of your frames, such as anamorphic lens, macro lens, 14mm wide angle, hero shot, or Dutch angle. ### Guide: How To Go From Screenshot To Bug Fix With Cursor Mobile > In this guide, you will learn how to screenshot a bug on your website from your phone, send it to a Cursor Cloud agent, and review the resulting pull request in GitHub before you make it back to your desk. - Canonical page: https://app.therundown.ai/guides/how-to-go-from-screenshot-to-bug-fix-with-cursor-mobile - Machine-readable page: https://app.therundown.ai/guides/how-to-go-from-screenshot-to-bug-fix-with-cursor-mobile.md - Difficulty: beginner - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-07-05 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to screenshot a bug on your website from your phone, send it to a Cursor Cloud agent, and review the resulting pull request in GitHub before you make it back to your desk. The new Cursor mobile app might be the best way to maintain codebases on the go when the issue is visible, scoped, and ready for a quick agent pass. #### Who This Is Useful For - Founders who notice broken pages or checkout issues while away from their laptop. - Developers who want to kick off a first-pass fix from a customer screenshot or mobile QA note. - Anyone who maintains a live site and wants a faster path from visible bug to reviewable PR. #### What You Will Build You will set up a mobile bug-fix loop: screenshot the issue, start a Cursor Mobile thread on the right repo, attach the screenshot, ask for a small fix, and review the PR from GitHub Mobile. ![Cursor Mobile PR review screen](https://tru-images.b-cdn.net/guide-assets/local-300/how-to-go-from-screenshot-to-bug-fix-with-cursor-mobile/01-cursor-mobile-pr-review-guide.png) By the end, you should have a repeatable workflow for turning mobile QA moments into GitHub pull requests with changed files, validation notes, and a clear review path. #### What You Need To Get Started - A Cursor Pro plan. - The official Cursor iOS app from Anysphere (https://apps.apple.com/us/app/cursor/id6767085653). - The official GitHub Mobile app (https://apps.apple.com/us/app/github/id1477376905). - A GitHub repo connected to Cursor. - A website, app screen, or customer-reported issue you can safely screenshot. #### Going Further On desktop, enable Remote Agents so Cursor can work on your machine and prep local changes before you get back to your desk. That is useful when a mobile bug report turns into a deeper fix that needs your local environment, existing branches, or desktop-only testing. You can also turn the prompt into a reusable incident template: ```text Investigate this bug on [page or route]. Use the attached screenshot and note to find the relevant component, make the smallest safe fix, and open a PR. Include changed files, validation notes, and anything I should review before merging. ``` Use that template for support tickets, QA screenshots, and urgent UI regressions. Keep the task small, keep the screenshot clean, and use GitHub review as the checkpoint before anything ships. For more context on the mobile release, Cursor's official iOS launch post (https://cursor.com/blog/ios-mobile-app) and launch thread (https://x.com/cursor_ai/status/2071641103191998810) show how Cursor expects people to start agents, track work, and move PRs forward from a phone. ### Step 1: Install The Right Mobile Apps Start by installing the official Cursor iOS app from Anysphere (https://apps.apple.com/us/app/cursor/id6767085653). Cursor's iOS launch post (https://cursor.com/blog/ios-mobile-app) describes the app as a way to start agents from your phone, track work, review artifacts, and move PRs forward while you are away from your computer. You will also want GitHub Mobile (https://apps.apple.com/us/app/github/id1477376905). Cursor can create and update the PR, but GitHub is where you will review the diff, approve it, merge it, or leave follow-up comments from your phone. > Pro tip: Use the Anysphere App Store link above. There are fake Cursor apps in the App Store, and grabbing the wrong one will break the workflow before you even get to the repo. You also need a Cursor Pro plan for this workflow. Get that handled before you try to start a mobile agent so you are not troubleshooting plan access in the middle of a bug report. ### Step 2: Capture The Bug From Your Phone Open the website or app screen you are working on from your phone. Then take a screenshot of the bug, broken layout, or UI change you want fixed. The screenshot should show the problem clearly. If the issue is a broken CTA, bad spacing, hidden button, weird modal, or mobile layout regression, keep the important area visible. If the screenshot includes private customer data, tokens, internal URLs, or anything sensitive, clean it up before sending it to the agent. Add a short note with the page, feature, or route and the expected behavior. Keep it plain: ```text Page: [page or route] Issue: [what is visibly broken] Expected: [what should happen instead] Context: attached screenshot from my phone ``` > Pro tip: Cursor Mobile works best when the bug is visible and scoped. Save broad redesigns, security-sensitive changes, unclear repro steps, and heavy local debugging for desktop. ### Step 3: Start A Cursor Mobile Thread Open Cursor Mobile and go to the workspace or repo list. Choose the repo that owns the page you just screenshotted. ![Cursor Mobile workspaces screen](https://tru-images.b-cdn.net/guide-assets/local-300/how-to-go-from-screenshot-to-bug-fix-with-cursor-mobile/02-cursor-mobile-workspaces-guide.png) If you are using All Repos, pick the correct project before you start the agent. This is a small step, but it prevents the agent from wasting time in the wrong codebase. ![Cursor Mobile repo list](https://tru-images.b-cdn.net/guide-assets/local-300/how-to-go-from-screenshot-to-bug-fix-with-cursor-mobile/03-cursor-mobile-repo-list-guide.png) Tap the plus button to start a new thread. Attach the screenshot before you send the prompt so the agent has visual context from the beginning. ![Cursor Mobile new thread composer](https://tru-images.b-cdn.net/guide-assets/local-300/how-to-go-from-screenshot-to-bug-fix-with-cursor-mobile/04-cursor-mobile-new-thread-guide.png) Use a short prompt that names the target page and asks for the smallest useful fix: ```text Investigate this bug on [page or route]. Use the attached screenshot, find the relevant component, make the smallest safe fix, and open a PR. ``` If you know the likely file, component, or recent change, add one short sentence after the prompt. Do not turn the phone prompt into a full spec. The goal is to give the agent enough context to start, then let the PR and validation notes carry the details. ### Step 4: Track The Agent And Review The Evidence After the task starts, Cursor will show progress in the mobile thread. You can leave the app and come back when the agent needs input or finishes. ![Cursor Mobile agent progress](https://tru-images.b-cdn.net/guide-assets/local-300/how-to-go-from-screenshot-to-bug-fix-with-cursor-mobile/05-cursor-mobile-agent-progress-guide.png) When the agent returns, look for evidence before you approve anything. A useful result should include the changed files, a root-cause summary, validation notes, and a PR you can inspect. Screenshots, logs, preview links, or test output are even better. If the agent says the bug is fixed but gives no evidence, ask for a tighter summary before reviewing the PR: ```text Before I review this, summarize the files changed, why those files were chosen, how you validated the fix, and what I should check first in GitHub. ``` This keeps the mobile workflow useful without turning it into blind trust. The phone is great for starting the fix and triaging the result. The PR still deserves a real review. ### Step 5: Review The PR In GitHub Mobile Cursor will send you a notification when the work is ready. Tap through to the PR and review it in GitHub Mobile. Start with the file list and diff. Confirm the change is scoped to the bug you reported, then check the validation notes from the Cursor thread. If the fix is clean, approve and merge it from GitHub. If it is close but needs edits, leave a PR comment and tag @Codex with the exact change you want. Good follow-up comments are specific: ```text @Codex tighten the mobile spacing only on the checkout CTA. Do not change the desktop layout. Add a short note explaining how you verified the mobile view. ``` Save the Cursor thread, PR link, screenshot, and final summary if this is part of a customer issue or incident. That gives you a clean audit trail when you are back at your desk. ### Guide: How To Let Claude Use Your Computer Remotely With Dispatch > This guide teaches you how to use Dispatch to hand Claude a task from your phone or desktop and have it finish the work on your computer. - Canonical page: https://app.therundown.ai/guides/let-claude-use-your-computer-remotely-with-dispatch - Machine-readable page: https://app.therundown.ai/guides/let-claude-use-your-computer-remotely-with-dispatch.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Automation & Agents, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-03-24 - Updated: 2026-07-23 ### Overview #### The Rundown This guide teaches you how to use Dispatch to hand Claude a task from your phone or desktop and have it finish the work on your computer. You will set up Dispatch, pair it with your phone, and run a simple remote task so small admin work can keep moving while you are away from your desk. #### Who This Is Useful For - Operators and chiefs of staff who want inbox cleanup, file organization, or quick recaps handled while they are moving between meetings - Founders and solo operators who switch between phone and desktop and want one continuous thread instead of restarting the task every time - Anyone testing Cowork seriously who wants a safe first workflow for remote computer use before handing Claude bigger jobs #### What You Will Build ![image-2142](https://tru-images.b-cdn.net/guide-assets/69/69c3087d9b6fb4f9889a5557/image-2142.png) You will build one repeatable Dispatch workflow for remote computer use. The workflow can be reused for tasks like email cleanup, file organization, or an end-of-day work summary. - A working Dispatch setup in Claude Desktop - A paired Claude mobile app connection - A tested first remote task Claude can run on your computer - A repeatable pattern for reviewing what Claude did and where it saved results #### What You Need - A Claude Pro plan or higher - The latest Claude Desktop app - The latest Claude iOS app - The Claude in Chrome extension installed and logged in - Optional: a dedicated local project folder for your first run #### Going Further - Once the basic workflow is working, reuse Dispatch for practical situations where you are away from your desk, such as opening your calendar, meeting notes, and Zoom before a virtual meeting, or preparing a report of the emails and Slack messages you missed during lunch and opening it on your desktop. ### Step 1: Update Claude and start Dispatch Start on your computer. Open Claude Desktop and make sure it is current before you do anything else. On Mac, click Claude > Check for Updates in the menu bar. Update the Claude mobile app on your phone too. Once both apps are current, open Cowork in Claude Desktop, click Dispatch, and click Get started. ### Step 2: Set permissions and prepare a safe workspace During setup, Claude will ask for permissions. Turn on file access if you want Claude to read and write inside a local folder. Turn on keep-awake if you want the task to keep running while you step away. ![image-2143](https://tru-images.b-cdn.net/guide-assets/69/69c308899b6fb4f9889a5dde/image-2143.png) Make sure you have the Claude in Chrome extension installed and logged in. Optional: Set up a dedicated local project folder before your first run. It is easier to monitor what Claude touched, easier to review outputs, and much safer than giving access to your whole machine. ### Step 3: Pair mobile and run a first remote task After the desktop setup, you will be asked to pair Dispatch in the Claude mobile app. Once the connection is live, send Claude a simple task from your phone and let it continue the work on your computer. Good first tasks are: - Triage, label, and archive all your emails from today - Organize files in a project folder - Compile a quick summary from files already on your machine - Generate an end-of-day report from the work you finished today Start with something reversible and easy to review. Dispatch is strongest when the task is useful but not high risk. Remember, you do not want to create more work for yourself when you come back to your desk. To make the review easier, tell Dispatch to write a log file of what it did on your computer. ### Step 4: Review the result and reuse the workflow When Claude finishes, check the output on desktop and in the same synced conversation on mobile. You should be able to see what it did, where it saved the result, and what still needs your input. Dispatch currently keeps everything inside one continuous thread on your phone. That makes it easy to pick back up later, but it can also feel messy if you expect every mobile request to behave like a clean new chat. In the desktop app, Dispatch will start new threads under Cowork, and they will be labeled as Dispatch threads. ![image-2144](https://tru-images.b-cdn.net/guide-assets/69/69c3089c9b6fb4f9889a6679/image-2144.png) Dispatch gets more valuable when it becomes a habit instead of a stunt. The core pattern is simple: send a task from your phone, let Claude do the desktop work, then come back to something cleaner and more organized than you left it. ### Pro tip > If you go to Settings > General > Computer Use, you can toggle on an experimental feature that allows Claude to use your keyboard and mouse to use any app on your computer. This is worth experimenting with, but leaving it off by default can reduce workflow slowdowns and avoid giving Claude too many options, as well as limit security concerns. ### Guide: How to One-Shot a Landing Page With Replit Design Mode > Learn how to create a unique, high-converting landing page for your business in under 5 minutes using Replit Design Mode. - Canonical page: https://app.therundown.ai/guides/how-to-one-shot-a-landing-page-with-replit-design-mode - Machine-readable page: https://app.therundown.ai/guides/how-to-one-shot-a-landing-page-with-replit-design-mode.md - Difficulty: beginner - Featured tool or company: Replit - Categories: Coding & App Development, Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2025-12-04 - Updated: 2026-07-23 ### Overview #### The Rundown Learn how to create a unique, high-converting landing page for your business in under 5 minutes using Replit Design Mode. You’ll produce a detailed landing page prompt, refine it with a visual style reference, and use it to generate a custom single-page site in Replit. No coding required. #### Who This Is Useful For - People who are tired of generic AI designs - Solo founders who need to launch fast - Experimenters who want to test ideas with real landing pages - Anyone who needs a beautiful landing page without touching code #### What You Will Build You’ll build a custom single-page marketing site for your business that can be refined or published in Replit. ![replitdesignmodelandingpageexample](https://tru-images.b-cdn.net/guide-assets/69/6931a733aaa981a77eb55cab/replitdesignmodelandingpageexample.png) - A reusable prompt tailored to your company, audience, offer, and conversion goal - A more specific design direction based on a reference landing page - A generated landing page in Replit Design Mode #### What You Need - Your company or business name, audience, offer, and conversion goal - An AI chat tool, such as Claude or another AI agent - A Replit account with access to the Design tab - Optional: a landing page reference screenshot from Google, Behance, or Dribbble #### Going Further - Update your social proof, testimonial, and offer sections via Replit Agent - Ask Replit Agent to embed a popup form or lead capture from your CRM or form builder - Keep the page simple and lean - Buy a custom domain and deploy the site via Replit - Deploy the Replit to an existing site’s subdomain if the landing page is for paid ads or another outbound campaign ### Step 1: Build your prompt Open an AI chat tool, such as Claude, and paste in the prompt formula below. ```text Use this formula to create a prompt for my company [company name] with [audience] and [offer]. The goal is to get the user to [conversion goal]. This is a single page site. [OFFER & CONVERSION GOAL] What you're offering: [lead magnet/product/service/event] Target action: [signup for emails/register for webinar/make purchase/schedule demo/etc.] What visitor gets: [specific benefit they receive for taking action] [BUSINESS & AUDIENCE] Build a single-page marketing site for [your business/product/offer]. Target audience: [who they are + specific pain point they're experiencing] [PAGE SECTIONS] Include these sections: - Hero: [headline that speaks to pain point, clear CTA, supporting visual] - [Value proposition/Problem-Solution] - [Features/Benefits] - [Social proof/Trust signals] - [Final conversion section] [VISUAL STYLE & MOOD] Style: [modern/minimal/bold/playful/professional/luxury] Colors: [brand colors or color mood] Overall vibe: [how should it feel - urgent/calm/exciting/trustworthy] [CONVERSION ELEMENTS] - Primary CTA: [exact button text + where it appears] - Form fields needed: [name/email/phone/company/etc.] - Trust signals: [testimonials/logos/guarantees/stats] ``` Edit the initial paragraph with your business details. You can let AI fill in the rest of the formula based on what you’ve provided. Generate your initial prompt, then refine it in the next step. ### Step 2: Add specificity with a reference design If you only use the prompt from Step 1, you may get a generic landing page design. LLMs need specificity to be creative; otherwise they can default to vague style instructions like “bold, modern, edgy.” ![replitgenericlandingpageexample](https://tru-images.b-cdn.net/guide-assets/69/6931a94e92e5a39431f2fe70/replitgenericlandingpageexample.png) The secret sauce is to use a visual reference. Open Google or a design database like Behance or Dribbble and search for “high converting landing pages” or “best agency landing pages.” Screenshot your favorite and drop it into Claude or another AI agent. ```text Rewrite the prompt to be in the style of the attached screenshot. Replit can generate images so I'd like it to generate some relevant clip art like on this landing page example. ``` > 💡 Pro Tip: You don’t have to include the part about clip art. But to get consistent, unique designs, it’s good to instruct Replit to generate assets rather than just searching the web for images. ### Step 3: Put the prompt in Replit Sign up or sign in to Replit and click the Design tab. Paste your prompt there. Because the prompt is long, it will attach as a text file. ![replitpromptexample](https://tru-images.b-cdn.net/guide-assets/69/6931a9d43542292b7b4d9ed6/replitpromptexample.png) ```text Generate a single page landing page site based on the attached instructions. Generate assets as needed. ``` Hit the start button. > 💡 Pro Tip: You don’t need to upload your reference photo from Step 2 into Replit. Let the prompt do the talking. In just a few moments, Replit will create a custom landing page that you can further refine or publish. ### Resources - [How to One-Shot a Landing Page With Replit Design Mode PDF](https://tru-images.b-cdn.net/guide-assets/69/6931b64cc0c1bb8aac56efce/9d4679de-f0ef-47b7-973a-91f552fd40a72fhow-to-one-shot-a-landing-page-with-replit-design-mode-compressed.pdf) ### Pro tip > To get consistent, unique designs, instruct Replit to generate assets rather than just searching the web for images. ### Guide: How To Pick a Dedicated AI Device > In this guide, you will learn which hardware devices are good for AI tinkering and which one is right for you. - Canonical page: https://app.therundown.ai/guides/how-to-pick-a-dedicated-ai-device - Machine-readable page: https://app.therundown.ai/guides/how-to-pick-a-dedicated-ai-device.md - Difficulty: beginner - Categories: AI - Instructors: Billy Howell - Published: 2026-09-01 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn which hardware devices are good for AI tinkering and which one is right for you. We cover what each option is good for and which one we’d recommend for most people. #### Who This Is Useful For - Creators who want to turn an AI idea into a small physical gadget without buying a full computer. - Operators who want an always-on agent, dashboard, or home automation service on their own network. - Anyone who wants a separate computer for AI coding, local models, or experimental apps. #### What You Will Build By the end, you will have a simple way to choose between an ESP32, Raspberry Pi, and dedicated Mac based on your time, budget, and what you actually want to build. ![ESP32, Raspberry Pi, and Mac device options compared side by side](https://from-ava.b-cdn.net/guide-assets/local-467/how-to-pick-a-dedicated-ai-device/00-three-device-comparison.png) #### What You Need To Get Started - A rough idea of what you want to build - The amount of time you want to spend setting it up - Your full budget, including accessories - The product page for any board or computer you are considering #### Going Further Whatever you buy, copy the product link into ChatGPT and tell it what you want to build. Ask it to help with setup, code, and troubleshooting. You can start with this prompt: ```text I am setting up [DEVICE] for [FIRST PROJECT]. Help me identify the required power, storage, cooling, cables, and software. Then give me one setup step at a time. Use the documentation for this exact model, explain how to verify each step, and stop when you need information from the device or from me. ``` Keep it focused on the exact model in front of you. When something works, save the commands and fixes so you do not have to solve the same problem twice. ### Step 1 Answer Three Questions Before you compare specs, answer three questions. - Time commitment: Do you want an afternoon project, a weekend project, or something longer? An ESP32 gets you there fastest. A Raspberry Pi takes more setup, and a Mac fits a longer-term workflow. - Budget: Do you want to spend less than $50, less than $300, or $500+? Price the whole setup, including storage, power, cooling, a case, and cables. - Use case: Do you want to run local AI, use AI to tinker, or leave OpenClaw or Hermes Agent running on a dedicated device? Those three answers do most of the work. Now we can look at what each device is actually good for. ### Step 2 Pick an ESP32 for One Physical Gadget An ESP32 is basically a simple IoT device that usually has a pretty screen or an OLED touch screen. It can connect over Wi-Fi or Bluetooth and control buttons, microphones, lights, sensors, and other physical parts. It is not a normal computer, so the AI work usually happens through a cloud service or a stronger computer while the ESP32 handles the interface. We tested a Waveshare ESP32-S3 4.3-inch LCD board as the main example. We also looked at a much smaller touch-screen device with physical buttons. These boards make sense for projects such as: ![The Waveshare ESP32-S3 4.3-inch display board](https://from-ava.b-cdn.net/guide-assets/local-467/how-to-pick-a-dedicated-ai-device/01-esp32-display-board-options.png) - A button that starts a voice conversation with AI - A light that turns red while you are recording - A small display for weather, agent status, or AI usage - A sensor or home automation controller Choose an ESP32 when the project has one clear job. The limitations are part of the fun. You have to pick one thing, which makes it much easier to finish. ### Step 3 Pick a Raspberry Pi for an Always-On Service Our favorite option is the Raspberry Pi. It is a real Linux computer with a file system and a desktop, but you can also run it headlessly and connect from another computer. You get a lot of bang for your buck, especially for agents, dashboards, local network services, cameras, and home automation that need to stay on. We run two or three Hermes agents on an 8 GB Raspberry Pi 4. It has stayed up for months, and it also connects to a 3D printer. Because it lives on the home network, it keeps working without tying up a daily laptop. The Raspberry Pi Zero W we showed is much more limited. Treat small Zero models as an option for lightweight services and simple hardware projects. For a general always-on AI server, use a more capable Pi with enough memory for the job. The $350 listing in the screenshot is a retailer price, not official Raspberry Pi pricing. Check the current board price on the Raspberry Pi website (https://www.raspberrypi.com/products/raspberry-pi-5/) before you buy. ![A retailer's Raspberry Pi 5 listing beside a smaller Raspberry Pi board being held up](https://from-ava.b-cdn.net/guide-assets/local-467/how-to-pick-a-dedicated-ai-device/02-raspberry-pi-device-comparison.png) > Pro tip: Budget for the full setup, not only the board. Storage, power, cooling, a case, and cables can change the real cost. You may need: - A power supply - A microSD card or external storage - A case, fan, or heat sink - Cables, adapters, or a display for initial setup The tradeoff is tinkering. If an agent stops, you may need to connect through the terminal, inspect the service remotely, or power-cycle the device. It takes more work than opening a laptop, but that is exactly why you learn so much from a Pi. ### Step 4 Pick a Mac for Full-Computer Work Move up to a Mac when you want serious local AI or a full computer for coding agents, creative apps, and daily work. You do not need to start here. A MacBook gives you a built-in screen and keyboard, so it is easier to open and troubleshoot anywhere. A Mac mini makes more sense when the device will stay at a desk and run as a permanent workstation or server. Remember to include the display, keyboard, and mouse when you compare the total setup. We used a MacBook Neo for Codex projects and Hermes agents. We also accessed Grok Bot from it, but Grok Bot keeps working on a persistent cloud computer rather than using the Mac as its dedicated compute. ![The MacBook Neo product page beside the laptop being held up](https://from-ava.b-cdn.net/guide-assets/local-467/how-to-pick-a-dedicated-ai-device/03-macbook-neo-dedicated-ai-device.png) An 8 GB MacBook Neo can get slow when Codex, another agent, and many browser tabs are open together. Memory matters if you plan to run several agents or local models at the same time. Check the requirements for the specific model and apps you want to use before choosing a configuration. ![Mac mini memory options shown on Apple's configuration page](https://from-ava.b-cdn.net/guide-assets/local-467/how-to-pick-a-dedicated-ai-device/04-mac-mini-memory-options.png) A dedicated Mac also keeps all of this tinkering off your real work computer. You can play with agent permissions, terminal changes, and test accounts without worrying that an agent will touch your private files or break the machine you use every day. You still need sensible account, permission, remote-access, and backup rules. ### Step 5 Match Your Answers Here is the simple version: | Choose | When it fits | | :--- | :--- | | ESP32 | You want an afternoon project, have less than $50 to spend, and want to build one focused gadget. | | Raspberry Pi | You want a weekend project, have less than $300 to spend, or want an always-on OpenClaw or Hermes Agent. | | Mac | You can spend $500+, want serious local AI, or need a computer separate from your work laptop. Pick a MacBook Neo for portability or a Mac mini for your desk. | For most people, we would start with the cheapest Raspberry Pi we can find. Learn the setup, ask ChatGPT for help, and upgrade once you know exactly what you need. Go with an ESP32 if you already have one small gadget in mind. Go with a Mac if you need serious local performance or want to keep agent work off your main computer. ### Guide: How To Replace Siri With a Free Local Model > In this guide, you will download a free AI model to your iPhone and connect it to your phone’s Action Button like Siri. - Canonical page: https://app.therundown.ai/guides/how-to-replace-siri-with-a-free-local-model - Machine-readable page: https://app.therundown.ai/guides/how-to-replace-siri-with-a-free-local-model.md - Difficulty: beginner - Featured tool or company: Locally AI - Categories: AI Foundations & Prompting - Instructors: Billy Howell - Published: 2026-04-30 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will download a free AI model to your iPhone and connect it to your phone’s Action Button like Siri. After setup, the model runs locally, so you can use it without an internet connection and without sending private prompts to a cloud service. The initial setup requires internet access and may take a few minutes while the model files download. #### Who This Is Useful For - iPhone users with an Action Button who want a better assistant shortcut than Siri - Privacy-focused users who want more AI tasks to run on-device - People curious about local AI who want an easy first model to test - Anyone who wants a quick voice assistant for explaining concepts, translating text, or working through small tasks #### What You Will Build You will turn your iPhone’s Action Button into a shortcut for an AI assistant that runs on your phone without needing internet or cellular connection after the model is downloaded. ![IMG_3010](https://tru-images.b-cdn.net/guide-assets/69/69f8bf86e22e540d39ec24ca/img-3010.jpg) - A Locally AI model downloaded onto your iPhone - Locally AI Voice Mode connected to the Action Button - A local voice assistant you can test with practical prompts #### What You Need - An iPhone with an Action Button, such as an iPhone 15 or newer - The Locally AI app - 3+ gigabytes of storage space - Internet access for the initial model download - Wi‑Fi recommended for downloading several-gigabyte models #### Going Further - Once the newest Gemma model is working, download a bigger model and run the same prompt through both. The point is not just to pick the “best” model in theory, but to find the model that is useful enough, fast enough, and private enough for the tasks you actually want a phone assistant to handle. - Compare how fast each model responds - Compare how much storage each model takes - Compare how accurate the answers are - Compare which one sounds better out loud ### Step 1 Download Locally AI Open the App Store and download Locally AI. Once it installs, open the app and continue through the welcome screen. ### Step 2 Choose Your First Model When the app asks you to choose a model, start with the latest Gemma model available in the app. > Pro tip: Do the downloads on Wi‑Fi before you need the assistant offline. Most models are several gigabytes. ![image-2194](https://tru-images.b-cdn.net/guide-assets/69/69f8c1fb269ba686b1828bdb/image-2194.png) Gemma is Google’s open-source model family and is a strong first pick for this setup. It is good enough for normal assistant tasks and gives you a clean baseline before you start testing bigger models. > Pro tip: Model names usually include sizes like 4B, 6B, or 8B. That number is roughly the model’s parameter count. Bigger usually means more capable, but it also means a bigger download and slower performance on your phone. ### Step 3 Download and Test the Model Tap download and keep the app open while it finishes. This part may take a few minutes. That is normal. You are downloading the actual model file onto your phone, so it can be a few gigabytes. Once the model is installed, try one quick prompt before you connect it to the Action Button. ```text Give me a quick plan for the three most important things I should do before a 30-minute client call. ``` ![image-2198](https://tru-images.b-cdn.net/guide-assets/69/69f8c94b9ead7f9c51d26369/image-2198.png) This test gives you a feel for speed and answer quality. If it responds well enough for quick assistant tasks, you are ready to bind it to your phone. ### Step 4 Open Action Button Settings Swipe down on your home screen and search for Action Button. ![image-2199](https://tru-images.b-cdn.net/guide-assets/69/69f8c9d49ead7f9c51d26e6a/image-2199.png) Open the Action Button settings. Swipe through the options until you see Shortcut. ### Step 5 Choose Locally AI Voice Mode Tap whatever is currently set as the shortcut, then search for Locally AI. ![image-2200](https://tru-images.b-cdn.net/guide-assets/69/69f8ca959ead7f9c51d2792d/image-2200.png) Select Voice Mode. Your Action Button should now be set to Locally AI Voice Mode. ![image-2201](https://tru-images.b-cdn.net/guide-assets/69/69f8cab6e87a3b5bdb1edc99/image-2201.png) Press the Action Button and wait for the chime. The first time you open Voice Mode, Locally AI may ask you to download a speech-to-text model. Download it. That speech-to-text model is what lets the app handle voice input locally. It is a separate download from the AI model itself. > Pro tip: If Voice Mode does not launch right away, go back into the Action Button settings and confirm the selected shortcut says Locally AI Voice Mode, not New Chat. ### Step 6 Test Your Local Assistant Press the Action Button and ask a question out loud. Try something practical: ```text Explain the difference between local AI and cloud AI in simple terms. ``` ```text Translate "I will send the document this afternoon" into Spanish. ``` ```text Give me three ways to prepare for a quick client call. ``` In testing, the local model was useful but not perfect. One smaller model missed a couple of basic factual questions. Local models are getting better fast, but they can still be less reliable than huge cloud models like those used in ChatGPT or Claude. For everyday assistant tasks, that tradeoff can still be worth it. Explaining concepts, translation, quick planning, and simple rewrites are all good fits. > Pro tip: After the model and speech-to-text model are downloaded, turn on airplane mode and ask another question. If it still responds, you have proved the important part: the assistant is running locally on your phone. ### Pro tip > Do the model downloads on Wi‑Fi before you need the assistant offline. Most models are several gigabytes. ### Guide: How To Run the Latest Google AI Models on Your Phone for Free > Learn how to download and locally run Google’s Gemma AI model on your phone for free. - Canonical page: https://app.therundown.ai/guides/how-to-run-the-latest-google-ai-models-on-your-phone-for-free - Machine-readable page: https://app.therundown.ai/guides/how-to-run-the-latest-google-ai-models-on-your-phone-for-free.md - Difficulty: beginner - Featured tool or company: Google - Categories: AI Foundations & Prompting, Coding & App Development - Instructors: Billy Howell - Published: 2026-04-13 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to download and locally run Google’s Gemma AI model on your phone for free. You will set up a local AI chat that runs on your phone’s GPU, with no account, subscription, API key, or internet required after setup. The source guide frames this as a three-step setup. #### Who This Is Useful For - Travelers who want a working AI on a plane, in another country, or anywhere without reliable wifi or cellular service - Privacy-focused users who do not want their prompts sent to a cloud server - Anyone curious about local AI who wants a simple first step into running models on their own hardware #### What You Will Build You will build a local AI chat running Google’s Gemma model directly on your phone with Thinking mode enabled. It runs without cloud access, an API key, or ongoing cost. - A downloaded local Gemma model in Google AI Edge Gallery - An offline-capable AI chat on your phone - Thinking mode enabled for chat responses #### What You Need - An iPhone running iOS 17 or later, or an Android phone running Android 12 or later - An internet connection for the initial download only #### Going Further - After setting up chat, you can explore more local AI uses. The app also includes audio transcription, image recognition, games, and Agent Skills such as a password generator, QR code creator, Wikipedia lookup, and interactive map. You can also search “Local AI” in Rundown University for more advanced guides on running your own AI models. ### Step 1 Download the app Open the App Store or Google Play Store and search for Google AI Edge Gallery. Download it. It is free and does not require a sign-in. ![125-app-home-screen](https://tru-images.b-cdn.net/guide-assets/69/69dd65111b1ffe9f1c9971f0/125-app-home-screen.png) ### Step 2 Download a model Open the app and tap AI Chat. You will see a few model options. Choose the recommended option. In the source example, the recommended model was Gemma 4 E2B, about 2.5 GB. Tap to download it and give it a minute. If you want something more capable, you can choose a bigger model. Tap Learn more next to any option to see what it does differently. There are also smaller models around half a gigabyte if you want a lightweight backup that takes up less space. Once the download finishes, you do not need internet again. The model lives on your phone. ![125-model-selection-download](https://tru-images.b-cdn.net/guide-assets/69/69dd65281b1ffe9f1c99722b/125-model-selection-download.png) ### Step 3 Enable Thinking and start chatting Before you send your first message, tap the settings icon in the top right of the toolbar and enable Thinking. This lets you see the model’s step-by-step reasoning before it gives you the final answer, while running entirely on your phone. You will also see other developer settings, such as max token limits and temperature. Since the model is running locally, you can experiment without usage cost. Start a conversation. You can try a travel-planning prompt like the example below. ```text Plan a two-week road trip down Route 66 with landmarks, lodging, and food for each day. ``` ![125-chat-response-thinking-mode](https://tru-images.b-cdn.net/guide-assets/69/69dd65361b1ffe9f1c997257/125-chat-response-thinking-mode.png) One important caveat: this app is built for developers to test Google’s latest models before putting them in their apps. It does not save your chat history. If you close the app, the conversation is gone. You can tap the + button to see messages you have sent, but there is no way to pick up where you left off. Treat it as a scratch pad for one-off questions, not a notebook. ### Pro tip > Put your phone on airplane mode and run the same prompt again. It works. That is the whole point. ### Guide: How to Self-Host n8n (5-Minute Setup) > In this guide, you’ll set up your own n8n automation server on Railway, create a password-protected n8n instance, and import your first AI-powered automation from a JSON file. - Canonical page: https://app.therundown.ai/guides/how-to-self-host-n8n-5-minute-setup - Machine-readable page: https://app.therundown.ai/guides/how-to-self-host-n8n-5-minute-setup.md - Difficulty: intermediate - Featured tool or company: n8n - Categories: Automation & Agents, Coding & App Development - Instructors: Billy Howell - Published: 2026-02-22 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you’ll set up your own n8n automation server on Railway, create a password-protected n8n instance, and import your first AI-powered automation from a JSON file. The setup can be completed in under 10 minutes, and the source notes that a $5 virtual server can run thousands of automations per month. #### Who This Is Useful For - Anyone paying $20 a month for AI automations - Consultants running automations for multiple teams or clients - Intermediate n8n users #### What You Will Build You’ll build a live, password-protected n8n instance running on your own Railway server. In one session, you’ll deploy it, log in, and import your first AI-powered automation from a JSON file. - A Railway-hosted n8n server - A password-protected n8n login - An imported AI-generated workflow - Saved credentials for any connected services #### What You Need - A Railway account - A GitHub account, since Railway uses it to sign in - Claude or ChatGPT to generate your automation as a JSON file - Optional: API keys for any services you want to connect, such as OpenAI, Google Sheets, or Slack #### Going Further - After your n8n instance is live, you can invite other users by email for team or client work, set up separate Railway instances per client for isolation, and create a Google Cloud project if you want to connect Gmail, Google Sheets, Google Calendar, or other Google services. For a deeper dive, the source recommends the AI Automations Course for learning the full workflow design process from scratch. ### Step 1 Deploy n8n on Railway Go to railway.com/deploy/n8n and click Deploy Now. Sign in with your GitHub account if you haven’t already. On the next screen, you’ll see a configuration panel with environment variables. You don’t need to change any of these; the defaults work out of the box. Click Deploy. ![image-2073](https://tru-images.b-cdn.net/guide-assets/69/699b625ccbdb1862bdf56721/image-2073.png) Railway will spin up two containers: a PostgreSQL database to store your workflows and credentials, and the n8n application itself. This takes about 2–3 minutes. You’ll see both containers appear in your Railway dashboard once they’re live. ### Step 2 Open Your New n8n Instance and Create an Account Once the deployment finishes, click the n8n container in your Railway dashboard. You’ll see a generated URL near the top of the page, such as n8n-production-xxxx.up.railway.app. Click it. The first time you visit this URL, n8n will ask you to create an account. Enter your email and choose a password. From this point on, your instance is password-protected. Anyone who visits that URL will hit a login screen. That’s it. You now have a fully functional n8n server running at your own URL for a fraction of the cloud price. ### Step 3 Import an AI-Generated Workflow Click Create Workflow in your n8n dashboard. Instead of building your automation node by node, you can have Claude or ChatGPT design the entire workflow for you as a JSON file. Open Claude or ChatGPT and tell it what you want your automation to do. Be specific about the trigger, the steps, and the output. ```text Build me an n8n workflow that runs every morning at 8am, uses OpenAI to search for the latest AI news, writes a short summary memo, and sends it to my email. ``` The AI will generate a JSON file that defines the entire workflow: nodes, connections, credentials, and configuration. Save that JSON file to your computer. Back in n8n, open your new workflow, click the three-dot menu in the top right, and select Import from File. Choose your JSON file. The entire workflow will appear on your canvas, fully wired up and ready to run. ![image-2074](https://tru-images.b-cdn.net/guide-assets/69/699b6270cbdb1862bdf56f69/image-2074.png) From here, add your credentials: API keys for OpenAI, Gmail, Slack, or whatever services the workflow uses. Click any node that needs a credential, hit Create New Credential, paste in your key, and save. n8n stores these permanently so you only do this once per service. ### Step 4 Test and Activate the Workflow Click Test Workflow to do a manual run and make sure everything fires correctly. You’ll see each node light up green as it executes. If something breaks, click the failed node to see the error message. Usually it’s a missing credential or a formatting issue in the prompt. You can paste the error message into Claude or ChatGPT for troubleshooting. Once it runs cleanly, toggle the workflow to Active in the top right. It will now run on its own based on whatever trigger you set, such as a schedule, webhook, or new row in a spreadsheet. Your server runs 24/7 on Railway. You don’t need to keep a browser tab open; the workflows fire in the background automatically. ### Pro tip > Make sure you bookmark this URL. It’s your new automation dashboard. ### Guide: How to Set Up ChatGPT to Write in Your Voice > In this guide, you’ll build a simple ChatGPT writing setup that turns rough ideas into natural-sounding drafts. - Canonical page: https://app.therundown.ai/guides/how-to-set-up-chatgpt-to-write-in-your-voice - Machine-readable page: https://app.therundown.ai/guides/how-to-set-up-chatgpt-to-write-in-your-voice.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: AI - Instructors: Billy Howell - Published: 2026-08-17 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you’ll build a simple ChatGPT writing setup that turns rough ideas into natural-sounding drafts. This setup will save you hours of copy-editing tweaks. #### Who This Is Useful For - Writers and content teams turning rough ideas or transcripts into publishable drafts. - Consultants and marketers who want ChatGPT to keep their voice in emails, memos, and tutorials. - Anyone tired of removing filler from AI copy. #### What You Will Build You will set up ChatGPT to give you a better first draft. You will start with a quick Humanize Writing pass, test it on a simple writing exercise, and try a Technical English rule when the copy feels too fluffy. ![humanized writing output](https://tru-images.b-cdn.net/guide-assets/local-425/how-to-set-up-chatgpt-to-write-in-your-voice/01-humanized-writing-output.png) The Humanizer made this generic Instagram tutorial more specific and useful. #### What You Need To Get Started - The ChatGPT desktop app with access to Work and Skills #### Going Further Turn your best results into a small voice guide. Save the before-and-after examples that show what sounded right, what felt generic, and what you changed. Put those examples in your project instructions along with a short list of words and phrases you never want ChatGPT to use. Keep updating the voice guide whenever a draft misses the mark. That gives ChatGPT a better target before you run any humanizing pass. ### Step 1 Get a Quick Win With Humanize Writing Open ChatGPT Work, then go to Settings → Plugins → Skills and turn on Humanize Writing. If you do not see it, your workspace may not have the skill yet. ![humanize writing skill enabled](https://tru-images.b-cdn.net/guide-assets/local-425/how-to-set-up-chatgpt-to-write-in-your-voice/02-humanize-writing-skill-enabled.png) Turn on Humanize Writing before you run it on your first draft. Write a first draft. Type the command below and compare the result with the original: ```text /humanize-writing ``` Keep both versions and read them side by side. Look at the opening, sentence length, transitions, and word choice. If the original says something better or gets a fact exactly right, keep it. ### Step 2 Run a Simple Writing Exercise Next, give ChatGPT a clear writing exercise. For example: ```text Write me a memo about the history of the moon. ``` This gives you a clean sample you can run through each method. Keep the assignment the same so you can judge what actually changed. If you want the draft to sound more like you, give ChatGPT more of your own words to work with. You can dictate the request or add a transcript. For example: ```text Use this transcript to draft a [memo/email/tutorial] for [AUDIENCE]. Preserve every source fact and the way I naturally explain the topic. Make the structure easy to scan, but do not replace my voice with generic business language. ``` Now ChatGPT has your vocabulary, rhythm, examples, and priorities to work with. Check the source first and remove anything private or irrelevant before you add it. > Pro tip: We always recommend using dictation, especially for writing. Meeting transcripts or audio notes work well too. ### Step 3 Test the Open-Source Humanizer Skill For level two, install Blader's open-source Humanizer skill (https://github.com/blader/humanizer). Press Command-J to open ChatGPT's terminal, then run: ```text npx skills add blader/humanizer --global ``` The installer may ask where you want to add the skill. Choose the agent you plan to use, then confirm the installation. ![humanizer install confirmation](https://tru-images.b-cdn.net/guide-assets/local-425/how-to-set-up-chatgpt-to-write-in-your-voice/03-humanizer-install-confirmation.png) The terminal confirms that the open-source Humanizer is installed. Create another copy of the untouched draft and run: ```text /humanizer ``` In the recorded test, the Humanizer took a generic Instagram tutorial and gave it a useful niche formula: topic + audience + outcome or angle. Use the same assignment for every test and change one method at a time. You will see the differences right away. ### Step 4 Try Technical English for More Direct Copy Technical English is not only for technical work. Try it whenever AI writing feels too fluffy or indirect: ```text Write in Technical English. Use ASD-STE100 as a style reference. ``` ASD-STE100 Simplified Technical English (https://asd-ste100.org/about.html) was created for clear technical documentation. You are borrowing the style here, not claiming that the result formally follows the standard. ![technical english rule applied](https://tru-images.b-cdn.net/guide-assets/local-425/how-to-set-up-chatgpt-to-write-in-your-voice/04-technical-english-rule-applied.png) Add the Technical English rule to your project instructions when you want it to persist. Compare the result with the original and decide whether you like the more direct style. If you want shorter replies across a whole project, put the same rule in the project instructions or AGENTS.md. You do not need to call another skill every time the agent responds. ### Guide: How To Take AI Notes on Phone Calls > This guide shows you how to set up Granola as an AI notetaker for iPhone calls, including outbound calls and a workaround for inbound calls. - Canonical page: https://app.therundown.ai/guides/how-to-take-ai-notes-on-phone-calls - Machine-readable page: https://app.therundown.ai/guides/how-to-take-ai-notes-on-phone-calls.md - Difficulty: beginner - Featured tool or company: Granola - Categories: Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-04-05 - Updated: 2026-07-23 ### Overview #### The Rundown This guide shows you how to set up Granola as an AI notetaker for iPhone calls, including outbound calls and a workaround for inbound calls. You will create a repeatable setup for transcribing, summarizing, naming, and organizing phone call notes. The setup only takes a few minutes to verify your phone number. > Before recording calls, check the recording consent laws for your state. Some states require both parties to consent to being recorded. #### Who This Is Useful For - Sales reps and account managers who make 10+ calls a day and need to remember which prospect said what before updating the CRM - Founders and operators who take calls on the go, such as vendor negotiations, investor updates, or partnership conversations - Consultants and freelancers who want a record of what was agreed on during client calls #### What You Will Build ![image-2164](https://tru-images.b-cdn.net/guide-assets/69/69d1da36d9a25514b69d05c5/image-2164.png) You will build a repeatable setup where phone calls made from your iPhone get transcribed, summarized, and organized automatically. The final output is a set of named, searchable call notes that sync to your desktop and can optionally sync to your CRM, Notion, or Slack. - Transcribed phone calls - Clean AI-generated summaries - Action items and important details from each call - Named, searchable call notes - Optional sync to your CRM, Notion, Slack, or other connected tools #### What You Need - An iPhone - The Granola app; the free plan works - A few minutes to verify your phone number - Optional: iPhone contacts connected to Granola - Optional: Granola desktop app if you want call notes to appear on your computer - Awareness of your state’s recording consent laws before recording calls #### Going Further - After your call notes are working in Granola, connect integrations in Granola settings to store them somewhere more permanent. You can connect Notion, Zapier, Slack, HubSpot, or your CRM so notes land where you will actually use them. ### Step 1: Set up phone calls in Granola Download Granola from the iOS App Store and open the app. Tap the phone icon in the bottom-left corner to start phone call setup. ![image-2163](https://tru-images.b-cdn.net/guide-assets/69/69d1da1dd9a25514b69cfd02/image-2163.png) Enter your phone number and follow the verification instructions. Granola will call you to confirm, and you will enter a code to complete the pairing. Once verified, you can optionally connect your iPhone contacts. This makes it faster to call people directly through the Granola app instead of typing numbers every time. ### Step 2: Make an outbound call and review the notes Tap the phone button in Granola, pick a contact or type a number, and call. The call will work like a normal phone call. The difference is that Granola listens in the background and transcribes the conversation. When you hang up, wait a minute or two. Granola generates a clean summary with action items and anything worth remembering. You do not need to stay on the screen while it processes. > Pro tip: If you create section headings in your notes before or during a call, the AI will automatically organize your notes into those headings at the end of the call. Try headings like “Follow-up” or “Background Info.” ### Step 3: Handle inbound calls Granola works best with outbound calls. If you open a new note on an inbound call, it will only be able to transcribe your voice. If you have your laptop, you can open a new note there and take your call on speakerphone. ### Step 4: Organize and sync your call notes Name every call note with a templated title like [name] @ [company] so they are easy to scan later. Create separate folders for different call types, such as sales calls, personal calls, vendor conversations, or whatever categories make sense for you. Your call notes sync to Granola’s desktop app automatically, so everything you capture on your phone shows up on your computer too. ### Pro tip > Before you start recording calls, check the recording consent laws for your state. Some states require both parties to consent to being recorded, so confirm whether you are in a one-party or two-party consent state. ### Guide: How To Test Multiple AI Models With the Same Prompt, FAST > In this guide, you will learn how to use OpenRouter Fusion to test the same prompt across multiple AI models at once. - Canonical page: https://app.therundown.ai/guides/how-to-test-multiple-ai-models-with-the-same-prompt-fast - Machine-readable page: https://app.therundown.ai/guides/how-to-test-multiple-ai-models-with-the-same-prompt-fast.md - Difficulty: beginner - Featured tool or company: OpenRouter - Categories: AI Foundations & Prompting - Instructors: Billy Howell - Published: 2026-05-08 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use OpenRouter Fusion to test the same prompt across multiple AI models at once. Instead of opening five apps and guessing which answer was best, you can compare outputs side by side and build a quick cheat sheet for your real work. #### Who This Is Useful For - Consultants and operators who use several AI tools and want a faster way to pick the best model for client work - Builders and coders who want to test models before wiring them into an app, agent, or coding workflow - AI power users who want one place to compare quality, speed, and cost without paying for every subscription separately #### What You Will Build You will build a quick model-comparison workflow in OpenRouter Fusion. ![fusion side by side model comparison](https://tru-images.b-cdn.net/guide-assets/local/175/how-to-test-multiple-ai-models-with-the-same-prompt-fast/01-fusion-side-by-side-model-comparison.png) By the end, you will have: - one place to run the same prompt across multiple models - a repeatable way to compare writing, coding, and research prompts - a fused answer that combines the strongest parts of the model outputs - a simple way to check how much the test cost - the start of a personal model cheat sheet for your work #### What You Need - A free OpenRouter (https://openrouter.ai/) account - Access to OpenRouter Fusion (https://openrouter.ai/labs/fusion) - Either OpenRouter credits or your own provider API keys - Two or three prompts you regularly use in real work > Pro tip: Fusion is useful because it keeps the conditions the same. Same prompt, same context, same moment. That is much better than comparing one ChatGPT answer from Monday to one Claude answer from Thursday. #### Going Further Once you have a few results, go to the Models tab and compare price, speed, and categories. OpenRouter often has free or preview models available, and those can be useful for experimentation. Just remember that popular models can change quickly, especially when a new research preview launches. The best next step is to create your own model-routing cheat sheet: - Run one prompt for business writing - Run one prompt for coding or debugging - Run one prompt for research, planning, or summarization - Record the winner for each category - Re-test when a major new model comes out That turns model choice from a guessing game into a two-minute habit. ### Step 1 Set Up OpenRouter Billing or Bring Your Own Keys Start by creating a free OpenRouter account. OpenRouter acts like a single interface for many AI models, so you can test models from different providers without jumping between separate apps. There are two common ways to use it: - Add credits to OpenRouter and let OpenRouter handle usage across models - Use BYOK, which means bring your own key, and connect provider API keys you already have ![openrouter bring your own key settings](https://tru-images.b-cdn.net/guide-assets/local/175/how-to-test-multiple-ai-models-with-the-same-prompt-fast/02-openrouter-bring-your-own-key-settings.png) The simple version: - Use OpenRouter credits if you want one balance and one place to test models - Use BYOK if you already pay providers directly and want OpenRouter to route through those keys ### Step 2 Create a Custom Fusion Test Open OpenRouter Fusion (https://openrouter.ai/labs/fusion). You will see options like Quality, Budget, and Custom. Choose Custom when you want to control the comparison yourself. ![openrouter fusion custom model setup](https://tru-images.b-cdn.net/guide-assets/local/175/how-to-test-multiple-ai-models-with-the-same-prompt-fast/03-openrouter-fusion-custom-model-setup.png) We used a mix of models like Claude, OpenAI GPT, Grok, Perplexity/Sonar, and free or budget models. The exact model names will change over time, so do not overthink the starting set. A good first test is: - one model you already trust - one model you are curious about - one cheaper or free model - one fuse model that combines the final answer after the other models respond The important part is that all of them receive the exact same prompt. > Pro tip: Do not try to crown one permanent winner. The model that writes the best client email may not be the model that catches the most coding edge cases. ### Step 3 Run a Business Writing Benchmark Start with a prompt that creates a useful business output. We used a business decision prompt because it shows differences in structure, clarity, and judgment quickly. Use this prompt: ```text You are advising a 20-person SaaS company deciding whether to replace its weekly status meeting with an async written update. Write a recommendation memo with: 1. a clear recommendation 2. 3 benefits 3. 3 risks 4. a 2-week experiment plan 5. a short message the CEO can send to the team Keep it concise and practical. ``` After you run the prompt, Fusion sends it to each selected model. ![business memo model comparison](https://tru-images.b-cdn.net/guide-assets/local/175/how-to-test-multiple-ai-models-with-the-same-prompt-fast/04-business-memo-model-comparison.png) Open each response and compare: - which model gives the clearest recommendation - which one writes in a style you would actually send - which one adds useful detail without getting wordy - which one skips constraints or produces generic output Then look at the fused result. ![fused business memo result](https://tru-images.b-cdn.net/guide-assets/local/175/how-to-test-multiple-ai-models-with-the-same-prompt-fast/05-fused-business-memo-result.png) Fusion's final answer is useful because it can combine the best structure, examples, and wording from the model set. It is not always perfect, but it gives you a stronger starting point than any single answer when the models each did one thing well. ### Step 4 Run a Coding or Debugging Benchmark Next, test something more technical. This is where model differences get obvious because weak answers often sound confident while missing edge cases. Use this prompt: ```text Here is a Python function and a failing test. Function: def parse_price(value): return float(value.replace("$", "")) Failing cases: - "$19.99" - " $12 " - None - "" - "free" Explain the bug, rewrite the function safely, and provide 5 pytest test cases that cover normal and edge cases. ``` Look for whether the model notices: - None - empty strings - whitespace - invalid text like free - realistic tests instead of vague suggestions ![fused code debugging result](https://tru-images.b-cdn.net/guide-assets/local/175/how-to-test-multiple-ai-models-with-the-same-prompt-fast/06-fused-code-debugging-result.png) We noticed that different models had very different formatting and depth. Claude had easier-to-parse code formatting, while the fused response from Grok became too wordy. That is exactly the kind of signal you want from this workflow. ### Step 5 Check Cost Before You Keep Testing After a few runs, open Activity in OpenRouter. ![openrouter activity cost breakdown](https://tru-images.b-cdn.net/guide-assets/local/175/how-to-test-multiple-ai-models-with-the-same-prompt-fast/07-openrouter-activity-cost-breakdown.png) This shows spend, request count, token usage, and model-level breakdowns. We used roughly 10 comparisons and spent about 40 cents. That is the real advantage. Instead of paying for several subscriptions and guessing which tool to open, you can run a cheap comparison, see the differences, and build evidence for your own work. Use a simple cheat sheet: | Task type | Winning model | Notes | | --- | --- | --- | | Business memo | [model] | Best structure and tone | | Coding/debugging | [model] | Catches edge cases | | Search/planning | [model] | Better factual coverage | | Fused response | [model] | Best final synthesis | You only need a few tests to learn more than most people know from months of switching between apps. ### Guide: How To Turn One Thumbnail Into Five Different Social Posts > This guide shows you how to create a YouTube thumbnail in Canva, customize it with your own photo and text, then use Canva’s AI resize feature to duplicate it into social media formats. - Canonical page: https://app.therundown.ai/guides/how-to-turn-one-youtube-thumbnail-into-5-different-social-posts - Machine-readable page: https://app.therundown.ai/guides/how-to-turn-one-youtube-thumbnail-into-5-different-social-posts.md - Difficulty: beginner - Featured tool or company: Canva - Categories: Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2026-02-16 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to create a YouTube thumbnail in Canva, customize it with your own photo and text, then use Canva’s AI resize feature to duplicate it into social media formats. You’ll produce a set of ready-to-post assets from one original design, with Canva handling the layout for each size. #### Who This Is Useful For - Content creators who post across multiple platforms and spend too long manually resizing the same graphic - Marketing teams managing brand presence across five or more social channels - Freelancers and agencies delivering client assets in multiple formats #### What You Will Build You’ll build a YouTube thumbnail from a Canva template using your own photo, then resize it into major social platform formats. The finished designs will live together in one Canva folder, ready to download and post. ![image-2064](https://tru-images.b-cdn.net/guide-assets/69/6993a82c8ee4bba267608e19/image-2064.png) - A customized YouTube thumbnail - Resized versions for social platforms such as Twitter/X, Facebook, and Instagram - A single Canva project folder containing the resized assets #### What You Need - Canva Pro account ($15/month) - A photo of yourself to upload into the thumbnail - Optional for the advanced Going Further workflow: Chrome, for Canva’s color dropper tool #### Going Further - After resizing, download each post from Canva using Share > Download. If you want to move beyond Canva thumbnail templates, you can create your own thumbnail, upload it into Canva, recreate the background and text with the color dropper tool, and use AI background removal plus cropping to bring in AI-generated clip art or people before resizing the design like a Canva template. ![image-2067](https://tru-images.b-cdn.net/guide-assets/69/6993ac28aef176cdd4518653/image-2067.png) ### Step 1 Make a YouTube thumbnail from a template From the Canva home screen, click Create, search for “thumbnail,” and select YouTube Thumbnail to open a new project at the correct dimensions. In the left sidebar, browse Templates, choose one that fits the vibe you want, and click it to load it into your canvas. ![image-2065](https://tru-images.b-cdn.net/guide-assets/69/6993a8438ee4bba267609668/image-2065.png) Replace the template’s placeholder copy with your own headline and subtitle. Upload a photo of yourself by going to Uploads in the left sidebar and dragging your photo in. You can use a quick Photo Booth shot, a phone photo, a reaction face, or a simple smile. Drag your uploaded photo over the person in the template so Canva swaps it in. Then click your photo and use Canva’s AI background removal tool to cut yourself out and blend the image into the design. ![image-2066](https://tru-images.b-cdn.net/guide-assets/69/6993ac1baef176cdd4517e1b/image-2066.png) ### Step 2 Resize into every social format In the top bar, click Resize to open Canva’s list of available social media formats. - Twitter Post - Facebook Post - Facebook Story - Instagram Post - Instagram Story Click Copy & Resize. Canva will duplicate your project, resize it, and use AI to lay out the text, photo, and background for each format. Find each resized design in one folder in your projects sidebar. Open each version, then use Share > Download to export the posts. ### Guide: How to Upgrade Your AI Coding Workflow With This Free Context Tool > This guide teaches you how to make an AI coding agent smarter with Marksnip by downloading only the docs pages you actually need and saving them inside your repo. - Canonical page: https://app.therundown.ai/guides/how-to-upgrade-your-ai-coding-workflow-with-this-free-context-tool - Machine-readable page: https://app.therundown.ai/guides/how-to-upgrade-your-ai-coding-workflow-with-this-free-context-tool.md - Difficulty: intermediate - Featured tool or company: Marksnip - Categories: Coding & App Development, Automation & Agents - Instructors: Billy Howell - Published: 2026-03-31 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches you how to make an AI coding agent smarter with Marksnip by downloading only the docs pages you actually need and saving them inside your repo. You will create a reusable context folder your agent can inspect before coding, which helps reduce repeated copy-pasting and keeps important docs from disappearing in a chat thread. The workflow is designed to be quick and lightweight. #### Who This Is Useful For - Anyone coding with Codex, Claude Code, Cursor, or another agent who is tired of re-pasting the same docs into every thread - Operators or founders shipping integrations who want the agent focused on the exact setup docs and API references that solve the problem - Developers building over time in one repo who want a context folder future agents can inspect instead of relying on chat history #### What You Will Build You will build one reusable agent-context/ folder inside your project with the exact Markdown docs your coding agent should read first. In the source example, that context helped Codex add a custom scrolling animation to a website using the React Motion docs, but the workflow works for any library or integration. - A stable folder such as agent-context/ or project-docs/ inside your repo - Markdown copies of the docs pages that matter for the current task - A repeatable prompt pattern that tells the agent to read those files first ![image-2157](https://tru-images.b-cdn.net/guide-assets/69/69cc3e77b990d2c36f4bcebd/image-2157.png) #### What You Need - A Chromium browser with Marksnip installed - A project repo with a folder like agent-context/ or project-docs/ - One docs page you actually need, such as setup docs, SDK docs, API reference, or a changelog #### Going Further - Reuse the same clipping workflow for articles and Twitter threads when an AI cannot browse well or wastes time navigating junk and ads. - Clip additional docs into your context folder as the project grows, especially setup pages, API references, and changelogs that affect the implementation. - Pair this manual context system with a docs-aware tool like Context7 when you want automatic docs access alongside a folder of exact pages you chose on purpose. ### Step 1: Install Marksnip and pick the right page Install Marksnip from the Chrome Web Store, then open the docs page you actually need for the task in front of you. - Setup guides - SDK docs - API reference pages - Changelogs - Technical blog posts with implementation details The mindset shift is to avoid downloading an entire library just because you can. Choose the pages you want to flag for the agent’s attention so both of you are working from the same source of truth. > If you only need part of a page, Marksnip lets you clip either a selection or the full document. This is useful when a docs page is huge but only one section matters for the task. ### Step 2: Clip the page into Markdown Click the Marksnip icon in the browser toolbar and let it process the page. Marksnip runs the current page through Mozilla Readability and converts the extracted content to Markdown, which makes the output cleaner than a normal copy-paste. ![image-2158](https://tru-images.b-cdn.net/guide-assets/69/69cc3ef0b990d2c36f4bd859/image-2158.png) ### Step 3: Save it in a stable context folder Drag the Markdown file into your working repo and save it in a folder like agent-context/ or project-docs/. That folder is the point: when you or a future coding agent opens the repo, there is a visible packet of trusted context sitting next to the code. - The file stays in the project - The source doc is obvious - You can update the file when the docs change - Future agents can inspect the same context without you re-explaining everything If the task is more involved, add two to five files that cover the setup, the main API surface, and any recent changelog that affects your implementation. A good default habit is to make the folder first, then keep feeding it the pages that actually matter. ### Step 4: Tell the agent to read that folder first Once the file is in the repo, point your coding agent at the folder and tell it to use those files as the source of truth. ```text Read the files in agent-context first and use them as the source of truth for this task. ``` For example, if the agent-context folder has docs on how to install and use the React Motion library, you can tell Codex you want to add a scrolling animation into the site and have it work from those files first. In the source test, the animation was added correctly using the docs that had just been clipped into the project. The same pattern works for integrations like payments APIs, auth providers, analytics tools, or front-end libraries. Manual selection matters because sometimes you want to tell the agent: these are the exact pages that solve the problem, so start here. ### Step 5: Customize the workflow to match your setup Once the basic flow works, Marksnip has settings that can make the system more reusable. You do not need all of them for the core workflow, but they are useful to know about. - Templates let you prepend or append text to every file, which is useful for YAML front matter or a standard source block. - Title Template controls the filename, so you can make your saved docs more consistent. - Downloads Subfolder helps route files into a cleaner structure when you download often. - Markdown Options let you change heading style, code block style, link style, image style, and other output details. - Images can download images alongside the Markdown when you are using Downloads API mode. - Send to Obsidian works if you use the Advanced Obsidian URI plugin and want clipped pages to become notes instead of repo files. - Batch Processing lets you convert multiple URLs at once if you are building a bigger context pack. One especially useful toggle is Template. If you turn it on, Marksnip can add front matter for Obsidian. There are also options for saved location, default title, and other file-level customization if you want your clipping workflow to be more organized. ### Pro tip > If you only need part of a page, Marksnip lets you clip either a selection or the full document. That is useful when a docs page is huge but only one section matters for the task. ### Guide: How To Use Claude To Find What's Actually Eating Space on Your Computer > Learn how to use Claude as a storage-cleanup copilot for your Mac so you can identify what is actually taking up space before deleting random folders. - Canonical page: https://app.therundown.ai/guides/how-to-use-claude-to-find-whats-actually-eating-space-on-your-computer - Machine-readable page: https://app.therundown.ai/guides/how-to-use-claude-to-find-whats-actually-eating-space-on-your-computer.md - Difficulty: intermediate - Featured tool or company: Claude - Categories: Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-03-23 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use Claude as a storage-cleanup copilot for your Mac so you can identify what is actually taking up space before deleting random folders. You will investigate large directories, review common developer clutter, and turn the cleanup process into a reusable monthly checklist. #### Who This Is Useful For - Operators, founders, and freelancers who use AI tools on their main computer and keep running into storage warnings - Developers and vibe coders with old repos, large node_modules folders, Homebrew clutter, or app caches they do not fully understand - Mac users who want to free up space without blindly deleting files and breaking a working setup #### What You Will Build You will build a Claude-assisted storage audit workflow that ranks the biggest storage offenders on your machine, separates low-risk cleanup targets from items that need review, and becomes a reusable monthly cleanup checklist. - A ranked view of large folders and likely storage offenders - A review of common cleanup categories like node_modules, Homebrew cache, Docker clutter, and Claude desktop cache - A repeatable monthly checklist or Claude skill for future storage audits ![image-2140](https://tru-images.b-cdn.net/guide-assets/69/69c1bbee74eb348bf3631490/image-2140.png) #### What You Need - Access to Claude or another LLM that can help with coding - A Mac you want to audit for storage usage - Terminal access so you can run Claude’s suggested investigation commands and paste the results back in - Optional: development folders, node_modules directories, Homebrew, Docker, or Claude desktop cache to review #### Going Further - After the first audit, reuse the workflow monthly and expand it to other non-obvious storage eaters that often build up when you code with AI. - Python environments like .venv folders and old conda environments - Xcode simulators and derived data if you do Apple-device development - Downloads filled with installers, ZIP archives, models, and exports - Screen recordings and generated media from Loom, CleanShot, or AI video tools - Package-manager caches from npm, pnpm, yarn, or pip ### Step 1 Investigate Before You Delete Anything Start a new Claude session and make the safety rule explicit: you want help investigating first, not deleting first. The point is to have Claude explain what is large, what it likely contains, and what is usually safe to remove before any destructive command gets suggested. ```text Help me investigate what's taking up storage on this Mac. Start by identifying the biggest directories in my home folder. For each large folder, explain what it likely contains, whether it's usually safe to delete, and what questions I should answer before removing anything. Do not suggest destructive commands until we've reviewed the findings. ``` Once Claude gives you commands, run them in Terminal and paste the results back in. This gives Claude real local information to work from instead of vague guesses. > Pro tip: The built-in Mac storage screen is useful for broad categories, but it often misses the real story. The problem is usually not one huge file. It is a pile of medium-size directories, old dependencies, and app caches that add up fast. ### Step 2 Find Old node_modules and Other Reproducible Bloat If you do any kind of web development, node_modules is usually one of the first places to check. Old experiments, abandoned repos, and half-finished client work can each carry hundreds of megabytes or several gigabytes of dependencies. ```text Scan my dev folders for large node_modules directories. Show me the largest ones, where they live, and when they were last modified. Separate them into: 1. likely safe to remove because the project looks old or inactive 2. review manually because the project may still be active Explain your reasoning for each one. ``` This is one of the easiest cleanup wins because you are usually deleting installed dependencies, not your source code. If a project is old and you need it later, you can reinstall everything with npm install. > Pro tip: If you do not know which projects are still active, ask Claude to interview you. It can walk through the list one folder at a time and help you decide what is stale. ### Step 3 Clean Up Homebrew and Other Tooling Cache Carefully After node_modules, move on to package-manager and tooling clutter. Homebrew is a common offender because it accumulates old downloads, outdated package versions, and dependencies you no longer need. Docker can do the same thing with images, containers, and build cache. ```text Explain how to safely clean up Homebrew on this Mac. Tell me what old versions, caches, or unused dependencies are taking space, what the normal cleanup commands would do, and which ones are low-risk versus which ones need manual review first. ``` ```text Help me find Docker files, images, and caches that are taking significant space on this Mac. Show me what is low-risk to remove, what needs review first, and how much space each cleanup might save. ``` The habit that matters is asking Claude to preview the impact first. Do not run cleanup commands just because they look standard. ### Step 4 Clear Claude Cache Without Deleting the Wrong Thing App cache is another easy place to reclaim space, and the Claude cache can get surprisingly large. During setup for this guide, the cache had grown to around 15 GB, much of it tied to Claude Cowork usage. Open the Claude desktop app and click Help > Troubleshooting > Clear Cache and Restart. You can also click Delete Cowork VM Bundle and Restart. That deletes virtual coding sandboxes that Claude Cowork has used. > Do not click Delete Cowork VM Sessions and Restart, as that will delete all your Claude Cowork threads. ![image-2141](https://tru-images.b-cdn.net/guide-assets/69/69c1bc0474eb348bf3631d6c/image-2141.png) > Pro tip: This same logic applies to other local AI apps too. If you use ChatGPT desktop, Codex, or similar tools heavily, their local caches may be much larger than you think. ### Step 5 Turn This Into a Reusable Monthly Cleanup Checklist The bigger win here is not just getting space back once. It is creating a repeatable cleanup workflow so you do not end up rage-deleting files the next time your disk fills up. When you finish the audit, ask Claude to turn your process into a reusable Claude skill. You can also ask Claude to create a checklist in a markdown skill file so you can manually track the same process every month. ### Pro tip > Ask Claude to preview the impact before running cleanup commands. Do not run cleanup commands just because they look standard. ### Guide: How To Use Google Stitch To Redesign Your Entire Website (For Free) > This guide teaches you how to use Google Stitch to redesign an existing web page by uploading screenshots of the page you already have. - Canonical page: https://app.therundown.ai/guides/how-to-use-google-stitch-to-redesign-your-entire-website-for-free - Machine-readable page: https://app.therundown.ai/guides/how-to-use-google-stitch-to-redesign-your-entire-website-for-free.md - Difficulty: beginner - Featured tool or company: Google - Categories: Content & Creative - Instructors: Billy Howell - Published: 2026-03-22 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches you how to use Google Stitch to redesign an existing web page by uploading screenshots of the page you already have. You will generate improved layout options, create variations, and produce a design direction you can hand off to a developer or coding AI. The source describes this as a very fast way to improve weak layouts without starting from a blank canvas. #### Who This Is Useful For - Marketers and founders who know a page on their site feels weak but do not want to redesign it from scratch in Figma - Anyone working with a developer or coding agent who wants a better visual starting point before handing off the build - People managing older websites who want to improve one page at a time instead of committing to a full redesign #### What You Will Build You will create a redesigned version of an existing web page using screenshots of the page you already have. From there, you will choose a stronger variation and export it as a prototype, a Figma file, or code for a developer or coding AI. - A redesigned page direction based on your existing page - Multiple layout variations from the same starting point - An exportable prototype, Figma file, or code output ![image-2137](https://tru-images.b-cdn.net/guide-assets/69/69c046edbf269886385956d4/image-2137.png) #### What You Need - A Google account - A web page you want to improve - A screenshot of that page, or the page URL if you want to test both approaches #### Going Further - After generating a redesign, you can export to AI Studio to turn it into a live prototype using Google’s vibe coding workflow. You can also use Stitch’s design system controls to adjust fonts and visual direction before generating more versions, then hand the design direction and components to your developer or coding agent. ### Step 1: Pick one page and define what needs to change Start with a single page on your site that you want to improve. Do not try to fix your whole site at once; this works better one page at a time, or even one component at a time. Take a screenshot of the page and identify what needs to improve. Look for issues like too much dead space, unclear value, or a layout that makes the page feel empty. The more clearly you can describe the problem, the better Stitch can respond. > Pro tip: You can also paste a URL into Stitch and ask it to redesign the page that way. Try both the screenshot method and the URL method to see which gives you the better starting point. ### Step 2: Upload the screenshot and prompt the redesign Go to Google Stitch and drop the screenshot into the chat box. Then give Stitch a prompt focused on the design problem you want solved. ![image-2138](https://tru-images.b-cdn.net/guide-assets/69/69c04710bf26988638595f97/image-2138.png) ```text Improve the layout of this page so the user sees more content and perceives more value. Make the content more accessible and reduce the dead space. ``` This works because it tells Stitch the outcome you want, not just the tool you are using. You are giving it design direction. > Pro tip: Include the reason behind the redesign. Yours might be to increase conversion, make the page feel more valuable, improve readability, or guide the user to the next action more clearly. ### Step 3: Generate variations until one clicks Once Stitch gives you the first redesign, do not assume version one is the final answer. Click into the generated page and go to Generate > Variations > Generate Variations. ![image-2139](https://tru-images.b-cdn.net/guide-assets/69/69c04726d86a6542d6a4c9d7/image-2139.png) Stitch should generate multiple new versions of the page. Instead of starting over, you can explore different directions from the same visual starting point and quickly see which layout feels strongest. In the source test, this was the fastest way to move from “better” to “actually usable.” ### Step 4: Export the version you want to build Once you get a variation you like, click More > Export. From there, you can move the design into Figma or download the code and hand it off to your developer or coding AI. Do not treat the exported page like a perfect drop-in replacement every time. Stitch is useful for generating the layout, components, and design direction, but your engineer may still want to rebuild the best parts using your existing styles and structure. > Pro tip: This is especially useful if you work with coding agents. Let Stitch do the design work first, then tell your coding AI to rebuild the components and layout using your current CSS or component system. ### Step 5: Reuse the same canvas for more pages Repeat the process with more pages inside the same canvas so Stitch keeps working from the same design system as you go. Instead of redesigning every page from scratch, you can slowly build a more consistent visual system across your site. ### Pro tip > Try both the screenshot method and the URL method in Stitch to see which gives you the stronger starting point. ### Guide: How To Use Grok for Free Automated Research > This guide teaches you how to use Grok’s Tasks feature to create free automated research briefings from live X data. - Canonical page: https://app.therundown.ai/guides/how-to-use-grok-for-free-automated-research - Machine-readable page: https://app.therundown.ai/guides/how-to-use-grok-for-free-automated-research.md - Difficulty: beginner - Featured tool or company: xAI - Categories: Data, Research & Analysis, Automation & Agents, Marketing & Growth - Instructors: Billy Howell - Published: 2026-03-16 - Updated: 2026-07-23 ### Overview #### The Rundown This guide teaches you how to use Grok’s Tasks feature to create free automated research briefings from live X data. With a free X.com account, you can set up two automated daily tasks that research topics you care about and deliver the results to your email and phone. #### Who This Is Useful For - Anyone with an X account who wants free automated research without paying for another AI subscription. - Marketers and founders who want daily or weekly competitor and industry monitoring. - Sports fans, news junkies, and hobbyists who want timely updates on topics X is good at tracking. #### What You Will Build You will build one or two scheduled Grok tasks that automatically research topics you care about and deliver the results to your email and phone. - A daily research briefing task. - An optional second daily task for another topic. - Email delivery and optional phone push notifications for the results. ![image-2122](https://tru-images.b-cdn.net/guide-assets/69/69b87d8ed57955fcee1c396f/image-2122.png) #### What You Need - A free X.com account. - Optional: the Grok mobile app if you want push notifications on your phone. #### Going Further - Use your weekly and monthly limits. You get two daily tasks for free, but you can also schedule up to 10 weekly or monthly tasks. For example, you could set up a weekly competitor roundup on Mondays, a monthly industry trends summary, or a weekly research brief on a topic you are learning about. - Stack keywords across tasks. Use your two daily tasks for topics that change fast, like breaking news, sports, or market moves, and use weekly tasks for slower-burn research like industry trends, competitor deep dives, or product launches. - If you need more tasks, the source notes that unlimited tasks require SuperGrok or X Premium+. Most people can still get a lot done with the free limits. ### Step 1 Open Grok Tasks Go to grok.com and sign in with your X account. On desktop, you have two ways to find Tasks: - Go directly to grok.com/tasks. - Click your profile picture in the bottom-left corner, then click Tasks in the popup menu. ![image-2123](https://tru-images.b-cdn.net/guide-assets/69/69b87d9cd57955fcee1c41f0/image-2123.png) ### Step 2 Create Your First Scheduled Research Task Click Create Task. You will see schedule options: daily, weekly, monthly, or yearly. Pick daily for your first task, then write a prompt based on what you want Grok to research. Here are a few examples you can copy and adapt: ```text Search X for the top trends in [your niche] from the last 24 hours. Summarize the top 3 and flag anything gaining traction. ``` ```text Search X for any news, product updates, or mentions about [competitor name] from the past 24 hours. Summarize the key findings. ``` ```text Give me a daily update on local news, events, and weather forecasts in [your city]. ``` ```text Search X for injury updates and any roster changes for [your team or league]. Flag anything that happened in the last 24 hours. ``` Focus on topics where X has an edge: timely, breaking, real-time information that traditional search engines may be slower to pick up. ### Step 3 Test and Refine the Task Before you let the task run on schedule, click into it and hit Test. This runs it once so you can see what the output looks like. - Test results will not show up in your sidebar automatically. Go back to the tasks screen, click the task, then click Chat to see the output. - If the results are too broad or too narrow, tweak your prompt and test again. ![image-2124](https://tru-images.b-cdn.net/guide-assets/69/69b87dafd57955fcee1c4a7c/image-2124.png) ### Step 4 Choose How You Receive Your Briefings Once your task is running, Grok will deliver results in two ways: - Email to the address tied to your X account. - Push notifications through the Grok mobile app, if you have it installed. You can toggle email and app notifications on or off depending on how you want to receive your briefings. If you do not want another app on your phone, email works fine on its own. To view results on desktop anytime, navigate back to grok.com/tasks and click into the task. ### Pro tip > You can include multiple keywords in one task. If you are tracking your industry, ask Grok to search for news on several terms in a single prompt so one well-written task can cover more ground. ### Guide: Launch an AI-Powered Outbound Calling Agent in 15 Minutes > In this guide, you will build an AI agent that makes real outbound sales calls from a contact list. - Canonical page: https://app.therundown.ai/guides/launch-an-ai-powered-outbound-calling-agent-in-15-minutes - Machine-readable page: https://app.therundown.ai/guides/launch-an-ai-powered-outbound-calling-agent-in-15-minutes.md - Difficulty: beginner - Featured tool or company: ElevenLabs - Categories: Sales, Consulting & Customer Success, Automation & Agents - Instructors: Billy Howell - Published: 2026-02-15 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will build an AI agent that makes real outbound sales calls from a contact list. You will create the agent, get a phone number, upload contacts from a CSV, and launch a calling batch. The setup takes about 15 minutes and costs $0 to test. #### Who This Is Useful For - Sales teams running outbound campaigns who want to qualify leads before a human gets on the phone - Small business owners who do not have time to make 50 cold calls a week but need to keep pipeline moving - Consultants and agencies who want to automate appointment booking, customer check-ins, or event RSVPs without hiring a VA #### What You Will Build You will build a fully functioning AI phone agent that calls a list of contacts, has a natural conversation based on your prompt, handles objections, and logs full transcripts from a simple CSV upload. - An ElevenLabs AI calling agent - A connected Twilio phone number - A CSV-based outbound calling batch - Call transcripts you can review after the calls #### What You Need - ElevenLabs account, free plan or $9 - Twilio account, free to sign up with $15 in credits included - A list of phone numbers to call - An idea of what you want the agent to say #### Going Further - After the basic outbound flow is working, you can add live call transfers, connect scheduling and CRM tools, add qualifying questions, or reuse the same setup for reminders, feedback calls, RSVPs, renewal check-ins, and other repetitive calling workflows. ### Step 1: Create your AI agent in ElevenLabs Sign in to ElevenLabs and go to the Agents section. Create a new agent. Choose a voice from the voice library that fits the tone you want for your outbound calls. > Pro tip: You can paste your business’s URL into the agent’s knowledge base so it can pull context from your site. This helps the agent answer questions about your company instead of making things up. Write the system prompt. Tell the agent who it is, what company it represents, and the goal of the call. ```text You are a friendly sales rep for [your company]. You're calling to introduce [your product/service] and see if the prospect would like to schedule a 15-minute meeting this week. Be conversational, not pushy. If they're not interested, thank them and end the call politely. If they want to schedule, confirm their preferred day and time. ``` Set the conversation starter. This is the first thing the agent says when someone picks up, so make it natural. ```text Hey, this is [Agent Name] calling from [Company]. I wanted to quickly share something we've been working on — do you have two minutes? ``` Hit Publish when you are done. ![image-2058](https://tru-images.b-cdn.net/guide-assets/69/69923dbae89d1ad4712bca59/image-2058.png) > Pro tip: Keep the system prompt short. A few sentences covering who, what, and the goal is all the agent needs. It handles the back-and-forth naturally, so you do not need to script every possible response. ### Step 2: Set up Twilio and get a phone number Go to twilio.com and create a new account. Twilio gives you $15 in free credits to start, which is more than enough for testing. Buy a phone number. It costs about $1.20 to purchase and $1.20 per month to keep. Browse the available numbers and pick one. ![image-2059](https://tru-images.b-cdn.net/guide-assets/69/69923dc5e89d1ad4712bd2a1/image-2059.png) Once you have your number, go to your Twilio dashboard and grab three things: - Your phone number, the one you just bought - Account SID, which is on the main dashboard - Auth Token, which is right below the SID; click to reveal it ![image-2060](https://tru-images.b-cdn.net/guide-assets/69/69923dcfe89d1ad4712bdae4/image-2060.png) That is everything you need from Twilio. > Pro tip: Twilio is pay-as-you-go, at fractions of a cent per minute of call time. The $15 in free credits covers a lot of test calls. You will not need to add a payment method until you are ready to scale. During the free trial, when someone answers a call from your Twilio number, they will hear a short message from Twilio saying “press any key to connect.” That message goes away once you upgrade, but it does not affect testing. ### Step 3: Connect Twilio to ElevenLabs Go back to ElevenLabs. In the left sidebar, find Phone Numbers and click Create New. Paste in three things: - Your Twilio phone number - Your Account SID - Your Auth Token ![image-2061](https://tru-images.b-cdn.net/guide-assets/69/69923deae89d1ad4712be343/image-2061.png) That is the entire connection. Your ElevenLabs agent now has a real phone number it can make calls from. ### Step 4: Upload contacts and start calling In ElevenLabs, look in the bottom-left corner for Outbound. Click it and create a new batch. Select Telephony as the channel. Download the CSV template that ElevenLabs provides. It has several columns; you can remove everything except the phone number column. Add your list of numbers and save the file. Upload the CSV back into the outbound batch. From here, you have three options: - Send immediately: starts calling your list right now - Test: calls your own number so you can hear how it sounds first - Schedule: sets the batch to call at a specific time, like 9am tomorrow morning Select your phone number and agent, then launch. ![image-2062](https://tru-images.b-cdn.net/guide-assets/69/69923dffe89d1ad4712bebb3/image-2062.png) > Pro tip: If you want to override the system prompt or conversation starter for a specific batch without changing your main agent settings, enable that in your agent’s Security tab by toggling on the override options. This lets you customize messaging per batch while keeping your default agent intact. ### Step 5 Extend the workflow Once the basic outbound flow is working, you can build on top of it. ![image-2063](https://tru-images.b-cdn.net/guide-assets/69/69923e0be89d1ad4712bf420/image-2063.png) Enable call transfers. In your agent settings, go to Tools and toggle on the Transfer to Number system tool. Enter your phone number and add a condition to the instructions, such as “if the prospect says yes to a demo, transfer the call.” The agent will call you and patch the lead through live. Connect scheduling and CRM tools. ElevenLabs integrates with Cal.com, HubSpot, Salesforce, and more. You can also build custom integrations using webhooks and n8n to route call outcomes into whatever system you use. Qualify leads automatically. Add qualifying questions to your prompt, such as budget, timeline, and decision-maker status, so the transcript becomes a lead scorecard you can sort through without listening to every call. Use the same setup beyond sales. It also works for appointment reminders, customer feedback calls, event RSVP collection, renewal check-ins, or any scenario where you would normally make repetitive phone calls. Swap the prompt, upload a new list, and let the agent handle it. ### Pro tip > Keep the system prompt short: a few sentences covering who the agent is, what it represents, and the goal of the call is enough. ### Guide: Learn Any Subject With Custom NotebookLM Podcasts > Learn how to use Google’s NotebookLM to generate a custom podcast on any topic you want to study. - Canonical page: https://app.therundown.ai/guides/learn-any-subject-with-custom-notebooklm-podcasts - Machine-readable page: https://app.therundown.ai/guides/learn-any-subject-with-custom-notebooklm-podcasts.md - Difficulty: beginner - Featured tool or company: Google - Categories: Productivity & Knowledge Work, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-01-22 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use Google’s NotebookLM to generate a custom podcast on any topic you want to study. You’ll turn your own sources into an Audio Overview, which can take 5–10 minutes to generate once your notebook is set up. #### Who This Is Useful For - Entrepreneurs who want to stay on top of the latest news in their sector - Newsletter readers who are a couple of issues behind - Anyone who wants a process for learning about new topics - Audio-first learners who want a better way to process incoming information #### What You Will Build ![00_what_you_will_build-2](https://tru-images.b-cdn.net/guide-assets/69/69725224f0549b4aeb49089a/00-what-you-will-build-2.png) You will build a NotebookLM notebook with learning tools tailored to your sources and prompting. - A topic-specific NotebookLM project - A set of selected sources, such as links, files, Google Drive documents, or YouTube videos - A generated podcast-style Audio Overview #### What You Need - A Gmail or Google Workspace account - Sources for the topic you want to learn about, such as websites, files, Google Drive documents, or YouTube videos - Optional: Google AI Plus, which provides more sources per notebook and more daily generations #### Going Further - Use the same workflow to catch up on Rundown AI newsletters by pasting newsletter links as sources and generating a podcast. - Select a single newsletter and ask NotebookLM to do a deep dive on the companies and news mentioned in it, then generate an Audio Overview for that content. ### Step 1: Choose a topic and gather sources Pick a topic you want to learn about. You can use any website, topic, or file as a source. It’s a good idea to start with at least 2–3 sources, and you can also attach documents from Google Drive. ![01_doc_upload](https://tru-images.b-cdn.net/guide-assets/69/697252533847849bc74642e0/01-doc-upload.png) > 💡 Pro tip: You can link to any YouTube video and NotebookLM will use it as a source. Good use cases to try: - Earnings call transcripts to learn about stocks - Summarizing YouTube tutorials from your watchlist - Learning the capabilities of coding documentation - Summarizing last month’s client deliverables from your Google Drive ### Step 2: Create a NotebookLM notebook and add sources Go to notebooklm.google.com and create a new notebook. Add your links and documents as sources. You can move sources in and out of the AI’s context window by checking or unchecking them in the left toolbar. ![02_source_toggle](https://tru-images.b-cdn.net/guide-assets/69/6972526a3847849bc74642ee/02-source-toggle.png) ### Step 3: Generate your audio overview Click Audio Overview on the right under Studio. The audio can take 5–10 minutes to generate. You can also generate other learning assets from the same notebook: - Flash cards - Slide decks - Videos, including podcast plus PowerPoint slides - Quizzes - Mind maps - Reports, including briefing docs - Infographics - Data tables ### Step 4: Listen, download, or make it interactive Once your podcast is generated, click the three dots to download it and AirDrop the MP3 to your phone, or press play to listen in your browser. You can also use the NotebookLM mobile app. > 💡 Pro tip: Hit the “Interactive” button to call in to the podcast and ask the hosts specific questions. ### Pro tip > You can link to any YouTube video and NotebookLM will use it as a source. ### Guide: Make AI Note-Taking Easier With This ChatGPT + Granola “Captain’s Log” Setup > In this guide, you will learn how to stop losing ideas and follow-ups by building an AI “Captain’s Log. - Canonical page: https://app.therundown.ai/guides/make-ai-note-taking-easier-with-this-chatgpt-granola-captains-log-setup - Machine-readable page: https://app.therundown.ai/guides/make-ai-note-taking-easier-with-this-chatgpt-granola-captains-log-setup.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-07-20 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to stop losing ideas and follow-ups by building an AI “Captain’s Log.” Capture quick Granola notes, then use ChatGPT to turn them into a daily summary and task list automatically. #### Who This Is Useful For - Busy operators and managers who need to capture follow-ups before they disappear between meetings. - Founders and creators who want a lightweight record of daily progress, ideas, and open tasks. - Anyone with scattered voice notes who wants to turn them into a useful daily summary and running task list. #### What You Will Build You will build a voice-to-action system that turns quick Granola recordings into a daily summary and a running task ledger. Capture the thought while it is fresh, then let ChatGPT organize the notes and keep open work visible. ![Captain's Log active task list](https://tru-images.b-cdn.net/guide-assets/local-388/make-ai-note-taking-easier-with-this-chatgpt-granola-captains-log-setup/01-captains-log-active-task-list.png) #### What You Need To Get Started - Granola (https://www.granola.ai/) installed on your phone and computer. - A ChatGPT (https://chatgpt.com/) account. Optional: - An iPhone for one-button capture with the Action Button. Otherwise, open the Granola app directly. - A paid ChatGPT plan for Scheduled Tasks. #### Going Further Use the same task ledger to start the next day with a focused plan. Tell ChatGPT: ```text Read my current Captain’s Log task list. Give me the three highest-priority open tasks for today, show stale items that need a decision, and flag only deadlines that came directly from a log. ``` You can reuse the pattern for recurring meeting follow-ups, phone-call notes, walking brainstorms, and end-of-day reviews. Capture first, let Granola preserve the context, and tell ChatGPT to turn that context into the next action. ### Step 1 Install Granola and Set Up One-Button Capture Install Granola on your phone and computer, then sign in to the same Granola account and workspace on both devices. Notes recorded on your phone will appear in the desktop app, where you can review the enhanced note and raw transcript. On a supported iPhone, open Settings → Action Button → Shortcut, search for Granola, and choose New Note. Press and hold the Action Button to test it. Apple’s Action Button guide (https://support.apple.com/guide/iphone/use-and-customize-the-action-button-iphe89d61d66/ios) confirms that the Shortcut option can open an app or run a shortcut. If your phone does not have an Action Button, open Granola and start a new note manually. You can also use Apple Watch as another quick capture point when the Granola shortcut is available there. The goal is to make starting a note easy enough that you use it before an idea disappears. ### Step 2 Connect Granola to ChatGPT Open ChatGPT and go to Settings → Plugins. Search for Granola, select it, and complete the connection. Granola’s official setup guide (https://docs.granola.ai/help-center/sharing/integrations/mcp) confirms that ChatGPT can search notes, read meeting details, and—on paid Granola plans—read raw transcripts. ![ChatGPT Granola plugin](https://tru-images.b-cdn.net/guide-assets/local-388/make-ai-note-taking-easier-with-this-chatgpt-granola-captains-log-setup/02-chatgpt-granola-plugin.png) Create a ChatGPT project called Captain’s Log, then tell ChatGPT how to organize the notes you will send it: ```text Set up this project as my Captain’s Log. When I tell you to process my logs, use Granola to find notes whose opening transcript begins with “Captain’s Log.” Organize them by day and create: - a short daily summary - new and completed tasks - open loops and follow-ups Maintain a running task list across days. Only use details stated in the notes. Do not invent owners, deadlines, or completion states. ``` > Pro tip: Run this workflow in ChatGPT, not Codex. If ChatGPT starts reaching for browser control instead of Granola, type @Granola and select the connected plugin directly. ### Step 3 Record Consistent Captain’s Log Entries Start every recording with the exact words “Captain’s Log.” That opening phrase gives ChatGPT a reliable way to separate these entries from ordinary Granola meetings. Keep each log short and natural. Record one at the beginning or end of the day, after a recurring meeting, after finishing a large deliverable, or whenever an idea hits. Use this spoken structure when you want a little guidance: ```text Captain’s Log. Today I [WHAT HAPPENED]. I still need to [OPEN TASKS]. I need to follow up with [PEOPLE], and [CONTEXT OR DEADLINE]. ``` When you stop recording, Granola automatically creates an enhanced note while keeping the raw transcript available. You do not need to format the thought perfectly before speaking. ![Granola enhanced note and transcript](https://tru-images.b-cdn.net/guide-assets/local-388/make-ai-note-taking-easier-with-this-chatgpt-granola-captains-log-setup/03-granola-enhanced-note-and-transcript.png) For more structure on desktop, open Manage Templates in Granola and create a Captain’s Log template with sections such as Daily Summary, New To-Dos, and Closed To-Dos. ![Granola Captain's Log template](https://tru-images.b-cdn.net/guide-assets/local-388/make-ai-note-taking-easier-with-this-chatgpt-granola-captains-log-setup/04-granola-captains-log-template.png) If that template does not appear as a phone or watch Action Button option, keep the shortcut tied to New Note and say “Captain’s Log” first. The spoken label lets the workflow work without depending on a specific template. ### Step 4 Create the Daily Summary and Task Automation Test the workflow manually before scheduling it. Tell ChatGPT: ```text Use @Granola to read my [NUMBER_OF_RECENT_NOTES] most recent notes. Only include notes whose opening transcript begins with “Captain’s Log.” Use them to create today’s summary, list my open tasks and follow-ups, and update the running task list in this project. Do not invent owners, deadlines, or completed work. ``` Review the first result. Make sure ChatGPT found the right entries, carried forward unfinished work, and did not turn a passing comment into a task. Once the manual run looks right, tell it to automate the same process: ```text Turn this process into a reusable Captain’s Log skill and schedule it to run every day at [TIME]. On each run, use @Granola to check the last [LOOKBACK_DAYS] day(s) for notes whose first spoken sentence begins with “Captain’s Log.” Create or update that day’s summary and a running Markdown task list. Add explicit new tasks, update matching tasks, and mark an open task stale after [STALE_AFTER_DAYS] days without a mention. Never mark a task complete unless the log says it is complete, and do not invent owners or deadlines. ``` For a daily end-of-day run, set [LOOKBACK_DAYS] to 1 and choose a time after you normally finish work. Then open Scheduled in the ChatGPT sidebar, open the new task, and confirm the project, frequency, and instructions. ![Daily Granola Captain's Log automation](https://tru-images.b-cdn.net/guide-assets/local-388/make-ai-note-taking-easier-with-this-chatgpt-granola-captains-log-setup/05-daily-granola-captains-log-automation.png) > Pro tip: Tag the actual Captain’s Log skill in Scheduled instead of leaving a text reference to its file. In our test, ChatGPT initially inserted only the skill’s location, so the scheduled task needed that manual correction. OpenAI notes that scheduled tasks cannot read ordinary files stored inside a ChatGPT project. Keep every instruction the run needs inside the scheduled task or the tagged skill, then check the first scheduled result before relying on it. ### Guide: Make Any Complex Topic Click With the Claude ELI5 Skill > In this guide, you will learn to make Claude a better teacher with the ELI5 skill. - Canonical page: https://app.therundown.ai/guides/make-any-complex-topic-click-with-the-claude-eli5-skill - Machine-readable page: https://app.therundown.ai/guides/make-any-complex-topic-click-with-the-claude-eli5-skill.md - Difficulty: beginner - Categories: AI, Claude, skills, Claude Code - Instructors: Billy Howell - Published: 2026-09-14 - Updated: 2026-09-14 ### Overview #### The Rundown In this guide, you will learn to make Claude a better teacher with the ELI5 skill. Then you will learn to test it to see the improvement yourself. #### Who This Is Useful For - **Sarah in ops** who needs to explain a new process to a new hire and a director without rebuilding the prompt each time. - **Marcus in consulting** who wants to check a skill before using it on a client deliverable. - **Anyone installing skills in Claude Code or claude.ai** who wants to see whether a skill changes the output. #### What You Will Build You will install ELI5, use it to explain a hard topic for one audience, then test the same prompt with the skill on and off. ![Terminal output showing a PASS RATE SUMMARY comparing an ELI5 skill run against a no-skill baseline, both at 100% with a 0.0% delta](https://tru-images.b-cdn.net/guide-assets/local-480/make-any-complex-topic-click-with-the-claude-eli5-skill/01-eval-pass-rate-summary.png) #### What You Need To Get Started - [Claude Code](https://claude.com/product/claude-code) with terminal access - A [claude.ai](https://claude.ai/) account with Code execution and file creation turned on in **Settings > Capabilities** if you also want to use the skill outside the terminal (any plan works, including Free) - Git and Python installed - The [DreambigOu/ELI5](https://github.com/DreambigOu/ELI5) GitHub repository #### Going Further Give Claude any skill's GitHub repo and ask it to build an evaluation that checks whether the skill changes the result. Use the same pattern for every skill you install. ### Step 1: Install the ELI5 Skill ELI5 is a public Claude skill. It gives Claude instructions for explaining a hard topic in simpler language for a particular audience. ![GitHub README for the ELI5 repo showing the Installation section with a git clone command and a link to the eval-system guide](https://tru-images.b-cdn.net/guide-assets/local-480/make-any-complex-topic-click-with-the-claude-eli5-skill/02-install-eli5-skill-github.png) Open a terminal in a folder you don't mind using for this, then clone the repo and copy the skill folder into Claude Code's skills directory: ```text git clone https://github.com/DreambigOu/ELI5.git cp -r ELI5/skills/eli5 ~/.claude/skills/eli5 ``` Restart Claude Code, then run `/skills` to confirm ELI5 appears in the list. For claude.ai, skip the terminal. Open the `eli5` skill folder in the GitHub repo, download `SKILL.md`, and drag it into a chat. First turn on Code execution and file creation in **Settings > Capabilities**. An Enterprise admin can also turn on skill and plugin scanning in Organization settings. If that setting is off, read the file yourself before you trust it. > **Pro tip:** Skim `SKILL.md` before trusting any skill with real prompts. It's a plain-text file, so reading it takes a minute and tells you exactly what the skill will and won't do. ### Step 2: Run the Skill on a Real Prompt Try it once before doing anything else: ```text Explain like I'm five what skills are and how you can create one. ``` ![Claude Code terminal output where the ELI5 skill explains what a Claude skill is using a recipe card cookie analogy](https://tru-images.b-cdn.net/guide-assets/local-480/make-any-complex-topic-click-with-the-claude-eli5-skill/03-eli5-explains-what-a-skill-is.png) You can also point the skill at a document, codebase, or error message. The useful part is getting an explanation that fits the person reading it. ### Step 3: Ask Claude to Build Its Own Test Cases Trying a skill once tells you what it sounds like. A quick evaluation tells you whether it changes anything. The ELI5 repo includes an evaluation setup. Ask Claude to use its format in `eli5-workspace/evals.json`: ```text Set up two test cases for me from this repo. You'll find the format in the GitHub instructions. Make one about the NBA playoffs and make one about golf. ``` Claude adds the prompts and pass/fail checks to `evals.json`. You do not have to build the evaluation system yourself. ![Diff view of evals.json with two new test cases added: an NBA-playoffs prompt for a 10-year-old and a golf-scoring prompt for a parent, each with scoring assertions](https://tru-images.b-cdn.net/guide-assets/local-480/make-any-complex-topic-click-with-the-claude-eli5-skill/04-define-eval-test-case-evals-json.png) ### Step 4: Run the Eval and Read the Score, Not the Vibe Ask Claude for the command for each new test, then run it from the cloned repo folder. The command uses the test id in `evals.json`. In our example, NBA was test 4 and golf was test 5: ```text python3 eli5-workspace/run-evals.py --test=4 python3 eli5-workspace/run-evals.py --test=5 ``` Keep `--test=N`. Without it, the script reruns every case in `evals.json`. ![Terminal running run-evals.py, showing the with-skill and baseline configs kicking off for the NBA-playoffs test case](https://tru-images.b-cdn.net/guide-assets/local-480/make-any-complex-topic-click-with-the-claude-eli5-skill/05-run-eval-comparison-terminal.png) Each test answers the prompt with ELI5 and without it, then grades both against the checks and prints a pass-rate summary. We tested NBA playoffs and golf, and both passed with ELI5 and without it. Those topics were easy enough that baseline Claude explained them well too. If that happens to you, pick a harder topic before deciding the skill is not helping. > **Pro tip:** If both versions pass, try something harder or more specific, such as a piece of your own code or a dense internal process. The ELI5 repo's published evals cover tougher tests, including explaining a database index to a 5-year-old, a codebase to a manager, and a git merge conflict to a 5th grader. That is the repo's earlier benchmark, not a result we reproduced in this guide. ### Guide: Make Claude an Expert at Anything With This Skills Repo > This guide shows you how to supercharge Claude Code by adding a marketplace of skills from the CLI. - Canonical page: https://app.therundown.ai/guides/make-claude-an-expert-at-anything-with-this-skills-repo - Machine-readable page: https://app.therundown.ai/guides/make-claude-an-expert-at-anything-with-this-skills-repo.md - Difficulty: intermediate - Featured tool or company: Claude - Categories: Coding & App Development, Automation & Agents - Instructors: Billy Howell - Published: 2026-01-21 - Updated: 2026-09-11 ### Overview #### The Rundown This guide shows you how to supercharge Claude Code by adding a marketplace of skills from the CLI. You’ll quickly browse, install, and use skills that help Claude work with documents, frontend design, and project files. #### Who This Is Useful For - Website developers using Claude Code - Anyone working with file types like PDFs, DOCX files, Excel spreadsheets, and more - Anyone who wants to save time building custom sub-agents in Claude Code #### What You Will Build You’ll set up Claude Code with reusable skills, then use those skills to create practical outputs from your own project context. - A Claude Code setup with skills installed from a marketplace - A project folder containing source context, such as a Markdown file or CSV - A PDF explaining your Markdown file - A simple landing page based on your Markdown document ![claude_skills_terminal-1](https://tru-images.b-cdn.net/guide-assets/69/69712e4885954edd92f3371d/claude-skills-terminal-1.png) #### What You Need - Claude Pro plan or above - Claude Code installed on your computer - Git, optional #### Going Further - After you finish, you can explore more marketing and business operations skills by adding another marketplace with /plugin marketplace add alirezarezvani/claude-skills. You can also use skills.sh to read what’s inside each skill and better understand how it works. ### Step 1: Add the Claude skills marketplace Open your terminal, type claude, and press Enter. Then add the Claude skills marketplace by running: /plugin marketplace add anthropics/skills. ### Step 2: Browse and install useful skills Type /plugins and press Enter. Use Tab to toggle between menus, then press Tab twice to go to Marketplace. Highlight claude-plugins-official, press Enter, and select “Browse plugins.” - Install context7. - Install frontend-design. - Add anthropic-agent-skills as a marketplace. - Install document-skills. ### Step 3: Prepare your project context Create a new folder and open it in your code editor. Add project context, such as a Markdown file with a project plan or a CSV with data. In the example, the folder includes a Markdown file with stats and news about the current NBA season. ### Step 4: Generate a PDF from your project file Optionally, initialize Git by running git init in your terminal. Run claude /init and press Enter. This tells Claude to create a Claude.md instructions file that it can update as it works. Ask Claude to create a PDF explaining the Markdown file you added. Explicitly tell it to use the PDF doc tool. ```text Create a PDF explaining the Markdown file in this project. Use the PDF doc tool. ``` ![04_pdf_creation](https://tru-images.b-cdn.net/guide-assets/69/69712e9e6630912af37e1e33/04-pdf-creation.png) ### Step 5: Create a landing page with frontend-design Use the frontend-design tool to create a simple landing page based on your Markdown document. ```text Use frontend-design to create a simple landing page based on my Markdown document. ``` In the example, Claude was asked to create an early 2000s-style sports website. ![05_frontend_design](https://tru-images.b-cdn.net/guide-assets/69/69712ea86630912af37e1e65/05-frontend-design.png) ### Guide: Make Claude Code Smarter With the Context7 MCP > Learn how to connect Claude Code to the Context7 MCP so it can pull current coding documentation instantly and make fewer mistakes. - Canonical page: https://app.therundown.ai/guides/make-claude-code-smarter-with-the-context7-mcp - Machine-readable page: https://app.therundown.ai/guides/make-claude-code-smarter-with-the-context7-mcp.md - Difficulty: intermediate - Featured tool or company: Claude - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2025-12-18 - Updated: 2026-09-11 ### Overview #### The Rundown Learn how to connect Claude Code to the Context7 MCP so it can pull current coding documentation instantly and make fewer mistakes. You will set up Context7, point Claude Code to the right docs, and test the workflow by planning and building a PokéAPI HTML, CSS, and JavaScript site. In the source example, the working site was built in 3 minutes. #### Who This Is Useful For - Anyone building a website on top of an external API library - Anyone coding with an unfamiliar coding language or integration - Coders using the latest frameworks and libraries #### What You Will Build You will create a Claude Code project connected to Context7, add reusable project rules in Claude.md, and test the setup by generating a plan for a simple PokéAPI site. - A Context7 MCP connection for Claude Code - A Claude.md file that tells Claude Code when and how to use Context7 - A PokéAPI project plan generated with current documentation - A functioning HTML, CSS, and JavaScript PokéAPI app as the example build ![1_what_you_will_build](https://tru-images.b-cdn.net/guide-assets/69/69443fdeea127e50240f20a2/1-what-you-will-build.gif) #### What You Need - Claude Pro plan for Claude Code - Cursor or another IDE - An API library with public documentation - A free Context7 account - A Context7 API key - A new project folder #### Going Further - After the setup works, verify requests in your Context7 dashboard, confirm the documentation source and version, then ask Claude Code to build a fully functional V1 of the app from its plan. You can reuse the same Context7 and Claude.md workflow for other libraries, APIs, frameworks, and integrations with public documentation. ### Step 1: Set up your project and terminals - Create a new project folder and open it in Cursor. - Open a new terminal by clicking Terminal in the top bar, then New Terminal. - Type “Claude” to start Claude Code. - Open a second terminal. This is where you will paste your Context7 install script. ### Step 2: Create your Context7 API key and install script - Create a free Context7 account. - Click the “Claude Code” tab and copy the installation script for “Remote.” Save it in your notes. - Click Create API Key and copy the key. - Paste the key into your installation script where it says YOUR_API_KEY. Pro tip: Never share your API keys with anyone, even for free platforms like PokéAPI. ![2_context7_screenshot](https://tru-images.b-cdn.net/guide-assets/69/69443fc4ea127e50240f2024/2-context7-screenshot.png) Use Context7’s full installation guide if you need more help with the setup. ### Step 3: Install and verify the Context7 MCP Go back to Cursor and paste the installation script into your regular terminal, not the Claude Code terminal. ![31_claudecode](https://tru-images.b-cdn.net/guide-assets/69/69443fc3ea127e50240f201c/31-claudecode.png) Check the install with the command claude mcp list. ![32_claude_code](https://tru-images.b-cdn.net/guide-assets/69/69443fc3ea127e50240f2014/32-claude-code.png) ### Step 4: Add Context7 rules to Claude.md Create a Claude.md file. Use it to set rules for how Claude Code works in this project. In this case, specify that Claude Code should always use Context7 when it needs to read documentation. ```text #Workflow Always use context7 when I need code generation, setup or configuration steps, or library/API documentation. This means you should automatically use the Context7 MCP tools to resolve library id and get library docs without me having to explicitly ask. #PokéAPI The pokeapi docs can be found with context7 at . Make sure you're using v2 of the docs. ``` You can also specify the source of the docs. Verify the source by searching on Context7. For this PokéAPI example, v2 of the API is the version you should use. ### Step 5: Test Context7 with a PokéAPI planning prompt Now test Context7 with PokéAPI, which is free. Send a planning prompt to Claude Code in your terminal. You may need to press Enter afterward to approve the tool use. You can arrow down and select “Yes, and never ask again for Context7” so Claude Code does not need to ask for each document it pulls. ```text I want to build a simple html site that creates random teams of 6 of pokemon. you can “lock” Pokemon into a team to reshuffle the remaining team members or select a specific type and/or generation to filter the random selection by. Write a quick plan in markdown using the pokapi docs. ``` ![41_claude_code](https://tru-images.b-cdn.net/guide-assets/69/69443fc3ea127e50240f200c/41-claude-code.png) You can verify that the requests went through in your Context7 dashboard. ![42_context7](https://tru-images.b-cdn.net/guide-assets/69/69443fc3ea127e50240f2004/42-context7.png) ### Step 6: Build the first functional version After Claude Code lays out the plan, ask it to build a fully functional V1 of the app. Pro tip: Press ctrl+o in a Claude Code terminal to see its full thoughts. It is good practice to understand how it thinks. For the demo, Claude opened the app and it worked in one shot. ![1_what_you_will_build](https://tru-images.b-cdn.net/guide-assets/69/69443fdeea127e50240f20a2/1-what-you-will-build.gif) ### Resources - [Make Claude Code Smarter With the Context7 MCP Full Guide](https://tru-images.b-cdn.net/guide-assets/69/69444bc41f9bea17659c4a02/make-claude-code-smarter-with-the-context7-mcp-full-guide.pdf) ### Pro tip > This is especially important for newer versions of coding frameworks and libraries that were released after Claude Code’s knowledge cutoff. ### Guide: Mine Reddit Complaints Into Business Ideas With an AI Pain Radar > In this guide, you will learn how to use Codex and Airtable to build an infinite business idea generator. - Canonical page: https://app.therundown.ai/guides/mine-reddit-complaints-into-business-ideas-with-an-ai-pain-radar - Machine-readable page: https://app.therundown.ai/guides/mine-reddit-complaints-into-business-ideas-with-an-ai-pain-radar.md - Difficulty: beginner - Featured tool or company: Codex + Airtable - Categories: Data, Research & Analysis, Business Operations & Leadership, Automation & Agents - Instructors: Billy Howell - Published: 2026-06-20 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Codex and Airtable to build an infinite business idea generator. If you build websites, apps, or automations with AI, this will help you find the painful problem people may pay you to fix. #### Who This Is Useful For This is useful if you can already build with AI, but you are not sure what painful problem someone might pay you to fix. The goal is not to ask AI for random startup ideas. The goal is to create an AI Pain Radar: a repeatable system that finds public complaint signals, stores the raw evidence, and turns repeated pain into business ideas that stay linked to the posts that inspired them. It is especially useful for: - solo builders looking for validated app ideas - freelancers who want productized service ideas - agency owners looking for repeatable client problems - creators who want newsletter, landing page, or prototype ideas based on real demand #### What You Will Build You will build a two-table Airtable base and connect it to recurring Codex automations. The first automation mines public complaint signals and turns them into linked business ideas. The second automation, in the Going Further section, looks at the top-rated idea from the day before and creates a simple landing-page proof of concept in a local folder. ![Proof-of-concept landing page generated from a top-rated Airtable idea.](https://tru-images.b-cdn.net/guide-assets/local-272/mine-reddit-complaints-into-business-ideas-with-an-ai-pain-radar/01-freelancer-payment-reminder-landing-page.png) By the end, Airtable becomes your evidence database. Codex becomes the research and build agent that keeps adding new ideas and tests. #### What You Need To Get Started - A free Airtable (https://airtable.com/) account - The Codex desktop app (https://openai.com/codex/) - The Airtable plugin connected inside Codex - Three subreddits or public communities you want Codex to monitor - A local folder where Codex can save proof-of-concept landing pages #### Going Further: Turn Top Ideas Into Landing Pages The next step is to make the system build a tiny proof of concept for the best idea. Create a standalone folder inside the guide or project: ```text proofs-of-concept/ ``` Then add a few tracking fields to your Business Ideas table: - prototype_status - prototype_path - prototype_generated_at Now set up a second Codex automation. This one should run after the idea miner, look at the top-rated business idea from the previous day, and create a single static landing page in a new local folder. Use this prompt: ```text Set up a recurring Codex automation called Daily Airtable Idea Landing Page POC. Every morning, look at the Business Ideas table in Airtable and find the top-rated idea created during the previous calendar day. Rank ideas by opportunity_score, then confidence, then evidence strength. Prefer ideas that do not already have prototype_status = generated. For the selected idea, create a new local folder inside: [PROJECT_FOLDER]/proofs-of-concept/ Slug the folder from the idea name. Inside that folder, create: - index.html - styles.css - README.md Build a simple static landing page using the Airtable fields for the headline, target persona, problem, proposed solution, evidence summary, validation metric, next action, and weakness or risk. The page should include: - a hero - a product workflow or product preview - an evidence/proof section - a waitlist or interview CTA stub After the files are created, update the selected Airtable record: - prototype_status = generated - prototype_path = the local index.html path - prototype_generated_at = the current local ISO timestamp If generation fails after selecting an idea, set prototype_status = needs_review and explain the issue in reviewer_notes. If there are no qualifying ideas from the previous day, do not create a project. ``` This turns the Pain Radar into a full idea loop: find pain, store evidence, score opportunities, and create something small enough to test. You can use the landing page as a waitlist, a cold outreach asset, a tweet thread anchor, or a quick prototype brief. The important part is that the landing page still traces back to the Airtable idea, and the Airtable idea still traces back to raw evidence. That chain is what keeps the system useful. ### Step 1: Connect Airtable To Codex Start by signing in to Airtable. If you are new to Airtable, think of it like Google Sheets with stronger field types, views, AI fields, and automations. Then open the Codex desktop app, go to Plugins, search for Airtable, and add it. ![Adding the Airtable plugin inside Codex.](https://tru-images.b-cdn.net/guide-assets/local-272/mine-reddit-complaints-into-business-ideas-with-an-ai-pain-radar/02-codex-airtable-plugin.png) After the plugin is connected, start a new Codex thread and mention Airtable directly with @Airtable. Before you ask it to create anything, ask it to inspect the connection. > Pro tip: Have Codex inspect a new plugin before it builds with it. This prevents a lot of failed prompts where the model guesses what the connector can do. Use this prompt: ```text Check my Airtable connection. List the bases I can access, identify the most recently used base, and tell me whether you can create tables and records in it. Do not change anything yet. ``` This gives Codex a chance to learn what the Airtable connector exposes in your account. Once Codex can see Airtable, Airtable becomes the database layer for the research agent. ### Step 2: Have Codex Create The Airtable Schema The clean version of this system uses two tables: - Raw Reddit Posts - Business Ideas This matters because raw posts are evidence. Business ideas are conclusions. Keeping them separate makes the system harder to fool. Ask Codex to create the schema for you instead of pasting a long list of columns manually. Use this prompt: ```text Create an Airtable schema for an AI Pain Radar. Use two tables: 1. Raw Reddit Posts - one row per raw public complaint signal - fields for source, URL, thread title, post/comment text, quote, persona, company stage, job to be done, pain summary, pain cluster, intensity, urgency, willingness to pay, evidence quality, discovery date, research status, duplicate notes, and reviewer notes 2. Business Ideas - one row per synthesized opportunity - fields for idea ID, idea name, status, pain cluster, target persona, problem summary, proposed solution, supporting posts, evidence summary, best quotes, scoring rubric, confidence, business model, landing-page headline, landing-page test, validation metric, next action, weakness/risk, and reviewer notes Link Business Ideas back to Raw Reddit Posts through a supporting_posts linked-record field. Seed a few fictional demo rows so I can verify the relationship. ``` After Codex finishes, open the Airtable base and check the two tables. The raw table should hold the source evidence. ![Raw Reddit Posts table with seeded public complaint evidence.](https://tru-images.b-cdn.net/guide-assets/local-272/mine-reddit-complaints-into-business-ideas-with-an-ai-pain-radar/03-raw-reddit-posts-table.png) The ideas table should hold the synthesized opportunities. ![Business Ideas table shown as a gallery view.](https://tru-images.b-cdn.net/guide-assets/local-272/mine-reddit-complaints-into-business-ideas-with-an-ai-pain-radar/04-business-ideas-gallery.png) The important field is the linked-record field from Business Ideas back to Raw Reddit Posts. Every idea should point back to the raw posts that support it. If an idea cannot point back to evidence, it should stay marked as weak evidence or stay out of the ideas table entirely. ### Step 3: Make The Base Easier To Review You can use Airtable as a plain database, but a few views make review much faster. For Raw Reddit Posts, ask Codex to create or suggest views like: - Ready for Review - High Intensity - Needs Source Check - Linked to Ideas For Business Ideas, create views like: - Top Opportunities - Landing Page Tests - Weak Evidence - Rejected or Parked Use this prompt: ```text Suggest useful Airtable views and AI classification fields for this AI Pain Radar. For Raw Reddit Posts, suggest views for: - Ready for Review - High Intensity - Needs Source Check - Linked to Ideas For Business Ideas, suggest views for: - Top Opportunities - Landing Page Tests - Weak Evidence - Rejected or Parked Also suggest Airtable AI field instructions that classify pain clusters, evidence quality, and idea quality without inventing facts. ``` Airtable AI can help classify and score rows after the evidence is already in the base. ![Airtable AI field agent setup for automated classification.](https://tru-images.b-cdn.net/guide-assets/local-272/mine-reddit-complaints-into-business-ideas-with-an-ai-pain-radar/05-airtable-ai-field-agent.png) The key limitation: Airtable AI is not the recurring web researcher here. Use it to classify, summarize, and score the records Codex adds. ### Step 4: Set Up The Reddit Pain Radar Automation Now have Codex set up the recurring research job. This automation should check public communities, find complaint signals, dedupe against Airtable, write raw evidence first, and only create business ideas when there is enough support. Use this prompt: ```text Set up a recurring Codex automation called Reddit Pain Radar. Every weekday morning, find 10-20 fresh public complaint signals from Reddit-style discussions, public forums, product reviews, and founder/operator communities. Write accepted evidence into Raw Reddit Posts first. Only create or update Business Ideas when there is enough evidence or a clearly high-intensity workflow pain. Every Business Ideas record must link back to supporting Raw Reddit Posts. Respect source terms, avoid private or login-gated communities, do not store personal identifiers, dedupe before writing, and label weak evidence instead of overstating it. ``` You can swap in your own sources: ```text Every weekday, check [SUBREDDIT_1], [SUBREDDIT_2], and [SUBREDDIT_3] for complaints about slow, expensive, confusing, or repetitive work. ``` Do not make the job create an idea for every post. That defeats the point. Most posts should stay as evidence until there is a pattern. A good business idea record should include: - the pain cluster - the target persona - a plain-English problem summary - a proposed solution - linked supporting posts - evidence summary - opportunity score - weakness or risk - next landing-page test ### Step 5: Score Ideas Without Letting AI Drift Once Codex has added raw evidence and a few candidate ideas, use Airtable AI or another model to score the ideas. The rule is simple: the score has to come from linked evidence. Use this prompt on a Business Ideas record: ```text Classify and score this idea using only linked supporting posts. Return: 1. pain cluster 2. buyer/persona 3. evidence summary 4. best customer quotes 5. intensity score 1-5 6. frequency score 1-5 7. specificity score 1-5 8. willingness-to-pay score 1-5 9. buildability score 1-5 10. 100-point opportunity score 11. one weakness or risk 12. next landing-page test If the linked evidence is weak, say weak evidence instead of improving the idea. ``` This keeps the system honest. AI is allowed to score the idea, but it is not allowed to make the evidence sound better than it is. If the model starts inventing details, use this recovery prompt: ```text Pause and revise using only linked raw post evidence. For each claim, point to the supporting post ID or quote. If there is not enough evidence, mark it weak evidence. Keep the output to: 1. pain cluster 2. evidence quotes 3. score 4. weakness 5. next landing-page test ``` ### Step 6: Run A Small Test Before You Trust The System Before you let the automation run every day, ask Codex for one bounded test run. Use this prompt: ```text Run a small test of the Reddit Pain Radar. Find 5 public complaint signals across the target communities, add them to Raw Reddit Posts, and create Business Ideas only when the evidence supports them. After writing to Airtable, read the records back and confirm that each idea links to its supporting raw posts. ``` In our test, Codex added five raw complaint signals and three business ideas. The strongest idea was a freelancer payment reminder assistant, supported by two posts about the same pain: freelancers need to get paid without sounding pushy or damaging client relationships. That test became the landing-page proof of concept shown at the top of this guide. ### Guide: Never Lose Your Vibe Coding Progress Again With Git > Learn how to save your vibe coding progress with Git so you do not lose your work. - Canonical page: https://app.therundown.ai/guides/never-lose-your-vibe-coding-progress-again-with-git - Machine-readable page: https://app.therundown.ai/guides/never-lose-your-vibe-coding-progress-again-with-git.md - Difficulty: intermediate - Featured tool or company: Git - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-01-12 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to save your vibe coding progress with Git so you do not lose your work. You’ll practice eight practical Git commands, create your first commit, use a feature branch, merge changes back into main, and add Git rules for your AI coding assistant. #### Who This Is Useful For - Vibe coders who skipped CS class - Claude Code users who want to save tokens - Cursor users who want to save their progress #### What You Will Build You will build a Git-tracked coding project that can preserve progress through commits, feature branches, and merges. You will also create reusable Git rules for your AI coding assistant. - A project initialized as a Git repository - A first commit saved to the main branch - A feature branch called feature-1 - A merge from the feature branch back into main - A reusable AI context prompt for Git workflow rules ![what-you-will-build](https://tru-images.b-cdn.net/guide-assets/69/6966a7098a2739575d875be0/what-you-will-build.png) #### What You Need - A computer with terminal or command line access - A text editor or IDE, such as Cursor - Git installed on your computer - A basic HTML file or the downloadable example file #### Going Further - After you have the local Git workflow working, you can save your code to GitHub, create a new repository, and push branches to the cloud with git push origin [branch name]. ### Step 1: Install Git Go to git-scm.com. - On Windows, download the installer. - On Mac, follow the instructions to install Git with Homebrew. Homebrew and Git are both safe, open source libraries that have been around for decades. Verify the installation by pasting this into your terminal and pressing return: git --version ### Step 2: Add Git to your project Open your code editor and create a new folder with a basic HTML file in it. You can download the example file from the resources section if you’d like. Open a terminal session in your code editor. Type this command and press return: git init You can now track changes in this repository. > 💡 Pro tip: Ask your AI coding assistant to create a README.md that explains your project. This will come in handy if you upload the project to GitHub. ### Step 3: Write your first commit A commit is a snapshot of your code at a specific moment. - A unique ID, such as abc1234def5678 - A message describing what changed - An author - A timestamp Before writing your commit, add your changes to the commit: git add . Now write your first commit: git commit -m 'first commit!' ![first-commit](https://tru-images.b-cdn.net/guide-assets/69/6966a6e58a2739575d875b9e/first-commit.png) Your changes are now saved and added to your main branch. ### Step 4: Create a feature branch When you’re coding, it’s important to create new branches for new features. This practice can save headaches and improve your coding agent’s context. Create a new branch called “feature-1”: git checkout -b feature-1 Any commits you make now will be saved to the feature-1 branch. To see a list of your branches and your current branch, run: git branch ![git-branch](https://tru-images.b-cdn.net/guide-assets/69/6966a7188a2739575d875bfe/git-branch.png) To switch branches, use: git checkout [branch name] ### Step 5: Add a feature to your main branch Now save your new progress to your main branch. First, add some changes to a file in your new branch. Save them with a commit. Switch to your main branch: git checkout main Merge your feature branch into main: git merge feature-1 ![git-merge](https://tru-images.b-cdn.net/guide-assets/69/6966a72c8a2739575d875c25/git-merge.png) Your changes are now saved to your main branch. > 💡 Pro tip: Every commit and merge has a unique identifier that you can see with git log. To undo a merge, you can use git reset [commit hash]. ### Step 6: Make AI use Git for you When you start a new project, add this simple prompt into your context file. For example, add it to CLAUDE.md if you are using Claude Code. ```text # Git Rules 1. Create feature branch: `git checkout -b feature-[name]` 2. Commit after each logical change: `git commit -m "Add [what you did]"` 3. When done, ask: "Ready to merge feature-[name] into main?" 4. Wait for approval before merging 5. Never commit directly to main ``` ### Step 7 Optional: Save your code to GitHub - To save your code to GitHub, install it to your terminal and then log in with gh auth login. - Create a new repository on github.com. Copy the commands into your terminal and press return. ![going-further](https://tru-images.b-cdn.net/guide-assets/69/6966a74669b5ffdd7edf5cef/going-further.png) Any time you want to save a branch in the cloud, use this command: git push origin [branch name] ### Resources - [Git visualizer example HTML](https://tru-images.b-cdn.net/guide-assets/69/6966a7b8dd067f1607c14f08/git-visualizer-index.html) ### Pro tip > Ask your AI coding assistant to create a README.md that explains your project. This will come in handy if you upload the project to GitHub. ### Guide: OpenClaw/Moltbot/Clawdbot Installation Guide & How To Use It > This guide teaches you how to install and set up your own autonomous agent using OpenClaw, formerly called Clawdbot and Moltbot. - Canonical page: https://app.therundown.ai/guides/moltbot-clawdbot-installation-guide-and-how-to-use-it - Machine-readable page: https://app.therundown.ai/guides/moltbot-clawdbot-installation-guide-and-how-to-use-it.md - Difficulty: advanced - Featured tool or company: Moltbot - Categories: Automation & Agents, Coding & App Development, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-01-28 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches you how to install and set up your own autonomous agent using OpenClaw, formerly called Clawdbot and Moltbot. You will produce a working OpenClaw bot that can be configured with search, an AI provider, memory hooks, and Telegram messaging. The setup involves collecting API keys, installing the CLI, running onboarding, and pairing the bot with Telegram. #### Who This Is Useful For - Anyone with a laptop and an AI subscription - Tinkerers trying to learn more about custom agents - People who want a proactive AI assistant #### What You Will Build You will build and configure your own OpenClaw bot. The example bot in the source, named Botthew, designed and deployed a status website and represented its sub-agents “Sandstorm” and “Crawler.” - A local OpenClaw/Moltbot setup - A connected AI provider key - Brave search access - A Telegram bot channel for messaging your agent - Memory hooks selected during onboarding #### What You Need - ChatGPT Plus is recommended, or Google AI Pro for Gemini + Nano Banana, or Claude Pro. - A MacBook. Linux and Windows may work, but the source notes that Moltbot is tailored to Mac. - A backup or wiped computer is strongly recommended because this is experimental, open-source code. - If you install on your primary MacBook, complete a full Time Machine backup first. - Node version 22 or higher. - A Brave API key for search. - A fresh API key from OpenAI, Claude, or Gemini. - A Telegram bot key created through BotFather. #### Going Further - Give Moltbot a Gemini key from AI Studio and ask it to use Nano Banana Pro to generate a Telegram avatar. - Ask it to send a 7AM report each morning with the weather and its goal for the day. - Connect Gmail so OpenClaw can monitor your inbox, or use a free agentmail inbox and forward mail to it. - After your Moltbot has an email address, sign it up for GitHub so it can deploy websites through GitHub Pages. ![son-of-botthew](https://tru-images.b-cdn.net/guide-assets/69/697a869c78b0ac9a05734ac3/son-of-botthew.png) ### Step 1: Gather your keys and create a Telegram bot Before you start setup, collect the keys you will need and save them in a notes file. - Register a free Brave account and get a Brave API key so your bot can use search. - Get a fresh API key from your AI provider: OpenAI, Claude, or Gemini. - In Telegram, search for the user “BotFather” and start a new chat. - Click “Create a New Bot,” then give it a name and a handle that ends with “bot.” - Copy the unique Telegram bot key and save it to your notes for onboarding. ![botfatherpng](https://tru-images.b-cdn.net/guide-assets/69/697a863978b0ac9a057349d2/botfatherpng.png) ### Step 2: Install the CLI and complete onboarding Make sure Node is installed and up to date. You need Node version 22 or higher to use the Moltbot CLI. Install the Moltbot CLI with: curl -fsSL https://openclaw.bot/install.sh | bash Run the onboarding wizard with: sudo openclaw onboard --install-daemon ![moltbot_onboarding](https://tru-images.b-cdn.net/guide-assets/69/697a865e78b0ac9a05734a3b/moltbot-onboarding.png) - Select “Quickstart.” - Select your AI provider and paste in your token. - Say yes to Brave search when prompted and paste your API key. - Select Telegram as your channel. This is where you will message your bot. - Skip the Google Places API information if you do not need it. The source notes that it is for things like weather updates. - Select all the hooks so the bot can develop its memory each day. > Pro tip: Don’t get distracted. Skip all the frills and only install one skill at a time. ### Step 3: Start the introduction wizard After setup, you will be taken into the introduction wizard. At this point, you are talking with your OpenClaw bot. ![moltbot-setup](https://tru-images.b-cdn.net/guide-assets/69/697a868278b0ac9a05734a93/moltbot-setup.png) ### Step 4: Pair your bot with Telegram - Search for your bot’s username in Telegram. - Send it a message. - Copy the pairing code it gives you. - Open Terminal and run this command, replacing with your pairing code: openclaw pairing approve telegram You can now message your bot through Telegram. > Pro tip: You may have to try the clawdbot or moltbot command if openclaw does not work. That happened when this guide was recorded. ### Pro tip > Don’t get distracted. Skip all the frills and only install one skill at a time. ### Guide: Optimize Your Prompting With this Markdown Strategy > This guide teaches you how to take a task you’ve already completed with AI and extract it into reusable Markdown instructions. - Canonical page: https://app.therundown.ai/guides/optimize-your-prompting-with-this-markdown-strategy - Machine-readable page: https://app.therundown.ai/guides/optimize-your-prompting-with-this-markdown-strategy.md - Difficulty: beginner - Featured tool or company: Google - Categories: AI Foundations & Prompting - Instructors: Billy Howell - Published: 2026-01-19 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches you how to take a task you’ve already completed with AI and extract it into reusable Markdown instructions. You’ll turn a valuable AI workflow into a portable, repeatable project template that can be used across AI tools. #### Who This Is Useful For - Marketers creating high volumes of custom marketing materials - Entrepreneurs using multiple AI tools for different tasks - Anyone who wants to A/B test their prompting strategies #### What You Will Build You will build a reusable set of Markdown system instructions that captures a completed AI task and turns it into a repeatable template for future prompts. - Markdown instructions for a completed AI workflow - Rules for generating consistent outputs, such as the number of variations to create - Follow-up questions the AI should ask before starting the task - A reusable project template that accepts new input variables ![00_what_you_will_build_1456x816](https://tru-images.b-cdn.net/guide-assets/69/696e9afb14f769507a7d81e5/00-what-you-will-build-1456x816.png) #### What You Need - A Google account for NotebookLM - An example task, such as creating email template variations - Context documents for the task - Task inputs such as tone, target audience, word count, brand name, value proposition, and specific copy rules #### Going Further - Reuse the template by starting a new chat with custom input variables to reproduce the same type of output. - Build a task database in Notion, Airtable, or a Google Drive folder. - Each time you complete a valuable AI task, create Markdown instructions and save them so teammates can reuse the workflow regardless of which AI tool they use. ### Step 1 Complete a representative task in NotebookLM Create a new Notebook in NotebookLM and do your task as you normally would. In this guide, the example task is using existing email copy to create custom email sequence variations. Upload context documents and talk the AI through creating your deliverables. Specify the important inputs that should shape the output: - Tone - Target audience - Word count or character count - Name of your brand - Product you are selling or value proposition of the content you’re creating - Specific copy rules, such as “don’t use em-dashes” ### Step 2 Extract the workflow into Markdown instructions After you finish the task, paste this prompt into NotebookLM: ```text Take the task we’ve just completed and extract the relevant context into a set of system instructions for AI in markdown format. The goal is to drop these instructions into any AI with different input variables and receive polished output with one prompt. Specify rules like number of variations to generate and any follow-up questions the AI should ask before starting the task. ``` ### Step 3 Turn the instructions into a reusable project template Create a new project in NotebookLM and paste the Markdown instructions in as text. ![03_new_NotebookLM](https://tru-images.b-cdn.net/guide-assets/69/696e9b1c14f769507a7d8251/03-new-notebooklm.png) - For ChatGPT, click New Project → three dots → Edit Instructions - For Claude, click Projects → New Project → Create Project → Set Project Instructions - For Perplexity, click Spaces → New Space → Add Instructions - For Grok, click Projects → Create Project You’ve now turned hours of prompting and refining into a versatile project template. Start a new chat with custom input variables to achieve the same valuable outputs as your original chat. ### Guide: Our Claude Cowork System for Video Clipping and Editing > Learn how to use Claude Cowork to replace expensive video clipping tools. - Canonical page: https://app.therundown.ai/guides/our-claude-cowork-system-for-video-clipping-and-editing - Machine-readable page: https://app.therundown.ai/guides/our-claude-cowork-system-for-video-clipping-and-editing.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Content & Creative, Automation & Agents - Instructors: Billy Howell - Published: 2026-01-30 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use Claude Cowork to replace expensive video clipping tools. You will use the included prompt files with a video and timestamped transcript to generate your first clips in under 5 minutes. #### Who This Is Useful For - Social media marketers - Podcasters and video editors - Anyone with lots of longform videos sitting on their hard drive #### What You Will Build You will build a Claude Cowork video clipping workflow that reviews longform video, creates a clip-review.md file of clip ideas, and generates selected clips into a /clips subfolder. - A project folder containing SOP prompt files, a video, and a transcript - A clip-review.md file with scroll-stopping clip ideas - Generated video clips saved in the /clips subfolder ![05 - Smart Content System 3-Step](https://tru-images.b-cdn.net/guide-assets/69/697d2219fa86390bd53f93f7/05-smart-content-system-3-step.jpeg) #### What You Need - Claude Pro plan or higher - A video and transcript with timestamps, such as the attached examples - The attached prompt files #### Going Further - After generating clips, you can reuse Claude Cowork to combine, crop, cut, or mute clips. You can also create your own timestamped .srt transcriptions locally with Claude Code and openai-whisper when online transcript tools are slow or inaccurate. ### Step 1: Create your project folder Create a folder for the clipping project and add the required files. - Our prompt files - The video file - The video transcript ### Step 2: Open the folder in Claude Cowork Open the Claude app on your computer and select the Cowork tab. Click the folder icon and select the folder you created. ![coworkui](https://tru-images.b-cdn.net/guide-assets/69/697d225965f4172890f4bdb2/coworkui.png) ### Step 3: Generate clip ideas to review Ask Claude to read the main SOP, scaffold the directory, and find clip ideas. ```text Read the main SOP and scaffold our directory as described. Then find me 10 scroll-stopping clip ideas to review. ``` Claude will create a clip-review.md file where you can review each clip idea. ![03_terminal-1](https://tru-images.b-cdn.net/guide-assets/69/697d226865f4172890f4bde2/03-terminal-1.png) ### Step 4: Generate and edit selected clips Tell Claude which clips you want generated. ```text Generate clip 1, 4 and 6. ``` The generated clips should appear in the /clips subfolder. Try follow-up commands to combine, crop, cut, or mute clips. ```text Combine clips 4 and 5. ``` ```text Crop clip 1 into a vertical video. ``` ```text Cut 00:08:00:15 from clip 3. ``` ```text Mute clip 5. ``` ### Step 5 Optional: Create transcripts locally with Claude Code If generating transcripts with online tools is time-consuming or inaccurate, you can create transcriptions locally using Claude Code. - Install openai-whisper with pip install -U openai-whisper, or pip3 if you are on Mac. - Open Claude Code. ```text Use openai_whisper to transcribe this video into a .srt file with timestamps. ``` ### Resources - [00-MAIN-SOP.md](https://tru-images.b-cdn.net/guide-assets/69/697d238b65f4172890f4bf02/00-main-sop.md) - [01-CLIP-FINDING-SOP.md](https://tru-images.b-cdn.net/guide-assets/69/697d239665f4172890f4bf2c/01-clip-finding-sop.md) - [02-CLIPPING-SOP.md](https://tru-images.b-cdn.net/guide-assets/69/697d239e65f4172890f4bf60/02-clipping-sop.md) - [transcript.srt](https://tru-images.b-cdn.net/guide-assets/69/697d24c9779af18d8a6c2352/transcript.srt) - [Shia LaBeouf Just Do It Motivational Speech (Original Video by LaBeouf, Rönkkö & Turner) - MotivaShian (480p, h264, youtube).mp4](https://tru-images.b-cdn.net/guide-assets/69/697d24d459967088bb076edb/shia-labeouf-just-do-it-motivational-speech-original-video-by-labeouf-r-nkk-turner-motivashian-480p-h264-youtube.mp4) ### Guide: Perform Real-Time Market Research Using Grok > In this guide, you will learn how to track Twitter trends and news stories from the last 24 hours with xAI’s Grok. - Canonical page: https://app.therundown.ai/guides/perform-real-time-market-research-using-grok - Machine-readable page: https://app.therundown.ai/guides/perform-real-time-market-research-using-grok.md - Difficulty: beginner - Featured tool or company: Grok - Categories: Data, Research & Analysis - Instructors: Billy Howell - Published: 2025-12-27 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to track Twitter trends and news stories from the last 24 hours with xAI’s Grok. You will connect Make.com and Notion so Grok can turn that research into daily memos about a trend or competitor. #### Who This Is Useful For - Business owners - Startup employees - People looking for new AI workflows #### What You Will Build You will build a reusable market research dashboard and automation that creates research memos in Notion using Grok, Make.com, and an xAI API key. ![00_what_you_will_build](https://tru-images.b-cdn.net/guide-assets/69/694af09be0e753eabd9779ba/00-what-you-will-build.png) - A Notion Market Research Dashboard database - A Make.com scenario that receives a Notion webhook - A Grok-powered research flow that writes a full report back into Notion - Database properties for status, sources, research output, action items, and citations #### What You Need - X Premium or Premium+ account ($3) - Notion Plus plan ($10) - Make.com account; the free tier includes 1,000 operations/month - xAI API key from console.x.ai - At least $5 to fund the xAI developer account - Expected monthly cost: $13–$30, depending on usage #### Going Further - Set up a trigger that creates a new research memo on the same topic each day. - Send the webhook from that trigger each day, or manually turn it on each day to save tokens. ![trigger2config](https://tru-images.b-cdn.net/guide-assets/69/694feee1a95b89c155b76031/trigger2config.png) ### Step 1: Copy the Notion database template - Open the Market Research Dashboard template. - Click Duplicate in the top-right corner. ![01_duplicate_notion_template](https://tru-images.b-cdn.net/guide-assets/69/694af09dac16d19889c7f29d/01-duplicate-notion-template.png) - Open the database and verify these properties exist: - Research Topic (Title) - Industry/Competitor (Select) - Sources (Multi-select: X, Web, News, RSS) - Status (Select: New → Report Ready → Reviewed → Archived) - Date - Research Output - Action Items - Citations ### Step 2: Get your xAI API key ![02_grok_API_key](https://tru-images.b-cdn.net/guide-assets/69/694af09cac16d19889c7f295/02-grok-api-key.png) - Go to console.x.ai and sign in. - Click API Keys in the sidebar, then Create new secret key. - Name it “Market Research” and copy the key immediately. - Save it somewhere safe. You will not see it again. Note: When you sign up for an xAI developer account, you will need to fund it with at least $5. ### Step 3: Import the Make.com blueprint ![03_make_flow](https://tru-images.b-cdn.net/guide-assets/69/694af09cac16d19889c7f28d/03-make-flow.png) - Download the blueprint from the resources section. - Log in to Make.com and go to Scenarios → Create a new scenario. - Click the three dots (⋯) in the top right, then choose Import Blueprint. ![makecmblueprint2](https://tru-images.b-cdn.net/guide-assets/69/694fee482bb95ce0c46bf5d3/makecmblueprint2.png) - Upload the JSON file, then click Save. After importing, you will see four modules: - Webhook: triggers from Notion. - xAI: calls Grok. - Notion: Append Page Content: writes the full report. - Notion: Update Database Item: updates properties. ### Step 4: Connect your accounts Connect xAI: - Click the xAI module, which is Module 2. - Click Add next to Connection. - Paste your API key from Step 2, then click Save. Connect Notion: - Click the Notion: Append Page Content module, which is Module 3. - Click Add, then authorize access to your Market Research Dashboard. - Use the same connection for Module 4, Update Database Item. ### Step 5: Set up the Notion webhook trigger This is how Notion tells Make.com to run the automation. If you duplicated the provided Notion template, skip to sub-step 4. - Open your database in Notion, then click ⚡ Automate → + New automation. - Set the trigger: When: “Page is added to database”; Database: Market Research Dashboard; Filter: Status is “Start Research.” - Add the action: click New action → Send webhook. - Get your webhook URL: go to Make.com, click the Webhook module, which is Module 1, and copy the webhook URL. - Back in Notion, paste the webhook URL into the URL field. - Make sure the Research Topic, Date, Industry/Competitor, and Sources properties are selected to be sent to Make.com. ![webhooktriggerconfig](https://tru-images.b-cdn.net/guide-assets/69/694feeb2a95b89c155b75fb9/webhooktriggerconfig.png) ### Step 6: Test the workflow Time to see it work. - In your Notion database, click New. - Fill in the page with: Research Topic: “Customer complaints about ChatGPT”; Industry/Competitor: “OpenAI”; Sources: X, Web; Status: Start Research. - Create the page by clicking away or pressing Enter. - Wait 30–60 seconds, then refresh your database. - Open the page you created. - Ask Notion’s AI assistant to reformat the markdown for you. You can also ask Claude or ChatGPT if you do not have Notion AI. ![06_results](https://tru-images.b-cdn.net/guide-assets/69/694af09bac16d19889c7f285/06-results.png) ### Resources - [Grok market research Make.com blueprint](https://tru-images.b-cdn.net/guide-assets/69/6952fd7030b5c50fbf407897/grok.json) ### Pro tip > To save tokens, you can manually turn on the daily trigger instead of sending the webhook automatically every day. ### Guide: Plan Your Weekly Content Calendar in Minutes With Manus > In this guide, you will learn how to build a custom content calendar in Manus that plans, generates, and updates each week of social posts for you. - Canonical page: https://app.therundown.ai/guides/plan-your-weekly-content-calendar-in-minutes-with-manus - Machine-readable page: https://app.therundown.ai/guides/plan-your-weekly-content-calendar-in-minutes-with-manus.md - Difficulty: beginner - Featured tool or company: The Rundown AI - Categories: Marketing & Growth, Automation & Agents - Instructors: Billy Howell - Published: 2026-06-04 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to build a custom content calendar in Manus that plans, generates, and updates each week of social posts for you. The finished system gives you a simple one-page calendar you can review in Manus, while the generated posts and assets are saved back into Google Drive. #### Who This Is Useful For - Marketers and creators who want a real planning system instead of another spreadsheet full of half-finished ideas. - Founders and operators who need social posts, captions, and assets organized somewhere a team can actually use them. - Manus users who want to turn a good one-off workflow into a reusable skill and weekly automation. #### What You Will Build ![Manus weekly content calendar preview](https://tru-images.b-cdn.net/guide-assets/local-232/plan-your-weekly-content-calendar-in-minutes-with-manus/232-NL-thumb.png) You will build a simple one-page HTML content calendar in Manus. The calendar will: - plan one week of content at a time - show each generated post inside the calendar - include titles, captions, channels, and asset notes - save the generated posts and assets back into Google Drive #### What You Need To Get Started - A Manus account - The Manus desktop app (recommended) - A Google Drive folder for the project - A few brand inputs, such as a brand statement, campaign brief, example posts, product notes, or past newsletter copy > Pro tip: If you use Instagram, connect it too. Manus can use it for brand context and, depending on your setup, help with posting. #### Going Further Once the skill works, ask Manus to turn it into a weekly automation. Use this: ```text Create a weekly automation that runs this skill. Every week, generate the next seven days of social posts from the brand documents in Google Drive. Update the HTML content calendar and save the generated posts and assets back into Google Drive. Before scheduling it, show me: - when it will run - which Google Drive folder it will use - what it will update in the calendar - where the generated posts and assets will be saved ``` That is the bigger pattern: Google Drive holds the brand context and finished assets, Manus runs the weekly planning process, and the HTML calendar gives you a visual dashboard for what is coming next. ### Step 1 Set Up the Project Folder Start by creating a Google Drive folder for this content calendar. Add any brand documents Manus should learn from. You do not need a giant strategy archive. A small input pack is better: - brand voice notes - campaign brief - content pillars - example posts - product notes - landing page copy - recent newsletter copy The goal is to give Manus enough context to sound like your brand and plan around a real campaign. ![Google Drive input docs](https://tru-images.b-cdn.net/guide-assets/local-232/plan-your-weekly-content-calendar-in-minutes-with-manus/01-drive-input-docs.png) If you already have clean brand docs, use those. If you do not, add the best materials you have and let Manus tell you what is missing. ### Step 2 Connect Google Drive to Manus Open Manus and connect Google Drive from the Connectors page. Choose Google Drive, authorize your account, and then create a new Manus project for this calendar. ![Google Drive connector in Manus](https://tru-images.b-cdn.net/guide-assets/local-232/plan-your-weekly-content-calendar-in-minutes-with-manus/02-google-drive-connector.png) Once Drive is connected, Manus can read the files in your project folder and save generated posts or assets back into Drive. That is the part that makes this workflow useful. The calendar lives in Manus, but the actual work files live somewhere your team can access. > Pro tip: If you connect multiple services, keep the task focused. Toggle on the connectors Manus actually needs for this project so it does not wander into unrelated context. ### Step 3 Build the One-Page Calendar Now ask Manus to build the simple calendar interface. Use this prompt: ```text Build a simple one-page HTML content calendar that I can use as a visual planning dashboard for one week of content at a time. Use the brand documents in my Google Drive folder to generate post titles, captions, and needed assets for [Instagram, LinkedIn, X, email]. Show each post and asset in the calendar page, and save the generated posts and assets back into Google Drive. ``` Be specific about the channels you actually use. If you only post on LinkedIn and Instagram, remove X and email from the prompt. > Pro tip: Say simple one-page HTML content calendar, not full app or SaaS app. In the demo, Manus started going bigger and tried to turn the workflow into a full app with more moving parts. That looked impressive, but it also created more debugging. For this guide, stay simple. You want a calendar you can open, review, and refresh. You do not want to spend the rest of the afternoon programming. ### Step 4 Generate the First Week of Posts Once the page works, ask Manus to generate the first week of posts based on the brand assets in Google Drive. Use this prompt: ```text Generate the first week of posts for this content calendar. Base the ideas on the brand documents and assets in my Google Drive folder. For each post, include: - channel - post title - caption - CTA - asset needed or generated - status Show every post in the HTML calendar page, and save the generated posts and assets back into Google Drive. ``` After Manus finishes, check two places: ![Generated calendar posts and assets](https://tru-images.b-cdn.net/guide-assets/local-232/plan-your-weekly-content-calendar-in-minutes-with-manus/03-calendar-posts-and-assets.png) - Open the calendar page and make sure the posts are visible there. - Open Google Drive and make sure the generated posts and assets were saved there too. That second check matters. A calendar preview is nice, but the real payoff is having the actual files in Drive where you can use them, review them, or share them with a team. ### Step 5 Turn the Process Into a Skill When the first week looks right, turn the working process into one Manus skill. Use this: ```text Turn this planning and generation process into one reusable Manus skill. The skill should: - read the brand documents from this Google Drive folder - generate one week of social posts - update the one-page HTML content calendar - save the generated posts and assets back into Google Drive - keep the workflow lightweight and avoid turning this into a full SaaS app ``` This is the point where the system becomes more than a one-off prompt. Next time, you should be able to run the skill instead of rebuilding the calendar from scratch. Keep the skill focused on the repeatable part: plan one week, update the calendar, save the outputs. Do not make it repeat one-time setup steps like connecting Google Drive or creating the project. ### Guide: Quickly Iterate on Sora Videos With This Simple, Three-Step Automation > In this guide, you will learn how to generate Sora videos in bulk with a simple Make. - Canonical page: https://app.therundown.ai/guides/quickly-iterate-on-sora-videos-with-this-simple-three-step-automation - Machine-readable page: https://app.therundown.ai/guides/quickly-iterate-on-sora-videos-with-this-simple-three-step-automation.md - Difficulty: beginner - Featured tool or company: OpenAI - Categories: Content & Creative, Marketing & Growth, Automation & Agents - Instructors: Billy Howell - Published: 2025-12-16 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to generate Sora videos in bulk with a simple Make.com automation. You will produce a workflow that takes prompts from a Google Sheet, generates videos with Sora, uploads them to Google Drive, and writes the Drive link back to the sheet. Once the automation runs, the sheet should update with the video link within a few minutes. #### Who This Is Useful For - People who want to save time generating and tweaking video ideas - Marketers who want to generate ads with Sora - Content creators generating B-roll #### What You Will Build You will build a reusable Make.com scenario connected to a Google Sheet. Each row can contain a Sora prompt, and the automation will generate the video, save it to Google Drive, and update the sheet with the video’s Drive link. - A Google Sheet for Sora video prompts - A Make.com scenario with a Google Sheets trigger - An OpenAI Sora module connected with an API key - A Google Drive upload step for generated videos - A Google Sheets update step that returns the Drive web view link ![01_what_you_will_build](https://tru-images.b-cdn.net/guide-assets/69/6941d9de5ccb42ed4c429c8c/01-what-you-will-build.png) #### What You Need - Make.com account; a free trial works - OpenAI account - Billing enabled - Verified OpenAI organization #### Going Further - After the basic automation is working, you can make it more responsive by using a Google Sheets webhook and the Make.com add-on so row updates are sent to Make.com automatically instead of waiting for the 15-minute trigger interval. ### Step 1: Duplicate the Google Sheet - Duplicate the Google Sheet template. ![06_example_sheet](https://tru-images.b-cdn.net/guide-assets/69/6941d9df5ccb42ed4c429ca1/06-example-sheet.png) ### Step 2: Set up the Google Sheets trigger - Sign in to Make.com and create a new scenario. - Add the Google Sheets Watch Rows trigger. - Connect your Google Sheets credentials. - Select the sheet you created in Step 1. ![02_google_sheets_trigger](https://tru-images.b-cdn.net/guide-assets/69/6941d9e05ccb42ed4c429ca9/02-google-sheets-trigger.png) ### Step 3: Add the OpenAI Sora module - Add an OpenAI Sora module. - Create an OpenAI API key. - Verify your organization if you have not already. - Add the key to the module in Make.com. ![03_sora_module](https://tru-images.b-cdn.net/guide-assets/69/6941d9e05ccb42ed4c429cb6/03-sora-module.png) ### Step 4: Upload generated videos to Google Drive - Create a new folder in Google Drive for your videos. - Add a Google Drive Upload a File module. - Select the folder you created. - Click Save. ![04_google_drive_module](https://tru-images.b-cdn.net/guide-assets/69/6941d9e05ccb42ed4c429cbe/04-google-drive-module.png) ### Step 5: Write the Google Drive link back to the sheet - Add a Google Sheets Update a Row module. - Select your sheet. - Set it to update the Drive link column with the “Web View link.” ![05_update_row1](https://tru-images.b-cdn.net/guide-assets/69/6941d9e05ccb42ed4c429cce/05-update-row1.png) ![05_update_row2](https://tru-images.b-cdn.net/guide-assets/69/6941d9e05ccb42ed4c429cc6/05-update-row2.png) ### Step 6: Run the automation and confirm the output Write a prompt in your Google Sheet and make sure your row has an ID number. Wait for the time trigger to run, or click the “run once” button in the Make.com scenario. Within a few minutes, you should see the Google Sheet update with a link to the video in your Google Drive. ![07_soratestvideo_in_drive](https://tru-images.b-cdn.net/guide-assets/69/6941e5975ccb42ed4c42a289/07-soratestvideo-in-drive.png) ### Step 7 Optional: Trigger the scenario with a Google Sheets webhook Instead of waiting every 15 minutes for the automation to fire, you can set up Google Sheets to fire a webhook straight to Make.com every time a row is updated. - Create a new scenario. - For the trigger, click Google Sheets → Watch Changes module. - Copy the webhook you created in the module. - In Google Sheets, go to Extensions → Add-ons → Get add-ons and search “make.com.” - Add the add-on, then open it by clicking Extensions → Make → Settings. - Paste in the webhook. - Now, each time you add a row, that row’s data will be sent to your Make.com scenario. ### Resources - [Quickly Iterate on Sora Videos With This Simple Three-Step Automation](https://tru-images.b-cdn.net/guide-assets/69/6941e5e105bfa5e91f1afafa/quickly-iterate-on-sora-videos-with-this-simple-three-step-automation.pdf) ### Guide: Redline Any Contract With Claude and Microsoft Word > In this guide, you will learn how to use Claude directly in Microsoft Word to redline legal documents before you sign or send them. - Canonical page: https://app.therundown.ai/guides/redline-any-contract-with-claude-and-microsoft-word - Machine-readable page: https://app.therundown.ai/guides/redline-any-contract-with-claude-and-microsoft-word.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Productivity & Knowledge Work, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-08-04 - Updated: 2026-08-04 ### Overview #### The Rundown In this guide, you will learn how to use Claude directly in Microsoft Word to redline legal documents before you sign or send them. #### Who This Is Useful For - Founders and operators who need a faster first pass on contracts, vendor agreements, or RFPs. - Marketers and project leads who need to understand an agreement before routing it to the right reviewer. - Anyone reviewing documents who wants clear, reviewable redlines without moving the file into another chat. #### What You Will Build ![Claude’s suggested contract edits shown as tracked redlines in Microsoft Word](https://tru-images.b-cdn.net/guide-assets/local-378/redline-any-contract-with-claude-and-microsoft-word/01-word-contract-tracked-redlines.png) You will build a first-pass contract review workflow inside Word. Claude will surface suggested edits as tracked changes, you or a teammate will make the final calls, and Doc Check will give the near-final document one more pass. #### What You Need To Get Started - A paid Claude account, such as Claude Pro. - A Microsoft 365 license with Microsoft Word. - A contract, agreement, RFP, or other document you are allowed to review. - The Claude for Microsoft 365 Marketplace listing (https://marketplace.microsoft.com/en-us/product/office/WA200010725?tab=Overview). This is a first-pass review workflow, not legal advice. Keep a qualified legal reviewer in the loop whenever the document or decision calls for one. #### Going Further After reviewing a few documents, open Claude chat’s three-dot menu and download the transcripts. Save a few real review transcripts in a Claude Project, then ask Claude to turn the common process into a reusable skill your teammates can use. ![Claude turns the review process into a reusable contract review skill](https://tru-images.b-cdn.net/guide-assets/local-378/redline-any-contract-with-claude-and-microsoft-word/05-claude-create-contract-review-skill.png) That gives your team a starting point that reflects how you actually review agreements. If your teammates are on the same Claude plan, share the skill with them. Otherwise, download the skill file and send it over so they can add it to Claude themselves. ### Step 1 Install Claude for Microsoft 365 Open the Claude for Microsoft 365 Marketplace listing (https://marketplace.microsoft.com/en-us/product/office/WA200010725?tab=Overview), click Get it now, and sign in. Microsoft will ask for permission to let Claude work in Word. Approve the add-in, then choose Open in Word. ![Microsoft Word asks you to allow the Claude for Microsoft 365 add-in](https://tru-images.b-cdn.net/guide-assets/local-378/redline-any-contract-with-claude-and-microsoft-word/02-claude-for-word-add-in-permission.png) Word will open the add-in and ask you to sign in with Claude. For most people, choose Log in. If your company uses a provider or has locked down add-ins, ask the Microsoft 365 admin who manages your Word setup to help with the connection. ![Claude for Word sign-in screen](https://tru-images.b-cdn.net/guide-assets/local-378/redline-any-contract-with-claude-and-microsoft-word/03-claude-for-word-log-in.png) Once you are connected, you should see Claude in the top-right of Word. You can work with a DOCX directly. If your document is a PDF, open it with Word first so you can review and edit the converted document there. ### Step 2 Ask Claude for a First-Pass Redline Open the contract you want to review in Word. Then ask Claude: ```text Do a first read of this contract and redline anything that needs attention or clarification. ``` Claude may ask who you represent or offer a neutral read. Give it the lens you want it to use, then choose Tracked changes before it starts applying edits. ![Claude asks which side of the agreement you represent before it starts the redline](https://tru-images.b-cdn.net/guide-assets/local-378/redline-any-contract-with-claude-and-microsoft-word/04-claude-redline-contract-party-selection.png) > Pro tip: Ask Claude to use tracked changes so you can review its edits just like you would a human’s work. Claude will return issue cards and suggested edits in the sidebar. Open the reasoning on a card when you want to see why it made the recommendation, then use Apply all only when you are comfortable with the suggestions. If Claude cannot redline a document with bookmarks, remove those bookmarks or have it return the recommended edits in the chat instead. That is a limitation worth checking before you rely on the workflow for an important document. ### Step 3 Review the Tracked Changes Now you can review Claude’s recommendations and accept or reject them. Open Word’s Review tab to move through the tracked changes one at a time. This is the step where you keep the judgment. You can revise a suggestion, tell Claude what you want changed instead, or send the document to a teammate so they can review the same tracked changes in context. Claude applying an edit in its sidebar does not automatically finalize it in Word. Use the Word review controls to accept or reject each redline before treating the document as ready. ### Step 4 Run Doc Check Before You Send It When you think you’re ready to sign or send a document, type / in Claude and run the Doc Check skill. Doc Check makes a final pass and may surface new items that need attention. Review those findings the same way you reviewed the earlier redlines. The goal is a cleaner document and better questions for the right reviewer—not a substitute for legal approval. ### Guide: Run an LLM on Your Laptop for Free With Ollama > This guide teaches you how to install Ollama and chat with a real AI model that runs entirely on your laptop. - Canonical page: https://app.therundown.ai/guides/run-an-llm-on-your-laptop-for-free-with-ollama - Machine-readable page: https://app.therundown.ai/guides/run-an-llm-on-your-laptop-for-free-with-ollama.md - Difficulty: beginner - Featured tool or company: Ollama - Categories: AI Foundations & Prompting - Instructors: Billy Howell - Published: 2026-04-16 - Updated: 2026-07-23 ### Overview #### The Rundown This guide teaches you how to install Ollama and chat with a real AI model that runs entirely on your laptop. You will produce a local AI chat setup for drafting, summarizing, and brainstorming, with no account, subscription, or cloud processing required. The source frames this as a lightweight setup you can get installed tonight. #### Who This Is Useful For - Consultants and agency owners who handle client information they do not want sent to a third-party server. - Marketers, writers, and operators who want to iterate on prompts without watching a usage meter. - Anyone curious about local AI who does not want to sign up for another subscription just to try it. #### What You Will Build You will build a working local AI chat setup using the Ollama desktop app. The model lives on your hard drive, runs on your CPU or GPU, and works offline once it has downloaded. ![CleanShot 2026-04-16 at 16.38.38@2x](https://tru-images.b-cdn.net/guide-assets/69/69e148d71f8716153f0348d0/cleanshot-2026-04-16-at-16-38-38-2x.png) #### What You Need - A reasonably modern laptop running Mac, Windows, or Linux. - At least 8 GB of RAM is ideal, though 4 GB works with a smaller model. - At least 3 to 8 GB of free disk space, depending on the model you pick. - No account or subscription is required. #### Going Further - Wire up a coding agent. Ollama can launch Claude Code, Codex, or OpenCode against the model you downloaded using ollama launch claude, ollama launch codex, or ollama launch opencode. - Give your model tools and web search. Ollama exposes a local API at http://localhost:11434 and supports tool calling, so you can connect it to web search, file readers, and other utilities. - Move it to a dedicated machine. An old Mac mini, a recent used M-series Mac mini, or a repurposed PC can become an always-on local AI box that other devices on your home network can reach. ### Step 1: Install Ollama Go to ollama.com/download and download the installer for your operating system. No account, sign-in, or setup is required first. - Mac: open the downloaded file and drag the Ollama icon into your Applications folder. - Windows: run the installer and click through the wizard. - Linux: open a terminal and run curl -fsSL https://ollama.com/install.sh | sh . Once it is installed, open Ollama from your Applications folder or Start menu. > Pro tip: Mac users who prefer Homebrew can run brew install ollama instead of using the .dmg installer. ### Step 2: Pick a Model That Fits Your Machine Click New Chat in the Ollama app, then click the model dropdown at the bottom of the chat window. The right model depends on how much RAM your machine has. The source tested gemma3:4b on a 16 GB MacBook and found it ran comfortably. ![CleanShot 2026-04-16 at 16.39.40@2x](https://tru-images.b-cdn.net/guide-assets/69/69e149171f8716153f034984/cleanshot-2026-04-16-at-16-39-40-2x.png) - 4 GB RAM: gemma3:1b, qwen3:0.6b, or tinyllama. These are tiny models: fast, but with limited reasoning, best for short rewrites and simple Q&A. - 8 GB RAM: gemma3:4b, llama3.2:3b, or phi4-mini. This is the everyday tier for chat, drafting, and summarizing. - 16 GB RAM: gemma3:12b, llama3.1:8b, or qwen3:8b. This is the sweet spot for most laptops and is noticeably smarter for real writing work. - 32 GB+ RAM or a discrete GPU: gemma3:27b, gpt-oss:20b, or qwen3:32b. These are heavyweight models and the closest you will get to cloud quality on a single machine. Apple Silicon Macs share memory between CPU and GPU, so a 16 GB M-series Mac generally runs models that a 16 GB PC struggles with. Pick a model and Ollama downloads it the first time you select it, then caches it after that. > Pro tip: Browse ollama.com/search to see every model Ollama supports, including specialized models for coding, vision, and tool use. ### Step 3: Start Chatting in the App Once the model finishes downloading, type a prompt and hit enter. You are now chatting with an AI model running entirely on your laptop, with no internet call, API key, or per-token cost. The first response is usually slower because the model has to load into memory. After that, replies stream in faster. > Pro tip: If you prefer the terminal, run ollama run gemma3:4b instead. It is the same model and the same chat, without the GUI. ### Step 4: Test That It Runs Locally Turn on airplane mode or unplug your ethernet cable, then send another prompt. It should still work, because the model is running locally instead of hitting a server. This is the point of running local AI: nothing is hitting a server, nothing is being stored anywhere, and you are not paying for each response. If you handle client NDAs, internal financials, or anything you would not paste into a hosted AI tool, this distinction matters. ### Pro tip > If you prefer the terminal, you can run ollama run gemma3:4b instead of using the desktop app. ### Guide: Run Better 1-on-1s With Employees Using AI > In this guide, you will learn how to run better 1-on-1s with employees or mentees using Claude with Cowork turned on. - Canonical page: https://app.therundown.ai/guides/run-better-1-on-1s-with-employees-using-ai - Machine-readable page: https://app.therundown.ai/guides/run-better-1-on-1s-with-employees-using-ai.md - Difficulty: beginner - Categories: Business Operations & Leadership, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-07-07 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to run better 1-on-1s with employees or mentees using Claude with Cowork turned on. You will set up a Claude project from past call transcripts, create your own 1-on-1 template and evaluation rubric, and use it to help the people you manage make progress more consistently. #### Who This Is Useful For - Managers who run recurring 1-on-1s and want a more consistent structure for coaching, follow-up, and progress tracking. - Founders, team leads, or operators who manage researchers, contractors, creators, or analysts and want better visibility into work quality over time. - Anyone mentoring multiple people who wants to turn scattered call notes into a reusable template, rubric, and tracker. #### What You Will Build ![A Claude dashboard showing team rubric scores over time.](https://tru-images.b-cdn.net/guide-assets/local-306/run-better-1-on-1s-with-employees-using-ai/01-team-performance-dashboard.png) You will build a Claude project that stores your 1-on-1 history, creates a reusable call template and rubric, and tracks progress over time. The finished workflow can also create a simple dashboard from your rubric CSV so you can see how the team is trending. #### What You Need To Get Started - Claude desktop app (https://claude.com/download) - Previous 1-on-1 transcripts, reading notes, or call summaries (completely optional) #### Going Further Once the tracker works, the next move is turning it into a live dashboard artifact. In the video guide, we have Claude create a CSV to track transcripts, rubrics, scores, and follow-ups, then use that CSV as the source of truth for a dashboard you can reopen as the team changes over time. Use this prompt after Claude creates the tracker: ```text Make an artifact that visualizes the team's performance over time. Base it on the one-on-one tracker CSV. Include: - each person's latest average rubric score - category-level trends across the team - open follow-up items - a short summary of each person's most recent performance - any missing data that makes the dashboard less reliable ``` The first dashboard may be simple if you only have one rubric per person. That is fine. The important part is that the dashboard is built from the CSV, so every new transcript and rubric update can feed the same progress view. The simplest upgrade is to have Claude tighten the rubric and call template as more transcripts come in. After every few 1-on-1s, ask: ```text Review the latest 1-on-1 transcripts and tell me whether one-on-one-template.md or evaluation-rubric.md should be updated. Only suggest changes that are supported by repeated patterns in the transcripts. Keep the template short enough to use during a live call. ``` You can also use the same project when you schedule the next call. If Claude has access to your calendar, ask it to draft the agenda from the tracker before the invite goes out: ```text Schedule my next 1-on-1 with [Employee]. Look at our most recent 1-on-1 transcript, completed rubric, and one-on-one tracker CSV. Create a short agenda for the calendar invite that includes: - the main follow-up from last time - one progress question - one blocker question - one commitment to revisit ``` The main idea is simple: your 1-on-1s should not disappear after the call ends. With a Claude project and Cowork turned on for the organizing work, every transcript can make the next conversation easier to run. Keep the system human. Use Claude to organize the history, draft better questions, and spot patterns. You still decide what matters, what is fair, and how to coach the person in front of you. ### Step 1 Create A New Claude Project Start by creating a new project in the Claude desktop app for this workflow. Claude and Cowork now live in the same app, so the project is the workspace and Cowork is the mode you turn on when you want Claude to work across the project files. Keep the project narrow. You can create one project for your whole team, or one project per employee or mentee. For a first pass, a team-level project is usually enough. Name it something obvious, such as: ```text 1-on-1 Coaching System ``` Inside that project, add the transcripts or notes from previous 1-on-1s. In the video guide, we use a physical project folder with a few Markdown transcripts from previous calls. Use simple filenames so Claude can tell what each file is: ```text 2026-07-01-maya-1on1.md 2026-07-02-devon-1on1.md 2026-07-03-rina-1on1.md ``` Each transcript should include the date, the person, and the rough topic of the conversation. If you do not have a transcript, use meeting notes instead. A short summary is better than nothing as long as it captures what was discussed, what was promised, and what needs follow-up. If you do not already have transcripts in a folder, use Claude's connector directory to pull them from your note-taking tool. In the video guide, we call out Granola, Fireflies, and Otter as examples of tools that can hold past call transcripts. ![Claude's connector directory can help you pull call transcripts from connected note-taking tools.](https://tru-images.b-cdn.net/guide-assets/local-306/run-better-1-on-1s-with-employees-using-ai/02-connect-transcript-sources.png) Before asking Claude to build the system, select Cowork in the message box and confirm it can see the files: ```text Before we start, confirm which transcript or note files you can read in this project. Only list the files you can actually access. Do not summarize them yet. ``` This quick check prevents a common failure mode: Claude writes a generic template because it cannot actually see the source material. ### Step 2 Create A Team Reference File Before creating the template and rubric, give Claude a compact reference file for the team. This makes the rest of the workflow less generic because Claude can see who is on the team, what they do, and what kind of calls you are running. Use this prompt: ```text Take a look at the call transcripts I have in this project. These are from my 1-on-1s with [team or employee type]. Use the transcripts to create a quick team.md file that describes: - who is on the team - what kind of work they do - what the 1-on-1s are usually about - any recurring themes Claude should remember for future calls Important: - Treat this as coaching support, not an HR performance review. - Ground the summary in the transcripts. - Do not infer private motives, health, personality traits, or personal circumstances. ``` In the video guide, Claude creates a team.md file for a fictional AI research team. That file becomes the project reference point for the next prompts. ![Claude created a team reference file from the previous 1-on-1 transcripts.](https://tru-images.b-cdn.net/guide-assets/local-306/run-better-1-on-1s-with-employees-using-ai/03-generate-team-file.png) If the team changes later, you can ask Claude to update team.md instead of rebuilding the whole project. ### Step 3 Ask Claude To Build The Template And Rubric Once Claude understands the team, ask it to create the two core files: a standard call template and a standard evaluation rubric. The template helps you run the live call. The rubric helps you review each call consistently afterward. Use this prompt: ```text Given what you know about the team and the information in these transcripts, create a 1-on-1 call template that I can use for monthly progress check-ins with my team. Then create an evaluation rubric that I can use to grade my team members after the call or during the call. Make both the rubric and the template custom to my team and how the transcripts go, but make sure they are two files that are standardized across my team. When you are done, tell me: - what files you reviewed - how you used those files to create the documents - which decisions came from the transcripts or team.md Ground every decision in the transcripts or the team.md file. ``` Replace the team description and cadence with your own context. For an AI research team, the template might include research focus, proof of work, demo readiness, and next artifacts. For a customer success team, it might include account risk, follow-up quality, escalation blockers, and renewal readiness. ![Claude generated a reusable 1-on-1 template and an evaluation rubric from the transcript patterns.](https://tru-images.b-cdn.net/guide-assets/local-306/run-better-1-on-1s-with-employees-using-ai/04-create-template-and-rubric.png) Review both files before you use them. A good result should feel specific to your team. It should mention patterns from the transcripts, not just generic management advice. If the output is too vague, use this recovery prompt: ```text This is too generic. Revise it so every major point includes: - the specific transcript it came from - the observed behavior - a short quote or paraphrased moment - why it matters for future 1-on-1s - one coaching question I can ask Remove any point that is not supported by the transcripts. ``` If Claude returns the template and rubric as chat text instead of saved files, ask it to create the files directly: ```text Create two saved project files from that output: 1. one-on-one-template.md 2. evaluation-rubric.md Keep both practical, short enough to reuse, and specific to the transcript patterns you saw. ``` ### Step 4 Use The Rubric After Your Next 1-on-1 Use the template to structure the next live call. Open it before the conversation and use it as a checklist for what to cover: progress, blockers, commitments, coaching questions, and next actions. After the call, you have two options. You can fill out the rubric yourself, or you can ask Claude to create a first pass from the transcript. Use this prompt: ```text Take a look at [Employee]'s transcript from [Date]. Create a new rubric for that call and fill it out objectively based on that transcript. Use evaluation-rubric.md as the structure. For each category: - give the score - cite the transcript evidence - say when a category was not discussed enough to score - avoid HR, compensation, promotion, or disciplinary recommendations ``` In the video guide, Claude fills out Maya's July 1 rubric and marks categories that were not evidenced by the transcript. That is the right behavior. It is better for Claude to say "not discussed" than to invent a score. ![Claude filled out a rubric from Maya's 1-on-1 transcript and flagged categories that were not evidenced.](https://tru-images.b-cdn.net/guide-assets/local-306/run-better-1-on-1s-with-employees-using-ai/05-filled-rubric-from-transcript.png) Your judgment should stay in the loop. Treat Claude's completed rubric as a draft you review, edit, and approve. ### Step 5 Create A Tracker Once Claude has at least one completed rubric, ask it to create a tracker. In the video guide, we ask for a CSV that tracks 1-on-1s, team member scores, and performance over time. Use this prompt: ```text Make a CSV to track 1-on-1s, team member scores, and performance over time. Include: - employee or mentee name - date of each 1-on-1 - rubric category scores - average score - short evidence notes - commitments made - follow-up items ``` That tracker becomes the memory layer for future check-ins. After each new 1-on-1, drop in the transcript and ask Claude to update the rubric and tracker. You can use the same project to prep future calls. Before your next 1-on-1, ask: ```text Look at my most recent 1-on-1 transcript and scoring rubric for [Employee]. Use them to create a short agenda for our next call. Include: - what to follow up on - what improved - what still needs evidence - questions I should ask - commitments to revisit ``` This is the loop: use the template, run the call, record or summarize the conversation, fill the rubric, update the tracker, and prep the next call from the same history. ### Guide: Run Your Own Coding Agent on Your Laptop (for Free) > This guide teaches you how to download a coding LLM to your laptop with Ollama and wire it into Claude Code or Codex with a single command. - Canonical page: https://app.therundown.ai/guides/run-your-own-coding-agent-on-your-laptop-for-free - Machine-readable page: https://app.therundown.ai/guides/run-your-own-coding-agent-on-your-laptop-for-free.md - Difficulty: advanced - Featured tool or company: Ollama - Categories: Coding & App Development, Automation & Agents - Instructors: Billy Howell - Published: 2026-04-17 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches you how to download a coding LLM to your laptop with Ollama and wire it into Claude Code or Codex with a single command. You will end up with a local coding agent that can handle simple tasks for free while keeping proprietary code on your machine. #### Who This Is Useful For - Solo developers and indie founders paying for Claude Code Max or Codex Pro on work that does not need frontier reasoning. - Students, hobbyists, and new coders who want a working agentic setup to practice on for free. - Anyone working on proprietary or NDA code who wants the whole loop to stay on their laptop. #### What You Will Build ![CleanShot 2026-04-17 at 11.13.50@2x-1](https://tru-images.b-cdn.net/guide-assets/69/69e2548b0abaf596b5d4bac8/cleanshot-2026-04-17-at-11-13-50-2x-1.png) You will build a local Claude Code, Codex, or OpenCode session pointed at a free Ollama model running on your own hardware: the same agent-style interface, zero per-token cost, and nothing leaving your machine. - A local Ollama coder model selected for your hardware. - A coding-agent session launched against that local model. - A larger context window configured for more practical agentic coding. - An optional OpenCode setup if Claude Code feels too heavy for the model. #### What You Need - Ollama installed. If you do not have it, follow the Ollama install guide first. - Claude Code signed in and working, or OpenCode from the installer in the OpenCode step. - A terminal and a project folder you actually want to work in. - 16 GB of RAM is ideal. 8 GB works with smaller models, and 32 GB+ is better. #### Going Further - Once the basics are working, treat local models as a practice environment, a privacy layer for sensitive projects, and a budget hedge for tasks that do not need frontier reasoning. A high-leverage next move is a hybrid setup: keep Claude Code or Codex as your main agent, but use your local model as a cheap subagent for boilerplate while the frontier model handles architecture and planning. - Remote-control your local session from your phone by running claude remote-control inside the session and scanning the QR code with the Claude mobile app. - Put the model on a dedicated machine, such as an old Mac mini or repurposed PC, and use it as an always-on local inference box over your network. ### Step 1 Check your hardware and ask an LLM what to run Before you pull anything, check what your machine can handle. - On Mac: Apple menu > About This Mac, then screenshot the specs panel. - On Windows: Settings > System > About, then screenshot the specs panel. Drop that screenshot into Claude, ChatGPT, or any LLM, and ask: ```text Which Ollama coder models can I realistically run on this machine with Claude Code or Codex? ``` It will read the RAM and chip from your screenshot and give you a short list. Use this rough guide for what fits where: - 8 GB RAM: 3B parameters or smaller, such as qwen3-coder:3b. - 12 GB RAM: 4–7B parameters, such as gemma4:e2b or qwen3-coder:7b. - 16 GB RAM: 7–12B parameters, such as qwen3-coder:7b or gemma4:e4b. - 32 GB+ RAM or GPU: 20B+ parameters, such as qwen3-coder:32b, gpt-oss:20b, or gemma4:26b. > Pro tip: Download the biggest version you can reasonably fit. Bigger means more reliable tool calls, which is what actually breaks small models inside Claude Code. ### Step 2 Pick a coder model that supports agentic tools Browse ollama.com/search?q=coder and open the page for a model your LLM recommended. On the model page, scroll to the Applications section and confirm it lists Claude Code, Codex, OpenCode, or OpenClaw. If it lists none of them, the model does not support the tool calls agentic coding requires, so skip it. The three models that held up best in testing were: - qwen3-coder: the purpose-built coder pick, with strong raw code generation at its size. - gemma4: explicit tool-use and thinking training, and more reliable multi-step tool chains. - gpt-oss: OpenAI’s open-weights MoE model with strong agentic support. > Pro tip: If you cannot choose between two candidates, pull both. Disk is cheap, and you can swap between them with the --model flag. Keep the one that holds up in your workflow. ### Step 3 Launch your coding agent with the local model From the model’s Ollama page, copy the launch command. It looks like: ollama launch claude --model gemma4:e4b ![CleanShot 2026-04-17 at 10.48.39@2x](https://tru-images.b-cdn.net/guide-assets/69/69e2547956e3009fffba9b36/cleanshot-2026-04-17-at-10-48-39-2x.png) Open a terminal in your actual project folder, paste the command, and hit Enter. Confirm the download when prompted, then wait for the weights to pull the first time. After that, Ollama drops you into Claude Code pointed at the local model instead of Anthropic’s API. Inside the session, type /model to confirm which model is wired in. Every response from here costs zero tokens. > Pro tip: Run ollama ps in a second terminal to see what is actually running. It shows the active model, RAM in use, and GPU utilization. 100% GPU means you are fully accelerated. Anything lower means part of the model is spilling to CPU, and responses will be slower. ### Step 4 Bump the context window before serious work This is the single most important setting in the setup. By default, Ollama allocates only 4K of context per model, which is too small for agentic coding. Claude Code can read one file, fill the buffer, and immediately start forgetting the rest of the conversation. Fix it once: - Open the Ollama app. - Go to Ollama menu > Settings. - Find the Context slider. - Bump it to 32K to start, or higher if your specs allow. > Pro tip: Ask your LLM what context size your specs can safely handle. Maxing it out can push the model past your GPU’s limits and crash things. Start at 32K, verify with ollama ps, then raise if there is headroom. ### Step 5 Try OpenCode if Claude Code feels too heavy Claude Code is a sophisticated harness with a lot of tools exposed at once. Small local models sometimes get confused by the volume of choices. OpenCode is a lighter-weight coding agent built for this case, and it uses the same ollama launch pattern. Install it on Mac with one line: curl -fsSL https://opencode.ai/install | bash Then launch it the same way: ollama launch opencode --model gemma4:e4b > Pro tip: Smaller models work better when you ask them to think less. Hit Shift+Tab inside Claude Code or OpenCode to toggle plan mode, where the agent writes its approach before touching files. Lower the reasoning effort in Settings if the agent keeps over-thinking simple tasks. ### Pro tip > Download the biggest model version you can reasonably fit. Bigger means more reliable tool calls, which is what tends to break small models inside Claude Code. ### Guide: Run Your Workday by Voice With ChatGPT (Chat, Work & Codex) > In this guide, you will learn how to use ChatGPT Voice to turn an approved request into a reviewed PDF, save it to shared Google Drive, and prepare a handoff for your team without touching your keyboard. - Canonical page: https://app.therundown.ai/guides/run-your-workday-by-voice-with-chatgpt - Machine-readable page: https://app.therundown.ai/guides/run-your-workday-by-voice-with-chatgpt.md - Difficulty: beginner - Featured tool or company: OpenAI - Categories: Productivity & Knowledge Work, Automation & Agents - Instructors: Billy Howell - Published: 2026-08-02 - Updated: 2026-08-02 ### Overview #### The Rundown In this guide, you will learn how to use ChatGPT Voice to turn an approved request into a reviewed PDF, save it to shared Google Drive, and prepare a handoff for your team without touching your keyboard. #### Who This Is Useful For - Founders and operators who need to turn an approved request into a finished internal deliverable. - Consultants and client-service teams who build repeatable reports and handoffs from shared context. - Anyone who wants to replace scattered files and app switching with a voice-first workday loop. #### What You Will Build You will build a private-to-shared reporting loop that produces top-guides-video-engagement-brief.md, converts the approved Markdown into a PDF, and prepares a reviewable handoff for a teammate. #### What You Need To Get Started - The ChatGPT desktop app with Voice. - Some approved data and a reporting need. - Google Drive for the finished file and Slack or Teams for the handoff. #### Going Further Once one report type works, turn the tested flow into a reusable skill. Keep the report-specific source, fields, review gate, destination, and messaging rule inside the skill, then run it with fresh data: ```text Turn this approved workflow into a reusable skill for [REPORT_TYPE]. On each run, read only [APPROVED_SOURCES] for the last [LOOKBACK_DAYS] days, create or update a Markdown report with [REQUIRED_FIELDS], pause for my approval, export the approved version to [SHARED_DRIVE_FOLDER], and prepare a draft handoff for [DELIVERY_DESTINATION]. Keep all external messages draft-only unless I explicitly approve the final send. ``` You can reuse the same loop for client updates, content reviews, operating reports, and other deliverables that need a source, a reviewer, a durable home, and a clear next decision. ### Step 1 Set Up a Private Project and Start Voice Open the ChatGPT desktop app and create a private project for this workflow. In the demo, the project is called University Demo. Keep the project as the working space for drafts, prompts, and review history. Start a new voice chat before you send your first message. This gives you a live conversation where you can interrupt, ask follow-ups, and change direction while work continues. You can also keep talking while ChatGPT works in the background and come back to check progress. ![ChatGPT voice project setup](https://tru-images.b-cdn.net/guide-assets/local-416/run-your-workday-by-voice-with-chatgpt/01-voice-chat-project.png) > Pro tip: Start a new voice chat before sending the first message. If you begin in another mode, the microphone works as dictation instead of a live voice conversation. ### Step 2 Connect Only the Sources and Destinations You Need Think of the workflow as three separate jobs: - Source: the approved folder, document, analytics view, or meeting note ChatGPT may read. - Storage: the shared Google Drive folder where the approved deliverable will live. - Handoff: the test Slack, Teams, or Gmail destination where a teammate will receive the link. Use the + control to add a missing connection, then check what each connection can actually do. A source that can be read is not automatically approved for writing files or sending messages. ![Connected ChatGPT sources](https://tru-images.b-cdn.net/guide-assets/local-416/run-your-workday-by-voice-with-chatgpt/02-connected-sources.png) For a safe first run, use fictional or demo data and a test destination. Tell ChatGPT the exact scope before it starts: ```text Work in the private project "University Demo." Read only these approved demo sources: - the connected Google Drive folder "Demo Guide Context" - the fictional meeting note "Weekly Content Review - demo notes" First, list the sources you can access and summarize the workday context in three bullets. If either source is missing, stop and ask me. Do not search other files, chats, accounts, or connected services. Do not create files, send messages, save to Drive, or change anything external yet. ``` Once the scope is correct, scaffold the project for repeatable work: ```text Scaffold this private project for a simple draft-to-review-to-final reporting flow. Keep working drafts in Markdown, make the review gate explicit, and document the approved sources, the shared Drive destination, and the draft-only messaging rule. Do not create files, save to Drive, or send messages until I approve the next step. ``` ### Step 3 Plan the Workday Before You Build the Deliverable Use Voice to ask for a short plan from the approved context. The goal is to make the next deliverable obvious before you ask ChatGPT to create anything. ```text Using only the approved demo sources, give me a concise plan for today. Return: 1. the three highest-priority actions in order 2. the one decision that needs human approval 3. the first private-project deliverable to make for [TEAMMATE_NAME] 4. the external action you will wait to take until I approve it Keep this to the demo context. Do not use traffic, newsletter conversion, revenue, customer, or other sensitive metrics. ``` If the plan points to a report, confirm its purpose, audience, source, required fields, and review gate in one sentence. That small pause keeps the tool from turning an ambiguous request into an external action. ### Step 4 Draft and Refine the Markdown Brief Now ask ChatGPT to create the working file in the private project. Markdown is a good intermediate format because you can scan the structure and ask for precise edits before creating a polished PDF. ```text Create `top-guides-video-engagement-brief.md` in this private project for [TEAMMATE_NAME]. The file must contain only a Markdown title and one table with exactly these columns: - Guide title - Video start rate - Video completion rate Use only the approved fictional demo values from the sources. If a value is missing, use `[demo value]` instead of guessing. Do not include traffic, newsletter conversion, revenue, audience, customer, or any other metric. Do not save to Google Drive or send a message yet. ``` The Markdown file is the automatic output of this request. You do not need to ask ChatGPT to create a separate “draft artifact” after it finishes. Read the file, then refine it by voice: ```text Read `top-guides-video-engagement-brief.md` and tighten it for a teammate who needs a fast review. Keep only the title and the exact three table columns: guide title, video start rate, and video completion rate. Preserve only approved fictional demo values, remove every other field or explanation, and show me the final Markdown for approval. Do not make external changes. ``` Approve the exact Markdown before you export it. If the source data is incomplete, keep [demo value] rather than letting the tool fill the gap with a guess. > Pro tip: Refine the Markdown before making the PDF. It is easier to catch extra fields or a wrong table structure in plain text than in a finished document. ### Step 5 Save the Approved PDF and Prepare the Handoff When the Markdown is correct, approve the external save explicitly. Ask ChatGPT to create one PDF and save an exact copy to the shared Drive folder: ```text The final brief is approved. Convert `top-guides-video-engagement-brief.md` to one PDF and save an exact copy to `[DEMO_SHARED_DRIVE_FOLDER]` in the connected shared Google Drive. Create no other files, overwrite nothing, and return the Drive link when the save is complete. Keep the private project as the working space and the Drive copy as the durable shared record. ``` Use the returned link to prepare the teammate handoff. Keep the message short and free of metric values: ```text Using only the approved demo meeting-note context, prepare one concise message to [TEAMMATE_NAME] for [DEMO_SLACK_CHANNEL] or [DEMO_GMAIL_THREAD]. Include: - the Google Drive link to the approved brief - that the brief contains the three agreed fields only - the next review choice this helps [TEAMMATE_NAME] make Do not include any metric values and do not contact anyone else. Show me the exact message for approval before it is sent. ``` Review the message in Slack, Teams, or Gmail and send it yourself. The durable pattern is simple: the private project holds the work in progress, Drive holds the approved record, and the message carries the record to the person who needs to act on it. > Pro tip: Make messaging draft-only by default. Review the populated message in the destination app and click Send yourself so a mistaken thread or extra detail does not go to the team. ### Guide: Save Hours On Meeting Prep With This Automated Google Gemini Setup > In this guide, you will learn how to use Google Workspace Studio to turn upcoming meetings into automatic prep time and a Gemini-written brief. - Canonical page: https://app.therundown.ai/guides/save-hours-on-meeting-prep-with-this-automated-google-gemini-setup - Machine-readable page: https://app.therundown.ai/guides/save-hours-on-meeting-prep-with-this-automated-google-gemini-setup.md - Difficulty: beginner - Featured tool or company: Google Gemini - Categories: Productivity & Knowledge Work, Automation & Agents, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-06-18 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Google Workspace Studio to turn upcoming meetings into automatic prep time and a Gemini-written brief. The key move is building the workflow inside Google so you get a real AI meeting-prep system without Claude, ChatGPT, or a separate automation tool. #### Who This Is Useful For - Anyone who keeps doing meeting prep five minutes before the call by searching email and skimming the calendar invite. - Operators, sales leads, and managers who want a repeatable pre-meeting workflow instead of ad hoc notes. - Teams already using Google Workspace that want a simple AI automation without standing up Zapier, Make, or custom code. #### What You Will Build You will build a Google Workspace Studio flow that starts before a meeting, asks Gemini to write a short prep brief, creates a Google Doc, and drops a prep block onto your calendar with the doc link inside it. ![Workspace Studio flow with meeting starter, Gemini, doc, and block time steps](https://tru-images.b-cdn.net/guide-assets/local-282/save-hours-on-meeting-prep-with-this-automated-google-gemini-setup/01-studio-flow-overview.png) The outcome here is not another chat response. It is protected time on your calendar and a brief waiting where you will actually use it. #### What You Need To Get Started - A Google Workspace account with access to Workspace Studio (https://studio.workspace.google.com/manage). - Google Calendar and Google Docs. - Gemini available inside Workspace. - One fake calendar event you can safely test against. If you can open studio.workspace.google.com/manage, you are far enough along to build this workflow. Availability depends on your Workspace plan and rollout. #### Going Further Once the base version works, the next move is not adding more AI. The next move is tightening the workflow for different meeting types. Good extensions: - Sales calls: emphasize account context, objections, next steps, and recent news. - Recurring team meetings: emphasize open loops, blockers, owners, and decisions needed. - Exec reviews: emphasize metrics, risks, unresolved questions, and proposed decisions. You can also add a Check if or Filter step later if you want to narrow the flow to specific meeting types instead of every event on your calendar. The reusable pattern is simple: trigger before the meeting, generate the brief, create the doc, and place the link inside the prep block. Once that works, you can adapt it to almost any meeting-heavy workflow inside Google Workspace. ### Step 1 Build The Base Flow In Workspace Studio Go to Workspace Studio (https://studio.workspace.google.com/manage) and create a new flow. Use Based on a meeting as the starter, then add these actions in order: - Ask Gemini - Create a doc - Block time ![Calendar prep block with linked Google Doc brief](https://tru-images.b-cdn.net/guide-assets/local-282/save-hours-on-meeting-prep-with-this-automated-google-gemini-setup/02-prep-block-with-brief-link.png) In the meeting starter, keep the setup simple: - Meeting: Every meeting - Time offset: 10 - Units: Minutes - Start: Before meeting That gives you a workflow that fires shortly before the meeting instead of at the exact start time, which is the whole point. If a brand-new calendar event does not show up right away during testing, wait a few minutes. Workspace Studio warns that calendar changes can take a little time to sync into the flow picker. ### Step 2 Write The Gemini Brief Prompt The Gemini step is where you decide whether this becomes useful or generic. Do not ask for a vague summary. Ask for a short brief that tells you what matters before the call. Use a prompt like this: ```text Search the web for any news that is relevant to my upcoming meeting, and summarize the results in a paragraph. Write a concise meeting prep brief with these sections: Context Summary, Open Loops, Questions To Ask, Suggested Agenda, and Source Notes. Use Workspace context when available, include a do-not-assume note if context is thin, and keep it short enough to read in under 10 minutes. Meeting title: [Meeting title variable] Meeting description: [Event description variable] ``` ![Gemini step configured with a meeting-prep prompt and meeting variables](https://tru-images.b-cdn.net/guide-assets/local-282/save-hours-on-meeting-prep-with-this-automated-google-gemini-setup/03-gemini-brief-prompt.png) This structure works because it forces Gemini to do a job, not just generate text. You want context, open loops, questions, and a suggested agenda. That gives you something you can actually use in the 10 minutes before the meeting starts. ### Step 3 Create The Doc And Add The Link To The Prep Block Next, configure the doc step so the output has somewhere to live. In Create a doc, use: - New doc name: Meeting Prep Brief - [Meeting title variable] - Content to add: [Content created by Gemini] Then configure Block time with the meeting variables instead of hard-coded values: - Title: [Meeting title variable] - Starts at: [Meeting start time variable] - Duration: 15 minutes - Repeat: Does not repeat - Event description: Meeting prep brief: [Link to doc] ![Block time step using the meeting start time and doc link variables](https://tru-images.b-cdn.net/guide-assets/local-282/save-hours-on-meeting-prep-with-this-automated-google-gemini-setup/04-block-time-variables.png) > Pro tip: Create the doc before the block-time step. That is what gives you the doc link variable you can paste into the calendar block description. This is the part most people miss. If the brief only lives in Docs, you still have to go hunt for it. Putting the link directly inside the prep block makes the workflow usable. ### Step 4 Test It With A Fake Meeting Before You Turn It On Before you enable the flow for real, create one fake meeting in Google Calendar at a convenient time. Then go back to Studio, click Test run, and select that meeting from the dropdown. That test matters because it shows you the full chain: - Gemini writes the brief. - Studio creates the Google Doc. - Studio adds the prep block before the meeting. - The prep block includes the brief link. Open the calendar event and confirm the link is there. Then open the doc itself and inspect the output. You want to see a short brief with the sections you asked for, not a wall of AI filler. ![Generated meeting prep brief with context summary, open loops, questions, agenda, and source notes](https://tru-images.b-cdn.net/guide-assets/local-282/save-hours-on-meeting-prep-with-this-automated-google-gemini-setup/05-generated-meeting-brief.png) If the brief is too generic, tighten the Gemini prompt and run the test again. Once the fake meeting test looks right, turn the flow on. ### Guide: Sell a High-Value AI Workflow Audit as a Consultant > In this guide, you will learn how to turn your AI knowledge into a focused consulting offer. - Canonical page: https://app.therundown.ai/guides/sell-a-high-value-ai-workflow-audit-as-a-consultant - Machine-readable page: https://app.therundown.ai/guides/sell-a-high-value-ai-workflow-audit-as-a-consultant.md - Difficulty: beginner - Categories: Sales, Consulting & Customer Success, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-07-21 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to turn your AI knowledge into a focused consulting offer. Using Claude, you will choose a narrow problem, collect client context, and create a one-page opportunity brief that can lead to an implementation project. #### Who This Is Useful For - The person friends or coworkers ask about AI who wants a simple way to test whether consulting could be a fit. - Creators, marketers, and operators who know one industry or repeated workflow well enough to run a useful audit. - Anyone curious about freelance consulting who wants to try a few free audits before deciding whether to sell one. #### What You Will Build You will build a repeatable AI consulting framework: a focused offer, a live assessment that scores prospects out of 100, a quick client audit, and a client-ready deck that recommends one high-return workflow to improve with AI. Run it free to test demand, or sell it as a fixed-price audit and scope implementation separately. ![Final assessment deck](https://tru-images.b-cdn.net/guide-assets/local-367/sell-a-high-value-ai-workflow-audit-as-a-consultant/01-final-assessment-deck.png) #### What You Need To Get Started - A Claude Pro subscription (https://claude.ai/upgrade) and the Claude desktop app (https://claude.ai/download). Pro is the minimum individual setup for the Project and Cowork workflow shown here. Claude Max works too. - One industry, audience, or recurring workflow you know reasonably well. #### Going Further Tell Claude to research the tools, templates, tutorials, and examples that apply to your niche before a prospect fills out the assessment. Keep the useful options in a local solution library, along with notes about when each one fits and what it needs. That lets you match a bottleneck to a starting point without rebuilding the research every time. ![Niche solution library](https://tru-images.b-cdn.net/guide-assets/local-367/sell-a-high-value-ai-workflow-audit-as-a-consultant/05-niche-solution-library.png) You can also build one reusable deck template, then replace the score, evidence, bottleneck, and recommendations for each new audit. For a more polished template, follow our Claude Design guide for decks (https://therundown.ai/guides/claude-design-branded-deck). ### Step 1 Find Three Places To Start Open a new Claude Project. Tell Claude to interview you about the industries, work, and AI tools you know, then have it suggest exactly three consulting angles. Treat the result as a starting point for an experiment, not a decision about your career. ```text I want to test whether a small freelance AI consulting offer could be viable for me. Interview me one question at a time about: - industries, teams, or communities I know - work I have done repeatedly - AI tools and workflows I can explain confidently - questions people already ask me - outcomes those people care about Then suggest exactly three consulting angles. For each, include: - the target audience - one repeated workflow or pain point - a simple assessment I could use - a small audit I could run for free - what I would need to learn before offering paid help Rank the angles by how close they are to my real experience and how easy they are to test. ``` Choose the angle that gives you the clearest sentence: “I help [this kind of person] improve [this repeated workflow] with AI.” The video explores AI implementation for video editors. AI social posting for local restaurants would be another focused example. ![Claude consulting angle interview](https://tru-images.b-cdn.net/guide-assets/local-367/sell-a-high-value-ai-workflow-audit-as-a-consultant/02-claude-consulting-angle-interview.png) > Pro tip: Start with work you already understand. You need enough context to run a helpful audit, not expertise in every AI tool. ### Step 2 Build a Live AI Assessment Tell Claude to turn your chosen angle into a live Artifact that a prospect can complete in a few minutes. The assessment gives them a useful 100-point score and gives you the context for a quick audit. Paste this prompt into Claude: ```text Create a live, shareable Claude Artifact called “[YOUR NICHE] AI Opportunity Assessment.” This is a friendly 3 to 5 minute quiz for prospective [CLIENT TYPE] clients. Collect: - business name and a required business website URL - the outcome they want - their current workflow and how often it happens - the people and tools involved - common delays, rework, or frustration - the result they want from AI When they finish, give them a score out of 100. Explain the score in plain English, identify the biggest opportunities, and create a short summary I can use for a quick audit. Publish the assessment as a live Artifact. Save completed responses in the Artifact dashboard. If Google Drive is connected, save a copy to a dedicated submissions folder there. ``` Test the Artifact with a made-up response before you share it. Check that the questions lead to useful answers, the score is clear, and the summary gives you enough material to do the next step. ![AI assessment artifact](https://tru-images.b-cdn.net/guide-assets/local-367/sell-a-high-value-ai-workflow-audit-as-a-consultant/03-ai-assessment-artifact.png) > Pro tip: Make the business URL required. It saves time later and gives Claude a clear public starting point for research. ### Step 3 Find the Highest-Return Bottleneck When someone completes the assessment, keep their response in the Artifact dashboard or Google Drive. Tell Claude to compare the answers with the business website and find one workflow worth improving first. Paste the assessment response and this prompt into the same Claude Project: ```text Act as my research assistant for a quick AI workflow audit. Use the public information at [BUSINESS WEBSITE] and the assessment response below. Identify exactly one high-return bottleneck or workflow that this business could improve with AI. Return: - the public signals and assessment answers that support the recommendation - the bottleneck in plain English - why it is the best place to start - what still needs to be confirmed with the client - one practical, human-reviewed solution direction Keep this focused on one workflow. Treat it as a hypothesis, not a diagnosis. Assessment response: [PASTE RESULTS] ``` Look for repeated, structured work with a clear owner and a simple outcome. For example, “turn unused client footage into a repeatable social-post workflow” is more useful than “use AI everywhere.” ![Assessment score and results](https://tru-images.b-cdn.net/guide-assets/local-367/sell-a-high-value-ai-workflow-audit-as-a-consultant/04-assessment-score-and-results.png) ### Step 4 Turn the Audit Into a Client-Ready Deck The deck helps you see whether your audit gives someone a clear, useful recommendation. You can walk through it on a call, send it after a review, or use it to decide whether this niche is one you want to keep testing. Tell Claude to turn the assessment, website research, and chosen bottleneck into a short deck: ```text Create a client-ready AI Workflow Audit deck for [BUSINESS NAME] using the assessment and research below. Include: - their assessment score and a short executive summary - the one workflow to improve first - the evidence behind that recommendation - practical AI tools, templates, or workflow approaches to consider - what needs to be confirmed before implementation - a simple next step, such as a small pilot, a follow-up review, or an implementation project Keep the deck practical, visual, and specific to this business. Do not promise ROI or present the solution as ready to deploy without validation. ``` Review the deck before you send it. Make the recommendation easy to understand, verify the resource links, and be honest about what is still unknown. Run your first few audits for free if you want market feedback. If people find the process useful and you want to keep doing it, you can later test a simple fixed-price audit, such as $500. ### Guide: Send a One-Page Proposal Minutes After Every Sales Call > In this guide, you will learn how to turn a Granola sales call note into a one-page proposal and a ready-to-review Gmail draft with Zapier. - Canonical page: https://app.therundown.ai/guides/send-a-one-page-proposal-minutes-after-every-sales-call - Machine-readable page: https://app.therundown.ai/guides/send-a-one-page-proposal-minutes-after-every-sales-call.md - Difficulty: beginner - Categories: AI - Instructors: Billy Howell - Published: 2026-09-14 - Updated: 2026-09-14 ### Overview #### The Rundown In this guide, you will learn how to turn a Granola sales call note into a one-page proposal and a ready-to-review Gmail draft with Zapier. You will build a Zap that pulls the call details, fills a branded template, and drafts the follow-up so a proposal is ready minutes after the call. #### Who This Is Useful For - Sales reps and account executives who lose momentum rewriting the same call notes into a proposal by hand instead of following up while the deal is hot. - Consultants and freelancers who quote scope and price differently for every client and want a consistent starting document instead of a blank page. - Anyone who takes AI meeting notes and wants those notes to trigger real follow-up work instead of just sitting in a folder. #### What You Will Build You will build a Zap that watches a Granola folder for finished sales-call notes, pulls the prospect's details out with AI, logs them to a sheet, fills a branded proposal template, exports it as a PDF, and drafts the follow-up email in Gmail, without sending anything until you check it. ![Gmail draft addressed to the prospect with the proposal PDF attached, unsent](https://tru-images.b-cdn.net/guide-assets/local-125/send-a-one-page-proposal-minutes-after-every-sales-call/01-finished-proposal-gmail-draft.png) #### What You Need To Get Started - Granola (https://www.granola.ai/) (or another AI notetaker Zapier supports, like Google Meet notes or Fireflies) sharing notes to a dedicated folder - A Zapier account on a plan that supports multi-step Zaps and AI by Zapier. The free plan caps monthly tasks, so check your limit before you build this - A proposal template in Google Docs, with double-bracket fields like {{prospect_name}} and {{budget_range}} where the AI-extracted data should land - Connected Google Sheets, Google Docs, Google Drive, and Gmail accounts inside Zapier #### Going Further - Add a validation step right after extraction that checks whether the note actually looks like a sales call (real prospect name, real needs, a discussed budget) before the Zap builds anything. That one gap is what let a non-sales meeting stall the workflow instead of failing cleanly. - Swap AI by Zapier for Claude, ChatGPT, or Gemini in the extraction step if you want steadier structured output, especially around budget and next steps. - Zapier has its own MCP server. Once this Zap is built, you can hand an agent like Claude Code or Codex the job of adjusting steps and field mappings for you. That's useful if you'd rather spend your own AI usage on higher-value work and let Copilot's tokens handle the Zap maintenance. ### Step 1: Set Up the Trigger and the Template Start in Granola. Pick (or create) a folder you can drop meeting notes and transcripts into after a call ends. In Step 2, you'll point Zapier's **When a note is shared to Zapier** trigger at this same folder, so sharing a note there becomes the intentional "go" signal for the whole workflow. Nothing fires until you decide a call is worth following up on. ![Granola app showing an empty "Sales Call (demo)" folder labeled "Where sales call recordings go"](https://tru-images.b-cdn.net/guide-assets/local-125/send-a-one-page-proposal-minutes-after-every-sales-call/02-granola-trigger-folder.png) Next, build a one-page proposal template in [your proposal template] tool of choice. Google Docs is what we used, since Zapier can generate a document straight from it. Add a double-bracket placeholder for every value you want filled in automatically: `{{prospect_name}}`, `{{company_name}}`, `{{key_needs}}`, `{{budget_range}}`, `{{next_steps}}`, and `{{decision_timeline}}`. > Pro tip: Those double brackets aren't decoration: that's how automation platforms like Zapier and Make.com know where to insert a variable. Anytime you want the Zap to drop a value into a document, wrap the field name in double brackets and the platform will find it. If you don't already have a proposal template, ask your AI assistant to draft one using the field names you plan to extract. If you do have one, just swap in the double-bracket placeholders where the old static text used to be. ### Step 2: Describe the Workflow to Zapier Copilot Open a new Zap and use Zapier's **Copilot** to build it conversationally instead of wiring up each step by hand. Tell it what you actually want, in plain language: ```text When I get off a sales call, I want the Granola notes to come here. Then AI will analyze it, pull out the important info, log it in Google Sheets, and use that to build a custom proposal based on an existing template. It'd be great if the template were a Google Doc, but it doesn't have to be. The end goal is a PDF emailed to the prospect from my email. ``` Copilot turned that into a six-step chain: **Granola** (When a note is shared to Zapier) → **AI by Zapier** (extract the fields) → **Google Sheets** (log a row) → **Google Docs** (create a document from the template) → **Google Drive** (export the doc as a PDF) → **Gmail** (create a draft with the PDF attached). ![Zapier Copilot chat summarizing the six-step Zap (Granola trigger, AI by Zapier extraction, Google Sheets log, Google Docs template fill, Google Drive PDF export, Gmail draft) next to the matching canvas](https://tru-images.b-cdn.net/guide-assets/local-125/send-a-one-page-proposal-minutes-after-every-sales-call/03-zap-workflow-overview.png) Copilot then asked follow-up questions to wire the details together: what the spreadsheet was called, and how to name each generated file. We told it: ```text Google Sheets: Match the AI-extracted fields to the fields they should go into, and put a date and run time into that first column. Google Docs: Do the same thing with the property matching, and also generate a document name based on what we extracted. It should follow this format: [Client Name] - [Company Proposal] [Clean Date]. ``` The first time you run this, Zapier will ask you to authenticate every connected app: Granola, Google Sheets, Google Docs, Google Drive, and Gmail. You only have to do this once. If Zapier finds a Google Doc or sheet you already created for this workflow, it can connect to it automatically; otherwise, you'll pick your Drive and the right file manually in each step. > Pro tip: If you're wary of connecting your real Gmail account, you don't have to grant full access. During the Gmail authentication step, you can toggle off the permission to send or read email and only allow the scopes needed to create drafts and look up contact addresses. ### Step 3: Check the Extraction Fields and Fix What's Missing Open the **AI by Zapier** step and look at the prompt Copilot wrote for the extraction: ```text Extract key information from this sales call to populate a proposal. Return the following fields: 1. prospect_name: Name of the person/decision maker 2. company_name: Company or organization 3. key_needs: Top 1-3 pain points or needs identified 4. budget_range: Budget mentioned (if any), or note if not discussed 5. next_steps: Agreed-upon next actions 6. decision_timeline: When they need a decision/proposal 7. call_summary: 2-3 sentence summary of the call Provide clear, concise, proposal-ready language. ``` ![Zapier AI extraction panel listing seven fixed output fields: prospect_name, company_name, key_needs, budget_range, next_steps, decision_timeline, call_summary](https://tru-images.b-cdn.net/guide-assets/local-125/send-a-one-page-proposal-minutes-after-every-sales-call/04-extract-info-prompt.png) This is the part worth getting right. Pulling every field explicitly, instead of asking the AI to "summarize the call," is what made the rest of the Zap reliable. Feed it the note, the call transcript, and the attendee list as input, so it has everything it needs to fill each field. Check the output against your own call before you trust it. We caught two gaps in testing: - The first pass didn't pull the attendee's email address, which you need if the Zap is going to email the proposal to anyone. Add it as its own field. - call_summary is useful for your log, but it doesn't belong on a client-facing proposal. Either drop that field from the document template, or make sure it only maps into the Google Sheets row and not the Google Doc. **What You Need To Get Started** already covers the accounts; from here, it's about verifying the fields, not the connections. ### Step 4: Test the Zap, Catch the Real Failure, and Review Before Sending Run the test built into each step. Zapier will step through Granola, extraction, the sheet row, the document, the PDF export, and the Gmail draft, and show you the output of each one as it goes. ![Google Docs "Sales Call Proposal" template filled with call summary, key needs, budget range, decision timeline, next steps, and a client e-signature block](https://tru-images.b-cdn.net/guide-assets/local-125/send-a-one-page-proposal-minutes-after-every-sales-call/05-proposal-template-filled.png) Watch the extraction output closely. This is where things go wrong first. In one test run, the AI returned a real budget figure but tacked its own editorial judgment onto it in parentheses, and that aside carried straight into the generated proposal. A real client would see the AI's opinion about their budget sitting right next to the number. > Pro tip: If the extraction step keeps producing judgment calls instead of clean data, swap AI by Zapier out for Claude, ChatGPT, or Gemini in that step. Their structured-output modules tend to hold to a schema more consistently, which matters most for a field like budget that should only ever contain a number or "not discussed." We also triggered the Zap with a note from a meeting that wasn't a sales call. It fired anyway, the extraction came back incomplete, and the Zap stalled instead of producing a usable proposal. A workflow like this needs a way to recognize when it's been handed the wrong kind of meeting. See Going Further below. However you generate the fields, keep the last step set to **Create Draft**, never **Send Email**. Before you send anything yourself, check: - Recipient: is the email address actually the prospect's, not a placeholder or a teammate's? - Scope and needs: does key_needs reflect what they actually said, not a guess? - Timeline: is decision_timeline real, or did the AI invent a date? - Price: is budget_range a clean number you'd put in front of a client, with no editorial aside attached? That review is the whole point of drafting instead of sending. The Zap gets you from call to reviewable proposal in minutes; you're still the one who decides it's ready to go. ### Guide: Set Up a Coding Factory With the New Codex App > This guide shows you how to use the Codex app as a “coding factory” for building a website brick by brick. - Canonical page: https://app.therundown.ai/guides/set-up-a-coding-factory-with-the-new-codex-app - Machine-readable page: https://app.therundown.ai/guides/set-up-a-coding-factory-with-the-new-codex-app.md - Difficulty: advanced - Featured tool or company: OpenAI - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-02-04 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to use the Codex app as a “coding factory” for building a website brick by brick. Instead of relying on one long vibe-coding session until you run out of tokens, you’ll plan a modular workflow where multiple Codex agents can work on separate parts of a React app in parallel. You’ll produce a reusable Codex workflow for building a professional SaaS-style website with parallel agents, feature branches, commits, previews, and automations that can keep working day by day. #### Who This Is Useful For - Entrepreneurs who need to ship a landing page and blog without hiring a developer - Semi-technical managers who are tired of wrestling with the command line and Git - Codex users ready to move from single prompts to automated assembly lines #### What You Will Build You will build a team of Codex agents that can work on a React SaaS website in parallel, with a PRD guiding separate agents across shared infrastructure, pages, reviews, and automations. - A new React app scaffolded with Vite - A modular PRD and execution plan for a professional boilerplate SaaS website - Parallel agents for pages such as landing, blog, about, contact, pricing, and privacy - A main “foreman” agent that can review and merge feature branches - Optional automations for feature building, commit review, and bug hunting #### What You Need - A Mac computer - ChatGPT Plus ($20/month) or higher - The Codex app - A terminal for scaffolding the initial React app - Optional: GitHub login through gh auth login if you want Codex to publish code to GitHub #### Going Further - After the core factory is running, you can reuse the PRD with Codex automations to schedule daily feature builders, commit reviewers, and bug hunters. You can also install additional Codex Skills from the sidebar to add useful tools in one click. ### Step 1: Create a New React App Start by scaffolding a new React app. Run this in your terminal to create a new React app with Vite: $ npm create vite@latest vite-react-app -- --template react > 💡 Pro tip: If you don’t use Vite, you can use Next.js, Create React App, or another React setup instead. If you’re not sure what to use, ask AI to help you choose. The Codex app runs in a sandbox by default, so creating the project in your terminal first keeps setup simple and helps keep your Codex sidebar tidy later. ### Step 2: Set Up Codex Get the Codex app, open it, and sign in. Click Open Folder, or use Cmd + O / Ctrl + O, and select the folder you just created. ![image-2030](https://tru-images.b-cdn.net/guide-assets/69/6983b63221ebfa1f40307c62/image-2030.png) Click “Create git repository” in the bottom right. This helps your agents keep track of what they are working on. > 💡 Pro tip: You can log in with GitHub by opening the terminal with Cmd + J / Ctrl + J and running gh auth login. This lets Codex publish code to GitHub for you. ### Step 3: Create the First Plan Have Codex make a plan for the app. Click the plus button and toggle on plan mode. ![image-2031](https://tru-images.b-cdn.net/guide-assets/69/6983b7443927ebd59b5d4bcd/image-2031.png) ```text Create a modular PRD (in markdown) and execution plan for a professional boilerplate SaaS website. The project must be built using a "Parallel Factory" workflow where multiple agents can work on separate pages simultaneously without file conflicts. Phase 0: The Designer Agent (Global Infrastructure). Then parallel agents should create components and pages for a landing page, blog, about page, contact us page, pricing page and privacy page. The only agent that can edit the main App.jsx is you. ``` > Pro tip: You can see your PRD and other files by clicking Open > Open in, then choosing your IDE. ### Step 4: Run Phase 0 Create a new thread and tell the agent to get started on Phase 0 of the app. You can tag the PRD file with the @ sign. ![image-2032](https://tru-images.b-cdn.net/guide-assets/69/6983b7753927ebd59b5d53f3/image-2032.png) > 💡 Pro tip: Click the commit button in the top right to save an agent’s progress. Commits act like checkpoints that help other agents understand what is going on, and Codex can auto-generate a commit description for you. ### Step 5: Start the Factory Start agents for each page. Open a thread for each page mentioned in the PRD and tell each agent to get to work. > 💡 Pro tip: Instruct your agents to open new feature branches for each page. When the page agents are done, have your main “foreman” agent review each feature branch and merge it into your codebase. ![image-2033](https://tru-images.b-cdn.net/guide-assets/69/6983b79021ebfa1f40308594/image-2033.png) You can launch a preview of your website from Codex. Click the play button in the top menu and enter: npm install npm run dev Then preview your app by pressing the play button. ### Step 6: Add Automations Click Automations to have agents review your work every hour, day, or on another schedule. You can schedule daily commit reviews, bug hunters, and feature builders. ![image-2034](https://tru-images.b-cdn.net/guide-assets/69/6983b7b821ebfa1f40308dd1/image-2034.png) Instead of vibe coding in a multi-hour sprint and running out of tokens, try dropping in your PRD and turning on automations like: - Daily feature builder agent that builds the next thing on the PRD - Commit review agent that checks the feature agent’s code each day - Bug hunter that looks for bugs each day and fixes them ### Step 7: Install Useful Skills Click Skills in the sidebar to install useful tools for Codex in one click. ![image-2035](https://tru-images.b-cdn.net/guide-assets/69/6983b7d021ebfa1f40309609/image-2035.png) ### Pro tip > Think modular and plan for parallel tasks so multiple Codex agents can work without file conflicts. ### Guide: Set Up Useful AI Teammates With New ChatGPT Workspace Agents > Learn how to set up a useful AI teammate in ChatGPT’s Workspace Agents tool. - Canonical page: https://app.therundown.ai/guides/set-up-useful-ai-teammates-with-new-chatgpt-workspace-agents - Machine-readable page: https://app.therundown.ai/guides/set-up-useful-ai-teammates-with-new-chatgpt-workspace-agents.md - Difficulty: intermediate - Featured tool or company: OpenAI - Categories: Automation & Agents, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-04-27 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to set up a useful AI teammate in ChatGPT’s Workspace Agents tool. You will produce one daily agent that owns a recurring morning workflow, such as reviewing a Notion database of published guides and generating new guide ideas, instead of relying on a prompt you have to remember to run yourself. #### Who This Is Useful For - Operators and founders who do the same review, research, or reporting task every morning and want it off their plate - Anyone managing a Notion database, lead pipeline, or shared folder who wants one agent to keep that system moving - Teams already using ChatGPT Business who want something more durable than one-off prompts #### What You Will Build ![CleanShot 2026-04-27 at 17.19.13@2x](https://tru-images.b-cdn.net/guide-assets/69/69efd2dcab745bc881a280bc/cleanshot-2026-04-27-at-17-19-13-2x.png) You will build a daily agent that owns one recurring workflow and runs it for you each morning. In the demo, the agent reviewed a Notion database of published guides, looked for useful patterns, and generated three new guide ideas every day. - a clear role and workflow - connected apps and memory - reusable starter actions - a daily 9AM schedule - one place to run, review, and refine the work over time #### What You Need - Access to ChatGPT Workspace Agents; OpenAI says workspace agents are in research preview for Business, Enterprise, Edu, and Teachers plans - One system you want the agent to own, like a Notion database, lead pipeline, or Google Docs folder - Any integrations that workflow depends on, like Notion, Gmail, or Google Calendar #### Going Further - If you do not want to start from scratch, use Browse Templates on chatgpt.com/agents. OpenAI has business-friendly templates for areas like software review, finance close, sales, data analysis, and support, and each template shows which apps it expects and what kind of work it is meant to handle. - copy a template that is close to your use case - swap in your own integrations - narrow it to one daily job - let it run every morning ### Step 1 Open the Agents Workspace and Start a New Agent Go to chatgpt.com/agents. From there, click Create agent. If you are on the wrong plan, ChatGPT will prompt you to sign up for a Business plan first. Once you are in, you will land on the new-agent screen with a single input box that asks what the agent should do. ### Step 2 Give the Agent One Clear System to Own Before you write anything, pick the system or process this agent should manage. - a lead pipeline - a Google Docs folder - a Notion database - a daily report process ```text Create an agent that manages [lead pipeline, Google Docs folder, Notion database, etc.]. It should do 3 main tasks that can run autonomously on schedules to save me time. ``` In our test, we used an AI workflow researcher that owned a Notion database and looked for patterns in our published guides. If you have a direct link to the database or tool you want it to manage, include that in the first setup message too. That gives the agent a cleaner starting point. > Pro tip: Point this at something annoying and sticky to maintain. A Notion database, Google Sheet, lead pipeline, task tracker, or reporting system is a much better fit than a vague brainstorming workflow. ### Step 3 Connect the Right Apps and Let the Agent Scaffold Itself After the first prompt, ChatGPT will walk you through the setup and ask for the tools it needs. ![02-connect-required-apps](https://tru-images.b-cdn.net/guide-assets/69/69efd2fa1e137a13aa468c7b/02-connect-required-apps.png) ChatGPT will fill in the rest of the details: role, workflow, apps, memory, and how the agent should behave on each run. ![04-agent-builder-thinking-through-setup](https://tru-images.b-cdn.net/guide-assets/69/69efd33b411882554f2418a7/04-agent-builder-thinking-through-setup.png) > Pro tip: Keep approval on for anything sensitive. OpenAI says workspace agents can ask for permission before steps like editing a sheet, sending an email, or creating a calendar event. That is the right default. ### Step 4 Add the Morning Schedule and Test It Once Once the agent is shaped correctly, use the Schedule control in the upper right. ![11-agent-home-with-completed-run](https://tru-images.b-cdn.net/guide-assets/69/69efd3549f9b9d9aac051122/11-agent-home-with-completed-run.png) From there, add a recurring run. In our test, we set it to run every day at 9AM. - a clear role and workflow - connected apps and memory - reusable starter actions - a daily 9AM schedule - one place to run, review, and refine the work over time > Pro tip: When you click on any of your agents, you should see three pre-made task types for the agent to do in one click. ### Pro tip > If you do not have ChatGPT Business yet, you can still use the same prompt and workflow design in another agent tool. The bigger lesson is how to define ownership and recurring tasks clearly. ### Guide: Stand Out in an AI Job Interview With the “Proof Project” Method > In this guide, you will learn how to turn an AI workflow you know into a “proof project” for a job application or interview. - Canonical page: https://app.therundown.ai/guides/stand-out-in-an-ai-job-interview-with-the-proof-project-method - Machine-readable page: https://app.therundown.ai/guides/stand-out-in-an-ai-job-interview-with-the-proof-project-method.md - Difficulty: beginner - Categories: AI - Instructors: Billy Howell - Published: 2026-09-02 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to turn an AI workflow you know into a “proof project” for a job application or interview. You will record a walkthrough, use its transcript to build a five-slide deck, and tailor the finished kit to the role. #### Who This Is Useful For - Job seekers who want to show practical AI skills instead of simply listing them. - People applying for AI-specific roles who need a clear example of how they use AI at work. - Anyone with a useful AI workflow who wants a concrete project to share in an application or interview. #### What You Will Build You will create a short video walkthrough and a five-slide “proof project” deck. Together, they show how you solve a real problem with AI, where your judgment comes in, and why that work matters for the role. ![A finished proof project deck built around a weekly AI reporting workflow](https://from-ava.b-cdn.net/guide-assets/local-444/stand-out-in-an-ai-job-interview-with-the-proof-project-method/01-proof-project-finished-deck.png) You can attach the deck to an application, send it to a recruiter, or walk through it live during an interview. #### What You Need To Get Started • One AI workflow you know well • A job description for a role you want • Loom (https://www.loom.com/) or another screen recorder that gives you a shareable link • ChatGPT (https://chatgpt.com/) or Claude (https://claude.ai/) to create the presentation #### Going Further Re-record the walkthrough for important applications so it names the company and role. Save each version in one ChatGPT Work or Claude Cowork project, then reuse your strongest proof for similar opportunities. This does not need to become a giant project. Our walkthrough was two minutes, so personalizing it took almost no extra time. That small bit of effort makes it obvious that you built the proof for this opportunity. You can also save several workflow videos in the same project. When a new role comes up, paste in the job description and ask ChatGPT or Claude to choose the best example and build a tailored five-slide deck around it. Start with this prompt: ```text Set up a project where I can give you job descriptions and links to short videos of AI workflows I have built. For each role, help me choose the strongest example and create a five-slide deck that shows the problem, the workflow, my judgment, and why it matters for the job. ``` ### Step 1 Pick One Useful AI Workflow Start with a real job in mind, then pick one AI workflow you know well. Keep it simple. The person reviewing your application should understand the problem and the value within a few minutes. Good options include: - Turning source data into a recurring report - Turning research into a brief - Turning meeting notes into action items - Turning raw content into a first draft We used our weekly reporting workflow. Codex pulls data from PostHog and builds the first version of the report. We check the data, improve the presentation, and update the reusable system when something needs to get better. ![The reporting workflow moves from source data to an AI draft, human review, system improvement, and a team update](https://from-ava.b-cdn.net/guide-assets/local-444/stand-out-in-an-ai-job-interview-with-the-proof-project-method/02-reporting-workflow-learning-loop.png) That one example proves several things at once. It shows that we can connect data to AI, build a repeatable system, review the result, and improve it over time. It is much stronger than saying, “I know how to prompt ChatGPT.” ### Step 2 Record a Five-Minute Walkthrough Use Loom to record your screen and face. Walk through the workflow, but spend most of your time explaining how you think. Cover three things: - The problem the workflow solves - How the workflow works and how you maintain it - What the AI handles and what you handle The third point is the big one. Companies want to see more than an AI output. They want to know where you check its work, how you make decisions, and how you keep the workflow useful. Our reporting walkthrough takes about two minutes. It shows the data source, the generated report, the review loop, the updates to the reusable workflow, and the final summary we share with the team. ![A Loom walkthrough shows the reporting workflow and the human review loop](https://from-ava.b-cdn.net/guide-assets/local-444/stand-out-in-an-ai-job-interview-with-the-proof-project-method/03-loom-workflow-walkthrough.png) > Pro tip: Loom has some light editing and AI features, but the goal is to record this in one take without wasting time on edits. Watch the walkthrough once when you finish. Make sure the audio works, the workflow is easy to follow, and the link is set so the recruiter can open it. Then copy the Loom link and transcript. ### Step 3 Turn It Into a Five-Slide Deck Paste the Loom transcript, Loom link, and job description into ChatGPT or Claude. Tell it: ```text Create a five-slide deck with the video embedded on one slide. Use the other slides to explain my thought process and how the workflow operates. ``` ![ChatGPT uses the Loom transcript to build a five-slide proof project deck](https://from-ava.b-cdn.net/guide-assets/local-444/stand-out-in-an-ai-job-interview-with-the-proof-project-method/04-chatgpt-builds-five-slide-deck.png) A simple deck can follow this structure: - The problem and why it matters - The workflow from start to finish - The Loom walkthrough - What the AI handles versus what you handle - Why this proves you can succeed in the role The deck should make the workflow easy to understand before someone watches the full video. It should also make your judgment visible. In our example, one slide says what AI handles and what we handle. Another connects the workflow to the skills the job needs. ![The finished proof project connects the workflow, human judgment, and job relevance](https://from-ava.b-cdn.net/guide-assets/local-444/stand-out-in-an-ai-job-interview-with-the-proof-project-method/05-proof-project-role-relevance.png) Check every slide before you share it. We tested this with ChatGPT, and the first generated version had visible layout errors. Fix anything that is hard to read, then open the shared deck and test the Loom link yourself. Google Slides works well when you want to present the deck live. A PDF also works if the Loom thumbnail links out to Loom. You can even build the same idea in Notion by adding each slide as an image and embedding the Loom directly. ### Guide: Stress Test Any Business Idea With One Perplexity Prompt (Runs for Free) > Use one Perplexity Deep Research prompt to stress test any business idea and turn it into a presentation with market research, a SWOT, key risks, and a final verdict. - Canonical page: https://app.therundown.ai/guides/stress-test-any-business-idea-with-one-perplexity-prompt-runs-for-free - Machine-readable page: https://app.therundown.ai/guides/stress-test-any-business-idea-with-one-perplexity-prompt-runs-for-free.md - Difficulty: beginner - Featured tool or company: Perplexity - Categories: Data, Research & Analysis, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-04-06 - Updated: 2026-07-30 ### Overview #### The Rundown Use one Perplexity Deep Research prompt to stress test any business idea and turn it into a presentation with market research, a SWOT, key risks, and a final verdict. Once the prompt is built, you can reuse it on future ideas for faster go or no-go decisions. A run takes about 5 to 6 minutes. #### Who This Is Useful For - Founders with a backlog of half-tested ideas - Consultants or operators who need a fast first-pass viability check - Anyone who wants a sharper go or no-go decision before building #### What You Will Build ![image-2165](https://tru-images.b-cdn.net/guide-assets/69/69d435f3553433d4ca6aacdb/image-2165.png) You will create one reusable Deep Research prompt that turns a rough idea into a 10-slide validation presentation. - TAM research - Top competitors and pricing - SWOT matrix - V1 build cost - Sourcing and tooling stack - Founding team considerations - GTM wedge for the first 100 customers - Top 3 risks - 30/60/90-day milestones - Final green, yellow, or red verdict #### What You Need - A Perplexity account; free is fine, with 5 Deep Research queries per day - One business idea written as a short paragraph - A customer and rough price point - Somewhere to save the prompt when it works #### Going Further - Once the main prompt works, adapt it into reusable variants for different validation workflows. - A shorter version for a co-founder pitch - A side-by-side version for comparing two ideas - A 90-day MVP version for ideas that already look promising - A Perplexity Space that stores the prompt so you can reuse it without pasting the full thing every time - A scheduled run that forces you to chew through your idea list every week ### Step 1: Write the Idea Clearly Before you open Perplexity, write the idea in one paragraph. Keep it concrete: name the customer, the problem, the offer, and why it is possible now. ```text Idea: An AI-powered [thing] that gives [target customer] a [job to be done] for [price point]. It [key behavior 1], [key behavior 2], and [key behavior 3]. It serves [customer] who can't afford [expensive alternative] but want more than [cheap alternative]. It is possible now because [the unlock, often “long context AI agents” or “the cost of inference dropped”]. ``` The price point matters more than most people think. It forces the research to stay grounded. ### Step 2: Run One Deep Research Prompt Open Perplexity and switch to Deep Research. A free plan includes 5 Deep Research prompts per day. Paste in your idea and ask for the full validation deck in the same run. Use this prompt: ```text I am evaluating this business idea: [paste your one paragraph idea here]. Build me a 10 slide presentation that visually SWOTs it and tells me if it is worth pursuing. Cover: TAM, top competitors and pricing, a SWOT matrix, v1 cost to build, sourcing and tooling stack, founding team, GTM wedge for the first 100 customers, top 3 risks, 30/60/90 day milestones, and a one paragraph verdict with a green/yellow/red signal. Cite every claim and prioritize sources from the last 12 months. ``` ### Step 3: Let It Run and Review the Key Slides Hit run and leave it alone for 5 to 6 minutes. Perplexity handles the research, structures the answer, and generates the presentation. When it finishes, you may see more than one version of the deck. In the source test, Perplexity generated a rough low-res version and a cleaner animated one. Open the cleaner one first if both appear. Go straight to the verdict slide and the SWOT slide. Those two slides tell you whether the rest of the research is a roadmap, a warning, or a clean kill. If the output looks thin, rerun the exact same prompt with a sharper idea paragraph. Weak input is usually the problem. ### Step 4: Save It and Reuse It Once you get a good run, save the prompt somewhere you will actually find it again. Rerun it on every idea you have written down. Change the paragraph at the top, keep the structure the same, and you get a fast validation pass every time. If you end up using this often, create a dedicated Space for business ideas and keep the prompt there as the default instruction. Future runs become much lighter. ![image-2166](https://tru-images.b-cdn.net/guide-assets/69/69d4360104b4b500170fd47b/image-2166.png) You can tell Perplexity to use the saved prompt on a new idea instead of rebuilding the workflow from scratch. The point is to stop treating idea validation like a two-week project: save one good prompt, run it every week, and let the verdict slide tell you what deserves more of your time. ### Pro tip > Regular Perplexity searches won’t create docs in one shot. You can ask Deep Research to make decks or memos and it will create them. ### Guide: Submit Your iOS App to the App Store With ChatGPT Work > In this guide, you will learn how to take a SwiftUI app from Xcode to App Review with ChatGPT Work and App Store Connect. - Canonical page: https://app.therundown.ai/guides/turn-your-codex-ios-app-into-a-paid-app-and-ship-it-to-the-app-store - Machine-readable page: https://app.therundown.ai/guides/turn-your-codex-ios-app-into-a-paid-app-and-ship-it-to-the-app-store.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: AI - Instructors: Billy Howell - Published: 2026-08-30 - Updated: 2026-08-30 ### Overview #### The Rundown In this guide, you will learn how to take a SwiftUI app from Xcode to App Review with ChatGPT Work and App Store Connect. You will prepare the listing, upload the build, set a one-time price, and submit it without hunting through every screen. #### Who This Is Useful For - Solo builders who have a finished SwiftUI app and need a clear path through App Store Connect. - Small product teams that want Codex to handle repetitive listing and upload work while they review each decision. - Anyone submitting an iOS app for the first time who does not want to hunt through every App Store screen. #### What You Will Build You will move a finished SwiftUI app from an Xcode archive into App Store Connect, complete its product page, set a one-time price, and submit it to App Review. ![App Store Connect confirming one item submitted](https://from-ava.b-cdn.net/guide-assets/local-429/turn-your-codex-ios-app-into-a-paid-app-and-ship-it-to-the-app-store/01-app-submission-confirmed.png) #### What You Need To Get Started - A finished SwiftUI app packaged as an Xcode project - An approved Apple Developer Program (https://developer.apple.com/programs/) membership with App Store Connect access - A Mac with Xcode and Codex installed - Public privacy-policy and support URLs, plus an optional marketing URL - App Review contact details and accurate age-rating and content-rights answers #### Going Further Turn the workflow into a reusable release checklist for future versions. Keep the listing assets in one folder, increment the version and build numbers, create a fresh archive, upload it, and ask Codex to stop before the final App Review action. ```text Prepare the next App Store release for [APP NAME]. Review the current listing assets, update only the fields that changed, verify the version and build numbers, archive and upload the new build, and return a final submission checklist. Stop before Add for Review or Submit for Review. ``` Use the returned checklist to verify the listing, build, pricing, and review answers before you submit. ### Step 1 Start With a Working App Start with a working SwiftUI app. If you still need to build, test, and prepare yours in Xcode, follow Part 1 (https://app.therundown.ai/guides/build-test-publish-ios-app-without-leaving-codex) first. Next, gather the product page assets. Ask Codex to run the app in the simulator, capture the main workflow, and draft the copy and screenshots for App Store Connect. ```text Prepare everything I need for this app's App Store listing. Open the simulator and capture six to ten screenshots that show the main workflow with safe demo data. Draft the description, subtitle, and keywords. Use my existing privacy-policy and support URLs, plus [MARKETING_URL] if available. Stop before uploading or submitting anything. ``` We used this workflow for OurDays, a weekend-planning app. The finished asset set included simulator screenshots, a description, keywords, and the URLs needed for the listing. Review the screenshots at full size before uploading them. Remove private names, contact details, notifications, or real account data. ![Codex preparing App Store copy and simulator screenshots](https://from-ava.b-cdn.net/guide-assets/local-429/turn-your-codex-ios-app-into-a-paid-app-and-ship-it-to-the-app-store/02-codex-app-store-prep.png) ### Step 2 Archive and Upload the First Build The app needs an App Store Connect record before Xcode can upload it. In our test, Xcode offered to create the record from the app name, SKU, primary language, and bundle ID. Review those identity fields before continuing. Open the project in Codex and give it a direct upload instruction: ```text Upload this app to App Store Connect for the first time. Create the build, archive it in Xcode, and push the first build. Stop if any app identity, signing, version, or upload decision needs my approval. ``` Codex created the build and opened Xcode's Archives organizer. From there, choose Distribute App > App Store Connect > Next. Keep Xcode visible so you can approve system prompts and inspect the archive before distribution. Several names were already taken. Codex updated the app identity to OurDays Planner and completed the upload. If you change the identity during this step, recheck the app name, bundle ID, App Store record, version, and build number before trying again. ![Xcode preparing the archive for App Store Connect](https://from-ava.b-cdn.net/guide-assets/local-429/turn-your-codex-ios-app-into-a-paid-app-and-ship-it-to-the-app-store/03-xcode-distribute-app.png) ### Step 3 Fill the Listing in Codex's Browser Open the Browser panel in Codex and sign into App Store Connect. Select the app, open version 1.0, and ask Codex to use the assets from Step 1: ```text In App Store Connect, fill in the description and the other listing fields, then upload the screenshots and other assets we prepared. Ask before changing pricing, release behavior, review answers, or any private contact information. ``` Codex filled the description, keywords, support URL, and screenshots. Read every entry and correct anything that exceeds its character limit or does not match the app. Choose whether the app should release automatically after approval or wait for a manual release. We chose automatic release for OurDays. > Pro tip: Sign into App Store Connect in Codex's built-in browser before starting the form. General computer use wandered in our test; the signed-in browser filled the listing and uploaded screenshots much faster. ![The OurDays screenshots uploaded to the App Store listing](https://from-ava.b-cdn.net/guide-assets/local-429/turn-your-codex-ios-app-into-a-paid-app-and-ship-it-to-the-app-store/04-listing-screenshots-added.png) ### Step 4 Set the Price and Submit for Review Open Pricing and Availability, confirm the app's tax category, and add its price. We used a $2.99 U.S. base price for OurDays. If you charge for the app, make sure Apple's Paid Apps Agreement (https://developer.apple.com/help/app-store-connect/manage-app-pricing/set-a-price/) has been accepted before submission. Return to version 1.0 and finish the required review fields: - Add the private App Review contact details yourself. - Answer the age-rating questions based on the app's real content. - Confirm whether the app uses third-party content and that you have the necessary rights. - Select the uploaded build and confirm the release behavior. Click Add for Review. App Store Connect caught two missing sections in our test: Age Rating and Content Rights. We completed them for the scheduling app, returned to the submission, and clicked Submit for Review. The successful result is a submitted item waiting for Apple to review it. It is not approval or a live App Store listing. ### Guide: Teach Your AI To Edit Like You (Codex/Claude Code System) > Build a Codex or Claude Code file system that turns finished human edits into reusable writing rules. - Canonical page: https://app.therundown.ai/guides/teach-your-ai-to-edit-like-you-codex-claude-code-system - Machine-readable page: https://app.therundown.ai/guides/teach-your-ai-to-edit-like-you-codex-claude-code-system.md - Difficulty: intermediate - Featured tool or company: Codex / Claude Code - Categories: Content & Creative, Coding & App Development, Automation & Agents - Instructors: Billy Howell - Published: 2026-05-28 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to teach Codex or Claude your editing style by turning finished edits into reusable writing rules. The real move is the loop: draft, snapshot, edit, compare, improve. Instead of asking an AI to "remember my style," you will give it a small file system that makes your style visible. It drafts an email, freezes the first version, waits for your final edit, then compares the before and after to improve the rulebook for next time. #### Who This Is Useful For - Creators and marketers who draft repeatable email, newsletter, or sales copy - Operators who want AI to improve from real work instead of one-off prompt tweaks - Anyone building agent workflows who needs a simple example of skills, files, statuses, and automation working together #### What You Will Build You will set up a copywriting project structure that gives your AI a place to draft, save final edits, and learn from the difference. The folder keeps your source material, draft copy, finished copy, editorial rules, reusable skills, and automation instructions separate so the workflow does not turn into one messy prompt thread. ![Copywriting project folder structure](https://tru-images.b-cdn.net/guide-assets/local-packages/225-for-jason/01-copywriting-project-structure.png) By the end, you will have a small working system: one skill drafts copy from your rules, another compares the draft against your final edit, and an automation can periodically update the rules from new examples. #### What You Need To Get Started - Codex, Claude, or another agent that can read and edit project files - A project folder for the email system - A few opinions about your writing style, or examples of copy you like - One email assignment you can use as the first test draft #### Going Further This same pattern works outside marketing emails. Use it for scripts, newsletters, landing page copy, sales follow-ups, social posts, or any format where you repeatedly edit AI drafts. The reusable loop is: ```text 1. Keep a rules file. 2. Create an AI draft. 3. Freeze the first version. 4. Edit the final yourself. 5. Compare before and after. 6. Update the rules. 7. Automate the next review. ``` After a few weeks, look for friction. If you keep forgetting to update the status, make the status easier. If the folder structure feels too heavy, simplify it. If the rulebook gets noisy, ask the agent to deduplicate and merge similar rules. The goal is not to build a perfect editorial operations system on day one. The goal is to make every finished edit teach your AI how to write one percent closer to your style next time. ### Create the Project Folder Start with a simple folder structure. You do not need a content platform or a database. You need a place where the agent can separate drafts, approved finals, rules, and reusable skills. Use this structure: ```text email-copy-style-demo/ emails/ draft/ final/ rules/ editorial-rules.md skills/ ``` The draft/ folder is for work that is still changing. The final/ folder is for approved emails that can teach the system something. The rules/ folder is the shared memory. The skills/ folder is where the reusable instructions live. You can make this more detailed later, but keep the first version boring. The point is to prove the loop before you design the perfect folder system. ### Build the First Editorial Rules File Before you create a drafting skill, give the agent a starting style guide. The fastest way is to have it interview you. ```text Interview me about the editorial rules I use for marketing emails. Ask about: - audience - tone - subject lines - claims or phrases to avoid - CTA style - examples I like Turn my answers into rules/editorial-rules.md. ``` ![Editorial rules interview](https://tru-images.b-cdn.net/guide-assets/local-packages/225-for-jason/02-editorial-rules-interview.png) This is usually easier than writing the rules from scratch. Let the agent ask multiple-choice questions, answer quickly, and then clean up the result. If you already have examples of good copy, add them too. Ask the agent to extract patterns from those examples and merge them into rules/editorial-rules.md. Good starter rules are specific enough to change the next draft: - lead with the concrete workflow, not the tool category - use short sentences - avoid hype words like "revolutionary" or "game-changing" - include one specific use case before making a broad claim - keep CTAs direct and low-friction ### Draft One Email and Freeze the Snapshot Now create the first real test. Ask the agent to draft one email using the rules file, then save two copies: a working draft and a snapshot. ```text Draft a marketing email using rules/editorial-rules.md. Save the working copy in emails/draft/. Save an identical snapshot copy so we can compare it against my edited final later. ``` The snapshot is the important part. It should never change. It gives the agent a clean before-and-after comparison once you edit the email. If you want the project to stay organized, have each email live in its own folder: ```text emails/ draft/ product-launch-email/ prompt.md draft.md draft-snapshot.md final.md meta.md ``` The meta.md file can track the status: ```text Status: drafted Rules Updated: no ``` That status becomes useful later when the automation needs to know which emails are ready to process. ### Turn the Draft Flow Into a Skill Once the first draft-and-snapshot flow works, turn it into a reusable skill. This keeps you from rebuilding the same instructions every time you want a new email. Use a prompt like this: ```text Create a skill where I can prompt with an idea for an email, promotion, or product launch. The skill should: 1. Read rules/editorial-rules.md. 2. Create a new numbered folder for the email. 3. Save the assignment in prompt.md. 4. Draft the first version in draft.md. 5. Copy that exact first draft into draft-snapshot.md. 6. Create final.md for my edits. 7. Create meta.md with Status: drafted and Rules Updated: no. Never overwrite draft-snapshot.md after it is created. Never update editorial-rules.md from the first draft alone. ``` ![Create email skill prompt](https://tru-images.b-cdn.net/guide-assets/local-packages/225-for-jason/03-create-email-skill-prompt.png) The skill should leave the email in a drafted state. Do not let the agent mark its own first draft as final. The whole system depends on the difference between the AI's first version and your edited version. After the skill exists, test it with a small email assignment: ```text Use the email drafting skill to create a launch email for [PRODUCT_OR_WORKFLOW]. Audience: [AUDIENCE] Goal: [WHAT_THE_EMAIL_SHOULD_GET_READERS_TO_DO] CTA: [CTA] ``` ![Run email draft skill](https://tru-images.b-cdn.net/guide-assets/local-packages/225-for-jason/04-run-email-draft-skill.png) Check that the skill created both draft.md and draft-snapshot.md. They should match exactly at the start. ### Edit the Email and Update the Rules Now do the human part. Edit the working draft into the final approved version. Tighten the hook, remove weak claims, add concrete examples, and make the CTA sound like you. When the edit is done, ask the agent to compare the snapshot against the final: ```text Compare draft-snapshot.md against final.md. Identify the durable editorial lessons from my edits. Update rules/editorial-rules.md with only the rules that will help future emails. Do not add one-off campaign facts. Do not duplicate existing rules. Prefer fewer, sharper rules over a long list. ``` This is where the system starts learning. You are not describing your taste in the abstract. You are showing it the exact edits you made. A useful rule update might say: ```text - When introducing a product, describe the specific use case before making a broad benefit claim. - Avoid salesy reassurance phrases. Let the concrete example carry the promise. ``` A weak rule would say: ```text - Make the email better. - Mention Purple Punch. ``` The first set changes future drafting behavior. The second set either says nothing or only applies to one campaign. ### Automate the Improvement Loop Once the two skills work, automate the review. The automation should scan for emails that are approved, compare the first draft with the final version, update the rulebook, and then mark the email as processed. Use a prompt like this for the scheduled task: ```text Check the emails folder for the oldest draft that has been marked final, scheduled, or sent. Use the editorial-rules-update skill on it. Compare the original snapshot with the approved final. Update rules/editorial-rules.md with durable lessons only. Then update the email status so this draft does not run twice. ``` ![Scheduled editorial rules task](https://tru-images.b-cdn.net/guide-assets/local-packages/225-for-jason/05-scheduled-editorial-rules-task.png) Daily scanning works well if you publish or approve email often. If you want a slower review cycle, run the deeper improvement process weekly. The key is that the automation should only process finished work. Do not let it learn from drafts that you have not edited yet. > Pro tip: Keep the rule update conservative. You want a rulebook, not a junk drawer. If the agent finds ten possible lessons, approve the two or three that would actually improve the next draft. ### Pro tip > Keep the rule updates conservative. The goal is a small rulebook of durable lessons, not a long memory file full of one-off campaign facts. ### Guide: The Best and Worst of Google’s New Gemini in Gmail Features > This guide shows how to use the new Gemini features in Gmail, what is available right now, and which parts of the initial rollout are most useful or disappointing. - Canonical page: https://app.therundown.ai/guides/the-best-and-worst-of-googles-new-gemini-in-gmail-features - Machine-readable page: https://app.therundown.ai/guides/the-best-and-worst-of-googles-new-gemini-in-gmail-features.md - Difficulty: beginner - Featured tool or company: Gemini - Categories: Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-01-14 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows how to use the new Gemini features in Gmail, what is available right now, and which parts of the initial rollout are most useful or disappointing. You’ll learn how to enable smart features, ask Gemini questions about your inbox, test natural language search availability, and use Gemini to draft or refine email replies. #### Who This Is Useful For - Gmail users with lots of emails - Anyone who wants to ask their inbox questions - Google Workspace or AI Pro users #### What You Will Build You will set up a repeatable Gemini-in-Gmail workflow for checking your settings, asking inbox questions, drafting replies, refining email tone or length, and trying scheduling assistance from an email thread. - Enabled Gmail smart features - Example inbox questions you can reuse with Gemini - A draft-reply workflow using Help me write, Recreate, and Refine - An optional scheduling workflow using Help me schedule #### What You Need - Access to gmail.com - A Google Workspace plan for a work email account, or - A personal Google account with a Google AI Pro membership #### Going Further - Try the “Help me schedule” button in a draft pane so Gemini can read the email thread, check your calendar, and suggest meeting times. - Use @ mentions to reference specific Google Drive files in Gemini, remembering that this access is read-only. - If you are considering upgrading to AI Pro mainly for natural language search in Gmail, consider waiting until Gemini in Gmail is fully implemented. ### Step 1 Enable smart features in Gmail Go to gmail.com and click the settings gear in the top right. Make sure all “smart features” are enabled. ![01_settings](https://tru-images.b-cdn.net/guide-assets/69/6967d0743839d57ca2865865/01-settings.png) ### Step 2 Talk to your inbox Back in your inbox, click the “Ask Gemini” button to the right of the settings gear. You can ask Gemini questions about your inbox, such as: ```text What hotel did I stay at in [city] last summer? ``` ```text What emails do I have about [subject]? ``` You can also use the @ symbol to mention specific Google Drive files to Gemini, but this access is read-only. ### Step 3 Check natural language search availability Google’s demo shows that asking a question in the Gmail search bar can pull relevant emails and return an AI summary. Google says this feature rolled out to Google AI Pro users, but the source notes it had not been noticed as implemented yet. If you’re thinking about upgrading to AI Pro mainly for this, it may make sense to wait for full implementation of Gemini in Gmail. ### Step 4 Use Help me write In any draft email, press Option + H on Mac or Alt + H on Windows, or click the “Help me write” text. Then write a prompt, and Gemini will help draft a reply based on the context of the email thread. ![04_help_me_write](https://tru-images.b-cdn.net/guide-assets/69/6967d21c88b0317376ac4c10/04-help-me-write.png) After Gemini writes a reply, click “Recreate” to regenerate it or “Refine” to change the tone or length of the reply. ![04_help_me_write2](https://tru-images.b-cdn.net/guide-assets/69/6967d2dd3839d57ca28658f7/04-help-me-write2.png) The source found this feature to be fast and accurate. ### Step 5 Try Help me schedule Look for the “Help me schedule” button in your draft pane. Gemini can read the email thread for context, check your calendar, and suggest meeting times. ![going_further_help_schedule](https://tru-images.b-cdn.net/guide-assets/69/6967d2f1a52ff5fa840e7072/going-further-help-schedule.png) ### Pro tip > If you’re thinking about upgrading to AI Pro mainly for Gemini in Gmail, it may make sense to wait for full implementation. ### Guide: Transcribe Any Video for Free With This Local AI > Learn how to transcribe and translate video or audio files for free by running Whisper locally on your computer. - Canonical page: https://app.therundown.ai/guides/transcribe-any-video-for-free-with-this-local-ai - Machine-readable page: https://app.therundown.ai/guides/transcribe-any-video-for-free-with-this-local-ai.md - Difficulty: intermediate - Featured tool or company: OpenAI - Categories: Content & Creative, Coding & App Development - Instructors: Billy Howell - Published: 2026-03-02 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to transcribe and translate video or audio files for free by running Whisper locally on your computer. You will create a local terminal-based transcription setup that avoids uploading files to third-party transcription sites. After about 5 minutes of initial setup, a gigabyte-sized, seven-minute video can take about two minutes to transcribe on a MacBook Pro. #### Who This Is Useful For - Video or podcast editors - Privacy-conscious meeting recorders - Anyone with hours of content to transcribe or translate #### What You Will Build ![image-2095](https://tru-images.b-cdn.net/guide-assets/69/69a615a07108ae18776a0457/image-2095.png) You will build a local transcription setup that lets you transcribe any video or audio file with a single terminal command. You will also be able to translate non-English audio into English automatically. - A reusable local Whisper install - A one-command workflow for transcribing files - Transcript and caption outputs such as .txt and .srt files - An optional translation workflow for non-English audio #### What You Need - A Mac or Windows computer - Python 3 - Homebrew on Mac or Chocolatey on Windows for installing ffmpeg - A video or audio file you want to transcribe - About 5 minutes for the initial setup #### Going Further - After transcribing a file, you can reuse the outputs as plain text transcripts, closed captions, or editor-ready subtitle files. You can also control the output folder and format, or batch process a folder of recordings to create transcripts for every file. ### Step 1 Install ffmpeg and Whisper Open Terminal. On Mac, press Cmd + Space, type “Terminal,” and press Enter. First, install ffmpeg. This is the command-line video tool Whisper needs to read your files. Mac: brew install ffmpeg If you do not have Homebrew installed, ask Claude or ChatGPT how to set it up on your device. Windows: choco install ffmpeg Check that Python is installed. On Mac, type: python3 --version If you see a version number, you are good. If not, ask Claude or ChatGPT how to install Python 3 on your device. Install Whisper: pip3 install -U openai-whisper That is it. You should only need to run these install commands once. > Pro tip: If pip3 is not recognized, try python3 -m pip install -U openai-whisper instead. If you want to double-check the install commands for your operating system, the Whisper GitHub page has everything. ### Step 2 Transcribe your first file Grab a video file you want to transcribe. Right-click it in Finder and copy the file path. You will need that for your Whisper command. Run this in Terminal: python3 -m whisper "[your file path]" --model base For example, if your video is in Downloads: python3 -m whisper "/Users/you/Downloads/my-video.mp4" --model base Whisper will detect the language automatically, then pull the words and timestamps into your terminal. A gigabyte-sized, seven-minute video takes about two minutes to transcribe on a MacBook Pro. - .txt for a plain text transcript - .srt for closed captions, which you can upload to YouTube or import into your editor - .vtt, .tsv, and .json for other formats To control where the files go, add --output_dir and --output_format: python3 -m whisper "[your file]" --model base --output_dir "/Users/you/Downloads" --output_format txt This guide uses the base model. There are several model sizes, and the GitHub page shows how much RAM each needs. You can go up to turbo, but base is the sweet spot for speed and accuracy. > Pro tip: The .srt file has timestamps baked in and imports directly into Premiere, DaVinci Resolve, Final Cut, and most other editors. If you are using the transcript for captions, use the .srt instead of the .txt. ### Step 3 Translate non-English audio Whisper can translate most non-English audio. Run the same command as before on a video or audio file and add the --task translate flag to the end. python3 -m whisper "[your file path]" --model base --task translate Without the --task translate flag, Whisper transcribes in the original language. With it, you get an English translation. ### Step 4 Batch process a folder of recordings If you have a folder of recordings to process, you can batch the whole thing in one line. Open Terminal, cd into the folder, and run: for f in *.mp4; do python3 -m whisper "$f" --model base; done Every file in the folder gets transcribed. Walk away, then come back to a folder full of transcripts. ### Pro tip > If pip3 is not recognized, try python3 -m pip install -U openai-whisper instead. ### Guide: Translate Your Videos Into Any Language With AI (Full HeyGen Tutorial) > This guide shows you how to use HeyGen to translate an existing video into another language with lip-synced audio. - Canonical page: https://app.therundown.ai/guides/translate-your-videos-into-any-language-with-ai-full-heygen-tutorial - Machine-readable page: https://app.therundown.ai/guides/translate-your-videos-into-any-language-with-ai-full-heygen-tutorial.md - Difficulty: beginner - Featured tool or company: HeyGen - Categories: Content & Creative - Instructors: Billy Howell - Published: 2026-02-24 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to use HeyGen to translate an existing video into another language with lip-synced audio. You’ll produce a localized version that looks and sounds like it was recorded natively, and once submitted, the translation is typically ready in 5–10 minutes. #### Who This Is Useful For - Anyone running video ads to a global audience - Course creators with international demand - YouTube creators targeting non-English markets #### What You Will Build ![image-2076](https://tru-images.b-cdn.net/guide-assets/69/699e2d106dd25300cfe663a2/image-2076.png) You’ll create a translated version of one of your existing videos with lip-synced audio in a new language. You’ll also set up an automation concept for translating future videos hands-free. - A localized, lip-synced video in a new language - A repeatable workflow for choosing which languages to translate into first - An optional automation plan for future translations #### What You Need - A HeyGen account. The free plan gives you 3 translations up to 3 minutes each. - A talking-head video with one speaker and clear audio. - Access to your Meta or Google Ads dashboard to find your top markets, optional. - Google Analytics or YouTube Analytics if you do not have ads data, optional. #### Going Further - After creating your first translation, reuse localized videos in paid ads, YouTube uploads, landing pages, or course platforms. You can also connect HeyGen with Make.com, n8n, or the HeyGen API so new videos can be translated and saved back to Google Drive automatically. ### Step 1 Find your top non-English markets Go into your Meta Ads Manager or Google Ads dashboard and find your top three non-English-speaking countries. Write those down. Those are the languages you’ll translate into first. No ads? Check Google Analytics or YouTube Analytics for where your international traffic comes from. The point is to translate where you already have an audience, not guess. ### Step 2 Upload your video to HeyGen Sign up at heygen.com. The free plan gives you three credits for videos under three minutes. - Click Translate in the sidebar. - Click Upload, or paste a YouTube or Google Drive link. - Use a talking-head video with one speaker and minimal background noise. The cleaner the input, the better the output. ### Step 3 Select languages and translate - Choose the target languages from your top-market research. You can select multiple at once. - Keep the default settings. They work out of the box. - Hit Translate. ![image-2077](https://tru-images.b-cdn.net/guide-assets/69/699e2d2744d8080b535d20d8/image-2077.png) Your video will be ready in 5–10 minutes. > Pro tip: If the output quality looks low, that’s the free plan. The $29/month paid plan exports at original quality and gives you unlimited translations, which is worth it for ads or client-facing content. ### Step 4 Review and deploy your translated video Play back the translation and compare it to your original. The lip sync is the part that will blow your mind. HeyGen re-animates your mouth to match the translated words, and it preserves proper nouns, URLs, and numbers. Use your translated videos in places like: - Paid ads: Run the same creative in different markets with the localized version. - YouTube: Upload as separate videos targeting each language. Separate uploads perform better than YouTube’s built-in dubbing, which only supports audio, not lip sync. - Landing pages: Swap the hero video based on visitor location. - Course platforms: Upload translated versions as separate modules. ### Step 5 Automate future translations Make.com and n8n both have HeyGen nodes. Set up a two-step automation: - Workflow 1: Watch a Google Drive folder. When a new video lands, trigger three HeyGen translation modules, one per language. - Workflow 2: Watch for completed HeyGen translations. When ready, download and save them back to Google Drive. ![image-2078](https://tru-images.b-cdn.net/guide-assets/69/699e2d526dd25300cfe66cb3/image-2078.png) Drop a video in, get three translated versions back without touching HeyGen. > Pro tip: HeyGen also offers a pay-as-you-go API plan. You could have Claude Code or Codex set you up a translating app or script fairly easily. ### Pro tip > If the output quality looks low, that’s the free plan. The $29/month paid plan exports at original quality and gives you unlimited translations, which is worth it for ads or client-facing content. ### Guide: Turn a Family Scheduler Idea Into a Working App With Astra > In this guide, you will learn how to turn an idea from our Workflow Hub into your own app with Astra in Codex. - Canonical page: https://app.therundown.ai/guides/turn-a-family-scheduler-idea-into-a-working-app-with-astra - Machine-readable page: https://app.therundown.ai/guides/turn-a-family-scheduler-idea-into-a-working-app-with-astra.md - Difficulty: beginner - Categories: AI - Instructors: Billy Howell - Published: 2026-09-08 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to turn an idea from our Workflow Hub into your own app with Astra in Codex. We’ll go from a product brief to design mockups and a working family planner, then test whether its calendar saves and shares events the way we expect. #### Who This Is Useful For - Parents who want school events, practices, appointments, and meals in one family planner. - Household organizers who want to connect dinner plans, groceries, and everyday tasks. - Anyone who finds a useful AI project and wants to adapt its underlying idea to their own needs. #### What You Will Build You’ll create a product requirements document, compare four visual concepts, and build a working family-planner prototype. Our version, Kinfolk, brings a calendar, tasks, groceries, and dinners into one place. We started with Shawn’s shared eldercare log (https://app.therundown.ai/community/posts/a1bf8b99-d97e-42ca-8fdb-25286ecdbbee) and adapted the coordination idea for everyday family life. The same process works with a club organizer, a household chore board, or another app you wish existed. #### What You Need To Get Started - The desktop app with Codex mode and Astra available in its model selector. We used Astra on High on a Mac. - A link to an idea you want to adapt and a folder for the new project. - A short list of the people, plans, and activities the app should organize. - A decision about data storage and user access. Our prototype uses local PostgreSQL; its local setup requires Docker Desktop. Astra is OpenAI’s GPT-6 model (https://developers.openai.com/api/docs/models/gpt-6-astra), with support for coding and multi-step work. The desktop app supplies the project tools. Check the model selector in your own account before starting; availability can differ across accounts. #### Going Further Bring your version back to the Workflow Hub (https://app.therundown.ai/community). Share the problem you adapted, credit the original creator, and explain one thing you learned while building it. If you want to deploy the app, make that a separate next step: choose its hosting and database, decide how household access should work, and add proper authentication before storing real private information. The local prototype gives you a concrete starting point for those decisions. ### Step 1: Find an Idea and Give It Your Own Direction Browse the Workflow Hub (https://app.therundown.ai/community) for a project that solves a recognizable problem. Open the post, read why its creator built it, and look at the comments for questions you might want your version to answer. Shawn’s project brought family caregiving information into one place. We kept that coordination pattern and changed the focus to groceries, appointments, practices, recitals, and dinners. Give the original creator credit when you share your adaptation. Try the starting example before writing requirements. Our first demo looked promising, but navigating to a future Tuesday exposed a calendar limitation. That gave us a concrete requirement for the next version: a calendar we could fully navigate and use. Keep an existing demo in its own folder while you build the new version. We used a separate v2 app so we could compare results without overwriting the original. ### Step 2: Write the Product Brief Before Building Switch to Codex mode, select Astra, and enter /plan. Give it the inspiration link and explain the outcome you want. Here is a reusable version of the request we used: ```text Write a new PRD—a product requirements document—that remixes this idea for [your audience and use case]: [source link]. Bring [the activities and information you want to organize] into one place. Make it usable on desktop and mobile, with a full calendar people can navigate, add to, and edit. For this local prototype, use PostgreSQL and username-only member entry in one shared household. Treat username entry as profile selection, not secure authentication. Use fictional data. Ask me multiple-choice questions about the decisions that affect the result. Write the PRD first. After I review it, we’ll create four visual concepts and choose one before implementation. ``` A PRD is the written agreement about what you’re building. Review who uses the app, which screens it needs, what its calendar supports, and how data should persist. Answer the questions Astra raises instead of letting those decisions stay implicit. ![Astra asks how the full calendar should work while reviewing the source idea.](https://tru-images.b-cdn.net/guide-assets/local-451/turn-a-family-scheduler-idea-into-a-working-app-with-astra/01-astra-family-planner-requirements.png) > Pro tip: If Astra goes straight to work without clarifying the important choices, ask it to interview you with multiple-choice questions. We used those questions to choose how member entry and the full planner should work. We chose local PostgreSQL, which added setup work during the build. If you already know your intended backend, specify it in the brief. We discussed Supabase as an alternative, but the prototype here uses the local database. ### Step 3: Compare Four Designs and Pick One After reviewing the PRD, ask Astra for four distinct image mockups. Keep the features consistent across concepts and request a desktop screen beside a mobile screen so you can compare both layouts. ```text Use the approved PRD to create four distinct visual concepts. Show the desktop calendar and a mobile layout side by side in each image. Keep the functionality consistent so I can compare the visual directions. Wait for me to choose a concept before building the app. ``` We reviewed Hearth, Daylight, Patchwork, and Sunday, then chose Patchwork. Its large calendar, plum controls, pastel categories, and family-member labels gave the build a clear visual target. ![The selected Patchwork concept shows desktop and mobile calendar layouts.](https://tru-images.b-cdn.net/guide-assets/local-451/turn-a-family-scheduler-idea-into-a-working-app-with-astra/02-patchwork-desktop-mobile-concept.png) This is the generated design concept, not an interactive app screenshot. Look at the layout and hierarchy rather than treating every detail in the image as finished behavior. Calendar positions, dates, and buttons still need to work correctly in the implementation. If you use design skills, choose ones that fit your stack and the job. An unrelated installed skill can pull the project toward the wrong framework or visual style. The useful instruction is to build from the design you selected. ### Step 4: Build the Prototype and Test Its Behavior Give Astra the approved concept and ask it to complete a working prototype. Our request also asked it to save progress with Git, set a goal, and use sub-agents for independent work. ```text Build a working prototype from the approved PRD and Patchwork concept in the new app folder. Preserve the earlier demo. Get it running in your browser so I can use it. Use Git to track and save progress. Set a goal for completing the prototype, and use sub-agents for independent tasks that can run in parallel. Test the app and fix problems before handing it back. ``` Once the app opens, test the behavior that matters to you. We added a Piano recital event, saved it, logged out, refreshed, and returned with another username. The event was still there. Switching back to Billy also showed it. That test confirmed persistence and revealed an important access decision: everyone belonged to the same household. Seeing the same event was expected for that design. > Pro tip: Define whether users share one household or get separate accounts before building. Username-only entry selects a member; it does not verify identity or protect private records. Use fictional data for this prototype. Try a complete action, such as adding and reopening an event, before spending more time polishing the layout. Then check the task list, groceries, and dinner screens against your own requirements. Ask for a targeted correction if something behaves differently from what you intended. ### Guide: Turn a Two Hour Excel Slog Into a 10-Minute Task With Claude in Excel > In this guide, you’ll learn how to use Claude as a spreadsheet architect. - Canonical page: https://app.therundown.ai/guides/turn-a-two-hour-excel-slog-into-a-10-minute-task-with-claude-in-excel - Machine-readable page: https://app.therundown.ai/guides/turn-a-two-hour-excel-slog-into-a-10-minute-task-with-claude-in-excel.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-02-05 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you’ll learn how to use Claude as a spreadsheet architect. You’ll take multiple messy CSVs and have Claude clean the data, format tables, color-code tabs, and turn the workbook into a useful dashboard. The source example used 7 unlabeled CSVs of SEO data and produced a dashboard in under 10 minutes. #### Who This Is Useful For - Consultants with clients who love spreadsheets - Team leads who want to standardize reporting processes across their organization - The “Excel person” on the team who spends the end of each month cleaning up formulas and tables #### What You Will Build You’ll build a cleaned, formatted Excel workbook that combines several raw CSVs into a master dashboard and can be reused as a monthly reporting template. - Renamed and reformatted worksheet tabs - Cleaned tables for each data source - A master report sheet that ties the data together - Charts for dashboard components - A README tab explaining how to use the template ![image-2036](https://tru-images.b-cdn.net/guide-assets/69/698513cf8369e6363e84a59e/image-2036.png) #### What You Need - Microsoft Excel - Any paid Claude plan - 3–10 CSVs of raw data - Optional: sales data, expense tracking data, SEO data, or another dataset you report on each month #### Going Further - Use Claude for Chrome to log in to your CRM and export the source data for you. - Set this up as a recurring workflow to run at the end of each month so you can see trends over time. ### Step 1: Install Claude in Excel and gather your CSVs Install Claude in Excel from the Microsoft marketplace. Find 5–10 CSVs of data. You can use sales data, expense tracking, or anything else you report on each month. > 💡 Pro tip: Make sure you choose the same timeframe for your data sources. ### Step 2: Import your CSVs and ask Claude to clean the workbook Open a new Excel workbook. Import each CSV as a new tab using File > Import > CSV. Make sure you select “comma” as the delimiter. Click the Claude logo in the top right, then start with this prompt: ```text I imported [data type] data from [source 1] and [source 2] from my website. Can you rename all the tabs and reformat all the data in the sheet so that it’s more readable? ``` Asking Claude if it can do the task can produce better results because it encourages Claude to explore the data, consider its capabilities, and come up with a plan. You can also explicitly ask Claude to make a plan. > 💡 Pro tip: You can tell Claude to color-code the tabs based on data source and create tables in each tab. ### Step 3: Ask Claude to recommend and build a master report Prompting Claude to build the master dashboard is important. The best results came from first asking Claude to look at all the data and make a recommendation. ```text What do you recommend doing to tie all this data together into a quick master report sheet? ``` Claude can break the data sources into tables and write formulas to bring the data together. ![image-2037](https://tru-images.b-cdn.net/guide-assets/69/698513dfbb943aaec181e181/image-2037.png) ### Step 4: Add charts and turn the report into a reusable template Claude can also make charts from the data. Ask it to visualize each dashboard component. ![image-2038](https://tru-images.b-cdn.net/guide-assets/69/698513ed5611c243592fdcd0/image-2038.png) Once you’ve created a report you like, ask Claude to turn it into a reusable template: ```text Adjust the Dashboard formulas to make this work as a template I can use each month by importing the data into each tab. Create a README tab explaining how to use the template. ``` ### Pro tip > Choose the same timeframe for your data sources before asking Claude to combine them. ### Guide: Turn Abandoned Twitter Bookmarks Into Something Useful With AI > Set up a Perplexity research system that reads through saved articles and bookmarked content, scores each find by usefulness, and logs everything into a Google Sheet. - Canonical page: https://app.therundown.ai/guides/turn-abandoned-twitter-bookmarks-into-something-useful-with-ai - Machine-readable page: https://app.therundown.ai/guides/turn-abandoned-twitter-bookmarks-into-something-useful-with-ai.md - Difficulty: intermediate - Featured tool or company: Perplexity - Categories: Data, Research & Analysis, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-02-25 - Updated: 2026-07-30 ### Overview #### The Rundown Set up a Perplexity research system that reads through saved articles and bookmarked content, scores each find by usefulness, and logs everything into a Google Sheet. The example topic is OpenClaw, but the same setup works for any tool, trend, or topic you want to track. Once scheduled, the system can run daily and update your sheet automatically. #### Who This Is Useful For - Anyone with a backlog of saved articles they never get around to reading. - Consultants and freelancers juggling multiple tools or clients who need to stay sharp without constant scrolling. - Founders and operators keeping up with a fast-moving tool or space. #### What You Will Build A Perplexity research pipeline with three connected pieces: a Space with custom instructions, a daily scheduled Task, and a Google Sheet that becomes your master research list. - A Perplexity Space that scores and organizes research finds. - A scheduled Task that searches for new content automatically. - A Google Sheet with usefulness ratings, implementation time estimates, cost breakdowns, and recommendations for every find. ![image-2086](https://tru-images.b-cdn.net/guide-assets/69/699f82b101937ac6f2081ad4/image-2086.png) #### What You Need - Perplexity account: free for the Comet browser, Pro or Max for scheduled tasks. - A blank Google Sheet. Perplexity can set up the headers. - Comet browser, available as a free download. #### Going Further - Expand your sources by adding Gmail under the task’s model options so Perplexity can pull in developer newsletters and update emails. - Clone the setup for other tools by creating a new Space and sheet, then swapping “OpenClaw” for whatever you’re tracking. - Experiment with automating the bookmark scan. Scheduled Tasks start without browser access, but you can click into the thread afterward and grant access. The manual trigger is more reliable right now, but it’s worth testing. ### Step 1 Set Up Your Research Space Open Comet and go to Spaces > New Space. Name it something specific, like “OpenClaw Research” or whatever topic you’re tracking. In the custom instructions, tell Perplexity what its job is: ```text Your job is to read Twitter threads, articles, and GitHub updates about OpenClaw. Rank findings by usefulness, estimate implementation time, estimate cost, and give a recommendation on whether I should use each one. Log everything into my connected Google Sheet. ``` Then connect your Google Sheet: click Add files from cloud > Google Drive, log in with Google, and select your sheet. This gives Perplexity ongoing access to update it. ![image-2084](https://tru-images.b-cdn.net/guide-assets/69/699f828701937ac6f20809bb/image-2084.png) > Pro tip: Let Perplexity choose the column headers for your sheet. It’ll set up columns like Date, Source, Description, Usefulness Score, Implementation Time, Cost Estimate, and Recommendation, and it stays consistent with them going forward. ### Step 2 Create a Daily Scheduled Task Go to Scheduled Tasks > New Task. Set it to run daily. Use this prompt: ```text Find the most interesting new OpenClaw use cases, skills, plugins, and integrations from the last 24 hours. Check Twitter, GitHub, Reddit, and blogs. For each find, tell me: - What it does - How much traction it has - Whether it’s stable or experimental - Who would benefit most - Your recommendation on whether I should implement it Add the results to my Google Sheet. ``` ![image-2085](https://tru-images.b-cdn.net/guide-assets/69/699f829801937ac6f208123d/image-2085.png) Under Model Options > Sources, make sure Web and Social are both selected. You could also add Gmail if you want it pulling from developer newsletters. Each morning, Perplexity runs this automatically and drops the results into your sheet. > Pro tip: When you’re in a thread, toggle on Control Browser at the bottom. This lets Perplexity navigate to your Google Sheet and update it directly instead of just telling you what it found. ### Step 3 Use the Bookmark Hack This is the part that makes the system useful day to day. As you’re scrolling Twitter and spot interesting content, such as new skills, workflow ideas, or upgrade guides, bookmark it into a dedicated folder. Call it something like “OpenClaw Guides & Skills” or whatever fits your topic. ![image-2087](https://tru-images.b-cdn.net/guide-assets/69/699f82c501937ac6f2082346/image-2087.png) When you’re ready, open that bookmarks folder in Comet and tell Perplexity: ```text Check my Twitter bookmarks and add any bookmarks we don’t have cataloged into our Google Sheet. ``` Perplexity will go through your bookmarks, check them against what’s already in the sheet, and add anything new with scores and recommendations. This is better than fully automated research because you’re choosing the inputs. Instead of Perplexity dumping random articles on you every day, it processes the content you already flagged as interesting. The AI organizes and scores; you decide what’s worth looking at in the first place. > Pro tip: Add a “User Usefulness Score” column at the end of your sheet. After you read each find, rate it yourself on a scale of 1–5. This keeps you from blindly trusting every AI recommendation and trains your eye for what actually matters. ### Pro tip > Add a “User Usefulness Score” column at the end of your sheet so you can rate each find yourself and compare your judgment against the AI recommendation. ### Guide: Turn an Investment Memo Into a Polished Slide Deck in 5 Minutes With Manus > Learn how to use Manus to turn an investment memo PDF into a professional, ready-to-present slide deck in Google Slides. - Canonical page: https://app.therundown.ai/guides/turn-an-investment-memo-into-a-polished-slide-deck-with-manus-in-5-minutes - Machine-readable page: https://app.therundown.ai/guides/turn-an-investment-memo-into-a-polished-slide-deck-with-manus-in-5-minutes.md - Difficulty: beginner - Featured tool or company: Meta - Categories: Content & Creative, Productivity & Knowledge Work, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-03-06 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use Manus to turn an investment memo PDF into a professional, ready-to-present slide deck in Google Slides. You’ll produce a complete deck from a structured memo, and the source example took only a few minutes to generate. #### Who This Is Useful For - VC analysts and associates who write investment memos and need to present them to partners. - Founders who have pitch documents and need a polished deck fast. - Consultants and biz ops teams who turn structured reports into presentations every week. #### What You Will Build A complete, branded slide deck generated from an investment memo PDF. In the source example, Manus turned Sequoia’s 2014 DoorDash memo into a full deck with market sizing, competitive landscape, traction metrics, and team slides, and it picked up DoorDash’s brand colors. ![image-2102](https://tru-images.b-cdn.net/guide-assets/69/69aa24dab2dba1db903193bf/image-2102.png) #### What You Need - Manus account: free 7-day trial, then $20/month. - An investment memo or structured document as a PDF. - A Google account, if you want to export to Google Slides. #### Going Further - Export the finished deck as PPTX, PDF, or Google Slides depending on how you need to edit or share it. - Attach an existing deck with your company’s branding alongside your memo and tell Manus to style the new deck the same way. - Try the same workflow with other structured documents, including quarterly business reviews, board updates, and client reports. ### Step 1: Get Your Memo Ready Start with a PDF that has clear, structured sections. Investment memos are ideal because they follow a formula: market analysis, team background, business diligence, traction, SWOT analysis, deal dynamics, and investment recommendation. - We used Sequoia's 2014 DoorDash investment memo. Any memo with named sections will work. - The more structured the document, the better the output. Manus reads the section headers and uses them to decide what goes on each slide. - This also works with quarterly reports, project briefs, or any document that has a repeatable format. > Pro tip: You don't need to clean up the PDF first. Manus reads the raw file. No copy-pasting, no text extraction. Just attach the PDF directly. ### Step 2: Create the Task in Manus Go to manus.im and sign up. Once you're in, click New task. - Select Create slides as the task type. - Attach your memo PDF. - Type your prompt: ```text Turn this investment memo into a slide deck. ``` That's it. No fancy prompting. Manus reads the memo, identifies the sections, and builds the deck. - Manus runs in the cloud, so this is a “walk away and come back” task. It took a few minutes for our DoorDash memo. - You can watch it work in real time if you want, but you don't need to babysit it. ![image-2103](https://tru-images.b-cdn.net/guide-assets/69/69aa24e9b2dba1db90319c31/image-2103.png) > Pro tip: There’s no official Mac app for Manus, but there is one for Windows and iOS. You can use the iPad version on your Mac if you really want to. ### Step 3: Review and Edit Manus will usually one-shot the layout. For the DoorDash memo, it created slides for every major section and pulled in the actual numbers from the document. It even used DoorDash brand colors. - You can edit slides directly inside Manus. It works like a basic deck editor. - You can also present from within Manus if you need to share it quickly. - If something needs fixing, tell Manus in the chat. ```text Make the background higher contrast. Add a slide summarizing the key risks. ``` > Pro tip: If you don't like the visual style, tell Manus what to change. We asked it to fix the background and it updated the whole deck in one pass. ### Step 4: Export Your Deck When you're happy with the result, you have three export options: - Download as PPTX if you need to edit in PowerPoint. - Download as PDF if you just need to share it. - Convert to Google Slides if you connect your Google account. This is the cleanest option. It saves the deck directly into your Google Drive with all the charts and formatting intact. The Google Slides export is a standout feature. One click and the deck is in your Drive, ready to present or share with your team. ![CleanShot 2026-03-05 at 16.09.15@2x](https://tru-images.b-cdn.net/guide-assets/69/69aa24fcb2dba1db90319c6f/cleanshot-2026-03-05-at-16-09-15-2x.png) ### Pro tip > You don't need to clean up the PDF first. Manus reads the raw file, so you can attach the PDF directly without copy-pasting or extracting text. ### Guide: Turn Any Case Study Into a Client-Ready Video With Synthesia > Turn a written case study into a one- to two-minute Synthesia video with an AI narrator, b-roll, animations, and a client-ready CTA. - Canonical page: https://app.therundown.ai/guides/turn-any-case-study-into-a-client-ready-video-with-synthesia - Machine-readable page: https://app.therundown.ai/guides/turn-any-case-study-into-a-client-ready-video-with-synthesia.md - Difficulty: beginner - Featured tool or company: Synthesia - Categories: Content & Creative, Sales, Consulting & Customer Success - Instructors: Billy Howell - Published: 2026-05-25 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to turn a written case study into a 1-2 minute video complete with an AI narrator, b-roll, and animations. #### Who This Is For - Consulting agencies with a library of case studies they want to turn into sales assets. - Marketers who want to turn customer stories into YouTube videos or Facebook ads. - Founders and operators who want a faster way to package proof for a specific client or prospect. #### What You Will Build You will create a one to two-minute case study video from a written case study file. The finished video can include AI avatar scenes, b-roll or background scenes the avatar narrates over, and animated project-flow scenes with narration. ![Final Synthesia case study video with an AI avatar scene](https://tru-images.b-cdn.net/guide-assets/local-packages/226-for-jason/01-final-video-ai-avatar-scene.png) #### What You Need To Get Started - A Synthesia account (free works). - Google Chrome. - A written case study as a .txt, .pdf, or .pptx file. - Optional: a short client background or project proposal if you want the video to feel specific to one prospect. #### Going Further The best version of this workflow is not one generic case study video. It is a repeatable case study engine for specific audiences. For an agency, that could mean one approved case study plus several client background memos. Each memo changes the opening, relevance, and CTA while the original proof stays the same. For content, the same source file can become a YouTube explainer, a Facebook ad, a Reels cutdown, or a short sales page embed. After generation, change the video dimensions for the channel you care about. Use a vertical format when you want the same case study video to work for Reels, TikTok, or Shorts. Start with the clean case study. Add the audience. Generate the first draft. Then edit the version for the channel where it will actually be used. ### Prepare the Case Study You need a written case study. Save it as a .txt, .pdf, or .pptx file. Good sections to include: - the client or project name - the problem - what changed - the results or transformational KPIs - one quote or proof point - the CTA The most useful case studies have clear before-and-after metrics: higher conversion, faster response time, lower cost, more qualified leads, hours saved, or revenue gained. If you want a client-specific version, make a second file with the prospect's background: - who they are - what they care about - the problem they are trying to solve - the CTA you want the video to use > Pro tip: Here's a hack to save tokens and edit time while still making client-specific videos. Each time you want a custom case study video, you can keep the main video the same and simply generate new intros and outros that are client-specific. ### Start With Assistant Open Synthesia: https://www.synthesia.io in Google Chrome. On the home screen, click Start with Assistant. ![Synthesia home screen with Start with Assistant](https://tru-images.b-cdn.net/guide-assets/local-packages/226-for-jason/03-synthesia-start-with-assistant.png) Start with Assistant is the fastest path for turning a written case study into a video. You give Synthesia the source file, audience, and objective, then it creates the first outline and script. > Pro tip: If your case study is already a slide deck, you can also use Import PowerPoint. That is useful when the deck already has the structure you want, but Start with Assistant is better when you want Synthesia to create the video arc from a memo or written case study. ### Upload the Case Study and Write the Prompt Choose a 1- or 2-minute video for the first draft. Then attach the case study file. For a generic case study video, use a prompt like this: ```text Turn this case study into a 1-2 minute video for [AUDIENCE]. The topic is [TOPIC]. The objective is [OBJECTIVE]. Preserve the facts, metrics, and quote exactly. ``` For a more targeted sales video, upload the client background file too. Then tell Synthesia how to use both files. ![Synthesia prompt with attached case study and client background files](https://tru-images.b-cdn.net/guide-assets/local-packages/226-for-jason/04-prompt-with-case-study-and-client-background.png) Here is the version we used for the prospect-specific test: ```text Read the case study and turn it into a 1.5-minute presentation. Read the client background that I have also attached and mention what parts of the case study are most relevant to this client. Make the video custom to them. The audience is [PROSPECT], a prospective client. The topic is [CASE_STUDY_CLIENT]. The objective is to show [PROSPECT] the best information from the case study so they will want to become our client. ``` Keep the prompt direct. The case study file gives Synthesia the facts, and the prompt tells it what kind of video to make. After the prompt is in place, select a visual style and click Create Video. > Pro tip: You can generate the case study file and client background file specifically for a client. For example, write one file with the approved case study facts, then write a second file with the prospect's business model, pain points, and the CTA you want the video to use. ### Review the Outline and Scripts After you submit the prompt, Synthesia will read the attachments and respond with a draft direction. Check that it understood the audience, the topic, and the objective before you create the video. ![Synthesia Assistant response after reading the uploaded files](https://tru-images.b-cdn.net/guide-assets/local-packages/226-for-jason/05-assistant-script-generation-response.png) Synthesia will generate a video outline with several short scenes and scripts. ![Animated checklist scene with narration](https://tru-images.b-cdn.net/guide-assets/local-packages/226-for-jason/02-animated-checklist-narration-scene.png) If a slide in the outline looks blank, that does not always mean something is missing. It can mean Synthesia is going to generate an animation or process visual for that scene instead of using an AI avatar. This is the most important review point. Script generation does not spend your final video generation pass. Review the outline before rendering so you can catch a weak scene or CTA early. Read each scene before you generate. Check: - whether the case study facts are correct - whether the audience is right - whether the CTA matches your goal - whether the video uses the client background as framing - whether the visuals match the story The client background should shape the opening, examples, and CTA. The case study should still supply the proof. ### Generate and Edit the Video Once the outline and scripts look right, click Generate. The video can take several minutes to render. In our run, the finished videos appeared in My recents, where we could open them, play them, share them, or click Edit. ![Recent generated Synthesia videos](https://tru-images.b-cdn.net/guide-assets/local-packages/226-for-jason/06-recent-generated-synthesia-videos.png) The generated video will mix togther these scene types: - AI avatar scenes where the narrator speaks directly to the viewer. - B-roll and background scenes that the avatar narrates over. - Animated checklist, infographic, and project-flow scenes that explain the process. You can open a recently generated video and click Edit. You can change the scene structure, edit the script, adjust visuals, and regenerate the video. You can also swap the presenter if the video should feel more personal. Synthesia supports personal and custom avatars, so a consulting agency could use the founder, strategist, or account lead as the AI avatar. If a scene needs stronger visuals, generate alternate b-roll, backgrounds, or images inside Synthesia. The built-in media tools include options such as Google Veo for short video clips and Nano Banana Pro for image generation. ### Resources - [Synthesia](https://www.synthesia.io) - [Google Chrome](https://www.google.com/chrome/) ### Pro tip > For client-specific versions, keep the main case study proof the same and regenerate only the intro, outro, and CTA around each prospect. ### Guide: Turn Any ChatGPT Image Into an Editable Canva Design > In this guide, you will learn how to turn a ChatGPT-generated image into a Canva project you can keep editing. - Canonical page: https://app.therundown.ai/guides/turn-chatgpt-images-into-editable-marketing-assets-with-canva - Machine-readable page: https://app.therundown.ai/guides/turn-chatgpt-images-into-editable-marketing-assets-with-canva.md - Difficulty: beginner - Featured tool or company: Canva - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-06-12 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to turn a ChatGPT-generated image into a Canva project you can keep editing. Instead of regenerating the same visual every time you need a new format, you will move the image into Canva, clean it up, resize it, and reuse it across marketing assets. #### Who This Is Useful For - Marketers who like the first AI-generated concept, but need a faster way to fix layout, text, color, or crop issues. - Founders and operators who need social posts, banners, ads, or event graphics without starting from a blank Canva file every time. - Content teams that want one strong visual direction to become several channel-ready assets instead of several separate prompt threads. #### What You Will Build You will build a repeatable ChatGPT-to-Canva workflow: generate a source image in ChatGPT, hand that exact image into a new Canva project, then use Canva to make the infographic editable and reusable. ![Infographic added to a new Canva project](https://tru-images.b-cdn.net/guide-assets/local-252/turn-chatgpt-images-into-editable-marketing-assets-with-canva/01-infographic-added-to-canva-project.png) The example here is an infographic, but the same pattern works for social posts, banners, ads, and slide visuals. The first output does not need to be perfect. The goal is to get a visual that is strong enough to keep working on. #### What You Need To Get Started - A logged-in ChatGPT account. A Free account is enough to try ChatGPT apps where they are available, but app access can vary by region, workspace settings, and plan. - A Canva account. Canva Free is enough to try the core ChatGPT-to-Canva handoff and Magic Layers in ChatGPT, subject to Canva's AI usage limits. - Canva Pro, Teams, Business, Education, or Nonprofits is optional. It is useful if you want higher AI allowances, Magic Resize, Brand Kit access, premium assets, or heavier campaign production. - The Canva app connected inside ChatGPT. - One campaign, post, infographic, ad, or slide idea to turn into a visual. ### Step 1: Connect Canva Inside ChatGPT Start by making sure Canva is available inside ChatGPT. In the ChatGPT desktop app, open: ```text Settings > Apps > Browse Apps ``` Search for Canva, then choose the add or install option. ChatGPT will send you through Canva authentication in the browser. Once that is done, return to ChatGPT and start a fresh chat. This setup step matters because the handoff only works when ChatGPT recognizes Canva as an app, not just as a word in your prompt. > Pro tip: If the app menu has moved, search ChatGPT settings for Apps or Canva. The exact UI can shift, but the goal is the same: connect Canva before you try to create the project. ### Step 2: Generate A Source Image Worth Editing Next, generate the image in ChatGPT first. This is the key difference from a normal Canva prompt. You are not asking Canva to invent the asset from scratch. You are creating the initial concept in ChatGPT, then using Canva as the editable workspace. Use a prompt that gives ChatGPT enough design direction, but also leaves room for Canva edits later. ```text Create a marketing image starter for [CAMPAIGN]. Audience: - [AUDIENCE] Goal: - [OUTCOME] Visual direction: - [STYLE] - [COLOR MOOD] - [FOCAL SUBJECT] Requirements: - make the core idea clear at a glance - leave room for headline or CTA text - avoid tiny decorative copy - avoid dense paragraph text inside the image - keep the composition flexible enough for square, vertical, and wide crops Goal: - create a strong starting image that can be moved into Canva and adapted into multiple marketing assets ``` For this guide, use an infographic-style prompt so the handoff has real structure to test: ```text Create a clean branded infographic for [BRAND] about [TOPIC]. Title: - [SHORT TITLE] Sections: - [SECTION 1] - [SECTION 2] - [SECTION 3] Style: - modern editorial infographic - clean hierarchy - large readable text - simple icons or callout boxes - one accent color Requirements: - make each section feel distinct - avoid tiny paragraph copy - leave enough structure that Canva can turn this into a workable project for follow-up edits ``` Before you move on, check the image with a simple filter: - Is the concept right? - Is the focal point clear? - Is there enough empty space for edits? - Would this still work if you resized it? If the concept is wrong, regenerate in ChatGPT. If the concept is right but the layout needs cleanup, move it into Canva. ### Step 3: Send The Image Into A New Canva Project Once you have a usable image, stay in the same ChatGPT thread and ask Canva to create a project from that exact image. Use this prompt: ```text Now add that image into a new @Canva project. ``` When you type @Canva, make sure ChatGPT confirms the app mention. In the working flow, Canva appears as an app mention inside the prompt, then ChatGPT creates a project and returns a button or link to open the design. This is the hidden move. The ChatGPT image stops being a static output and becomes the starting point for a Canva design. > Pro tip: Do not just say add that into Canva and assume ChatGPT will know what you mean. In testing, vague wording could trigger Canva to generate a different image instead of using the ChatGPT image you already liked. Mention @Canva directly and ask for a new Canva project. If the handoff fails, try these fixes: - Start a fresh ChatGPT thread and repeat the handoff prompt. - Try the same workflow with a simpler visual, like an infographic or social post. - Manually create a Canva project, give it a unique name, then ask ChatGPT to find and edit that Canva project by name. Some image types are more reliable than others. Highly stylized word art can be harder to hand off cleanly, while simpler post or infographic layouts tend to behave better. ### Step 4: Use Canva To Fix The Usual Re-Gen Triggers Open the Canva design from ChatGPT. Once the image is inside Canva, your job is to fix the issues that usually make teams regenerate from scratch. Common re-gen triggers include: - headline is in the wrong place - text is too small - color accents do not match the brand - crop does not fit the channel - layout needs more space for a CTA - one object needs to move, but the whole concept is still good If Canva gives you a Customize this design option, use it. Then ask Canva to run Magic Layers on the project: ```text Run Magic Layers on that project now. ``` ![Magic Layers creates an editable Canva project from the infographic](https://tru-images.b-cdn.net/guide-assets/local-252/turn-chatgpt-images-into-editable-marketing-assets-with-canva/02-infographic-magic-layers-project.png) > Pro tip: Magic Layers is Canva's feature for turning a flat image into separated design layers. In this workflow, it can pull parts of the infographic apart, like text, icons, callouts, and background elements, so you can edit the design instead of treating it like one locked screenshot. After that, make one specific edit at a time: ```text Change the main title to: [NEW HEADLINE] ``` ```text Change the background color to: [BRAND COLOR OR STYLE] ``` ```text Add more room for a short CTA at the bottom. ``` Keep the edits practical. You are not trying to rebuild the whole infographic in ChatGPT. You are using Canva to clean up the parts that are easier to adjust in a design tool. Magic Layers and text editing are useful, but they are not perfect. If Canva cannot edit a piece of text or separate a layer, use the Canva editor manually. The workflow is still valuable because you already have the source image inside a real design project. ### Step 5: Resize The Project Into Channel Variants Once the master Canva project is working, duplicate it and resize it for the next format. In ChatGPT, you can ask Canva for a new resized version: ```text Duplicate this project and resize it for a story post. ``` Then ask for the resized version to become its own Canva project: ```text Now make a new @Canva project of that. ``` You can repeat the same pattern for: - Instagram square post - Instagram Story - paid social ad - newsletter header - website banner - YouTube thumbnail - event promo graphic At this point, the creative work changes. You are no longer trying to invent a brand-new concept. You are adapting the same concept for each channel. Use this prompt when you need tighter copy for each version: ```text Using the same campaign angle, write three short headline, subhead, and CTA options for: - Instagram square post focused on awareness - paid social ad focused on click-through - newsletter banner focused on a quick launch announcement Keep each version direct, marketing-friendly, and short enough to fit inside a design. ``` Then bring the best headline into the right Canva version. The decision rule is simple: ```text Move it into Canva when the concept is right but the execution needs cleanup. Regenerate in ChatGPT when the concept, composition, or subject is fundamentally wrong. ``` That is where the time savings show up. One good image can become a small campaign kit instead of a single disposable output. ### Going Further The same workflow also works for slide decks. Instead of asking AI for a full deck and hoping every slide lands, start with the structure in ChatGPT: ```text Draft an outline for a five-slide deck on [TOPIC]. ``` Then ask Canva to turn that outline into a project: ```text Generate that slide deck in a new @Canva project. ``` Canva will show a presentation review flow where you can inspect the outline, adjust the style, choose a brand kit if you have one, and generate the design. You can also ask ChatGPT to revise the outline before generating: ```text Swap slides two and three, then add a final slide with sources. ``` For more control, generate one strong slide visual at a time, hand each one into Canva, then assemble the final deck from cleaned-up Canva slides. The repeatable pattern is: ```text 1. Generate the strongest starting visual in ChatGPT. 2. Move that exact image into a Canva project. 3. Use Canva to fix the edit problems. 4. Save the Canva project as the master. 5. Duplicate and resize it into channel-specific variants. ``` The useful part is not just getting a prettier AI image. It is turning that image into a working asset you can keep editing, reusing, and shipping. ### Guide: Turn Any Flat Image Into a Fully Editable Design in 60 Seconds > Learn how to use Canva’s Magic Layers feature to turn a flat AI-generated PNG or JPG into an editable Canva design. - Canonical page: https://app.therundown.ai/guides/turn-any-flat-image-into-a-fully-editable-design-in-60-seconds - Machine-readable page: https://app.therundown.ai/guides/turn-any-flat-image-into-a-fully-editable-design-in-60-seconds.md - Difficulty: beginner - Featured tool or company: Canva - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-04-02 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use Canva’s Magic Layers feature to turn a flat AI-generated PNG or JPG into an editable Canva design. You’ll produce a layered version of the image so you can fix small details without regenerating the whole thing. Processing takes about 30 to 60 seconds once you run Magic Layers. #### Who This Is Useful For - Anyone generating images with AI who gets close to the right result but needs to fix one wrong word, date, or detail - Marketers and content creators who generate ads, social graphics, or product images - Small business owners or freelancers making AI-generated infographics, slides, or promotional graphics #### What You Will Build ![image-2159](https://tru-images.b-cdn.net/guide-assets/69/69cee3b6bbc3afeabb7f48cc/image-2159.png) You’ll build a reusable workflow for taking a flat AI-generated image and splitting it into editable layers inside Canva, so you can change text, move objects, swap colors, and fix details without starting over. - A flat image converted into editable Canva layers - Editable text for fixing dates, taglines, typos, and other details - Separate object and background layers you can move, resize, delete, or update #### What You Need - A Canva Pro account; Magic Layers is a Pro feature - A flat PNG or JPG image generated with any AI tool, or an old graphic you want to update - Optional: a PDF converted into an image first if you want to edit PDF content #### Going Further - Use this workflow for AI-generated infographics, slides, t-shirt designs, or other graphics where the overall design is strong but specific details need editing. Generate first, then bring the image into Canva to edit the text, reposition elements, and keep the visual direction you already liked. ### Step 1: Generate or Grab Your Image Start with an AI-generated image that is close to what you want but has one detail you need to fix, like a wrong date, misspelled word, or outdated tagline. In the example workflow, a product launch ad for headphones looked good, but the launch date was wrong. Instead of re-prompting and risking lower image quality, you’ll use Canva to edit the existing image. > Magic Layers may struggle with overly complex patterns or very small text, so don’t get discouraged if your first try does not work perfectly. ### Step 2: Open Magic Layers There are two ways to open Magic Layers in Canva. - From the homepage: Go to canva.com, select Magic Layers from the homepage, click Select Media, and choose your image from your existing projects or uploads. - From inside an existing project: If your image is already in a Canva design, select the image on the canvas, click Edit in the toolbar, go to Magic Studio, and click Magic Layers. ![image-2160](https://tru-images.b-cdn.net/guide-assets/69/69cee40dbbc3afeabb7f52f5/image-2160.png) ### Step 3: Wait for the Layers to Separate Once you run Magic Layers, Canva will process the image for about 30 to 60 seconds. It reads the layout, identifies elements like headlines, subheadlines, product images, and background patterns, then separates them into individual layers. In the headphones ad example, Canva split the design into four clean layers: the main headline, the subheadline with the launch date, the product photo, and the background. It also matched fonts and colors so the text could be edited separately. ![image-2161](https://tru-images.b-cdn.net/guide-assets/69/69cee4255670e645b88bcf2a/image-2161.png) > Magic Layers only works with PNG and JPG files. If you want to edit a PDF, convert it into an image first, then upload it into Canva. ### Step 4: Edit the Pieces That Need Fixing Click any text layer to edit it like live text. You can change a date, rewrite a tagline, or fix a typo without regenerating the image. Click an object layer to move it, resize it, or delete it. You can also drop in a new product photo while keeping the existing layout. If you are using a real-life photo or a photorealistic generated image, Magic Layers may not work as well. You can try Canva’s background removal tool to separate the foreground and background. ### Pro tip > Magic Layers may struggle with overly complex patterns, very small text, real-life photos, or photorealistic generated images. It only works with PNG and JPG files, so convert PDFs into images before uploading them. ### Guide: Turn Any Idea Into a Working Website With ChatGPT Work and Codex > In this guide, you'll learn how to turn a website idea into a working prototype using the new ChatGPT Work and Codex app (which uses the new GPT-5. - Canonical page: https://app.therundown.ai/guides/turn-any-idea-into-a-working-website-with-chatgpt-work-codex - Machine-readable page: https://app.therundown.ai/guides/turn-any-idea-into-a-working-website-with-chatgpt-work-codex.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-07-10 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you'll learn how to turn a website idea into a working prototype using the new ChatGPT Work and Codex app (which uses the new GPT-5.6 models). #### Who This Is Useful For - Founders and creators who want to turn a rough website idea into a working prototype. - Operators and researchers who need to collect directory data while another agent builds the site. - Anyone with a content-driven project who wants to move from a plan to a usable website in one desktop app. #### What You Will Build We built an Astro.js directory prototype for roguelike and roguelite games on Nintendo Switch. It has a homepage, filters, detail pages, structured game data, and images that the coding agent can use to populate the site. ![Finished directory prototype with cards and filters](https://tru-images.b-cdn.net/guide-assets/local-361/turn-any-idea-into-a-working-website-with-chatgpt-work-codex/01-final-directory-site.png) The same setup works for tools, products, books, services, local businesses, or any collection with consistent fields. #### What You Need To Get Started - The updated ChatGPT desktop app with ChatGPT Work and Codex visible. - A new project folder for the PRD, research files, images, and Astro.js site. - One website-directory idea with a clear theme and audience. #### Going Further When the site is ready, check whether the new Sites feature is available in the app. Tell Codex: ```text Publish this finished website using the Sites skill. Deploy it and return the public URL. ``` If Sites is available, OpenAI will deploy and host the site for you. That is the natural next step for this workflow: plan the site, gather the content, build the prototype, then publish it from the same app. ### Step 1 Update The ChatGPT Desktop App Download the ChatGPT desktop app (https://chatgpt.com/download/) and make sure it is updated. If the app shows an Update button, click it, let the app reopen, and choose Get started if prompted. You'll know the update is ready when the top left says ChatGPT Work. Click ChatGPT Work or Codex to toggle modes. ![ChatGPT Work and Codex mode switcher](https://tru-images.b-cdn.net/guide-assets/local-361/turn-any-idea-into-a-working-website-with-chatgpt-work-codex/02-chatgpt-work-codex-mode-switcher.png) ChatGPT Work is the longer-running task mode for research, files, and sub-agents. Codex is the coding mode for setting up the project and building the site. For more context, read OpenAI's ChatGPT Work announcement (https://openai.com/chatgpt-work/). > Pro tip: If ChatGPT Work is not in the top left, update the desktop app before starting. The labels and available models can vary by account. ### Step 2 Create A Project And Write The PRD Create a new project folder on your computer, then open it as a project in the ChatGPT desktop app. Start a new task in that folder, click ChatGPT Work, and describe the directory you want to build. We used a directory of roguelike and roguelite games for Nintendo Switch. The category is not the important part. Pick a clear theme with a few fields that every entry should share. Use the Astro.js framework (https://astro.build/) for the first version. We recommend it for content-driven sites because each item can live in a simple Markdown file with front matter. Astro turns that folder into fast, SEO-friendly pages without requiring a separate database, and its straightforward structure gives an AI coding agent less to untangle. Use a prompt like this: ```text I want to build a website directory for [DIRECTORY_TOPIC]. Plan the site using the Astro.js framework. Define: - the target audience - the theme and subject matter - the fields each directory item should have - the first version of the site - an acceptance checklist Then write a one-page PRD and save it in this project folder. ``` The PRD is the handoff between planning and building. It should explain what the site is, who it is for, what each entry contains, and what needs to work in version one. > Pro tip: Keep the first version static and content-driven. We can add ratings, comments, or other interactive features after the directory itself works. ### Step 3 Research The Directory And Build It In Codex When the PRD is saved, keep ChatGPT Work on the research task. Ask it to task sub-agents with researching the directory items and gathering the data and images for each entry. ```text Use sub-agents to research the items in this directory. For each item, gather: - name - website - release date or launch date - price or pricing signal - short description - cover, logo, or other useful image Save each item in a consistent folder structure with a Markdown file and front matter. Mark anything you cannot verify instead of guessing. ``` While Work is running, open a new thread, click the dropdown, and switch to Codex. Point the Codex thread at the PRD and research folder. ```text Read the PRD and the research folder. Set up this Astro.js coding project and begin building the directory site. Use sub-agents to build the pages and components where helpful. Create a searchable homepage, filters, detail pages, and the content structure described in the PRD. Open the preview when the first version is ready and summarize what you built. ``` This is the useful split: Work gathers the raw material while Codex turns the PRD into a codebase and a visible website. ![First Codex site build](https://tru-images.b-cdn.net/guide-assets/local-361/turn-any-idea-into-a-working-website-with-chatgpt-work-codex/04-first-site-build.png) > Pro tip: Keep Git enabled while Codex works. Small commits are save spots we can return to when a change breaks the site, and connecting GitHub gives us a cloud copy and a clean handoff for another agent. ### Step 4 Review The Prototype And Choose A Model When Codex finishes the first pass, open the preview and check the actual site. Click into a detail page, test the filters, and look for missing CSS or layout problems. In our recording, the Astro build produced 58 pages, but the first pass still needed a visual check and a CSS correction. If the project needs a local build check, run: ```text npm install npm run build npm run dev ``` We used GPT-5.6 Terra at medium effort and standard speed. Luna is smaller when we want to save tokens, and Sol is the biggest model for the hardest reasoning. Read the GPT-5.6 announcement (https://openai.com/index/gpt-5-6/) for the model breakdown, and use the model picker available on the recording account. ### Guide: Turn Any Idea Into an AI-Powered Site With Lovable > In this guide, you will learn how to turn a rough app idea into a working AI-powered site without writing a giant technical prompt. - Canonical page: https://app.therundown.ai/guides/turn-any-idea-into-an-ai-powered-site-with-lovable - Machine-readable page: https://app.therundown.ai/guides/turn-any-idea-into-an-ai-powered-site-with-lovable.md - Difficulty: beginner - Featured tool or company: Lovable - Categories: Coding & App Development, Content & Creative - Instructors: Billy Howell - Published: 2026-07-31 - Updated: 2026-08-06 ### Overview #### The Rundown In this guide, you will learn how to turn a rough app idea into a working AI-powered site without writing a giant technical prompt. #### Who This Is Useful For - Founders and product builders who want to test an app idea before writing a technical specification. - Consultants and agencies who need a repeatable way to turn client ideas into useful first-pass prototypes. - No-code builders and small teams who want to create AI-powered tools without choosing the stack by hand. #### What You Will Build You will build Method, a recipe app that turns pasted recipe text into an editable visual card. The card lays out ingredients and cooking actions in a flow that is easier to scan than a normal recipe page, then saves the result to a library. ![Method brownie recipe card](https://tru-images.b-cdn.net/guide-assets/local-415/turn-any-idea-into-an-ai-powered-site-with-lovable/01-method-brownie-recipe-card.png) The broader pattern is useful beyond recipes: turn messy input into a structured, reviewable object that people can edit and reuse. #### What You Need To Get Started - A ChatGPT account and access to a new conversation. - A Lovable account with permission to create a project. - A visual reference that shows the information design you want to borrow. - A recipe or other source text with enough detail to test the conversion. - A Lovable project where Cloud AI can be enabled. #### Going Further Reuse the same process for any app that turns messy input into a useful object: - turn meeting notes into an editable action plan; - turn a workout description into a structured training card; - turn a client brief into a reviewable project plan; - turn study notes into a visual reference sheet. Set an AI-token limit before publishing. Add accounts, a database, and URL imports only after the core loop works. If you add URL import later, make sure users have permission to process the source and keep the same review step before saving the result. ### Step 1 Talk Through the Product Idea With ChatGPT Start with the idea in plain English. Explain the problem, the finished experience, and the visual reference you want to use. The goal is to give ChatGPT enough context to ask useful questions before you start building. For this example, attach the reference card and describe Method as a simple app that accepts a recipe, turns it into a visual card, lets the user review or edit the result, and saves it to a library. ![Banana nut bread visual reference](https://tru-images.b-cdn.net/guide-assets/local-415/turn-any-idea-into-an-ai-powered-site-with-lovable/02-banana-nut-bread-reference-card.png) Use a prompt like this: ```text I want to build a small app called [APP_NAME]. A user pastes [INPUT], and the app turns it into [OUTPUT] inspired by the attached reference. The user can review or edit the result, then save it to a library. Help me think through the idea. Ask only the questions that would materially improve the first version. Keep the product focused on one core journey, and do not write the build brief yet. ``` Answer the questions that change the product experience. For Method, those decisions included preserving the original recipe where possible, allowing the user to edit or reorder steps, and flagging guesses instead of silently presenting them as facts. Keep this conversation separate from the build brief. You are using it to find the important product decisions, not to produce a technical specification. > Pro tip: Start with an interview, then ask for the build brief. This gives you a chance to catch missing product decisions before the app builder starts making them for you. ### Step 2 Ask for Requirements and Brand Feel Once the core journey is clear, ask ChatGPT to compress the conversation into a short handoff. This is the document you will give Lovable. Ask for four things: - A one-paragraph product summary. - The first-version feature requirements and user flow. - The required screens and important states. - The brand feel, color palette, and suggested typography. ![ChatGPT visual direction brief](https://tru-images.b-cdn.net/guide-assets/local-415/turn-any-idea-into-an-ai-powered-site-with-lovable/03-chatgpt-visual-direction-brief.png) Use this prompt in the same conversation: ```text Based on our conversation, write the handoff I can give to an AI app builder. Include: 1. One-paragraph product summary 2. First-version feature requirements and user flow 3. Required screens and important states 4. Brand feel, color palette, and suggested typography Keep it concise and implementation-agnostic. Do not prescribe a framework, database schema, component library, or API architecture. Mention that the attached image is the visual and information-design reference. ``` Read the result once before copying it. Remove features that do not support the first loop. For Method, that loop is paste, convert, review, edit, save, and reopen. Keep the handoff focused on what the product should do and how it should feel. Lovable can choose the implementation details it is built to support. ### Step 3 Build the Complete App in Lovable Create a new Lovable project and attach the same visual reference. Paste the ChatGPT handoff with a second prompt that asks for the complete first version. Use an outcome-first request: ```text Build the complete first version of [APP_NAME] from the requirements and brand brief below. Use the attached image as inspiration for the information design, but create an original product and visual identity. Keep the first version focused on the core journey and make sensible product decisions when details are not specified. Use the AI capability available in this Lovable project to turn [INPUT] into the structured, editable [OUTPUT] described in the brief. Preserve stated values, flag ambiguity for review, and show a useful loading state and error state that preserves the original input. Build the complete experience in one pass, including the main screen, the generated result, editing, saving, and reopening the saved result. [PASTE THE CHATGPT REQUIREMENTS AND BRAND BRIEF] ``` Select Plan before you approve the build. Read the proposed screens, states, and core flow. This is the point where you can catch a missing feature or a wrong assumption while the product is still easy to change. ![Lovable Method build plan](https://tru-images.b-cdn.net/guide-assets/local-415/turn-any-idea-into-an-ai-powered-site-with-lovable/04-lovable-method-build-plan.png) When the plan is right, approve it and let Lovable build the first version. In Connectors, open Cloud AI and enable Lovable’s built-in AI. That is what the recorded workflow used, so the recipe conversion did not require a ChatGPT integration or an OpenAI API key. Lovable’s current documentation describes Cloud as its built-in backend with AI, Secrets, and server-side Edge Functions available from the project. You can review the current Lovable connectors (https://docs.lovable.dev/integrations/introduction) and Lovable Cloud (https://docs.lovable.dev/features/cloud) documentation as the labels change. > Pro tip: Attach the visual reference to Lovable before you approve the plan. The finished design is much closer when the builder can see the information hierarchy instead of trying to infer it from prose alone. Let the first build finish before adding refinements. Judge it against the core journey and the reference, not against an imagined technical stack. If one visible part misses, ask for one focused correction. ### Step 4 Test the Core Journey and Publish Test the product as a user: - Paste a recipe with a clear title, ingredient list, and ordered instructions. - Start the conversion and wait for the generated card. - Compare quantities, temperatures, timing, and other important source details. - Edit one label or operation. - Save the card and reopen it from the library. - Check the result at a narrow screen width. - Test blank input and an AI or provider failure. - Set an AI-token limit before publishing if you plan to make the app public. - Run Lovable’s security checks before publishing, then smoke-test the live URL. The recorded Method build showed why the review step matters. A first recipe paste could not be structured. A more clearly organized recipe produced a usable card. The app should give the reader a result to review, rather than forcing them to manually format every field before seeing the payoff. ![Method card editor](https://tru-images.b-cdn.net/guide-assets/local-415/turn-any-idea-into-an-ai-powered-site-with-lovable/05-method-card-editor.png) Use a reusable test prompt like this: ```text Test the complete [APP_NAME] journey in the browser: paste [TEST_RECIPE], convert it, review the source values, edit one label, save it, reopen it from the library, and check the result at mobile width. Also test blank input and an AI or provider failure. The pasted input must not disappear on error. Run the platform’s security checks and report any blocker before publishing. Fix only issues in the core flow; do not add features. ``` > Pro tip: If the input cannot be structured, improve the source text before changing the app. A clear title, ingredient list, and ordered steps give the converter enough structure to produce a useful first draft. Do not treat the generated card as authoritative just because it looks polished. Verify the source values first, then save the edited result. ### Guide: Turn Claude Code Into Your Personal Wall Street Analyst > In this guide, you will learn how to add Anthropic's financial-services plugin marketplace to Claude Code, install market research skills, and use them to create market research reports, equity analysis, earnings reviews, and Excel spreadsheets from your own prompts. - Canonical page: https://app.therundown.ai/guides/turn-claude-code-into-your-personal-wall-street-analyst - Machine-readable page: https://app.therundown.ai/guides/turn-claude-code-into-your-personal-wall-street-analyst.md - Difficulty: intermediate - Featured tool or company: Claude Code - Categories: Data, Research & Analysis, Coding & App Development - Instructors: Billy Howell - Published: 2026-05-13 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to add Anthropic's financial-services plugin marketplace to Claude Code, install market research skills, and use them to create market research reports, equity analysis, earnings reviews, and Excel spreadsheets from your own prompts. This is not financial advice. The useful version of this workflow is a research assistant that organizes public market information, creates structured reports, and helps you decide what to inspect next. #### Who This Is Useful For - Investors and market watchers who want a faster first pass on companies, sectors, earnings, and macro themes - Founders and operators who need lightweight market research without building every spreadsheet by hand - Analysts and consultants who want Claude Code to generate structured research files they can review, clean up, and reuse #### What You Will Build ![What you will build market research skills](https://tru-images.b-cdn.net/guide-assets/local-182/turn-claude-code-into-your-personal-wall-street-analyst/01-what-you-will-build-market-research-skills.png) You will set up Claude Code with Anthropic's financial-services marketplace, then install finance-focused skills like: - market-researcher - earnings-reviewer - equity-research - financial-analysis ![Financial services plugin discover list](https://tru-images.b-cdn.net/guide-assets/local-182/turn-claude-code-into-your-personal-wall-street-analyst/02-financial-services-plugin-discover-list.png) Once those are installed, Claude Code can help you create market research briefs, equity research reports, earnings reviews, comparable company analysis, and Excel spreadsheets. ![Comparable company analysis spreadsheet](https://tru-images.b-cdn.net/guide-assets/local-182/turn-claude-code-into-your-personal-wall-street-analyst/03-comparable-company-analysis-spreadsheet.png) The bigger idea is simple: instead of asking Claude for a generic market summary, you give it specialized finance skills that know how to structure the work. #### What You Need - Claude Code installed and available from your terminal - This GitHub marketplace URL: ```text https://github.com/anthropics/financial-services ``` - A market research task, ticker list, sector, or company you want to analyze - Optional: Microsoft Excel is nice to have but you can import the reports into Google Sheets too - Optional: Claude desktop app access if you want to use these skills from the app or set up routines Disclaimer: This is not financial advice. Use a sample watchlist or public company list while testing. Do not connect a brokerage account or expose sensitive holdings. #### Going Further Once the basic setup works, build a small research system around it. Create one saved prompt for each repeatable job: - daily market brief - weekly earnings scanner - sector comparison - company deep dive - comparable company spreadsheet - earnings call review Keep the prompts strict. Ask for sources, confidence labels, and a final section of questions to verify. The goal is not to let Claude make decisions for you. The goal is to make the first research pass faster, more structured, and easier to audit. Used well, this turns Claude Code into a personal Wall Street analyst: it gathers context, builds the spreadsheet, drafts the memo, and tells you what still needs a human check. ### Step 1 Open Claude Code and Launch the Marketplace Open a terminal and start Claude Code. Once Claude Code is running, type: ```text /marketplace ``` ![Claude code marketplace command](https://tru-images.b-cdn.net/guide-assets/local-182/turn-claude-code-into-your-personal-wall-street-analyst/04-claude-code-marketplace-command.png) That opens Claude Code's plugin manager. You will see tabs for Discover, Installed, Marketplaces, and Errors. The important part is that this command is not a normal chat prompt. It opens the interactive marketplace interface where you can add new plugin sources and install skills. ### Step 2 Add the Financial Services Marketplace Inside the plugin manager, go to the Marketplaces tab. Select Add Marketplace. ![Add financial services marketplace](https://tru-images.b-cdn.net/guide-assets/local-182/turn-claude-code-into-your-personal-wall-street-analyst/05-add-financial-services-marketplace.png) Paste this GitHub URL: ```text https://github.com/anthropics/financial-services ``` After you add it, Claude Code should show a new marketplace named claude-for-financial-services. ![Financial services marketplace details](https://tru-images.b-cdn.net/guide-assets/local-182/turn-claude-code-into-your-personal-wall-street-analyst/06-financial-services-marketplace-details.png) In our run, the marketplace showed 20 available plugins, with several already installed. ### Step 3 Browse and Install the Finance Skills Open the financial-services marketplace and choose Browse plugins. The skills you install will depend on the work you want Claude to do. For a personal market research setup, these are the strongest starting points: - market-researcher for sector research, industry overviews, competitive landscapes, and peer comparisons - earnings-reviewer for earnings calls, filings, and model update notes - equity-research for equity research workflows, earnings analysis, initiation-style reports, and research memos - financial-analysis for financial modeling, comparable company analysis, DCF-style analysis, LBO-style analysis, and spreadsheet outputs Use the keyboard controls shown at the bottom of the plugin manager. In the screenshots, Space toggles a plugin, Enter opens or selects, and Esc goes back. Install only what you need at first. It is easier to test one workflow cleanly than to install everything and forget which skill is doing what. ### Step 4 Ask Claude Code for a Market Research Output Once the skills are installed, start with a bounded research request. Do not ask for "the market" in general. Give Claude a sector, ticker list, company, or event. For a daily market brief, you can use a prompt like this: ```text Use the market research skill to create a pre-market research brief for this watchlist: NVDA, MSFT, AAPL, TSLA, AMZN, META, GOOGL. Use public sources only. Include: - Broad market snapshot - Biggest watchlist movers - Likely reasons for the moves - Upcoming earnings or events - Macro events to watch - Source links for important claims Do not give buy or sell recommendations. Separate facts from interpretation and flag uncertainty. ``` For a spreadsheet-style output, you can ask: ```text Use the financial analysis skill to create an Excel spreadsheet comparing these companies: Lumentum, Coherent, and IPG Photonics. Include operating metrics, valuation multiples, margin comparison, and a short summary of what stands out. Use public information and cite sources where possible. ``` In the test run, Claude produced a comparable company analysis spreadsheet with operating statistics, revenue, growth, gross margin, EBITDA, net income, EPS, valuation multiples, and summary statistics. That is the practical payoff. The plugin setup turns Claude Code into a research workspace that can produce structured files, not just paragraphs. ### Step 5 Use the Skills From Claude Desktop You can also use the newly installed skills from Claude Code inside the Claude desktop app. Open Claude desktop, go into Claude Code, and ask for the same type of finance task. The skills you installed give Claude a more specific way to handle the work, especially when you need reports, spreadsheets, or repeatable research formats. This is useful when you want the workflow to feel less like a terminal session and more like a recurring assistant. ### Step 6 Turn It Into a Routine Once the research prompt is reliable, turn it into a routine. For example, you can create an earnings scanner that runs every Monday morning: ![Earnings scanner routine](https://tru-images.b-cdn.net/guide-assets/local-182/turn-claude-code-into-your-personal-wall-street-analyst/07-earnings-scanner-routine.png) The routine in the screenshot uses this instruction: ```text Check to see what important earnings reports are coming up this week. Give me a report of guidance and use market research skill to give me context on each company. ``` You can adapt that into a daily or weekly market update: ```text Every weekday morning, use the market research skill to create a concise market update for my watchlist. Include major overnight moves, relevant headlines, upcoming earnings, macro events, and questions I should verify before taking action. Do not give financial advice or trade recommendations. ``` Start with a weekly routine before you automate anything daily. Weekly updates are easier to review for quality, and they give you time to catch weak sources, vague summaries, or missing citations. ### Guide: Turn Claude Sessions Into Skills With a Daily Audit > In this guide, you will learn how to make Claude Cowork look through dozens of session files and improve the skills you already use. - Canonical page: https://app.therundown.ai/guides/turn-claude-sessions-into-skills - Machine-readable page: https://app.therundown.ai/guides/turn-claude-sessions-into-skills.md - Difficulty: beginner - Featured tool or company: The Rundown AI - Categories: Automation & Agents, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-06-01 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to make Claude Cowork look through dozens of session files and improve the skills you already use. It will also suggest new skills when there is enough evidence, then help you turn the audit into a repeatable skill and automation. #### What You'll Build A weekly skill audit that reviews your recent AI work, recommends only useful personal skill changes, and skips anything without clear evidence. ![01-skill-opportunities-audit](https://tru-images.b-cdn.net/guide-assets/local-229/turn-claude-sessions-into-skills/01-skill-opportunities-audit.png) #### Who This Is For - Heavy Claude, Claude Code, or Codex users who keep repeating the same prompts - Operators and creators who want their AI workflows to improve over time - Anyone using skills who wants a cleaner way to decide what to create, update, or ignore #### What You'll Need - Claude Cowork, Claude Code, or Codex - A few recent sessions or work examples for the audit - Personal skills enabled or a place where your assistant can save them - A few minutes to review changes before anything gets created You can also run the same basic workflow in Codex if that is where your sessions and personal skills live. #### Going Further Run the same kind of audit once on your automations. Ask: ```text Audit my automations. Look for scheduled tasks that are noisy, stale, too expensive, unclear, or missing approval steps. Recommend only changes with clear evidence. It is okay to say nothing needs to change. ``` This helps keep your automation setup from turning into background noise. The goal is the same as the skill audit: improve the workflows that are actually being used, and leave everything else alone. ### Step 1 Start a Clean Audit Thread Open Claude Cowork and start a new task or chat thread. Keep this focused on one job: auditing your recent AI work for skill opportunities. The goal is not to make Claude invent a giant list of possible skills. The goal is to find repeated friction. If you keep correcting the same output, rewriting the same prompt, or explaining the same workflow, that is the kind of thing that might deserve a skill. > Pro tip: You can also run this in Claude Code or Codex. In Codex, start a new thread outside any specific project, then paste the audit prompt there. Keeping it outside a project helps make the audit personal instead of tied to one repo. ### Step 2 Run the Skill Audit Prompt Use this prompt: ```text Scan my sessions from the past 7 days. If any personal skills should be created or improved, suggest them with evidence. Personal skills only, no repo skills. Prefer no change unless there is a clear reason. Show me what you would create or update before making changes. ``` The key phrase is "prefer no change unless there is a clear reason." Without that, the assistant may try to be too helpful and suggest skills for one-off tasks. What you want back is a short audit with evidence. A useful answer should tell you what happened repeatedly, what skill might help, and whether the best move is to create something new, update an existing skill, or do nothing. In the example audit, the assistant found two concrete moves: - Update an existing weekly-marketing-review skill because the working drafts consistently required extra sections that the skill did not yet ask for. - Delete a duplicate quick-reference skill that competed with the canonical SKILL.md. It also skipped ideas that did not have enough evidence. That is a good sign. A skill audit should not create busywork. ### Step 3 Review the Recommendations Before Anything Changes Do not approve every recommendation automatically. Approve only the ones that meet at least one of these tests: - You repeated the same task several times. - You corrected the same kind of output more than once. - The assistant found a clear gap in an existing skill. - The workflow is likely to happen again. Reject anything that feels like a one-off task. A single unusual request does not need a skill. It might just need a note, a template, or nothing at all. Also watch for scope creep. A good skill is narrow. It should have a clear trigger and a clear job. If the recommendation sounds like "make a skill for all of my marketing work," ask Claude to make it smaller before you approve it. ### Step 4 Turn the Audit Into a Skill Once you like the audit process, ask Claude Cowork to turn the audit itself into a skill. Use this prompt: ```text Turn this skill audit process into a personal skill. The skill should: - scan recent sessions or recent work - look for repeated friction, repeated corrections, and repeated workflows - recommend personal skill updates only - avoid repo skills unless I explicitly ask - prefer no change unless there is clear evidence - show proposed changes before editing files - report what it created, updated, skipped, or rejected Keep the skill narrow. It should audit skill opportunities, not manage every part of my workspace. ``` This is the part that makes the workflow reusable. Instead of remembering the full audit prompt every week, you can trigger the skill and get the same conservative review process. If Claude suggests a broad skill, tighten it. The skill should not rewrite your whole setup. It should inspect recent work, recommend improvements, and wait for approval. ### Step 5 Schedule the Audit After the skill exists, set up an automation to run it weekly or daily. A weekly schedule is usually enough. Daily can work if you use AI heavily and want more frequent cleanup, but the audit should still be conservative. Use a prompt like this: ```text Set up an automation that runs this personal skill audit once a week. Rules: - Review recent sessions or recent AI work. - Recommend personal skill changes only. - Do not edit repo skills. - Do not create or update anything unless there is clear evidence. - Tell me what you recommend before making changes. - It is okay to report that nothing should change. ``` In Claude Cowork, you can open the Scheduled Task tab after the automation is created. From there, adjust the timing, change the cadence, or move the task to a lower-cost model like Sonnet if the audit does not need your strongest model every time. ![02-weekly-skill-audit-scheduled-task](https://tru-images.b-cdn.net/guide-assets/local-229/turn-claude-sessions-into-skills/02-weekly-skill-audit-scheduled-task.png) This is also where you can tune the task. If it runs too often, move it to weekly. If it produces too many recommendations, add a rule that it can recommend at most one new skill and one existing skill update per run. ### Guide: Turn One Messy Dataset Into a Strategy Deck People Will Actually Read With Claude Design > In this guide, you will learn how to use Claude Design to turn your raw data into a useful strategy deck complete with actual insights, not AI slop. - Canonical page: https://app.therundown.ai/guides/use-claude-design-slide-decks-like-a-pro-with-this-strategy - Machine-readable page: https://app.therundown.ai/guides/use-claude-design-slide-decks-like-a-pro-with-this-strategy.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Data, Research & Analysis, Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2026-05-07 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Claude Design to turn your raw data into a useful strategy deck complete with actual insights, not AI slop. Claude Design analyzes what is working and gives concrete recommendations your team can use. #### Who This Is Useful For - Marketers and creators who need to turn campaign or content data into a useful presentation - Founders and operators sitting on messy exports from Shopify, Facebook ads, Stripe, YouTube, or a CRM - Consultants and analysts who need a fast first draft of a deck before polishing it for a client or team #### What You Will Build You will build a strategy deck from one messy dataset. In the test run, Claude Design turned six months of YouTube channel data, plus matching thumbnails, titles, and descriptions, into a presentation with rankings, charts, thumbnail analysis, title patterns, and concrete next steps. The point is not just to make prettier slides. The point is to get a deck that helps you make strategy decisions from your own data. You can export the finished deck as a PDF, PowerPoint file, or whatever format you prefer, or import it into Google Slides for final polish. ![Google Slides strategy deck with recommendations slide](https://tru-images.b-cdn.net/guide-assets/189/01-what-you-will-build-google-slides-strategy-deck.png) #### What You Need - A paid Claude account - A CSV or spreadsheet of data that you want to analyze and visualize as a strategy deck #### Going Further Once you have a deck structure that works, duplicate the project and add more data later. You could add another month of campaign data, a second channel, a different product line, or a new batch of sales exports. Smaller focused decks usually work better than one giant project with every source at once. ### Step 1 Start With One Data Source Do not throw every file you have into Claude Design. That makes the project slower, burns through usage faster, and usually creates a bloated deck. Start with one clean spreadsheet or CSV. This workflow can work with YouTube channel data, Facebook ad data, Shopify data, sales exports, customer research, or any messy report you need to explain. If you have supporting assets, include only the ones that help explain the data. For a YouTube analysis, that might mean thumbnails, titles, and descriptions tied to each video. ### Step 2 Create a Slide Deck in Claude Design Open claude.ai/design, switch to Slide deck, and keep the setup simple. For this workflow, skip the design system unless your team already has one uploaded. Turn on speaker notes so Claude can keep the slides lighter instead of cramming every detail onto the page. ![Claude Design slide deck setup screen](https://tru-images.b-cdn.net/guide-assets/189/02-claude-design-slide-deck-setup.png) ### Step 3 Upload Your Data After the project opens, upload your spreadsheet and any supporting assets. If you have images or creative files, make sure the filenames or spreadsheet fields make the relationship clear. Claude needs some way to match each asset back to the right row. ![Claude Design project files upload area](https://tru-images.b-cdn.net/guide-assets/189/03-claude-design-upload-context-files.png) ### Step 4 Ask for Synthesis, Not Just Analysis The prompt is the important part. Do not ask Claude to simply analyze the data. Push it toward strategy analysis, clear decisions, and useful takeaways. ```text Turn these files into a strategy deck on performance. Analyze the results by item and extract best practices from the data and supporting assets. Use charts, rankings, and concrete recommendations. Match any images or creative files to the CSV using the filename or matching field. Keep it crisp and presentation-ready. ``` If you know the categories you want Claude to study, add them. For the YouTube test, the prompt asked it to look at thumbnails, titles, descriptions, and subject matter. Swap those categories for whatever matters in your dataset. ![Claude Design prompt for a strategy deck](https://tru-images.b-cdn.net/guide-assets/189/04-claude-design-strategy-deck-prompt.png) ### Step 5 Watch the First Pass Claude Design can take 10-15 minutes to build the deck. The first time you run this workflow, watch how it thinks through the data instead of sending the prompt and leaving. Check whether the deck is doing real work. A useful deck should not just repeat numbers back to you. It should show what is working, explain why it matters, and turn patterns into next steps. ```text Here is an updated CSV with the correct date range. Everything else should be the same. Rebuild the deck using this file instead. ``` ### Step 6 Fix Only the Specific Problem Do not restart the whole project if the first pass is close. Fix the specific issue. If the deck focuses on the wrong metrics, ask Claude to rebuild around the metrics that best explain performance and de-emphasize vanity metrics. ```text Rebuild this deck around the metrics that best explain performance. De-emphasize vanity metrics and explain why the chosen KPIs matter. ``` If the slides are too busy, ask for cleaner hierarchy, fewer words per slide, and stronger contrast. If the charts are weak, ask for comparisons between top and bottom performers, pattern clusters, and visualizations that connect choices to performance. ```text Tighten the visual style. Make this feel like a polished strategy presentation, not a generic AI deck. Use cleaner hierarchy, fewer words per slide, and stronger contrast. ``` ```text Add more useful charts. Compare top and bottom performers, show pattern clusters, and visualize which choices correlate with better performance. ``` In the test run, Claude Design produced a strong deck because the starting prompt asked for best practices and recommendations. It pulled in thumbnails, compared winners and underperformers, and created slides closer to do-this-not-that than a basic analytics recap. ![Generated strategy deck key findings slide](https://tru-images.b-cdn.net/guide-assets/189/05-claude-design-generated-deck-key-findings.png) ### Pro tip > If your team or company has a design system uploaded in Claude Design, you can use it here to automatically format the deck with your logos, fonts, and color scheme. ### Guide: Turn Product Photos Into Scroll-Stopping Video Content > This guide teaches you how to turn a single product photo into a cinematic social media clip using Runway’s video generation tool. - Canonical page: https://app.therundown.ai/guides/turn-product-photos-into-scroll-stopping-video-content - Machine-readable page: https://app.therundown.ai/guides/turn-product-photos-into-scroll-stopping-video-content.md - Difficulty: beginner - Featured tool or company: Runway - Categories: Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2026-02-18 - Updated: 2026-07-23 ### Overview #### The Rundown This guide teaches you how to turn a single product photo into a cinematic social media clip using Runway’s video generation tool. You will produce a short video with camera movement, depth, and atmosphere; once you generate it in Runway, the clip takes about 15–30 seconds to create. #### Who This Is Useful For - E-commerce brands and product marketers who have product photos but no budget for video shoots and need video content for social, ads, and product pages - Real estate agents who want to animate listing photos into walkthrough-style teasers they can send to clients or post on Zillow - Content creators and personal brands who need video for TikTok and Reels but only have still images to work with #### What You Will Build You will build a cinematic video clip from a single product photo that is ready to post on social platforms. You will also learn how to use an AI assistant to write Runway motion prompts so you do not have to start from a blank text box. - A reusable product-photo-to-video workflow - A Runway-ready motion prompt generated with Claude - A short social clip with subtle motion such as panning, zooming, dolly movement, atmosphere, or animated light effects #### What You Need - A Runway account; the free plan is fine, and Standard is $15/month - A product photo or another still image with a clear subject - Claude or another AI assistant to help write the video prompt - Optional: CapCut or another video editor for text overlays, music, or voiceover - Optional: Suno if you want to pair the clip with a custom jingle #### Going Further - After creating one clip, reuse it across Instagram Reels, TikTok, product pages, email campaigns, or longer videos. Once you are comfortable with the one-photo flow, batch 10–20 product photos in Claude Cowork, generate Runway prompts for all of them, and run them through Runway in one session to build a library of video content ready to post. ### Step 1 Choose a starting product photo Start with a product photo you already have. This guide is intended for physical product shots, but you can also experiment with headshots, food photography, interior photos, event venues, or anything else with a clear subject. > Pro tip: Runway uses the photo you upload as the first frame of the video, so pick an image that works well with zooms, pans, and animated light effects. ![image-2069](https://tru-images.b-cdn.net/guide-assets/69/69964f1af6570adbc67939f7/image-2069.png) In the example, the starting image is a coffee grounds product photo because the light beams and steam from the mug can be animated. ### Step 2 Generate a Runway prompt with Claude Instead of writing the motion prompt from scratch, use an AI assistant to generate a few concepts. The source guide recommends Claude because it understood what Runway excels at and wrote strong prompts. ```text Write me a brief video prompt to use in Runwayml.com to generate a social media video based on this product photo. Add animations and/or camera movement if applicable. ``` ### Step 3 Create the video in Runway Go to runwayml.com > Tool > Video, then add your starting frame and the prompt you generated. > Pro tip: Set the time to 5 seconds and use Gen-4 Turbo to save tokens. ![image-2068](https://tru-images.b-cdn.net/guide-assets/69/69964f0bf6570adbc6793188/image-2068.png) Hit generate. Runway will create a short video clip in about 15–30 seconds. Play it back and compare it to the original still photo. Look for subtle, realistic motion. The source notes that Runway keeps panning, zooming, and dolly moves clean, which makes the output usable for real content. > Pro tip: Try experimenting with different product photos rather than several prompts on the same photo. Some photos may work better than others. ### Step 4 Publish or polish the clip Once you have your clip, drop it straight into your content. Post it as an Instagram Reel or TikTok, use it on your product page or in an email campaign, or make it the opening shot of a longer video. If you want to polish it, bring it into CapCut or another video editor. Add text overlays, music, or a voiceover. You can also pair it with a custom jingle from Suno to create a more complete video ad. ### Pro tip > Runway uses your uploaded photo as the first frame, so choose an image that works well with zooms, pans, and animated light effects. ### Guide: Turn Prompts Into Content With Claude Code + Higgsfield > Connect Higgsfield to Claude Code with the Higgsfield CLI, then send one image prompt to several AI image models at once. - Canonical page: https://app.therundown.ai/guides/turn-prompts-into-content-with-claude-code-higgsfield - Machine-readable page: https://app.therundown.ai/guides/turn-prompts-into-content-with-claude-code-higgsfield.md - Difficulty: beginner - Featured tool or company: Higgsfield - Categories: Content & Creative, Coding & App Development, Marketing & Growth - Instructors: Billy Howell - Published: 2026-05-14 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to connect Higgsfield to Claude Code with the Higgsfield CLI, then use Claude Code to send one image prompt to several AI image models at once. #### Who This Is Useful For - Content creators who want to test a visual idea across several AI image models - Marketers who need fast campaign concepts, thumbnails, product shots, or social images - Claude Code users who prefer running creative workflows from the terminal - AI power users who want one agent to generate, download, name, and organize outputs #### What You Will Build ![Claude Code running six Higgsfield image models from one prompt.](https://tru-images.b-cdn.net/guide-assets/local-210/turn-prompts-into-content-with-claude-code-higgsfield/01-claude-code-running-six-higgsfield-models.png) You will set up the Higgsfield CLI, connect it to Claude Code, and run one image idea through six different image models. The finished output is a small local folder with generated images named by model, plus a simple comparison file you can use to pick the best direction. #### What You Need - Claude Code - A Higgsfield account, free or paid - Node/npm installed - A terminal - One image prompt you want to test You can run this directly in your terminal or from the Claude Code desktop app. The recording used Claude Code and the Higgsfield CLI because the MCP path had authentication friction, while the CLI worked right away. #### Going Further - Add this setup to OpenClaw or Hermes Agent. - Try the same model-comparison workflow with short clips using Higgsfield video generation. - Use Claude Code as the operator, Higgsfield as the creative engine, and your terminal as the place where prompts turn into organized content assets. ### Step 1 Install the Higgsfield CLI Go to higgsfield.ai/cli. Higgsfield gives you three setup commands. ![Higgsfield CLI install steps.](https://tru-images.b-cdn.net/guide-assets/local-210/turn-prompts-into-content-with-claude-code-higgsfield/02-higgsfield-cli-install-steps.png) Run them in a fresh project folder: ```text npm install -g @higgsfield/cli higgsfield auth login npx skills add higgsfield-ai/skills ``` The second command opens a browser so you can sign in with your Higgsfield account. A free account works for testing. Paid plans and extra credits are useful once you start generating more assets. When you run the skills command, choose Claude Code when prompted. You can install the skill for the current project or globally. For a first test, project-level install is cleaner because it keeps the experiment contained. Pro tip: if setup gets confusing, ask Claude Code to help you install it. It can check whether Node and npm are installed, run higgsfield --help, and explain any terminal errors. ### Step 2 Ask Claude Code to Inspect Higgsfield Once the CLI is installed, open Claude Code in the same folder. Ask it to inspect the Higgsfield CLI page and the local command: ```text Look at https://higgsfield.ai/cli and inspect the local Higgsfield CLI. Confirm the install worked, show me the command pattern, and list the available image models. Do not generate anything yet. ``` Claude Code should check the CLI before spending credits. ![Authenticated Higgsfield CLI model list in Claude Code.](https://tru-images.b-cdn.net/guide-assets/local-210/turn-prompts-into-content-with-claude-code-higgsfield/03-higgsfield-cli-authenticated-and-model-list.png) This is where the workflow starts to click. Claude Code can see the Higgsfield command, understand the available model names, and decide which model IDs to use in later commands. If Claude Code lists video models too, tighten the instruction and tell it to use image models only for this test. ![Claude Code listing Higgsfield image model options.](https://tru-images.b-cdn.net/guide-assets/local-210/turn-prompts-into-content-with-claude-code-higgsfield/04-claude-code-image-model-options.png) ### Step 3 Run One Prompt Through Six Models Now give Claude Code one image prompt and ask it to run a real comparison. Use this prompt: ```text Use the Higgsfield CLI to generate this prompt with six different image models: "A retro 1970s travel photo of Mount Fuji at sunrise, soft film grain, warm colors, vintage postcard composition." Create a `higgsfield-model-test` folder. Save each output with the model name. Create `comparison.md` with the model, command, output file, and quick notes. Use image models only. Download the final image files locally when each generation finishes. ``` That last line matters. In testing, the first generation returned a hosted image link. If that happens, just tell Claude Code to download the image when it is done. ![Higgsfield generated image download in the terminal.](https://tru-images.b-cdn.net/guide-assets/local-210/turn-prompts-into-content-with-claude-code-higgsfield/05-higgsfield-generated-image-download.png) The useful part is not just that Higgsfield can generate an image. The useful part is that Claude Code can run the commands, wait for results, save files, and keep the comparison organized. ![Generated images saved locally in a Higgsfield model test folder.](https://tru-images.b-cdn.net/guide-assets/local-210/turn-prompts-into-content-with-claude-code-higgsfield/06-generated-images-folder.png) Use the saved outputs to compare framing, texture, color, and realism across models before choosing a direction. ### Step 4 Pick a Winner and Iterate After the six images are saved, ask Claude Code to review the folder and summarize the differences. Use a prompt like this: ```text Review the images in `higgsfield-model-test`. Update `comparison.md` with: - the strongest image overall - the best image for social - the best image for a newsletter hero - what each model did well - what prompt change we should try next ``` Then run one more focused pass. For example, if one model nailed the color but another handled composition better, ask Claude Code to combine those notes into a tighter prompt and test two or three finalists. You can also use this as a cost-control workflow. Test rough concepts on cheaper models first, then send the strongest prompt to a higher-quality model once the direction is clear. ### Pro tip > Use cheaper models for rough concept testing, then send the strongest prompt to a higher-quality model once the direction is clear. ### Guide: Turn SOP Training Documents Into Talking-Head Training Videos > This guide shows you how to turn onboarding docs and SOPs into AI avatar training videos using Claude or ChatGPT and Synthesia. - Canonical page: https://app.therundown.ai/guides/turn-sop-training-documents-into-talking-head-training-videos - Machine-readable page: https://app.therundown.ai/guides/turn-sop-training-documents-into-talking-head-training-videos.md - Difficulty: intermediate - Featured tool or company: Synthesia - Categories: Business Operations & Leadership, Content & Creative - Instructors: Billy Howell - Published: 2026-02-12 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to turn onboarding docs and SOPs into AI avatar training videos using Claude or ChatGPT and Synthesia. You will convert written documents into video scripts, generate talking-head videos, and build the foundation for a reusable onboarding course library. Video generation takes about 10–25 minutes per video after setup. #### Who This Is Useful For - Ops managers who spend 4–6 hours every quarter walking new hires through the same processes. - Founders or team leads with SOPs in Google Drive that people do not read and recurring questions in Slack. - Course creators or consultants who want to turn written frameworks into video modules without investing in a production setup. #### What You Will Build ![image-2048](https://tru-images.b-cdn.net/guide-assets/69/698d36630c12468b0d66f8cc/image-2048.png) You will build a complete onboarding course library with 4–5 AI-narrated training videos. - Short training video scripts adapted from existing documents. - Synthesia talking-head videos narrated by an AI avatar. - MP4 backups or embed links for sharing the videos. #### What You Need - Synthesia account ($29/month). - 3–5 existing documents you want to convert. - Claude or ChatGPT to write scripts; a free account is fine. #### Going Further - Repeat the process with the other documents you pulled, then embed all of the videos into one Notion page or Google Doc to create a one-page onboarding hub. You can also use Synthesia templates to create interactive videos and quizzes, but those cannot be downloaded and must stay embedded, so you will need an active Synthesia subscription for them to keep working. ### Step 1 Build your script Pick 3–5 internal documents that new hires or clients always read. - Mission statement - Onboarding checklist - Tool setup guide - Team directory - Role-specific processes or workflows Drop each file into Claude or ChatGPT with this prompt: ```text Turn this into a two to three-minute training video script for an AI-generated avatar. Include timestamps and shots. Only include text overlays with bullets. The avatar can be seated, standing, head-on, etc. ``` > 💡 Pro tip: Break each document into simple concepts like “How we use Slack” or “Notion setup guide.” ### Step 2 Save the script Copy the final script from Claude or ChatGPT. Open a text editor, such as VS Code, Notepad, or TextEdit, and paste it in. Save the file as [your-topic]-script.txt. Synthesia requires either a TXT or PDF file, so do not skip this step. ### Step 3 Create the video in Synthesia Go to Synthesia.io and log in to your account. The source guide found that the Starter plan was good for 8–9 videos. - Click Create New Video. - Choose Create from AI. The source guide found this works better than using templates. - Drag and drop your .txt script file into the upload area. - Set the duration to match your script. Click the Objective and Audience buttons in the sidebar. This is important for getting a useful result. - Objective: Describe what the video teaches, such as “Onboarding training for new team members on our communication workflow.” - Audience: Describe who the viewer is, such as “New team members joining our marketing department.” - Language: English, or your chosen language. ![image-2049](https://tru-images.b-cdn.net/guide-assets/69/698d36860c12468b0d67014c/image-2049.png) ### Step 4 Review scenes and add your logo Pick a clean-looking template. Any professional-looking option should work, regardless of how it is named. Click Create Outline. Synthesia will lay out your scene outlines with descriptions and bullets. Review the outline carefully. Once the video is generated, you will not be able to edit it. ![image-2050](https://tru-images.b-cdn.net/guide-assets/69/698d369c0c12468b0d6709cf/image-2050.png) Click Continue when you are ready. You will see a detailed outline with preview images. ![image-2051](https://tru-images.b-cdn.net/guide-assets/69/698d36af0c12468b0d671211/image-2051.png) > 💡 Pro tip: If you click Open in Editor, you can swap in your own logo into the template. ![image-2052](https://tru-images.b-cdn.net/guide-assets/69/698d36cb0c12468b0d671a66/image-2052.png) ### Step 5 Generate and export the video When you are happy with the outline and scenes, click Generate. Remember: once you click Generate, those tokens are spent and you cannot edit the video anymore. ![image-2053](https://tru-images.b-cdn.net/guide-assets/69/698d36dd0c12468b0d672297/image-2053.png) The video will take 10–25 minutes to generate. In the source guide’s experience, generation time usually came in under Synthesia’s prediction. When your video is ready, you have two options: - Download the video as an MP4. - Create an embed link. Always download the MP4s as backups. Using MP4s instead of embed links also means you can cancel your subscription and save money. > 💡 Pro tip: If you are embedding in Notion, Synthesia videos show up with chapters and captions, so your team can watch right in Notion without leaving. ### Pro tip > Break each document into simple concepts like “How we use Slack” or “Notion setup guide.” ### Guide: Turn Your ChatGPT Subscription Into a Team of Useful Agents With Raft > In this guide, you will learn how to turn your scattered Claude and ChatGPT projects into a persistent workspace for AI agents. - Canonical page: https://app.therundown.ai/guides/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft - Machine-readable page: https://app.therundown.ai/guides/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft.md - Difficulty: beginner - Featured tool or company: Raft - Categories: Automation & Agents, Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-07-29 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to turn your scattered Claude and ChatGPT projects into a persistent workspace for AI agents. You’ll create a Raft server, connect its onboarding agent to your Codex or Claude Code subscription, then build a three-agent team that can tackle almost any project together. #### Who This Is Useful For - Anyone with useful Claude or ChatGPT projects scattered across separate chats, folders, or computers - Teams that want collaborative agents they can chat with in a Slack-like environment - People who want to create useful agents quickly but are intimidated by OpenClaw #### What You Will Build ![Raft CRO project channel with lead, researcher, and maker tasks visible](https://tru-images.b-cdn.net/guide-assets/local-401/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft/01-finished-cro-agent-team.png) The finished CRO channel keeps the three-agent team, task cards, and handoffs in one place. You will build one focused project channel with three persistent agents: - An orchestrator, or team lead, that turns your brief into clear tasks and keeps the project moving - A researcher that gathers source-backed inputs, examples, and constraints - A maker that turns approved inputs into the final deliverable #### What You Need To Get Started - A Raft account and server on the Free plan ($0). - One computer with either Codex CLI or Claude Code installed and signed in. - An existing project that you want to work on with the team of agents. #### Going Further Ask Cindy what else you can do with your current setup. You can explore automations, create more project channels, or work through the Raft documentation (https://docs.raft.build/welcome/) to learn the task, reminder, and agent features. We really like Raft and may make a part two that goes deeper into automations, tasks, and other ways to run an agent team. ### Step 1 Connect Raft to Codex CLI or Claude Code Raft is the workspace where you chat with and manage your agents. The agents run in the background on your computer, and Codex CLI or Claude Code powers them through your existing account. Go to raft.build (https://raft.build), create an account, and choose Create New Server. Give the server a recognizable name so you can tell which workspace and connected computers belong to it. ![Raft Choose server screen with the Create New Server button](https://tru-images.b-cdn.net/guide-assets/local-401/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft/02-server-create-form-cropped-final.png) Choose Create New Server to make the workspace where your agents will collaborate. Raft will show a Connect a computer screen with two commands. The first installs the Raft Computer service. The second connects that computer to your new server. Open Terminal on the computer that will run your agents, then paste and run both commands exactly as shown. Approve the device login in your browser if prompted. Wait until the computer appears online and Raft detects the installed runtime. ![Raft Connect a computer screen with install and setup commands](https://tru-images.b-cdn.net/guide-assets/local-401/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft/03-connect-computer-setup.png) Raft gives you one command to install Raft Computer and another to connect it to your server. Choose either Codex CLI or Claude Code as the runtime for Cindy, Raft's onboarding assistant. That runtime powers Cindy and the agents you create later. When the setup is working, you will see a green or yellow circle next to an agent's name in Raft. That means the agent is connected or working. > Pro tip: You can start, stop, or restart Raft Computer from Terminal at any time. Replace /your-server-slug with the slug shown in your server's setup command. ```text raft-computer start /your-server-slug raft-computer stop raft-computer restart /your-server-slug ``` If you already run a production Raft server on the same Mac, do not migrate, adopt, or reconnect its existing Computer connection. Use the new server's generated setup command to create a new, separate Computer connection for the second server. ### Step 2 Create One Channel for Each Project or Workstream For this guide, our project is a CRO audit. But you can take any project or workstream you are already working on and give it its own channel with a three-agent work team. Remember: - One channel per project or workstream - A three-agent team for each channel ### Step 3 Build the Orchestrator, Researcher, and Maker Team In #onboarding-owner, describe the project to Cindy and ask for one project channel with three agents. You can use this structure: ```text Create a public project channel called [project name]. Create a three-agent team for this project: - An orchestrator or team lead that understands the brief, scopes the work, and coordinates handoffs - A researcher that gathers the inputs, evidence, examples, and constraints - A maker that turns approved research into the requested deliverable Keep the human as the final reviewer. ``` If Cindy has admin permission, she can create the channel and agents. Otherwise, a server owner or admin must create them or approve the changes. You can also create an agent manually from the target computer's page by choosing Create, then setting its name, description, runtime, and model. ![Raft Create Agent dialog showing computer, name, description, runtime, and model](https://tru-images.b-cdn.net/guide-assets/local-401/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft/04-create-agent-dialog.png) Create each role on an online computer, then choose its runtime and model. Give every role a narrow job: - The orchestrator owns the plan, identifies missing information, assigns clear tasks, and brings decisions back to you. - The researcher owns inputs. It should return evidence, examples, and uncertainty, not quietly make the final deliverable. - The maker owns outputs. It should build from the approved brief and research instead of starting from a vague request. After the agents exist, open the project channel's member panel and add all three. New agents join #all automatically, but they will not see messages from every project channel until you add them. ### Step 4 Start With the Researcher Start with the researcher. Its job is to find everything the maker agent will need before it creates the deliverable. In our CRO project, the researcher reviews the website pages, chooses a CRO framework, and audits the pages. It then passes those findings to the maker agent. For your project, tell the researcher to study the process for building the deliverable and gather the inputs the maker will need. If you already have project files, examples, or a documented process, point the researcher to those and ask it to use them. Post that research request in the project channel and send it as a task. The researcher can claim the task, keep its work in the task thread, and move it to In review when the inputs are ready. ![Raft project channel task view with an in-progress agent task](https://tru-images.b-cdn.net/guide-assets/local-401/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft/05-project-channel-task.png) The task thread keeps the research request, progress, and finished inputs together for review. > Pro tip: Because your agents run through Codex CLI or Claude Code on your computer, they can use the project files and skills available to that runtime. Tell the researcher to review the relevant ones first and save what it learns to its agent memory for future work. Review the research before the maker starts. If something is weak or missing, reply in the same task thread so the correction stays with the project. ### Step 5 Create a Task for the Maker Once you approve the research, open the project channel's Tasks tab and click New Task. Tell the maker exactly what to create, point it to the approved research, and assign the task to the maker agent. > Pro tip: Cindy, the onboarding assistant, can work as your first orchestrator while you learn Raft. Tell Cindy what you want done and ask her to divide the work between your agents. You can also add your team lead to the project channel, explain the finished result you want, and ask it to divide the work between the researcher and maker. Open the channel's Tasks tab at any time to see the work arranged as a Kanban board under Todo, In progress, In review, and Done. ![Raft CRO channel Tasks board with Todo, In progress, In review, and Done columns](https://tru-images.b-cdn.net/guide-assets/local-401/turn-your-chatgpt-subscription-into-a-team-of-useful-agents-with-raft/06-agent-task-kanban-board-cropped.png) The Tasks tab gives you a Kanban view of every task the agents are working on. Raft will not automatically hold one task until another finishes. If the maker must wait for the researcher, say that in the task thread and leave the maker's task unclaimed until you approve the research. ### Step 6 Review the Team's Work Open each task in In review and check the research or deliverable. If something needs another pass, reply in that task's thread with the correction. When you approve the result, move the task to Done. Then tell the team lead what you want to tackle next: ```text Make us a plan for what to audit next and assign tasks to the appropriate teammates. ``` The team lead can plan the next round, but you still make the final decisions and approve any external action. ### Guide: Use 60% Less Fable Tokens With This Orchestrator Setup > In this guide, you will learn two ways to stop burning Fable tokens on grunt work in Claude. - Canonical page: https://app.therundown.ai/guides/use-60-less-fable-tokens-with-this-orchestrator-setup - Machine-readable page: https://app.therundown.ai/guides/use-60-less-fable-tokens-with-this-orchestrator-setup.md - Difficulty: beginner - Categories: Automation & Agents, Coding & App Development - Instructors: Billy Howell - Published: 2026-07-10 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn two ways to stop burning Fable tokens on grunt work in Claude. Set Fable up as the planner and reviewer, then send browsing, coding, and research to Codex or a lower-cost Claude model so premium tokens are saved for the calls that need stronger judgment. #### Who This Is Useful For - Developers and builders who want Fable's judgment on a project without spending premium tokens on every file read, retry, and terminal task. - Operators and researchers who use Claude for long, tool-heavy work and need a cleaner way to separate planning from execution. - Anyone with a Claude paid plan who wants a worker to keep moving on routine work and call Fable only when the decision is genuinely hard. #### What You Will Build You will build a simple model-routing setup: Fable owns the plan, risky calls, and final review, while a cheaper worker handles the token-heavy loop. The result is a self-checking AI worker that knows when to escalate instead of pulling Fable into every step. ![Fable coordinating Codex background agents while it keeps the planning and review work](https://tru-images.b-cdn.net/guide-assets/local-333/use-60-less-fable-tokens-with-this-orchestrator-setup/01-fable-delegates-to-codex-agents.png) #### What You Need To Get Started - A Claude paid plan and the Claude desktop app - Claude Code, updated to the current version - A ChatGPT plan for the Codex-plugin route, or a lower-cost Claude model for the /advisor route #### Going Further Save the delegation instruction as a Claude Code project instruction or reusable snippet. Then make a few worker templates for the work you repeat: a repo change, a research brief, a QA pass, or a content draft. Keep the same routing rule in each one: Fable plans, judges risk, and reviews the handoff; the worker does the bulk execution. For a quick, low-risk task, skip the orchestrator setup and use the cheaper model directly. Bring Fable in when the work needs stronger judgment, not simply because it needs more tokens. ### Step 1 Check Your Fable Usage Before a Long Task Open the Claude desktop app and make sure it is current before you start: go to Settings > Check for updates and install anything available. The current app has Home and Claude Code tabs. Then click your name and go to Settings > Usage. This screen shows your weekly limit and your Fable limit. Check it before you let Fable run a large task, especially one with browsing, coding, file reading, or repeated retries. In our test, Fable consumed the remaining allowance quickly during a short session. ![Claude desktop app Settings Usage page showing weekly and Fable usage limits](https://tru-images.b-cdn.net/guide-assets/local-333/use-60-less-fable-tokens-with-this-orchestrator-setup/02-claude-usage-settings.png) > Pro tip: Keep Fable out of routine retries. Use it for the plan, risky calls, and final review, then let the worker handle the token-heavy loop. ### Step 2 Install Codex as Your Execution Worker If you have a ChatGPT plan, you can use Codex as Claude's workhorse. This is the official OpenAI Developers (Codex) plugin for Claude Code—not a third-party integration. Open a terminal in a new project folder, start a Claude Code session with claude, then follow the official OpenAI plugin setup instructions (https://developers.openai.com/learn/developers-codex-plugin) to add its marketplace and install the plugin: ```text /plugin marketplace add openai/openai-developers-for-claude /plugin install openai-developers@openai-developers ``` Start a new Claude Code session after the install. If you had the Claude desktop app open, restart it too. Then run the Codex setup check from Claude Code to confirm the integration is available. Claude can now send companion tasks to Codex instead of spending Fable tokens on routine coding, browsing, and website extraction. ![Claude Code terminal showing the Codex plugin installation and reload steps](https://tru-images.b-cdn.net/guide-assets/local-333/use-60-less-fable-tokens-with-this-orchestrator-setup/03-codex-plugin-install.png) ### Step 3 Give Fable Clear Delegation Rules Fable should be the orchestrator, not the person doing every mechanical step. Ask it to inspect the available Codex skills and route the work that needs execution to the right one. A short instruction is enough to set the behavior: ```text Use your Codex skills to save your own tokens. Keep planning, complex decisions, and final review with Fable. Delegate coding, browsing, file extraction, routine checks, and retries to Codex. Escalate back to Fable only when the worker hits ambiguity, a high-risk decision, or a stuck loop. ``` You can see the resulting work in Codex as Codex companion task threads. For coding work, add the Codex-first routing skill to Claude through Customize Skills > Add > Upload Skill. It tells Fable which coding tasks belong with Codex and which are worth keeping in the expensive judgment lane. ![Fable's response and Codex background tasks showing the orchestrator delegating work to Codex agents](https://tru-images.b-cdn.net/guide-assets/local-333/use-60-less-fable-tokens-with-this-orchestrator-setup/04-fable-codex-agent-hand-off.png) ![Codex-first skill route document showing which tasks Claude should delegate to Codex and which it should keep](https://tru-images.b-cdn.net/guide-assets/local-333/use-60-less-fable-tokens-with-this-orchestrator-setup/05-codex-first-skill-route.png) Ask the worker to return evidence, changed files, checks performed, and open questions. That gives Fable a small review package instead of a long stream of work-in-progress commentary. ### Step 4 Use Advisor Mode When You Do Not Have Codex You can use the same pattern without a separate Codex subscription. /advisor is a built-in Claude Code skill, but it works when you launch Claude Code in your terminal—not from the Claude desktop app interface shown earlier. Run claude in a terminal, use /model to select a lower-cost worker model, then use /advisor to set Fable as the advisor. Anthropic's advisor tool documentation (https://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-tool) explains the same executor-plus-advisor pattern. The worker continues with the task until it reaches a complex decision, an ambiguous failure, or a loop that is not making progress. At that point, it asks Fable for guidance and resumes the work. The advisor runs server-side and uses additional tokens, so use it for decision points rather than every turn. The recorded setup used a lower-cost model as the main worker and Fable as the advisor. That is the useful split: the worker handles the volume, while Fable supplies the stronger judgment only when it can change the outcome. > Pro tip: Advisor mode is a lighter-weight routing option, not a guarantee of a fixed savings percentage. How much it saves depends on the task and how often the worker escalates. ### Guide: Use ChatGPT to Build a Custom Mac Shortcut System > In this guide, you will learn how to use ChatGPT Work to set up time-saving keyboard shortcuts and automations for you on your Mac. - Canonical page: https://app.therundown.ai/guides/how-to-build-a-custom-mac-shortcut-system-for-faster-workflows - Machine-readable page: https://app.therundown.ai/guides/how-to-build-a-custom-mac-shortcut-system-for-faster-workflows.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: AI, Productivity, Automation - Instructors: Billy Howell - Published: 2026-08-11 - Updated: 2026-08-11 ### Overview #### The Rundown In this guide, you will learn how to use ChatGPT Work to set up time-saving keyboard shortcuts and automations for you on your Mac. Don’t worry, you don’t need to install anything to get started. #### Who This Is Useful For - Anyone who opens the same apps all day and is tired of using Spotlight to hunt for them each time. - People who work in distinct modes like recording, analytics, or writing and want one key to open the right setup. - Mac users with multiple browser accounts who want a faster way into the right Safari or Chrome profile. #### What You Will Build ![Shortcut cheat sheet](https://tru-images.b-cdn.net/guide-assets/local-423/how-to-build-a-custom-mac-shortcut-system-for-faster-workflows/01-shortcut-cheat-sheet.png) This is the one-page reference you’ll end with. You will build a personal shortcut system in Apple Shortcuts, plus a one-page reference that tells you what every key does and where it lives. The finished setup can jump you into the apps, browser profiles, and work modes you use most, without opening Spotlight or hunting through browser tabs. #### What You Need To Get Started - A Macbook running macOS Monterey (version 12) or later. - The ChatGPT desktop app with access to "Work" mode. #### Going Further If you want to go further, ask ChatGPT Work for the top 10 time-saving automations it would set up for you. It can set them up in Automator and pull them into Shortcuts. It’ll just do it. ### Step 1 Build Your Shortcut Plan With ChatGPT Open the ChatGPT desktop app, start a new chat, and switch to Work. Then give it a prompt based on your real workflow: ```text Based on what you know about me and my workflow, suggest 10 time-saving hotkeys we can set up in Apple Shortcuts. I’m interested in opening specific apps and profiles based on what I need them for or what mode of work I’m going into. ``` Tell it the apps you use most, the browser profiles you need, and the work modes you go into. For example, a recording mode could open your microphone software, recording software, camera software, and notes. An analytics mode could open your regular dashboards with one keystroke. If you have URLs you check every week, paste those in too. That gives ChatGPT enough to recommend shortcuts for the apps, URLs, profiles, and work modes you actually use. ![ChatGPT shortcut plan](https://tru-images.b-cdn.net/guide-assets/local-423/how-to-build-a-custom-mac-shortcut-system-for-faster-workflows/02-chatgpt-shortcut-plan.png) > Pro tip: ChatGPT Work can set up shortcuts that open tabs in specific Safari or Chrome profiles, so you can jump straight into the right account. ### Step 2 Choose Your Shortcuts Start with the shortcuts that will save you the most time, then add more once the first set is working. Say: ```text Keep the highest-frequency shortcuts and avoid macOS or app conflicts. ``` > Pro tip: On most Mac keyboards, Control + Option is a great choice for custom shortcuts because it is rarely used by system or app hotkeys. You are less likely to overwrite an existing command. Focus on the apps, URLs, profiles, and modes that cost you a few seconds over and over again. The cheat sheet handles recall while you add more later. ### Step 3 Set Up ChatGPT Work Before you do anything, make sure Computer Use is on. Go to Settings > Computer Use, then toggle Any App on. Go back to the approved list and say: ```text Set these up in Apple Shortcuts. Do not change anything outside this list. Test each shortcut. ``` Only approve the shortcut list you actually reviewed, and grant the permissions the setup needs. You are asking it to make the shortcuts you approved, not to rearrange the rest of your computer. ### Step 4 Build the Shortcuts Ask ChatGPT Work to build the approved shortcuts in Apple Shortcuts. In our test, we started with the work modes, then tested them with keystrokes. Open All Shortcuts to see the system ChatGPT just made. ![Shortcuts library](https://tru-images.b-cdn.net/guide-assets/local-423/how-to-build-a-custom-mac-shortcut-system-for-faster-workflows/03-shortcuts-library.png) Open Apple Shortcuts when it finishes and press play on each shortcut to test it. Use Quick Actions to bind or change its keyboard shortcut. Open any tile to inspect the actions behind it. ![Shortcut setup](https://tru-images.b-cdn.net/guide-assets/local-423/how-to-build-a-custom-mac-shortcut-system-for-faster-workflows/04-shortcut-setup.png) Test each shortcut from another app. The result should be simple: press the key and land in the right app, browser profile, or work mode. If one lands somewhere unexpected, fix that shortcut before moving on. ### Step 5 Make Your One-Page Cheat Sheet Once the shortcuts work, ask ChatGPT for the cheat sheet: ```text Write me a one-page cheat sheet of all the shortcuts you set up, where they live, and what’s left to do. ``` Save the PDF somewhere easy to find. It is your quick reference while the key combinations become familiar, and it keeps any unfinished setup separate from the shortcuts that are already working. ### Guide: Use Claude Cowork + Obsidian To Triple Your Output > Learn a simple system for planning and managing your workdays with Obsidian and Claude Cowork. - Canonical page: https://app.therundown.ai/guides/use-claude-cowork-obsidian-to-triple-your-output - Machine-readable page: https://app.therundown.ai/guides/use-claude-cowork-obsidian-to-triple-your-output.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Productivity & Knowledge Work, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-03-01 - Updated: 2026-07-30 ### Overview #### The Rundown Learn a simple system for planning and managing your workdays with Obsidian and Claude Cowork. You will set up an Obsidian vault that Claude can read, then use it to create daily plans automatically so each morning’s plan is ready in Obsidian. #### Who This Is Useful For - Anyone drowning in notes apps who writes things down but never checks them again. - Founders and freelancers who need a system but dislike rigid project management tools. - Anyone who tried to-do apps and stopped using them because maintaining the app felt like more work than doing the work. #### What You Will Build You will build a daily planning system that runs on Obsidian for storing and viewing your notes, and Claude Cowork for reading your vault and writing daily plans automatically. - An Obsidian vault with daily-notes/, projects/, and templates/ folders. - A my-workflow.md file that tells Claude your schedule, priorities, work style, and tools. - A daily-note-template.md file that defines the format for each day’s note. - Project files that Claude can use to understand what is due soon. - Scheduled Claude Cowork tasks for morning planning and closing the day. ![image-2089](https://tru-images.b-cdn.net/guide-assets/69/69a484c919b41f285b3398f3/image-2089.png) #### What You Need - Obsidian, available as a free download. - Claude desktop app with a Pro plan or higher. - Google Calendar connected to Claude. - The starter files from this guide: my-workflow.md and daily-note-template.md. - At least one active project to track in a project markdown file. #### Going Further - Before your end-of-day task runs, ask Claude to interview you about your day and update your notes. This helps capture anything you forgot to jot down during the day. ### Step 1: Create your Obsidian vault Go to obsidian.md and download the free Obsidian app. It works well with Claude because it stores everything as plain markdown files in a folder on your computer, not in the cloud. Open Obsidian and click Create new vault. Pick a name and a location. This creates a real folder on your computer that you will point Claude Cowork to later. Inside your vault, create three folders: - daily-notes/ - projects/ - templates/ That is the whole structure. If you look in Finder or File Explorer, you will see that Obsidian created these folders on your computer. > This vault is for tracking your work, not doing your work. If you have a big project that needs its own space, like a course, product launch, or client engagement, create a separate vault for it. That way Claude Cowork does not get overwhelmed reading files it does not need. ### Step 2: Add your workflow, template, and project files Download the two starter files from this guide and drop them into your vault. Put my-workflow.md in the root of your vault. This is the file that turns Claude from a generic AI into your assistant. It tells Claude your schedule, how you work best, your current priorities, and what tools you use. Edit it to match your actual life. ![image-2090](https://tru-images.b-cdn.net/guide-assets/69/69a484d8dc8de198190f12f3/image-2090.png) Put daily-note-template.md in the templates/ folder. This tells Claude how to format each day’s note. ![image-2091](https://tru-images.b-cdn.net/guide-assets/69/69a484e9dc8de198190f1b67/image-2091.png) Then add at least one project file in projects/. Use one markdown file per active project. Keep it simple: what the goal is, what is done, and what is left. ![image-2092](https://tru-images.b-cdn.net/guide-assets/69/69a484fb19b41f285b33a182/image-2092.png) ### Step 3: Have Claude create your daily plan Open the Claude desktop app and start a Cowork session. When it asks which folder to work with, point it at your Obsidian vault folder. Now give Claude Cowork this morning planning prompt: ```text Read my vault. Look at my-workflow.md to understand my schedule and priorities. Look at my project files to see what's due soon. Then create today's daily note in daily-notes/ with a plan for the day. Put the most important thing first. Flag anything that's overdue or urgent. ``` From here, work through your day in Obsidian. Check off tasks, add notes in the Notes section, and jot down anything that comes up. It is just a markdown file. ### Step 4: Automate your morning and end-of-day planning Running the prompt manually every morning works fine, but it is better when it runs automatically. In Claude Cowork, click Scheduled in the sidebar and create a new scheduled task. Create a morning plan task with this prompt: ```text Read my vault. Look at my-workflow.md to understand my schedule and priorities. Look at yesterday's daily note to see what I got done and what carried over. Look at my project files to see what's due soon. Then create today's daily note in daily-notes/ with a plan for the day. Put the most important thing first. Flag anything that's overdue or urgent. Use my Google Calendar connection to check meetings. ``` ![image-2093](https://tru-images.b-cdn.net/guide-assets/69/69a4850f19b41f285b33aa10/image-2093.png) Create a Close Day scheduled task with this prompt: ```text Read today's daily note. Look at what got checked off and what did not. Add an “End of Day” section summarizing what got done, what's carrying over to tomorrow, and any new tasks that came up in my notes today. Update the project files if any tasks were completed. ``` ![image-2094](https://tru-images.b-cdn.net/guide-assets/69/69a4852019b41f285b33b291/image-2094.png) ### Resources - [my-workflow.md](https://tru-images.b-cdn.net/guide-assets/69/69a4849319b41f285b338fe5/my-workflow.md) - [daily-note-template.md](https://tru-images.b-cdn.net/guide-assets/69/69a4849e19b41f285b33902b/daily-note-template.md) ### Pro tip > Use this vault for tracking your work, not doing your work. If a big project needs its own space, create a separate vault so Claude Cowork does not get overwhelmed reading files it does not need. ### Guide: Use Claude’s Chrome Extension to Quickly Book Business Trips > Set up Claude to control your browser and automate repetitive hotel research for business travel. - Canonical page: https://app.therundown.ai/guides/use-claudes-chrome-extension-to-quickly-book-business-trips - Machine-readable page: https://app.therundown.ai/guides/use-claudes-chrome-extension-to-quickly-book-business-trips.md - Difficulty: beginner - Featured tool or company: Claude - Categories: Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2025-12-30 - Updated: 2026-07-30 ### Overview #### The Rundown Set up Claude to control your browser and automate repetitive hotel research for business travel. You’ll create a Claude in Chrome shortcut that searches current hotel pricing and availability, then returns recommended hotels in a Google Sheet. Once launched, the hotel research should take about 10–15 minutes to come back. #### Who This Is Useful For - Anyone tired of opening dozens of tabs to compare hotel options. - Business owners booking lots of work travel. - People frustrated by outdated pricing and travel recommendations from LLMs. #### What You Will Build You’ll build a reusable hotel research workflow that uses Claude in Chrome to search Booking.com, apply your preferred filters, and populate a Google Sheet with the best hotel options. - A /hotel-picker Claude shortcut. - A Google Sheet for hotel research results. - A ranked list of hotel recommendations with fields like name, price per night, rating, distance to downtown, key amenities, and booking link. ![CleanShot 2025-12-23 at 16.00.40@2x](https://tru-images.b-cdn.net/guide-assets/69/695451e2ff0b27ac3474b01a/cleanshot-2025-12-23-at-16-00-40-2x.png) #### What You Need - Claude account, Pro or above. - Claude in Chrome extension. - Claude Desktop app, optional. - Google Sheets. #### Going Further - After creating the workflow, you can reuse the same approach for other browser tasks, schedule workflows, teach Claude tasks with screen and voice recordings, or send results into Notion instead of Google Sheets. ### Step 1: Set up Claude in Chrome - Navigate to Claude for Chrome in the Chrome Web Store. - Click “Add to Chrome” to install the Claude in Chrome extension. ![CleanShot 2025-12-30 at 16.28.02@2x](https://tru-images.b-cdn.net/guide-assets/69/695451fe15aa67ae62437b65/cleanshot-2025-12-30-at-16-28-02-2x.png) ### Step 2: Create a hotel research shortcut Before running hotel research, create a shortcut that handles the filtering automatically. - In the Chrome extension, type /shortcuts and click “Create Shortcut.” - Name it /hotel-picker. - Add this prompt: ```text When searching for hotels on Booking.com: 1. Property Type: Always use the “Property type” checkbox on the left sidebar to select “Hotels” only. This filters out boutique hotels, Airbnbs, vacation rentals, and full properties. Only show traditional hotels. 2. Distance from City Center: If the user specifies a distance requirement, such as “within 1 mile of downtown,” use the “Distance from city center” checkboxes on the left sidebar to select the appropriate range. 3. Property Rating: Default to 4 stars and up by using the property rating filter unless the user explicitly requests a different star rating. This ensures you’re showing reputable, quality hotels. 4. Review Quality: Find the top 10 hotels based on the user’s criteria and user reviews. Check both Booking.com reviews and cross-reference with TripAdvisor to verify quality and reputation. Prioritize hotels with high ratings, such as 8.5+ on Booking.com and 4+ stars on TripAdvisor, and recent positive reviews. 5. Search for modern, clean properties: When reviewing options, prioritize hotels that appear modern and well-maintained in photos. Look for amenities like rooftop bars, pools, fitness centers, and modern decor when available. Return the top 10 options sorted by best overall value considering price, location, reviews, and amenities. ``` ![CleanShot 2025-12-30 at 16.30.36@2x](https://tru-images.b-cdn.net/guide-assets/69/6954522715aa67ae62437bad/cleanshot-2025-12-30-at-16-30-36-2x.png) - Set “Start from” to booking.com. - Save the shortcut. ### Step 3: Set up your Google Sheet - Go to Google Sheets. - Create a new spreadsheet with the headers shown below. ![CleanShot 2025-12-30 at 16.32.01@2x](https://tru-images.b-cdn.net/guide-assets/69/69545275945731025244187f/cleanshot-2025-12-30-at-16-32-01-2x.png) - Copy the spreadsheet URL. ### Step 4: Test your shortcut ```text I’m traveling to [location] from [date]. Find hotels within 1 mile of downtown under [price per night]. I already have a spreadsheet set up at [YOUR SPREADSHEET URL HERE] — navigate to it and add a row for each hotel with: name, price per night, rating, distance to downtown, key amenities, and booking link. Sort by best value. ``` It should take 10–15 minutes to get your list of hotel recommendations back. ### Step 5 Best practices - Create a Chrome profile just for Claude for Chrome so you can control which logins it has access to. - Try not to click into the page while Claude is active in the tab. It can throw Claude off. You can continue working as normal in other tabs. - If Claude triggers anti-bot protections on a site, ask it to slow down and type out each letter. It will likely paste full sentences all at once to start. - Be patient. Manual tasks can take a few minutes for Claude to complete, but it is inexpensive to run because it uses Haiku by default. - You can set workflows to run automatically on a schedule, such as inbox cleaning or tracking flight prices. ### Step 6 Record tasks to teach Claude You can record your screen and voice to teach Claude to do a task for you. The source notes that text instructions have worked better, and that teaching Claude via recording still creates a text prompt afterward. ### Step 7 Use Notion instead of Google Sheets If you prefer Notion, you can use Claude Desktop and the Notion connector instead of sending results to a Google Sheet. - Open Claude Desktop, not just the Chrome extension. - Enable the Notion connector in Claude Desktop settings. - Give Claude permission to access your Notion workspace. - Replace the Google Sheet URL in your prompts with “create this in my Notion workspace.” - Claude will create pages and databases directly in Notion. ### Resources - [Use Claude’s Chrome Extension to Quickly Book Business Trips](https://tru-images.b-cdn.net/guide-assets/69/695453d127f706b7fd9bf7f9/use-claudes-chrome-extension-to-quickly-book-business-trips.pdf) ### Pro tip > You can use the extension in Chrome or from your Claude desktop app. If you are using it in the desktop app, make sure you have the model set to Haiku or you won’t be able to enable it. ### Guide: Use Codex /goal To Build a Fully Functional Game In One Prompt > In this guide, you will learn how to use Codex /goal to build and test a small browser game without nudging the agent every few minutes. - Canonical page: https://app.therundown.ai/guides/use-codex-goal-to-build-a-fully-functional-game-in-one-prompt - Machine-readable page: https://app.therundown.ai/guides/use-codex-goal-to-build-a-fully-functional-game-in-one-prompt.md - Difficulty: beginner - Featured tool or company: The Rundown AI - Categories: Coding & App Development, Automation & Agents - Instructors: Billy Howell - Published: 2026-05-29 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to use Codex /goal to build and test a small browser game without nudging the agent every few minutes. What you're really learning is how to give Codex a measurable finish line it can work toward on its own. #### Who This Is Useful For - Builders and creators who want to turn a rough idea into a working demo without managing every small coding step - Operators and business teams who want to learn how to give AI long-running tasks without burning tokens on vague back-and-forth - Anyone learning agent workflows who needs a simple exercise for defining success criteria, tests, and measurable outcomes #### What You Will Build You will build a small browser game called Endless Surf Carver. The player moves through waves, jumps cleanly through channels, earns points, and loses when the run breaks down. ![Endless surf carver gameplay](https://tru-images.b-cdn.net/guide-assets/local-214/use-codex-goal-to-build-a-fully-functional-game-in-one-prompt/01-endless-surf-carver-gameplay.png) #### What You Need To Get Started - A ChatGPT Plus or higher subscription with Codex access - Optional: the Codex app, which is useful when you want Codex to test a browser game visually #### Going Further Once you understand /goal, stop thinking about it as a game-building trick. Think of it as a way to delegate any task with a measurable finish line. The easiest test is this question: What metric would prove this task improved? For SEO, the metric might be a content score, keyword coverage, title clickability, internal link count, or missing schema. A strong goal could ask Codex to audit five pages, score each one against your SEO checklist, improve the lowest-scoring page, and rerun the score. For writing, the metric might be how similar a draft sounds to generic AI writing. You could give Codex your house style rules, ask it to flag phrases that sound AI-generated, rewrite the weakest sections, and report the before-and-after changes. For operations, the metric might be time saved, error rate, number of unresolved items, support tickets answered, or files processed correctly. Codex can audit the current state, make a small improvement, and verify the output against that metric. Here is a reusable pattern: ```text /goal Audit and improve [PROCESS_OR_PROJECT]. Success metric: - [ONE_MEASURABLE_SCORE_OR_CHECK] Acceptance criteria: - Inspect the current state. - Identify the top 3 issues hurting the metric. - Improve the highest-impact issue first. - Verify the metric again. - Summarize what changed, what improved, and what still needs human review. ``` That is the real lesson. /goal works best when you stop asking AI to "help with a task" and start giving it a clear outcome, a measurable check, and room to iterate until the result is good enough to review. ### Step 1 Enable Goals Open Terminal and run: ```text codex features enable goals ``` ![Enable goals terminal](https://tru-images.b-cdn.net/guide-assets/local-214/use-codex-goal-to-build-a-fully-functional-game-in-one-prompt/02-enable-goals-terminal.png) This turns on the Goals feature for Codex. You can use /goal from Terminal or inside the Codex app. For browser games, we prefer the Codex app because Codex can open the app, click around, use the browser, and verify that the thing actually works on screen. The important shift is that /goal is not a normal prompt. A normal prompt asks Codex to do a task once. A goal gives Codex an objective, a checklist, and permission to keep working until the evidence says the job is done. > Pro tip: Do not use /goal just because a task feels big. Use it when you can define a finish line. "Make this better" is a weak goal. "Build a playable HTML demo, run it locally, test controls, and report what passed" is a strong goal. ### Step 2 Pick a Small Idea With a Real Finish Line Start with a game idea that can be described in a few sentences. You do not need a huge design document. You need rules Codex can verify. Here is the kind of prompt that works: ```text /goal I want a top-down endless surfing game where you gain points for jumping waves and lose by crashing into rocks. Use HTML and test it; I just want a playable demo. ``` That prompt is short, but it has the pieces Codex needs: - a visible format: a top-down browser game - a core action: jumping waves - a scoring rule: gain points for clean jumps - a failure rule: lose by crashing into rocks - an output bar: a playable demo - a verification instruction: test it If your idea is harder to explain, pause before you send it to Codex. Ask ChatGPT to rewrite the idea in 100 words or less with objective tests. If ChatGPT cannot do that cleanly, the task is probably too large or too subjective for one goal. > Pro tip: Good /goal prompts spend more words on success criteria than vibes. Codex can work toward "spacebar jumps, score increases, collision ends the game, restart button works." It cannot reliably work toward "make it addictive" without more concrete signals. ### Step 3 Turn the Idea Into a Goal Prompt You can paste a simple goal directly, but better prompts give Codex a sharper loop. Use this structure: ```text /goal Build a small browser game called Endless Surf Carver. Game idea: - Top-down endless surfing game. - The player moves left and right through wave lanes. - The player gains points for jumping waves cleanly. - The player loses by crashing into rocks or obstacles. Acceptance criteria: - The game runs locally in the browser. - Keyboard controls work. - Score increases when the player successfully jumps waves. - Collision or failed movement can end the run. - There is a clear game-over state and a restart action. - Test it in the browser and summarize what passed. Constraints: - Keep it as a playable demo. - Use simple HTML, CSS, and JavaScript unless the current project already uses a framework. - Do not add unrelated features. ``` This is more efficient than sending five follow-up prompts later. The goal already tells Codex what to build, what not to build, and how to decide whether it is finished. You can still keep creative room in the prompt. The point is not to micromanage every color or animation. The point is to make the core loop measurable. > Pro tip: Put constraints at the bottom. Codex tends to overbuild when a prompt sounds open-ended. A line like "Keep it as a playable demo" prevents the agent from turning a five-minute exercise into a full game engine. ### Step 4 Start `/goal` and Watch the Checklist Once you send the prompt, Codex translates the request into a plan. You will usually see a checklist with steps like inspecting the project, implementing the game, running it, testing it, and completing the goal when the evidence matches the requirements. ![Goal active checklist](https://tru-images.b-cdn.net/guide-assets/local-214/use-codex-goal-to-build-a-fully-functional-game-in-one-prompt/03-goal-active-checklist.png) This is the main advantage of the Goal feature. It gives Codex a reason to keep going after the first draft. If a test fails, the browser does not load, or the interaction is broken, Codex can keep working through the issue instead of waiting for you to notice and ask again. For a small browser game, the run should usually take a few minutes. You can follow along while Codex edits files, starts a local server, opens the browser, and checks whether the game works. ![Goal prompt sent](https://tru-images.b-cdn.net/guide-assets/local-214/use-codex-goal-to-build-a-fully-functional-game-in-one-prompt/04-goal-prompt-sent.png) You are still supervising. If Codex starts drifting into unrelated features, interrupt with a direct correction. If the project gets too broad, narrow the definition of done. > Pro tip: Watch for the first plan, not every line of code. The fastest way to save tokens is to catch a bad plan early. If the checklist does not mention testing, controls, scoring, and a failure state, correct the goal before Codex spends time building the wrong thing. ### Step 5 Test the Game and Give Follow-Up Goals When Codex says the goal is complete, test the game yourself. Move the player. Jump. Score points. Crash. Restart. Make sure the basic loop actually matches the prompt. ![Endless surf jump scoring](https://tru-images.b-cdn.net/guide-assets/local-214/use-codex-goal-to-build-a-fully-functional-game-in-one-prompt/05-endless-surf-jump-scoring.png) If something is broken, do not go back to vague prompting. Give Codex another goal with the exact failure or improvement. For example: ```text /goal Improve the Endless Surf Carver demo by adding distinct animations for every action the user can take. Acceptance criteria: - Moving left or right has a visible lean or carve animation. - Jumping has a clear lift and landing animation. - Crashing has a distinct impact or wipeout animation. - Restarting resets all animation state. - Test the game in the browser and summarize what changed. ``` That prompt works because it names the gap and gives Codex a way to prove the fix. "Make it feel better" would probably waste more tokens. "Add distinct animations for every action" gives the agent a checklist. ![Endless surf game over](https://tru-images.b-cdn.net/guide-assets/local-214/use-codex-goal-to-build-a-fully-functional-game-in-one-prompt/06-endless-surf-game-over.png) This same method works for non-game tasks. The key is to describe what success looks like in a way Codex can inspect, run, compare, or score. > Pro tip: Follow-up goals should be smaller than the original goal. Use the first goal to get a working version. Use later goals to improve one thing at a time: controls, visuals, tests, accessibility, performance, or copy. ### Guide: Use Gemini Canvas to Visualize Google Sheets Like a Pro > In this guide, you will learn how to use Gemini's new artifact feature to save time making Google Sheets presentation-ready. - Canonical page: https://app.therundown.ai/guides/use-gemini-canvas-to-visualize-google-sheets-like-a-pro - Machine-readable page: https://app.therundown.ai/guides/use-gemini-canvas-to-visualize-google-sheets-like-a-pro.md - Difficulty: beginner - Featured tool or company: Canva - Categories: AI, Productivity - Instructors: Billy Howell - Published: 2026-08-22 - Updated: 2026-08-22 ### Overview #### The Rundown In this guide, you will learn how to use Gemini's new artifact feature to save time making Google Sheets presentation-ready. #### Who This Is Useful For - Sarah, an operations lead, who wants a readable dashboard from a project tracker without rebuilding the tracker in another tool. - Marcus, a consultant, who needs to turn a client’s spreadsheet into a clear summary before a meeting. - Anyone who manages a structured Sheet, who wants to test a visual dashboard or presentation from the data they already maintain. #### What You Will Build A shareable Canvas dashboard that turns the data in a reviewed Google Sheet into charts, filters, and lists you can inspect before sharing. In the recorded workflow, the Canvas is a generated artifact, not a live mirror of the source Sheet. ![finished gemini canvas dashboard](https://tru-images.b-cdn.net/guide-assets/local-431/use-gemini-canvas-to-visualize-google-sheets-like-a-pro/01-finished-gemini-canvas-dashboard.png) #### What You Need To Get Started - A Google account signed in to Gemini Apps (https://gemini.google.com/). - A non-sensitive Google Sheet you can attach from Drive or upload. - A browser on a computer so you can review the source and Canvas side by side. - A few minutes to check the generated numbers, labels, and filters against the Sheet. - If you use a work or school account, Gemini Apps and Drive attachments must be enabled by your administrator. > Note: This video uses Gemini Canvas in Gemini with an attached Sheet. Google also has a separate Sheets Canvas (https://support.google.com/docs/answer/17035851?hl=en-4) feature that runs inside Sheets. Do not assume the two surfaces have the same data-sync or editing behavior. #### Going Further Once Gemini has correctly understood the source, try another format that serves the same job: a short slide deck for a meeting, a comparison table, or a focused report. The recording turns a content-calendar tab into a six-slide strategy deck. ![canvas content calendar deck](https://tru-images.b-cdn.net/guide-assets/local-431/use-gemini-canvas-to-visualize-google-sheets-like-a-pro/04-canvas-content-calendar-deck.png) Keep the same discipline: attach the source, have Gemini summarize it, state the format and audience, then inspect the result. The recording shows an Export menu for the generated deck; confirm the export destinations available in your account before promising a Google Slides or PDF handoff. ### Step 1 Start With a Source Sheet Start with a Google Sheet. A project tracker, content calendar, or sales pipeline are three good examples. If the sheet is not already in Drive, upload the spreadsheet from your computer. Before you open Gemini, make sure you know what each tab contains and remove any information you would not want to share with an AI tool. In the recording, the Sheet has fields such as opportunity, company, stage, owner, deal size, next step date, and notes. Your fields can be different; the important part is that the labels make the data easy to interpret. ![source google sheet](https://tru-images.b-cdn.net/guide-assets/local-431/use-gemini-canvas-to-visualize-google-sheets-like-a-pro/02-source-google-sheet.png) Open a new Gemini chat. Click the attachment control, choose your Sheet from Google Drive, or upload a copy of the file. Gemini’s current Canvas guidance also starts from the Add files control in Gemini Apps. Google’s file-upload help (https://support.google.com/gemini/answer/14903178?hl=en-CA) covers the supported source workflow. ![select sheet from drive](https://tru-images.b-cdn.net/guide-assets/local-431/use-gemini-canvas-to-visualize-google-sheets-like-a-pro/03-select-sheet-from-drive.png) ### Step 2 Give Gemini the Source Context First Before you ask for a dashboard, make Gemini read the file. Ask it to summarize the tabs, key columns, and any ambiguous values. This gives you a chance to catch a wrong attachment or a messy field before Canvas turns it into a polished-looking artifact. Use this prompt: ```text Read the attached spreadsheet before creating anything. For each tab, list: - what the tab is for - the key columns and their meaning - any blank, duplicate, or inconsistent values that could affect a dashboard Do not create a Canvas yet. Tell me what I should fix first if the source is ambiguous. ``` > Pro tip: If Gemini invents data or seems to miss your Sheet, start a new chat, attach the file again, and ask it to read the source before opening Canvas. Use Thinking when accuracy matters. ### Step 3 Build a Dashboard From the Reviewed Sheet After Gemini has described the source correctly, select Canvas from the attachment controls and ask for one focused dashboard. Name the tabs and fields you want it to use so the request does not turn into a generic business mockup. Use this adaptable prompt: ```text Using the spreadsheet you just reviewed, build an interactive dashboard from the [TAB 1] and [TAB 2] tabs. Requirements: - Use the actual values in the reviewed source. - Show the most useful totals, charts, filters, and lists for [AUDIENCE]. - Label each section clearly. - Do not invent numbers, rows, or metrics. - Make it easy to compare [PRIMARY MEASURE] and [SECONDARY MEASURE]. ``` The result may include charts, filters, lists, and separate views for the data you attached. Treat it as a working prototype: it can help you see the story in the Sheet quickly, but it still needs a source-data check. ### Step 4 Check the Artifact Before You Share It Open the Canvas and compare it with the source Sheet. Check a few values in every major chart or summary, scan the labels, and try the filters you expect your team to use. This is where you catch a wrong attachment, an unlabeled metric, or output that does not match the data you supplied. The recorded Canvas dashboard is a generated snapshot. When the source Sheet changes, return to the same chat with the updated data or rebuild the Canvas from a fresh, reviewed attachment. Do not tell teammates that it will update live unless you have tested the specific surface and account you are using. Google’s Canvas help (https://support.google.com/gemini/answer/16047321?hl=en-NZ) explains the current Canvas creation and sharing controls. Check that official page again before publishing or sharing a workflow based on a fast-changing feature. ### Guide: Use Google Stitch to Generate HeatMaps to Design High Performing Websites > Learn how to use Google Stitch to create a predictive attention heat map for a webpage, identify where visitors’ eyes are likely to go, and use those insights to optimize a hero section for click-through rate and conversions. - Canonical page: https://app.therundown.ai/guides/perform-an-instant-attention-audit-on-your-webpage-with-google-stitch - Machine-readable page: https://app.therundown.ai/guides/perform-an-instant-attention-audit-on-your-webpage-with-google-stitch.md - Difficulty: beginner - Featured tool or company: Google - Categories: Marketing & Growth, Data, Research & Analysis, Content & Creative - Instructors: Billy Howell - Published: 2025-12-11 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use Google Stitch to create a predictive attention heat map for a webpage, identify where visitors’ eyes are likely to go, and use those insights to optimize a hero section for click-through rate and conversions. #### Who This Is Useful For - Anyone with an outdated website - Designers looking to improve conversion rate optimization - Website owners with low click-through rates or sales #### What You Will Build You will create a predictive heat map and use it to generate an optimized hero section design that focuses attention on your primary call to action. ![07_cover_image_hat_heat_map](https://tru-images.b-cdn.net/guide-assets/69/693cd43653bf39ffb0b4bcdf/07-cover-image-hat-heat-map.png) - A screenshot-based attention audit of your landing page hero section - A generated predictive heat map - A redesigned hero section concept in Google Stitch #### What You Need - A Google Stitch session - Google AI Studio - A screenshot of your site’s above-the-fold hero section - Your primary CTA text, such as “Get Started” - Optional: a full-page PNG if you want to generate a heat map for the entire landing page #### Going Further - Download your selected design from Google Stitch. - Use the PNG and HTML file with your AI coding assistant, such as Replit, to redesign your hero section based on the optimized screenshot. - If Stitch updates your background hero image, use AI Studio to extract the image from the new design PNG. - Use the generate variations tool in Google Stitch to explore more options before settling on a final design. ### Step 1: Sign in to Google Stitch Sign in to Google Stitch. It’s free to use. ### Step 2: Get familiar with the Stitch workspace Familiarize yourself with the layout and tools in Google Stitch. ![03_google_stitch_toolbar](https://tru-images.b-cdn.net/guide-assets/69/693cd44d7f61d451bbbded86/03-google-stitch-toolbar.png) - Generate variations or a predictive heat map of a page. - Ask the agent to make edits via chat, change the color scheme, or annotate it directly with Nano Banana. - Change the preview layout between web, mobile, or tablet. - Export your designs to AI Studio, a zip file, or your clipboard. ### Step 3: Capture your hero section Grab a screenshot of your site. Focus on the above-the-fold area, especially the hero section. ![04_hero_before](https://tru-images.b-cdn.net/guide-assets/69/693cd468ab4b90ac97ff0601/04-hero-before.png) Pro tip: You can generate a heat map for your entire landing page by right-clicking your site and saving it as a PNG. ### Step 4: Generate a predictive attention heat map Go to Google AI Studio. Drop in your screenshot with the prompt below. Update the CTA text so it matches your site. ```text Use Nano Banana to generate a predictive attention heat map on top of this landing page. The audit should specifically highlight areas of high visual engagement and identify any elements that are distracting focus away from the primary 'Get Started' CTA. ``` Download the generated heat map image. ### Step 5: Redesign the hero section in Google Stitch Drop the heat map and your original screenshot into a new Google Stitch session. Make sure you start with a new web design, then use the prompt below. ```text Analyze the provided landing page screenshot alongside its predictive attention heat map. Generate a redesign of the hero section optimized specifically to maximize the click-through rate on the primary 'Get Started' button. The new design should use the heat map insights to eliminate distractions and focus user attention directly on the main conversion goal. ``` ![05_prompting_stitch](https://tru-images.b-cdn.net/guide-assets/69/693cd4bb53bf39ffb0b4bd07/05-prompting-stitch.png) Review the before-and-after versions of your optimized designs. Small visual changes can still have a meaningful impact on click-through rate and conversions. ![06_before_after_optimizations](https://tru-images.b-cdn.net/guide-assets/69/693cd4ce7eb16e6e15e25a4f/06-before-after-optimizations.png) Look for structural changes that guide attention. In the example, the text width on the left moves from small to medium-wide, steadily drawing the eye from left to right toward the lead form. ![06_before_after_optimizations_2](https://tru-images.b-cdn.net/guide-assets/69/693cd4e07eb16e6e15e25a7a/06-before-after-optimizations-2.png) Check how the redesign treats buttons and secondary actions. In the example, the heat map showed that the secondary “View Lookbook” button was drawing too much attention, so Stitch changed it into a smaller link underneath the primary button. ### Pro tip > Once you have a new site design in Google Stitch, try the generate → variations tool. Don’t settle for the first result. ### Guide: Use Google’s Design.md Standard to Build a Consistent Brand > In this guide, you will learn how to use Google's open-source DESIGN. - Canonical page: https://app.therundown.ai/guides/use-googles-design-md-standard-to-build-a-consistent-brand - Machine-readable page: https://app.therundown.ai/guides/use-googles-design-md-standard-to-build-a-consistent-brand.md - Difficulty: beginner - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-07-01 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use Google's open-source DESIGN.md format and Stitch skills to give an AI coding agent a reusable visual source of truth. You will build or extract a DESIGN.md, use it to create a local prototype, and learn how to improve the result without re-prompting brand style from scratch. #### Who This Is Useful For - Founders and marketers who want quick mockups that still feel tied to an existing brand. - Designers and product teams who need AI tools to follow colors, typography, components, motion rules, and "do not use" guidance. - Developers and AI builders who want reusable design context inside Claude Code, Codex, Cursor, or another coding agent. #### What You Will Build You will install Google's Stitch skills, create or extract a DESIGN.md file, and use it to generate a lightweight local prototype. In our test, that file turned a rough Tiki bar idea into a visual system with a palette, typography, component rules, banned patterns, motion guidance, and a working "Tom's Tiki Drinks" prototype. ![Tom's Tiki Drinks prototype homepage](https://tru-images.b-cdn.net/guide-assets/local-288/use-googles-design-md-standard-to-build-a-consistent-brand/01-toms-tiki-drinks-homepage.png) #### What You Need To Get Started - Claude Code, Codex, Cursor, or another supported coding agent. This guide uses Claude Code. - A project folder where the Stitch skills can install. - The Google Stitch skills repo (https://github.com/google-labs-code/stitch-skills). - A design idea, an existing website, or frontend source code to analyze. - Optional: a Google Stitch (https://stitch.withgoogle.com/) account and API key if you want the agent to push designs into Stitch. #### Going Further If you use Stitch directly, create an API key in Stitch settings and let the agent connect to your account. The Stitch skills can help push design systems and screens into Stitch so you can review them on a canvas and share them outside the coding agent. You can also keep DESIGN.md in your repo like any other source file. Review diffs before a big redesign, keep one file per app or product surface, and hand the same file to Claude Code, Codex, Cursor, Stitch, or another AI builder. Google's announcement (https://blog.google/innovation-and-ai/models-and-research/google-labs/stitch-design-md/) frames DESIGN.md as a portable way to import and export design rules, and the spec is still marked alpha. Treat it as a strong workflow pattern, not a finished replacement for designers, accessibility review, or brand approval. ### Step 1 Install The Stitch Skills Open a new project folder, then install the Google Stitch skills for Claude Code. ```text npx plugins add google-labs-code/stitch-skills --scope project --target claude-code ``` When the installer asks which plugins to install, you can select the full set for this workflow. The repo lists stitch-design, stitch-build, and stitch-utilities as the main plugin groups. After install, restart Claude Code or open a fresh Claude Code thread in the same folder. The skills need a new agent session before they are available. ![Google Stitch skills install readme](https://tru-images.b-cdn.net/guide-assets/local-288/use-googles-design-md-standard-to-build-a-consistent-brand/02-stitch-skills-install-readme.png) > Pro tip: --scope project keeps the skills in the current folder. Use the repo's Codex marketplace or global setup instructions if you want these available across projects. ### Step 2 Make The Agent Read Its New Tools Before asking for a design, make the agent inspect the skills it just received. This is a small step, but it prevents you from blindly calling tools the agent has not loaded into context. Use this prompt: ```text Take a look at the Google Stitch design tools I just installed. They should be in this project. Get familiar with them and give me a brief report on what you think of them. Keep it brief and informal. ``` The agent should report which skills are available and what each one is for. In our test, Claude found skills for generating designs, converting code to design, extracting static HTML, extracting DESIGN.md, and managing design systems in Stitch. ### Step 3 Create Or Extract A DESIGN.md Google's DESIGN.md repo (https://github.com/google-labs-code/design.md) describes the file as two layers: YAML design tokens at the top, then markdown rationale below. The tokens give the agent exact values. The prose tells it when and why to use them. You can make the file from a new idea: ```text Build me a DESIGN.md using the Stitch tools for a tiki bar called Tom's Tiki Drinks. We want the vibes to be tropical, maximalist, fun to use, and we want surprising but purposeful animations. I'm thinking lots of golds, yellows, greens, and colors of tropical birds. ``` You can also extract a file from an existing site: ```text Use Google Stitch skills to extract a DESIGN.md from this website: [WEBSITE_URL] ``` The extraction path is useful when you want to redesign an existing brand, build a product page that matches the current site, or turn a prospect's website into a design brief before pitching a new prototype. > Pro tip: The extraction skill may be strongest when it can inspect source code. If it struggles with a live site, give the agent local frontend files, a static HTML snapshot, or a few screenshots. ### Step 4 Review The File Before Building Open the generated DESIGN.md before you ask for UI. Check that it names the important design choices instead of only listing colors. In the Tiki bar example, the useful parts were specific: - a 1960s Polynesian rec-room atmosphere - saturated gold, yellow, green, and tropical accent colors - custom typography guidance for drink names and labels - component rules for cards, buttons, tabs, and navigation - motion notes for torch flickers, palm fronds, ice clinks, and toucan animation - banned patterns, including generic SaaS dashboard choices ![Generated DESIGN.md system](https://tru-images.b-cdn.net/guide-assets/local-288/use-googles-design-md-standard-to-build-a-consistent-brand/03-generated-design-md-system.png) This is the point of the workflow. You are not asking the agent to remember brand taste from a chat thread. You are giving it a file it can inspect, revise, and reuse. ### Step 5 Build A Lightweight Prototype Once the file looks useful, ask the agent to build something small with it. ```text Use that DESIGN.md to create a quick, lightweight local prototype so I can see what it looks like in action. ``` Keep the first build visual and fast. The goal is not production code yet. The goal is to see whether the design file produces a distinct direction. In our test, the first pass made a dark Tiki drinks site with a strong logo, menu cards, illustrated drink art, custom colors, and small animations. It was not perfect, but it felt much less generic than a normal "make it fun and tropical" prompt. ![Tom's Tiki Drinks menu prototype](https://tru-images.b-cdn.net/guide-assets/local-288/use-googles-design-md-standard-to-build-a-consistent-brand/04-toms-tiki-drinks-menu.png) Then review the output like a design pass. Ask for targeted changes instead of restarting the whole build: ```text Everything is great, but slow down the ice cube animation. It goes really fast and it is hard to see. ``` That kind of tight follow-up works because the agent already has the visual system in the file. You are editing a direction, not renegotiating the whole brand. ### Step 6 Combine Current Brand And New Direction The most useful advanced workflow is to use two design files. First, extract the current site's design system: ```text Use Google Stitch skills to extract a DESIGN.md from [CURRENT_WEBSITE_URL]. Name it current-site-DESIGN.md. ``` Then create the new direction: ```text Create a second DESIGN.md for where we want the brand to go. Name it new-direction-DESIGN.md. Focus on [NEW_BRAND_DIRECTION], [KEY_AUDIENCE], [DESIRED_FEEL], and [PATTERNS_TO_AVOID]. ``` Finally, make the agent compare them: ```text Look at current-site-DESIGN.md and new-direction-DESIGN.md. Preserve the core brand cues from the current site, then move the prototype toward the new direction. Return a short design migration plan first. Then build one lightweight page so I can review the direction visually. ``` This is better than asking for a redesign in one giant prompt. The agent has to reason about what stays, what changes, and what should never be lost. ### Guide: Use Manus To Write Better Expert LinkedIn Posts That Actually Sound Like You > In this guide, you will learn how to use AI to write better LinkedIn posts by finding content ideas you didn't even know you had. - Canonical page: https://app.therundown.ai/guides/use-manus-to-write-better-linkedin-posts-that-actually-sound-like-you - Machine-readable page: https://app.therundown.ai/guides/use-manus-to-write-better-linkedin-posts-that-actually-sound-like-you.md - Difficulty: beginner - Featured tool or company: Manus - Categories: Marketing & Growth, Content & Creative - Instructors: Billy Howell - Published: 2026-07-15 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to use AI to write better LinkedIn posts by finding content ideas you didn't even know you had. You will build a Manus (https://manus.im/) workspace that interviews you, extracts your expertise and voice, then turns its findings into drafted posts. #### Who This Is Useful For - Founders and operators who have useful stories buried inside their daily work. - Experts who want help writing without making every post sound polished, generic, or unlike them. - Anyone who wants a repeatable way to turn firsthand experience into better LinkedIn content. #### What You Will Build You will build a Manus workspace that acts like a personal reporter. It interviews you, saves your strongest ideas, learns how you sound, and turns those ideas into LinkedIn drafts you can review. ![The finished Manus content operating playbook, supporting files, and completed task checklist](https://tru-images.b-cdn.net/guide-assets/local-338/use-manus-to-write-better-linkedin-posts-that-actually-sound-like-you/01-finished-content-operating-playbook.png) #### What You Need To Get Started - A Manus account. Full Rundown University Pro (https://app.therundown.ai/plans) members can currently claim one year of Manus Starter free through University Perks. - Links to a few public writing samples, such as your LinkedIn profile, X profile, blog, or articles. If you do not have any, bring examples of writers whose style you like. - A few quiet minutes to dictate without sharing anything private. - Optional: the Manus desktop app if you want to keep the workspace in a local folder. - Optional: Manus Browser Operator (https://manus.im/features/manus-browser-operator) if you want Manus to inspect pages through Chrome. #### Going Further Once the workflow feels right, use Skill Creator to save it as a reusable Manus skill, then call it from /skills. Run one short voice interview each week and tell Manus to add the new idea cards, voice notes, and drafts to the same project. You can also point Manus to your published articles and analytics before it drafts. That gives it real performance data to pair with the story instead of forcing it to guess which ideas worked. The loop stays simple: Manus asks one good question, you dictate the messy truth, it finds the useful ideas, and you teach it what sounds like you. ### Step 1 Start a Project and Let Manus Learn How You Think Start a new project in Manus and tell it what you want to build. Give it links to your public writing samples, then ask it to interview you so it can learn your expertise and write better posts in your voice. You do not need a giant system prompt. Use a prompt like this to get the project moving: ```text Help me build a simple content workspace that turns my work experience into LinkedIn posts in my voice. Start by reviewing these public writing samples: [LINKS]. Tell me which pages and files you can and cannot access. Then interview me one question at a time so you can learn my expertise, opinions, and natural writing style. Keep everything inside this project, and do not publish anything. ``` Manus can use your public posts to spot recurring topics, phrases, opinions, and patterns in how you explain things. It will combine that context with the answers you give during the interview. ![Manus keeping the content system inside the local project after stopping an unintended Notion path](https://tru-images.b-cdn.net/guide-assets/local-338/use-manus-to-write-better-linkedin-posts-that-actually-sound-like-you/02-project-native-workspace-correction.png) > Pro tip: Review and disable unused connectors first. Manus may choose an enabled tool automatically. During our test, it started using Notion, so we stopped the run and corrected the destination. ### Step 2 Answer One Incident Question at a Time Manus will interview you with questions designed to uncover stories hiding inside your normal workday. It might ask what changed your mind recently, what worked much better or worse than expected, or what changed how you now do a real task. Here is an adapted version of the first question we used: ```text Think about the tools, workflows, or decisions you tested in the last 7–14 days. Tell me about one moment that changed your mind—something that worked much better or worse than expected, exposed a misleading popular belief, or changed how you now do a real task. As you ramble, include whatever you remember: what you were trying to accomplish, which tools were involved, what happened on the first attempt, what happened after iteration, what surprised or annoyed you, and who should care. If several examples come to mind, share all of them. ``` Dictate a brain dump in response. Do not worry about structure, grammar, or whether the story is ready to become a post. Talk through what you expected, what actually happened, what surprised or annoyed you, and why someone else might care. Answer any focused follow-up questions. If a story depends on a result, audience response, or number Manus cannot access, give it the source or flag that detail for verification. ![A changed-mind interview prompt paired with the generated Voice and Expertise Profile](https://tru-images.b-cdn.net/guide-assets/local-338/use-manus-to-write-better-linkedin-posts-that-actually-sound-like-you/03-incident-based-dictation-prompt.png) This surfaced two useful stories in our test. One was about a basic one-on-one workflow that became more popular than much flashier AI topics. The other came from an older Microsoft Copilot recording we abandoned when the tool struggled with basic Excel formulas and formatting. The product may have improved since then, but the broader problem remains: people still ask for help with software their employers require, even when an earlier test went badly. Neither idea started as a polished content pitch. They came out of talking through a real incident. ### Step 3 Let Manus Pull Out the Ideas and Your Voice Once you finish answering, Manus automatically pulls out your strongest ideas, the supporting evidence available in your answers and project, and exact phrases that sound like you. It saves promising stories as idea cards and uses the interview to build a voice and expertise profile. In this setup, you do not need to ask for each file separately. The opening project request tells Manus to standardize the process, so it can turn the messy interview into the supporting files while it works. The idea card shown below came from the older Copilot story. It captured the failed recording, who still needs help, and the boundary that product-specific advice would require a current retest. Manus also saved phrases from the dictation instead of smoothing everything into generic AI language. ![The Copilot idea card preserves the incident, audience, and current-retest boundary](https://tru-images.b-cdn.net/guide-assets/local-338/use-manus-to-write-better-linkedin-posts-that-actually-sound-like-you/04-copilot-idea-card.png) That last part matters. The rough edges are often where your voice lives. If you use a phrase you would actually say, keep it. If Manus turns it into something that sounds like corporate thought leadership, flag it during review. Manus also turns patterns across the interview into supporting files, such as content pillars and future interview exercises. Over time, the project becomes a library of your ideas instead of a folder full of disconnected drafts. Manus does this work on its own computer, building the files while the task runs. > Pro tip: Let Manus finish building the workspace before taking over its computer. Taking control interrupts the flow, so leave the task running in the background and check back later. ### Step 4 Review the Drafts and Teach Manus What Sounds Like You Manus automatically turns the strongest idea cards into LinkedIn drafts. In our test, it created two pilot posts from the stories in the interview without needing a separate drafting prompt. Your job is to review the drafts and give simple, specific feedback: - Which phrases sound exactly like you? - Which lines feel too polished or generic? - What important detail or opinion is missing? - Is Manus stating anything more confidently than the evidence supports? For example, you can respond naturally: ```text The opening sounds like me, but the middle is too polished. Keep my original phrase about [PHRASE], add the missing detail about [DETAIL], and make the conclusion more direct. Use this feedback to write version two and update my voice profile. ``` ![A LinkedIn draft with its evidence boundary and review checklist](https://tru-images.b-cdn.net/guide-assets/local-338/use-manus-to-write-better-linkedin-posts-that-actually-sound-like-you/05-linkedin-draft-and-review.png) Each review gives Manus a clearer record of your preferences. You are showing it where the first draft captured your voice, where it lost it, and what to change in version two. > Pro tip: Review and approve version two before posting. Check what sounds like you, what feels too polished, what is missing, and whether the evidence supports the draft. ### Guide: Use OpenAI’s New GPT Live to Plan Any Trip Fast > In this guide, you will learn how to use GPT-Live to plan a full weekend trip by talking it through. - Canonical page: https://app.therundown.ai/guides/use-openais-new-gpt-live-to-plan-any-trip-fast - Machine-readable page: https://app.therundown.ai/guides/use-openais-new-gpt-live-to-plan-any-trip-fast.md - Difficulty: beginner - Categories: Productivity & Knowledge Work - Instructors: Billy Howell - Published: 2026-07-16 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to use GPT-Live to plan a full weekend trip by talking it through. Interrupt and change direction while it researches, then turn the conversation into a plan you can use. GPT-Live is the new model powering ChatGPT Voice—not a separate app. For OpenAI's explanation of the release, read Introducing GPT-Live (https://openai.com/index/introducing-gpt-live/) and the ChatGPT release notes (https://help.openai.com/en/articles/6825453-what-is-chatgpt-plus). #### Who This Is Useful For - Travelers planning a weekend away who want a decisive itinerary without running separate searches for every choice. - Busy professionals coordinating a trip around work, budget, and changing conditions. - Anyone who wants to talk through a messy plan, refine the tradeoffs, and leave with an organized next-action list. #### What You Will Build You will turn a spoken brief into a structured weekend plan: a recommended area to stay, a rough itinerary, an order for booking the important pieces, and a short list of facts to verify before spending money. ![Initial Chicago itinerary](https://tru-images.b-cdn.net/guide-assets/local-350/use-openais-new-gpt-live-to-plan-any-trip-fast/01-initial-chicago-itinerary.png) The broader method is simple: Outcome → Optimize → Decide → Research → Refine → Save. It works for travel, but it is equally useful for a report, event, purchase decision, or project plan. #### What You Need To Get Started - A personal ChatGPT account in a supported region. GPT-Live is rolling out on ChatGPT.com and the iOS and Android apps: paid consumer accounts use GPT-Live-1, while Free accounts use GPT-Live-1 mini. It is not available in ChatGPT Business, Enterprise, or Edu workspaces at launch. - A destination or project to plan, plus the dates, budget, priorities, and constraints that matter. - A browser or the relevant booking sites to verify current prices, availability, hours, and reservations. #### Going Further Reuse the same format for projects beyond travel: state the outcome, name what to optimize for, give ChatGPT permission to choose, refine the first pass, and end with a saved deliverable. It is a practical way to turn a voice conversation into a plan you can act on. ### Step 1 Open Voice and Pick the Right Pace GPT Voice is usually used in the ChatGPT mobile app. To use GPT-Live on desktop, open ChatGPT.com in your browser instead of the desktop app, then start a Voice chat. Availability can still vary by account, region, and rollout. If you see the intelligence selector, use Medium for a trip with research and conflicting constraints. It gives ChatGPT more room to work through a plan. Use Instant when you mainly want a fast conversational exchange. > Pro tip: To use voice AI more often, set your iPhone Action Button to open a new GPT-Live thread. Choose the intelligence level before you start a complex brief; switching it mid-session is easy to miss. ### Step 2 Give It an Outcome-First Brief The key is to let GPT-Live research in the background instead of walking it through a trip one task at a time. Tell it the complete outcome, what a good answer looks like, and what it has permission to decide. Use this as a spoken or typed starter: ```text Plan the complete [CITY] weekend for [NUMBER OF TRAVELERS]. Optimize for [PRIORITIES]. Our constraints are [DATES], [BUDGET], and [NON-NEGOTIABLES]. Choose the best neighborhood, hotel direction, activities, and dining plan for us. Give one primary recommendation in each category, explain the important tradeoffs, and list what to book first. Separate facts that need verification from your recommendations. ``` In the demo, the goals were a walkable Chicago weekend, great food, a relaxed pace, and one memorable dinner. That gave ChatGPT enough context to return a coherent recommendation instead of a broad list of hotels and restaurants. Ask it to “decide for me” and “explain why” when you need a real recommendation. Put your constraints at the top of the brief so the first plan has a chance of protecting them. ### Step 3 Refine the Plan While You Talk Use the live conversation to make the plan more realistic. Interrupt GPT-Live as the plan takes shape, then add one constraint at a time: a work block gets longer, the flexible budget drops, or an outdoor activity needs a nearby rain backup. It can keep researching while you change direction. Useful prompts include: ```text Give me the single best tradeoff first and tell me why. Keep [NON-NEGOTIABLE] and find the savings elsewhere. Build a nearby indoor backup for each day in case the weather changes. Before rewriting anything, repeat the constraints you are solving for. ``` This is where Voice is most useful: you can correct the direction in the moment rather than opening a new thread or retyping the whole brief. ### Step 4 Verify Current Details Outside the Conversation GPT-Live can use web research to find useful pricing and availability information, but use the result as a planning lead—not as a confirmed booking. In the recording, it found a ballpark hotel price and showed that it had searched sources, but the presenter still opened Google Hotels to confirm the current rate and options. ![Booking order checklist](https://tru-images.b-cdn.net/guide-assets/local-350/use-openais-new-gpt-live-to-plan-any-trip-fast/02-booking-order-checklist.png) Check the original hotel, venue, airline, or booking page for current price, availability, cancellation rules, and hours. This is especially important when an event or a busy weekend could change the cost. > Pro tip: Ask ChatGPT to include sources and a “things to verify” section in the final plan. It keeps the useful research separate from facts that can change after the conversation ends. ### Step 5 Turn the Conversation Into a Memo You Can Use When the voice session is over, do not leave the useful work buried in the transcript. Close Voice, then send this text chat: ```text Give me your final research as a memo. ``` ![Travel plan memo](https://tru-images.b-cdn.net/guide-assets/local-350/use-openais-new-gpt-live-to-plan-any-trip-fast/04-travel-plan-memo.png) This step turns the research into something you can scan, share, or update. In the demo, the plan included a budget range and a final recommendation rather than leaving the traveler with a raw voice transcript. If conditions change, add the new constraint and ask it to update the same memo. Here, the presenter asked it to add nearby wet-weather alternatives so the plan stayed geographically coherent. ![Weather contingency plan](https://tru-images.b-cdn.net/guide-assets/local-350/use-openais-new-gpt-live-to-plan-any-trip-fast/05-weather-contingency-plan.png) ### Guide: Use Perplexity as Your Personal Shopper (Perplexity Computer Walkthrough) > This guide shows you how to use Perplexity’s agent search feature, Perplexity Computer, as a personal shopper. - Canonical page: https://app.therundown.ai/guides/use-perplexity-as-your-personal-shopper-perplexity-computer-walkthrough - Machine-readable page: https://app.therundown.ai/guides/use-perplexity-as-your-personal-shopper-perplexity-computer-walkthrough.md - Difficulty: beginner - Featured tool or company: Perplexity - Categories: Productivity & Knowledge Work, Data, Research & Analysis - Instructors: Billy Howell - Published: 2026-03-26 - Updated: 2026-07-30 ### Overview #### The Rundown This guide shows you how to use Perplexity’s agent search feature, Perplexity Computer, as a personal shopper. You will use it to find items that match your preferences, check what is in stock in your size, surface deals, and produce a detailed shopping report with direct links. Perplexity Computer is normally for Max accounts, but Perplexity Pro users can test it with free credits right now. #### Who This Is Useful For - Busy shoppers who want help narrowing products by size, color, fit, or brand without clicking through a dozen dead ends - Perplexity Pro users who have bonus Computer credits and want a concrete test before deciding whether it is worth spending more - Anyone who is overwhelmed by the number of products and shopping websites online #### What You Will Build You will create a detailed shopping report with links to the best deals and options that match your shopping criteria. ![image-2152](https://tru-images.b-cdn.net/guide-assets/69/69c59e2f35ee66f5ffd7a948/image-2152.png) - A shortlist of best overall options - A best budget option - Items that are actually in stock in your size - Direct links to buy #### What You Need - A Perplexity Pro account - The Perplexity or Comet app, optional - A difficult shopping decision with lots of potential options #### Going Further - After you build your shopping report, you can ask Perplexity Computer to set up scheduled tasks to monitor item prices and new listings each day or week. The guide notes that these scheduled tasks should not use your Perplexity Computer tokens. ### Step 1 Turn on Perplexity Computer Open Perplexity in the app or in your browser and look for the Perplexity Computer toggle. ![image-2151](https://tru-images.b-cdn.net/guide-assets/69/69c59e2535ee66f5ffd7a0a9/image-2151.png) You can test the same shopping query side by side in regular Perplexity and Perplexity Computer. Regular Perplexity Assistant is useful for a quick search, but Perplexity Computer is more thorough for shopping research. ### Step 2 Give it a simple shopping brief Keep the prompt simple. Give Perplexity the item type, your preferences, your size, and what kind of shopping output you want. ```text Help me shop for [item type]. I want: - brand preferences: [brand or "no preference"] - cosmetic style: [clean / modern / minimal / bold / etc.] - sizing: [size] Give me: 1. The best overall options 2. The best budget option 3. Items that are actually in stock in my size 4. Direct links to buy ``` That is enough to get good results. If you want to add color or budget, you can, but you do not need to write a giant prompt. ![image-2150](https://tru-images.b-cdn.net/guide-assets/69/69c59e1735ee66f5ffd79811/image-2150.png) > Pro tip: Ask Perplexity to “find you a deal,” and it is smart enough to look for sales and discounts. Perplexity behaves like a real search query. If you give it two brands, it sticks to those two brands instead of wandering all over the place. > Pro tip: Before you spend Computer credits, run the same shopping question in regular Perplexity Assistant so you can see the difference. ### Step 3 Manage your credits and follow-up research Use your credits wisely. Perplexity Computer uses credits; regular Perplexity search does not. If you want to save a little on tokens, switch the model to Claude Sonnet for the first pass. Then use Perplexity Computer when you are ready to do the real shopping work. If you end up using this a lot, you can upgrade to Max or buy more tokens à la carte. To stretch your tokens further, take your shopping hit list into a regular session in Perplexity’s web browser, Comet. It can control your browser and do slower follow-up research for you. > Pro tip: Perplexity shopping has an “Instant Buy” feature where you enter your address and credit card, and it will use them to check out for you. As of when this guide released, it was only available for specific retailers that had opted in. ### Step 4 Order and track your purchase Try ordering through the Comet or Perplexity app. You should see your order being tracked in the “Orders” tab. Make sure you connect your Gmail too. ### Pro tip > Ask Perplexity to “find you a deal” so it looks for sales and discounts before you buy. ### Guide: Use Replit’s New Tasks Feature To Improve Your Site Faster > This guide teaches you how to use Replit’s Tasks feature to improve an existing website or app without disrupting the version that already works. - Canonical page: https://app.therundown.ai/guides/use-replits-new-tasks-feature-to-improve-your-site-faster - Machine-readable page: https://app.therundown.ai/guides/use-replits-new-tasks-feature-to-improve-your-site-faster.md - Difficulty: intermediate - Featured tool or company: Replit - Categories: Coding & App Development - Instructors: Billy Howell - Published: 2026-03-19 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches you how to use Replit’s Tasks feature to improve an existing website or app without disrupting the version that already works. You will create isolated parallel tasks, review each task in its own preview, and merge only the changes you want back into your main version. #### Who This Is Useful For - Anyone intimidated by Git who wants a simpler way to review and merge isolated changes - People who get into doom loops while vibe coding and need a cleaner way to make progress without breaking what already works - Founders, freelancers, and operators who want to improve one part of an app at a time without juggling conflicting edits #### What You Will Build You will set up parallel sub-agents inside Replit, with each one focused on a specific improvement to your website or app. The reusable output is a safer workflow for testing and merging improvements one at a time. - A mobile responsiveness task - A second parallel improvement task, such as a landing page refresh, navigation cleanup, or bug fix - Separate task previews for reviewing work before merging - A main app version updated only with approved changes ![image-2133](https://tru-images.b-cdn.net/guide-assets/69/69bc65f1b968b6e084a9cf4a/image-2133.png) #### What You Need - A Replit account - Access to Replit Tasks; the source tested this on Replit Core and notes Tasks may require Replit Pro after the stated Core availability period - An existing app to improve, or a clear app idea if you are starting from scratch #### Going Further - To go further, use Plan Mode to outline all the improvements you want to make, then break that plan into separate parallel tasks instead of asking the main agent to do everything at once. Good candidates include mobile responsiveness, UI cleanup, bug fixes, onboarding flows, and landing page sections. ### Step 1 Start in Plan Mode Go to replit.com. If you are starting from scratch, enter your app idea and choose Plan first. If you already have an app, open it and review the plan Replit generates before you start building. The plan shows how Replit is thinking about the work. It gives you a task list, notes what should not be touched, and defines what “done” looks like before code changes begin. > Take a second to actually read the plan. In the source test, it was one of the most useful parts because it clearly defined the tasks and the finish line. ### Step 2 Create Your First Task Once you are in your Replit project, click the plus button on the far-left sidebar to open the task menu. This is where you can create a separate task for one focused improvement. ![image-2134](https://tru-images.b-cdn.net/guide-assets/69/69bc65feb968b6e084a9d7cb/image-2134.png) ```text Make this dashboard and all relevant components fully mobile responsive. Use different components for mobile if not possible. ``` That last line matters. Sometimes components do not collapse cleanly on smaller screens. Telling Replit it can use different mobile components gives it more room to solve the problem instead of squeezing the same layout into a phone screen. ### Step 3 Queue a Second Task in Parallel While the first task is planning, queue a second one in parallel. This could be something like modernizing the landing page, cleaning up the nav bar, or fixing a bug. This is the value of Tasks: instead of piling everything into one thread and hoping the agent keeps track of it all, you split the work into isolated jobs that can run separately. > Keep each task narrow. “Modernize the landing page” is okay. “Redesign the whole app, fix three bugs, and improve onboarding” is how you get back into a doom loop. ### Step 4 Review Each Task Before You Merge Let each task generate its own plan, then start them. ![image-2135](https://tru-images.b-cdn.net/guide-assets/69/69bc6611f42dbbb357d5a228/image-2135.png) Replit treats tasks like mini agents with separate previews, so you can inspect the work without affecting your main app. When a task finishes, it turns green. Click into it and open the preview tab to review the result. In the source test, the mobile responsiveness task was the clear win. The landing page refresh looked better too, but it also showed why vague prompts usually need a second pass. ### Step 5 Apply the Good Changes When you find a task you like, click Apply Changes to Main Version to merge that updated code back into your production app. ![image-2136](https://tru-images.b-cdn.net/guide-assets/69/69bc6623b968b6e084a9e08e/image-2136.png) This makes the workflow safer: you are not committing to every idea the agent had. You review each isolated result, keep the good work, and leave the rest behind. ### Pro tip > Keep each task narrow so the agent can solve one isolated problem at a time instead of pushing the project back into a doom loop. ### Guide: Use the Loop Method to Get Better Results From ChatGPT > In this guide, you will learn how to improve any repeated ChatGPT workflow with the "Loop Method". - Canonical page: https://app.therundown.ai/guides/use-the-loop-method-to-get-better-results-from-chatgpt - Machine-readable page: https://app.therundown.ai/guides/use-the-loop-method-to-get-better-results-from-chatgpt.md - Difficulty: beginner - Featured tool or company: ChatGPT - Categories: AI, Productivity - Instructors: Billy Howell - Published: 2026-08-19 - Updated: 2026-09-11 ### Overview #### The Rundown In this guide, you will learn how to improve any repeated ChatGPT workflow with the "Loop Method". We created it to improve our video editing and image generation skills but you can try it on anything! #### Who This Is Useful For - Creators improving a repeated writing, editing, design, or image workflow. - Operators who want a useful AI process to produce more consistent results. - Anyone tired of correcting the same mistakes every time a workflow runs. #### What You Will Build You will set up a panel of sub-agents to adversarially review three improvement loops. In every loop, the panel finds the biggest weakness, ChatGPT changes the workflow, and the next result is checked against a clear finish line. ![chatgpt iterative review panel](https://tru-images.b-cdn.net/guide-assets/local-437/use-the-loop-method-to-get-better-results-from-chatgpt/01-chatgpt-iterative-review-panel.png) ChatGPT created a sub-agent panel for art direction, workflow design, and location accuracy before changing the skill. #### What You Need To Get Started - The ChatGPT desktop app with Work or Codex and sub-agent delegation available in your workspace. - A skill, prompt, project folder, or repeatable process that already produces a result. #### Going Further Save the loop prompt beside the workflow. Reuse it when quality starts to drift, changing the sub-agent roles and definition of done for the job. Save the method as a small process-improvement skill: ```text Run the saved process-improvement skill on [WORKFLOW]. Use [2 OR 3] loops with a sub-agent panel that adversarially reviews every loop. Preserve earlier versions and return the improved workflow plus a change report for my review. ``` ### Step 1 Pick a Workflow That Needs Improvement Choose a workflow that works but needs the same cleanup each time. For example: - a writing skill with an inconsistent voice; - an editing process that misses obvious cuts; - an image workflow that produces an uneven series. Run it once without changing anything. Save the instructions and result as your baseline. Use this quick brief: ```text Workflow: [NAME OR FOLDER] Test input: [INPUT] Current result: [OUTPUT OR FILE] Problems I noticed: [LIST] Files or instructions to preserve: [LIST] ``` The example in this guide uses a location-based image generator skill. It takes a place name and creates three related illustrations. The first Nags Head result became the baseline for testing research, style consistency, and output format. ![location skill first result](https://tru-images.b-cdn.net/guide-assets/local-437/use-the-loop-method-to-get-better-results-from-chatgpt/02-location-skill-first-result.png) The first location image gave ChatGPT a concrete baseline to review. ### Step 2 Ask ChatGPT to Run Three Review Loops Paste the template below with your workflow attached or open in the project: ```text Improve [WORKFLOW] through exactly 3 loops. Create a panel of sub-agents to adversarially review every loop. Give the panel three roles: - an outcome or quality reviewer - a workflow and prompt reviewer - a reviewer with relevant subject knowledge For each loop: 1. Review the current workflow and result. 2. Identify the biggest remaining failure. 3. Change the smallest part of the prompt, script, checklist, or process needed to fix it. 4. Run the relevant part again. 5. Check the result against the definition of done. After each loop, report: - what the panel found - what changed - what evidence shows the change helped - what still needs work Preserve earlier versions. Stop after loop 3. Do not publish, overwrite source files, or take external actions without my approval. Do not begin until I provide the goal and confirm the run. ``` Change the three roles to fit the job. An editing workflow might use an editor, pacing reviewer, and visual QA reviewer. An image workflow might use an art director, prompt reviewer, and location critic. ### Step 3 Give the Run a Clear Goal Define what the improved workflow must do before you start the loops: ```text Goal: [CAPABILITY TO IMPROVE] Definition of done: - [CHECK 1] - [CHECK 2] - [CHECK 3] Confirm the goal and checks with me before starting loop 1. ``` Use checks you can verify in the output. For the location-image skill, the result needed three separate images with consistent dimensions, a shared style, and recognizable details from the named place. > Pro tip: If the finish line is fuzzy, ask ChatGPT Work to propose a goal and definition of done. Edit it until you can tell from the next result whether the workflow passed. ### Step 4 Run the Loops and Review the Reports After you approve the goal, start the run: ```text The goal and definition of done are approved. Begin loop 1. Run all 3 loops. After each loop, show the panel findings, the workflow change, the evidence, and the biggest remaining issue before continuing. ``` Check each report for four things: - Did the sub-agent panel find a specific failure? - Did ChatGPT change the workflow instead of only repairing the output? - Did the next result improve against the same checks? - Is each change saved separately so you can compare versions? In the location-image example, the panel found generic place details and weak consistency rules. A later two-panel result exposed an ambiguous instruction, which was corrected before the next loop. ![reviewed location image set](https://tru-images.b-cdn.net/guide-assets/local-437/use-the-loop-method-to-get-better-results-from-chatgpt/03-reviewed-location-image-set.png) The reviewed workflow produced a more deliberate three-image set with shared visual rules. If a result reveals an unclear instruction, correct the instruction before the next loop: ```text The last result exposed this ambiguity: [AMBIGUOUS INSTRUCTION]. Update the workflow so it means: [PRECISE REQUIREMENT]. Apply that correction in the next loop and verify it in the output. ``` > Pro tip: When a result goes sideways, name the ambiguity before you run another loop. A specific correction—such as “three independent artworks, not a two-panel set”—gives the next pass something it can actually test. ### Step 5 Test the Improved Workflow on a New Example Change the input without rebuilding the workflow: ```text Use the updated [WORKFLOW] on [SECOND INPUT]. Keep the reviewed instructions and definition of done. Save the result separately from the first example, then report: - whether it passed each check - what still needs my review - which manual judgments should become scripts, rules, or checks ``` ![second location reuse](https://tru-images.b-cdn.net/guide-assets/local-437/use-the-loop-method-to-get-better-results-from-chatgpt/04-second-location-reuse.png) A second location showed whether the revised research and series rules worked beyond the first example. If the second test fails, fix the workflow and repeat the test with another input. Do not keep patching only the original output. ### Guide: Use These AI Skills To Make Better Decisions Quickly > In this guide, you will learn how to give your AI skills that make every decision easier. - Canonical page: https://app.therundown.ai/guides/use-these-ai-skills-to-make-better-decisions-quickly - Machine-readable page: https://app.therundown.ai/guides/use-these-ai-skills-to-make-better-decisions-quickly.md - Difficulty: beginner - Categories: AI - Instructors: Billy Howell - Published: 2026-08-27 - Updated: 2026-08-27 ### Overview #### The Rundown In this guide, you will learn how to give your AI skills that make every decision easier. You'll use an /unstuck tool, a /decide tool, and more to turn problems into solutions. #### Who This Is Useful For - Operators who need to break a stalled project into a practical next move. - Consultants choosing between competing projects, clients, or priorities. - Anyone pressure-testing a business idea before investing more time or money. #### What You Will Build You will build a repeatable decision workflow that turns a stuck problem into focused questions, challenged assumptions, and a concrete action. Depending on the skill you choose, that output could be a five-post publishing plan or a project decision with a clear recommendation and revisit date. ![Unstuck five-post plan](https://tru-images.b-cdn.net/guide-assets/local-435/use-these-ai-skills-to-make-better-decisions-quickly/01-unstuck-five-post-plan.png) #### What You Need To Get Started - A Claude Code installation and access to Terminal. - The free makerskills repository (https://github.com/coreyhaines31/makerskills). - A real decision, stalled project, or business idea you can describe with specific constraints. #### Going Further /business-brainstorm is a useful extra skill when you have a promising idea but need to shape and pressure-test it. It is not part of the main decision framework, but it can push a loose concept toward a differentiated product: ```text /business-brainstorm I want to launch [PRODUCT OR BUSINESS] for [CUSTOMER]. I want to compete in [CATEGORY OR PRICE POSITION] without copying [REFERENCE BRANDS]. Help me develop a differentiated hero product, pressure-test the positioning, and outline likely pricing and cost considerations. ``` We tested this with a premium blanket idea. /business-brainstorm came back with a two-sided product, name ideas like Sunday Weight, and a hook around a blanket that stays in place. ![Business brainstorm Sunday Weight result](https://tru-images.b-cdn.net/guide-assets/local-435/use-these-ai-skills-to-make-better-decisions-quickly/05-business-brainstorm-sunday-weight.png) In the video, we followed up with: ```text Turn the recommendation into a one-page product brief and draft a sourcing RFQ. Use an initial order target of about [QUANTITY] units rather than assuming a large production run. ``` ### Step 1 Install the Makerskills Library Open Terminal, start Claude Code, and add the makerskills marketplace: ```text claude /plugin marketplace add coreyhaines31/makerskills ``` Then install the library: ```text /plugin install makerskills@makerskills ``` If the marketplace is already installed, Claude Code may simply confirm that nothing needs to change. You can also open the marketplace inside Claude Code, find makerskills, and install it from the menu. > Pro tip: Start Claude Code before pasting the /plugin commands. They are Claude Code slash commands, not shell commands. ![Makerskills install commands](https://tru-images.b-cdn.net/guide-assets/local-435/use-these-ai-skills-to-make-better-decisions-quickly/02-makerskills-install-commands.png) ### Step 2 Match the Problem to the Right Skill Open the repository's routing table before you start. It organizes the library by the job you are trying to do, which is faster than opening every skill one by one. Two useful starting points are: - /unstuck when you keep circling the same problem or assumption. - /decide when you have a real fork and need to choose. In Claude Code, type /, choose the skill, and describe the outcome you need. Start with the skill that matches the decision in front of you instead of trying to test the whole library at once. > Pro tip: Run the skill inside a project or session that already has context on your work. It will ask more specific questions and make more relevant recommendations. ![Makerskills routing table](https://tru-images.b-cdn.net/guide-assets/local-435/use-these-ai-skills-to-make-better-decisions-quickly/03-makerskills-routing-table.png) ### Step 3 Describe the Situation and Work Through the Questions Give the skill the real problem, the result you want, the constraint slowing you down, and any deadline. For example: ```text /unstuck I am stuck on [PROJECT OR PROBLEM]. I keep spending time on [ASSUMPTION OR BOTTLENECK] instead of [DESIRED ACTION]. I want to reach [CONCRETE OUTCOME] by [DEADLINE], with these constraints: [CONSTRAINTS]. Help me challenge what I am assuming and choose the lowest-friction next move. ``` Answer the follow-up questions plainly. The value comes from letting the framework test whether the apparent blocker is the real one. We saw this when social-fetch could retrieve a Hacker News post but could not pull an X post. Its next route required a paid scraping API, which would have turned the tutorial into a costly, gray-area demo. We used /unstuck to challenge our assumption that we needed the social-fetch demo. It pointed us toward the more useful decision workflow instead. The same skill helped with a local newsletter that had gone nearly three weeks without consistent social posts because the design never felt finished. It challenged the value of polish and proposed five low-friction formats: a plain-text post, a newsletter screenshot, a phone photo, a local list, and a neighborhood question. It then recommended batching all five into a scheduler. > Pro tip: If a tool sends you toward paid access, brittle scraping, or a workflow you do not want to own, treat that as new evidence. Revisit the original goal before adding more setup. ### Step 4 Apply the Framework to Another Real Job Once you understand the pattern, try a different skill on a second problem. /decide works well when both options are plausible but you only have room for one. Give it the options, the deadline, the cost of being wrong, and how reversible the choice is: ```text /decide I need to choose between [OPTION A] and [OPTION B] by [DECISION DATE]. I only have capacity for one. Option A offers [UPSIDE] but requires [COST OR RISK]. Option B offers [UPSIDE] but has [TRADEOFF]. Interview me, challenge the assumptions behind the choice, recommend the next move, and give me a date to revisit the decision. ``` We tested /decide with two hypothetical client projects. It asked how much the decision mattered, tested whether it was reversible, recommended the new work stream, and set a future revisit date. ![Decide client project result](https://tru-images.b-cdn.net/guide-assets/local-435/use-these-ai-skills-to-make-better-decisions-quickly/04-decide-client-project-result.png) ### Guide: Use This Hidden Feature To Make Your Notion Agents Autonomous > In this guide, you will learn a hidden workflow that makes Notion Agents more autonomous and powerful than they are by default. - Canonical page: https://app.therundown.ai/guides/use-this-hidden-feature-to-make-your-notion-agents-autonomous - Machine-readable page: https://app.therundown.ai/guides/use-this-hidden-feature-to-make-your-notion-agents-autonomous.md - Difficulty: intermediate - Featured tool or company: Notion - Categories: Automation & Agents, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-05-05 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn a hidden workflow that makes Notion Agents more autonomous and powerful than they are by default. You will create a recurring page that wakes up an agent, gives it a specific task, and leaves you with a report on what it did. #### Who This Is Useful For - Operators and founders already running custom Notion agents who have hit the limits of the built-in agent schedule. - Anyone with a tasks-and-reports system in Notion who wants daily summaries, planning pages, or recurring checks running on autopilot. - Teams trying to make Notion agents reliable without cramming every possible job into one giant set of standing instructions. #### What You Will Build ![image-2202](https://tru-images.b-cdn.net/guide-assets/69/69fa53ec107d28ec1ddd7ba1/image-2202.png) You will build a recurring database template that creates a new page on a schedule, auto-dates the title, tags your agent, and gives it the exact job to run that morning. In the source setup, the template is a daily summary that runs every weekday at 7 a.m. and writes a recap of completed tasks, new content, and what other agents did overnight. - A recurring database template for agent runs. - An auto-dated page title using @Today. - Pre-filled properties such as agent, task type, and status. - A reusable page-body prompt that @ mentions the agent. - A recurring schedule for the template. #### What You Need - A Notion workspace with a business plan or higher. - One database where the recurring pages should live, such as a Reports database. - A custom Notion agent, or the ability to create one from the Agents section in the sidebar. - At minimum, a title and status column in the database if you are building it from scratch. #### Going Further - Once one template works, you can reuse the same pattern for daily debriefs, weekly reports, and email automations. This lets you route recurring work through the same planning agent instead of building a new one every time you find a new recurring task. ### Step 1 Create the agent Look for Agents in the sidebar and create a custom agent if you do not already have one. ![image-2203](https://tru-images.b-cdn.net/guide-assets/69/69fa54a947f2cd9cd28ad47e/image-2203.png) > Pro tip: You will see a schedule-based trigger on the agent itself. The problem is that it wakes the agent and runs the same system instructions every time, which can fall apart for any agent that does more than one job. ### Step 2 Open the Reports database Open the database where the recurring pages should live. In the source setup, this database is called Reports. Each daily run creates one new page in this database, giving you a clean, auditable history of every agent run. If you do not have a database yet, build one. Two columns is enough to start: a title and a status. ### Step 3 Make a new template called Daily Summary @Today In the database, click the New dropdown in the top right and open Templates. ![image-2204](https://tru-images.b-cdn.net/guide-assets/69/69fa54c947f2cd9cd28add61/image-2204.png) Create a new template and name it Daily Summary @Today. The @Today makes every duplicate inherit the current date in the title, so you do not have to rename anything. ![image-2205](https://tru-images.b-cdn.net/guide-assets/69/69fa54de47f2cd9cd28ae64b/image-2205.png) ### Step 4 Pre-wire the template Inside the template, do three things: - Set the properties you want pre-filled, such as agent, task type, and status. - Write the actual job in the page body. - @ mention the agent so the duplicate triggers it. A clean prompt to drop into the page body: ```text Hey @[your-agent], write a daily debrief for me. - 2-sentence recap up top - Bullet points by section - Pull completed tasks from @[tasks-database] - Summarize new content created today - Roll up reports written by other Notion agents Keep it short and skimmable. ``` > Pro tip: When you @ mention the agent inside the new template, open the agent and stop it. Otherwise, it triggers right then on the half-built template and overwrites it. ### Step 5 Set the template to recur From the template, click New template, then Duplicate, then pick a cadence. In the source setup, the template runs every weekday at 7 a.m. Look for the small blue recurring icon next to the template name to confirm it is live. If you do not see it, the schedule did not save. ### Pro tip > When you @ mention the agent inside the new template, open the agent and stop it so it does not trigger on the half-built template and overwrite it. ### Guide: Use This Two-Step Dictation Strategy To Write Better Docs (Typeless Tutorial) > This guide teaches a two-step voice dictation system for writing better docs with AI. - Canonical page: https://app.therundown.ai/guides/use-this-two-step-dictation-strategy-to-write-better-docs-typeless-tutorial - Machine-readable page: https://app.therundown.ai/guides/use-this-two-step-dictation-strategy-to-write-better-docs-typeless-tutorial.md - Difficulty: beginner - Featured tool or company: Typeless - Categories: Productivity & Knowledge Work, Content & Creative - Instructors: Billy Howell - Published: 2026-04-22 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches a two-step voice dictation system for writing better docs with AI. You will create a repeatable workflow where AI drafts the first structure, Typeless captures your spoken edits, and an agent rewrites the working draft in your voice. The system also improves over time by saving a before-and-after trail of your drafts. #### Who This Is Useful For - Founders and operators who write a lot of memos, updates, and internal docs but do not want them to sound like generic AI - Creators and marketers who use AI for first drafts but still need the final version to sound like their actual voice - Anyone using Codex, Claude Code, Claude, or ChatGPT who wants a repeatable writing system instead of starting from scratch every time #### What You Will Build ![image-2180](https://tru-images.b-cdn.net/guide-assets/69/69e936d75f406836371a3fe1/image-2180.png) You will build a reusable writing workflow where AI creates the first structure, Typeless captures your spoken edits, and an agent rewrites the working draft in your voice. - An untouched initial draft - A working draft you can edit freely - A before-and-after trail that helps an agent extract the editorial rules you keep applying #### What You Need - Typeless installed for dictation; the free plan works - Codex, Claude Code, Claude, ChatGPT, or another AI tool that can draft and revise documents - A document you want to write, like a memo, project update, guide, proposal, or email - A simple place to save markdown files, docs, or drafts #### Going Further - Once you have an initial draft and a final working draft, you can ask an agent to compare them and extract editorial rules. You can also turn the process into a weekly automation that looks for new initial and final drafts, compares them, and updates your editorial rules file so future writing sessions have more context. - Look at the initial version and the final working draft. Pull out the editorial rules for future writing. Add them to an editorial-rules.md file. Keep the rules concise and de-duplicate anything repetitive. ### Step 1 Set Up Typeless Download Typeless and go through the setup flow. Give it the permissions it needs so it can use your microphone and paste text into any text box. ![image-2178](https://tru-images.b-cdn.net/guide-assets/69/69e936a85f406836371a2e6c/image-2178.png) Set a hotkey you can use without thinking about it. In the recording, the hotkey was Option + Space Bar. Typeless is helpful because you can talk faster than you type, and it cleans up stutters and rough phrasing better than normal dictation. It also has an AI search hotkey, so you can quickly ask something without switching over to ChatGPT or Gemini. > Use Typeless for comments, not just finished writing. The comments are where your voice comes through most clearly. ### Step 2 Have AI Draft the Outline Open a new Codex or Claude Code session, or use Claude or ChatGPT if that is where you write. Ask for a rough outline or first draft. Do not try to make it perfect yet. ```text Draft an outline for a short internal memo about [topic]. Keep it concise. ``` For the demo, the draft was a made-up project update for an AI software app. The first draft does not need to sound like you. It just needs to give you something to react to. AI is good at giving you structure. Your voice comes in during the edit pass. ### Step 3 Save the Initial Draft and Create a Working Draft Before you start editing, tell the agent to preserve the first version. ```text Save this as the initial draft and do not edit it. Now create a separate working draft that we can revise. ``` ![image-2179](https://tru-images.b-cdn.net/guide-assets/69/69e936b25f406836371a3717/image-2179.png) The initial draft gives you a clean before version. The working draft is where all the edits happen. Over time, that difference becomes useful. You can compare the generic AI version against the final version and see exactly what changed: shorter intros, fewer generic phrases, more bullets, different endings, stronger examples, or no em dashes. ### Step 4 Brain Dump Your Comments Read the working draft and add a comments section at the bottom. Then use Typeless to talk through everything you would change. Start with factual fixes first. Then add missing context. Then call out anything that sounds generic, too polished, or not like you. Your comments can be messy. That is the point. ```text The overview needs to be two sentences. Current progress needs to have bullet points. Product focus needs to say this version adds social functionality. Say this version is built for virality. Make the conclusion less wordy. Make sure everything is in my voice and verbiage. ``` This is faster than typing edits one by one. You are basically giving the model a live style guide based on your real reaction to the draft. ### Step 5 Rewrite the Working Draft in Your Voice Once your comments are in the working draft, send the final rewrite prompt. ```text Rewrite the draft using the comments I left in the draft. Write it in my tone. Use my verbiage. No em dashes. Preserve the core points, but cut anything that sounds generic. Make it concise. ``` The rewrite should look noticeably different from the original AI draft. In the recording, the final version was much less wordy and much closer to the actual direction from the comments. If it is close but still not right, do another quick Typeless pass. Say the exact issue, then tell the agent: ```text Make only those edits. ``` That keeps the model from over-rewriting parts that are already working. ### Pro tip > Use Typeless for comments, not just finished writing. The comments are where your voice comes through most clearly. ### Guide: Use This X + OpenClaw Setup To Write Viral Tweets > In this guide, you will learn how to connect X to OpenClaw so your agent can research who you follow, read ideas, draft posts, and tweet for you. - Canonical page: https://app.therundown.ai/guides/use-this-x-openclaw-setup-to-write-viral-tweets - Machine-readable page: https://app.therundown.ai/guides/use-this-x-openclaw-setup-to-write-viral-tweets.md - Difficulty: beginner - Featured tool or company: The Rundown AI - Categories: Marketing & Growth, Automation & Agents - Instructors: Billy Howell - Published: 2026-06-11 - Updated: 2026-07-23 ### Overview #### The Rundown In this guide, you will learn how to connect X to OpenClaw so your agent can research who you follow, read ideas, draft posts, and tweet for you. Once connected, OpenClaw can turn X activity into research, reply ideas, and posts. #### Who This Is For This is useful if you want an always-on agent that can watch your X context and help with posting without starting from a blank prompt every time. It works especially well for founders, creators, operators, and research-heavy teams who save ideas on X, follow niche accounts, or want a daily feed of posts worth reading, replying to, or turning into content. #### What You Will Build You will set up an X developer app that lets OpenClaw connect to your account through xurl. ![X App Permissions And Bot Type](https://tru-images.b-cdn.net/guide-assets/local-230/use-this-x-openclaw-setup-to-write-viral-tweets/01-x-app-permissions-and-bot-type.png) By the end, your OpenClaw agent can authenticate with X and use that connection for workflows like: - researching accounts or posts you follow - checking saved posts or lists for useful ideas - drafting replies or tweets from those ideas - turning one bookmarked post into a thread outline - optionally posting from the connected account after you approve the workflow #### What You'll Need - An OpenClaw agent - An X account - Access to the X Developer Console - Terminal access to the machine where OpenClaw runs - The xurl X skill #### Going Further Once X is connected, make the workflow more specific. Have OpenClaw maintain a list of accounts worth watching, build a daily "reply queue," summarize AI use cases from a saved list, or turn bookmarked posts into a weekly thread plan. If you use this for posting, keep a review checkpoint in the workflow. OpenClaw can do the boring research and drafting. You should still decide what represents you in public. ### Step 1: Start With OpenClaw You need an OpenClaw agent before you can connect X to it. You can run OpenClaw on your own computer, on another machine, or on a hosted server. For most readers, Hostinger OpenClaw is the simplest path because it gives you an always-on agent without putting the setup on your personal computer. ![Hostinger OpenClaw Landing](https://tru-images.b-cdn.net/guide-assets/local-230/use-this-x-openclaw-setup-to-write-viral-tweets/02-hostinger-openclaw-landing.png) > Pro tip: You can use a Hermes Agent too. The X setup is the same, but this guide uses OpenClaw as the default example. If you use Hostinger, make sure you can reach the VPS terminal. You will need that later when xurl gives you a login link for the X OAuth flow. ### Step 2: Create The X Developer App Open the X Developer Console, create a new app, and save the keys when X shows them. Do not overthink the word "app" here. You are not building a web app or mobile app for customers. The app is just the wrapper that lets your bot connect to X through the API. Set the app permissions based on what you want OpenClaw to do: - Use Read if you only want research. - Use Read and Write if you want OpenClaw to draft and post tweets from the account. - Do not enable Direct Message access unless you actually need it. For the app type, choose: ```text Web App, Automated App or Bot ``` That is the option made for this kind of bot connection. ### Step 3: Add The Callback URL In the app auth settings, add this callback URL: ```text http://localhost:8080/callback ``` Then set the website URL to: ```text https://x.com ``` ![X App Callback Url](https://tru-images.b-cdn.net/guide-assets/local-230/use-this-x-openclaw-setup-to-write-viral-tweets/03-x-app-callback-url.png) This callback URL matters because xurl runs the auth flow locally on the machine where the agent is installed. In the recording, the confusing part was not the X app setup. It was remembering that the OAuth login has to happen from the machine where OpenClaw or Hermes is actually running. ### Step 4: Install xurl In OpenClaw Before installing the X skill, ask OpenClaw to update itself. Then have it install xurl. Use this prompt: ```text Update yourself, install the xurl X skill, and tell me what I need to configure locally before you can use X. ``` OpenClaw should tell you what it needs next, usually the app name, client ID, client secret, and X handle. Do not paste secrets casually into an agent chat. Save the credentials securely and enter them through the local setup flow where xurl expects them. ### Step 5: Authenticate From The Right Machine This is the part that makes the setup feel more annoying than it should. The OAuth link needs to be generated from the machine where OpenClaw runs. If OpenClaw is on your laptop, run the xurl auth flow there. If OpenClaw is on Hostinger, open: ```text VPS > Manage > Terminal ``` Then run the xurl auth flow from that terminal. It should give you a login link. Click that link, log into X, authorize the app, and finish the setup. This means you do not need a full desktop GUI on the VPS. You just need terminal access and the link that xurl gives you. Once auth succeeds, test the connection with a tiny approved action. If you want the safest first test, ask OpenClaw to read account or timeline context before you ask it to post. ### Step 6: Turn The Connection Into A Useful Workflow The point is not just "OpenClaw can tweet now." The useful part is giving OpenClaw repeatable X jobs. Start with a simple daily research workflow: ```text Every day at 7am, review my saved X list or bookmarks. Pick one or two posts that are most useful for: - new AI use cases - OpenClaw upgrade ideas - reply opportunities - tweet ideas for my account For each one, explain why it matters, summarize the core idea, and draft either: - three possible replies - one standalone tweet - one short thread outline Do not post anything until I approve it. ``` You can also use the connection for more focused tasks: ```text Look at the accounts I follow in [TOPIC]. Find five patterns people are discussing this week. Turn those patterns into tweet ideas for [AUDIENCE]. ``` Or: ```text Review my X bookmarks from the last [LOOKBACK_DAYS] days. Group them by theme. Pick the strongest idea and draft a post in my normal tone. ``` The safest setup is to let OpenClaw research, draft, and package ideas first. Once you trust the workflow, you can decide when posting should be manual, approval-based, or fully automated. ### Guide: Vet Business Opportunities With This NotebookLM Research System > In this guide, you will learn how to turn a rough business idea into a source-backed research brief with NotebookLM. - Canonical page: https://app.therundown.ai/guides/vet-business-opportunities-with-this-notebooklm-research-system - Machine-readable page: https://app.therundown.ai/guides/vet-business-opportunities-with-this-notebooklm-research-system.md - Difficulty: beginner - Featured tool or company: The Rundown AI - Categories: Data, Research & Analysis, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-06-15 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you will learn how to turn a rough business idea into a source-backed research brief with NotebookLM. Start with any business decision and build a NotebookLM notebook with relevant sources, short memos for each potential option, and a final recommendation that makes the case using real evidence. #### Who This Is For This is useful if you need to make business decisions faster, but still want the answer to be grounded in real sources. Use it when you are: - comparing multiple vendors - evaluating a partnership opportunity - deciding whether a new market is worth exploring - researching software tools before a sales call - turning scattered assumptions into a source-backed memo - preparing a recommendation for a manager, client, partner, or team The workflow is especially helpful when the decision is too important for a quick gut check, but not important enough to justify a week-long research project. #### What You Will Build You will build a lightweight decision system inside NotebookLM. The point is not just to collect sources. The point is to make a better call on a real business decision without turning every question into a giant research project. The finished notebook should help you answer: - what options are worth considering - what evidence supports each option - where the research is still thin - which option fits the decision best - what to verify before committing time or money By the end, you should know which option to pilot first, what evidence supports that call, and what assumptions need verification before you spend time or money. ![Ai Receptionist Vendor Comparison Chart](https://tru-images.b-cdn.net/guide-assets/local-244/vet-business-opportunities-with-this-notebooklm-research-system/01-ai-receptionist-vendor-comparison-chart.png) In the recording, that decision was which AI receptionist vendor a fictional restaurant group should pilot first. The final output compared Goodcall, Smith.ai, and Slang.ai, then turned the research into a recommendation memo and comparison chart. For your own work, the options could be vendors, partners, markets, agencies, products, or any opportunity where the best answer is not obvious from one sales page. #### What You'll Need - A Google account with access to NotebookLM - ChatGPT, Claude, Gemini, or another chat tool to create the seed memo - A business decision with at least two or three options to compare - A few minutes to review and import sources from NotebookLM's research feature For the demo, the starting decision is: ```text Which AI receptionist vendor should Harbor House Hospitality pilot first: Goodcall, Smith.ai, or Slang.ai? ``` #### Going Further Once you have a final recommendation memo, you can turn it into other formats. You can ask NotebookLM for a briefing doc, infographic, or audio-style summary. You can export the memo to Google Docs and clean it up for internal sharing. You can also paste the finished memo into ChatGPT, Claude, or Gemini and ask it to turn the output into a cleaner executive memo, slide outline, or reusable template for the next notebook. The main thing is to keep NotebookLM focused on what it does best: building a source-backed research workspace. Use it to find sources, inspect gaps, synthesize options, and compare the evidence. Then use whatever tool gives you the cleanest final format. For everyday business decisions, that is the practical win. You get a lightweight research system you can actually use, without turning every decision into a giant research project. ### Step 1: Create A Seed Memo Outside NotebookLM Start somewhere other than NotebookLM. Open ChatGPT, Claude, Gemini, or whichever general chat tool you already use, and talk through the decision you need to make. The goal is not to get the final answer yet. The goal is to create a simple seed memo that explains the business context clearly enough for NotebookLM to start researching. In the recording, the seed memo covered: - the business decision - the buyer profile - the options being compared - the budget and implementation constraints - the must-have requirements - the current hypothesis - the research questions that still needed answers You can create that memo by talking naturally to the model. This is useful because you can do the pre-work quickly, even away from your desk. You do not need to gather every document first. Use a simple request like this: ```text Help me turn this business decision into a one-page research memo. Decision: [describe the decision] Options: [list the options] Context: [describe the business, customer, team, or buyer] Constraints: [budget, timing, technical requirements, risk tolerance] Current hypothesis: [what you are leaning toward, if anything] Write this as a clear memo I can upload into NotebookLM. Include the main research questions we still need to answer. ``` Then save the memo as a Google Doc, Markdown file, PDF, or plain text. NotebookLM can work with several source types, so use whatever format is easiest for your workflow. ### Step 2: Upload The Memo Into NotebookLM Open NotebookLM and create a new notebook. When NotebookLM asks you to add sources, upload the seed memo first. In the recording, this was a Markdown memo, but you can also add files, website links, YouTube videos with transcripts, Google Drive files, or pasted text. Once the memo is in the notebook, do not immediately ask NotebookLM to choose a winner. Start by having it read the memo and tell you what is missing. Use a short ask: ```text Read this memo and identify any research gaps. ``` NotebookLM should identify what the memo does not yet prove. In the AI receptionist example, it pointed out that the notebook still needed vendor capabilities, pricing models, integrations, and proof points before it could make a reliable recommendation. That is the right starting point. You want NotebookLM to help plan the research before it writes the analysis. > Pro tip: Make NotebookLM inspect the source set before it compares anything. Ask which sources are vendor-written, which are third-party, and what is still missing. ### Step 3: Use NotebookLM Research To Find Sources After NotebookLM identifies the gaps, use its research feature to fill them. In the recording, NotebookLM suggested starting Deep Research on the vendors. You can also trigger this manually from the source panel. Look for the research controls in the left column, choose Fast Research or Deep Research, and give it a focused research task. For this demo, useful searches included: ```text Goodcall AI receptionist pricing features integrations ``` ```text Smith.ai AI receptionist pricing human backup integrations ``` ```text Slang.ai restaurant AI receptionist pricing OpenTable SevenRooms ``` Fast Research is better when you want a quick set of candidate sources. Deep Research can take longer, but it can return a richer set of sources. In the recording, Deep Research took a few minutes and returned a list of discovered sources that could be reviewed before import. When the research finishes, do not import everything blindly. Review the source list. Select the sources that actually help the decision, deselect anything weak or repetitive, and then click import. For a vendor decision, make sure you have sources like: - official product pages - pricing pages - integration pages - help center docs - customer proof or case studies - third-party comparisons or reviews If NotebookLM misses something obvious, add it manually. In the recording, the important backup move was to make sure the vendors' pricing pages were included. The research feature is helpful, but you still need to use judgment. > Pro tip: For vendor or partnership decisions, manually check that pricing pages, integration pages, and support docs made it into the notebook. NotebookLM can find a lot of sources, but the boring operational pages are often what make the decision real. ### Step 4: Write Short Memos For Each Option Once the sources are in the notebook, ask NotebookLM to write short memos for each option. Keep this step simple. You are not trying to get the final report yet. You are trying to make each option legible on its own. Use: ```text Write a short opportunity memo for each vendor using only the sources in this notebook. ``` For a broader opportunity review, replace "vendor" with whatever you are comparing: ```text Write a short opportunity memo for each option using only the sources in this notebook. ``` A useful memo should cover: - what the option is best for - what the strongest source-backed evidence says - pricing or packaging notes - implementation requirements - risks - open questions - what still needs to be confirmed In the recording, NotebookLM created mini memos for Goodcall, Smith.ai, and Slang.ai. That gave the later comparison table better raw material because each vendor had already been summarized against the same decision context. You can pin useful answers as notes inside NotebookLM. In the recording, this was done with the save or pin control so the strongest outputs stayed available in the notebook instead of disappearing inside the chat thread. > Pro tip: Save the strongest memo outputs as notes before moving on. Once the notebook starts producing tables, reports, and charts, pinned notes make it much easier to keep the best source-backed reasoning in one place. ### Step 5: Generate The Comparison Table After the individual memos are done, ask NotebookLM to compare them. Use a short request: ```text Turn these three memos into a comparison table. ``` The table should use criteria from the seed memo, not random categories. In the AI receptionist example, the useful criteria included restaurant fit, integrations, human fallback, implementation speed, pricing clarity, and risk. The table is the point where the notebook becomes more than a source pile. It lets you see the tradeoffs clearly: - one option may be fastest to set up - one may have the best fit for the industry - one may have better human backup - one may have clearer pricing - one may be strongest overall but still require a sales call to verify assumptions If the table is useful, save it as a note. If NotebookLM also generates an infographic or chart, save that too. In the recording, the infographic version ended up being one of the strongest artifacts from the session. ![NotebookLM Comparison Chart View](https://tru-images.b-cdn.net/guide-assets/local-244/vet-business-opportunities-with-this-notebooklm-research-system/02-notebooklm-comparison-chart-view.png) ### Step 6: Create The Final Recommendation Memo Once you have the memos and the table, ask NotebookLM for the final recommendation. Use: ```text Write a final recommendation memo using the comparison table and the source-backed memos. ``` The final memo should include: - the recommended option - the runner-up - the option to avoid for now - the reason for the recommendation - what assumptions could make the recommendation wrong - questions to ask before buying or committing - the next validation step In the recording, NotebookLM produced the recommendation directly in the chat thread. It also offered report-style outputs like a briefing doc and an infographic. The report feature is worth trying, but the chat output may sometimes be better than the generated report format. That is a useful caveat: NotebookLM is strongest at source-grounded research and synthesis. If you need a polished final layout, you may still want to export the content to Google Docs and clean it up, or paste the research into ChatGPT or Claude and ask for a cleaner formatted memo. In the recording, exporting to Google Docs worked, but the table formatting needed cleanup. That is fine for internal research. The important part is that the hard work was done: the sources were gathered, the options were compared, and the recommendation was grounded in evidence. > Pro tip: Treat NotebookLM as the research workspace, not always the final design tool. If the exported memo is messy, paste the source-backed output into ChatGPT, Claude, Gemini, or Google Docs and clean up the formatting there. ### Step 7: Save The Workflow For The Next Decision The best part of this workflow is that it becomes reusable. After you finish one research notebook, save the structure: - seed memo - gap check - source research - one memo per option - comparison table - recommendation memo You can reuse the same system for: - vendor selection - partnership opportunities - agency evaluations - market research - product ideas - acquisition targets - sponsorship decisions The specific example here was AI receptionists, but the workflow is really about business judgment. Start with a decision, build the source pack, force the options into comparable memos, and then ask for the recommendation. ### Guide: Want To Create Skills in ChatGPT? Use This Workaround (Codex Guide) > In this guide, you’ll learn how to create repeatable “skill”-style workflows using your ChatGPT subscription by treating the Codex desktop app as a skills playground. - Canonical page: https://app.therundown.ai/guides/want-to-create-skills-in-chatgpt-use-this-workaround-codex-guide - Machine-readable page: https://app.therundown.ai/guides/want-to-create-skills-in-chatgpt-use-this-workaround-codex-guide.md - Difficulty: intermediate - Featured tool or company: OpenAI - Categories: Automation & Agents, AI Foundations & Prompting - Instructors: Billy Howell - Published: 2026-03-26 - Updated: 2026-07-30 ### Overview #### The Rundown In this guide, you’ll learn how to create repeatable “skill”-style workflows using your ChatGPT subscription by treating the Codex desktop app as a skills playground. You’ll create a reusable skill that can be run on demand to produce the same kind of structured output each time, with no coding required. #### Who This Is Useful For - ChatGPT users who want repeatable workflows without using custom GPTs - Anyone with a ChatGPT Plus or higher plan - Anyone curious about Codex but intimidated by coding and looking for a practical first use case #### What You Will Build ![image-2153](https://tru-images.b-cdn.net/guide-assets/69/69c5cb2b48e27867740ad2fa/image-2153.png) You’ll build a “Skills Playground” folder for Codex and create a reusable skill inside it. Once created, you can call the skill with / whenever you want to run the same framework or workflow again. - A dedicated skills-playground folder - A reusable Codex skill for a repeatable ChatGPT workflow - A repeatable way to trigger that skill on demand #### What You Need - A ChatGPT Plus or higher plan - A Windows or Mac computer - One repeatable workflow you want to turn into a reusable skill #### Going Further - After you have one skill working, build a small library of repeatable workflows around the ChatGPT tasks you do most often. You can keep them together in one playground folder or split them into separate folders for cleaner separation by project. - You do not need to build large automation chains. Even turning basic, repetitive research and analysis tasks into reusable skills can save time in the long run. ### Step 1 Use Codex as your skills playground If you have a ChatGPT subscription, you may already be doing repetitive workflows, but reusable skills are not available directly in the ChatGPT app. Use the Codex desktop app as a skills playground instead. This gives you a practical way to create reusable skills for structured tasks without needing to code. ### Step 2 Create or select a skills-playground folder Open the Codex desktop app and click Add a new project. You can also use the shortcut CMD + O. Create a new folder or select an existing one, then name it “skills-playground.” A dedicated folder keeps the skills you use repeatedly in one clean place. ### Step 3 Create a reusable skill Start a new thread in that project and describe the skill you want Codex to make. ```text /skill creator Make me a skill to use the AARR framework to evaluate existing startups or business ideas. ``` Type /skill creator first so Codex uses the skill creation flow directly. You can also specify whether the skill should be global and available in any folder, or limited to this project folder. ### Step 4 Restart Codex and trigger the skill If the new skill does not appear immediately when you type /, quit the app with CMD + Q and reopen it. Then type / again and scroll down or type the skill name to find it. ![image-2154](https://tru-images.b-cdn.net/guide-assets/69/69c5cb3b900c96d86177b2a1/image-2154.png) You now know how to create and trigger your own Codex skills with no coding required. ### Pro tip > Use existing frameworks like AARR, ICE, etc. for research and analysis skills. ### Guide: We Tested 20 Prompts for Generating AI Headshots: Here’s the Best Strategy > This guide teaches a workflow for turning selfies into professional-looking headshots with Google’s Nano Banana Pro. - Canonical page: https://app.therundown.ai/guides/we-tested-20-prompts-for-generating-ai-headshots-heres-the-best-strategy - Machine-readable page: https://app.therundown.ai/guides/we-tested-20-prompts-for-generating-ai-headshots-heres-the-best-strategy.md - Difficulty: beginner - Featured tool or company: Google - Categories: Content & Creative, Business Operations & Leadership - Instructors: Billy Howell - Published: 2026-02-02 - Updated: 2026-07-30 ### Overview #### The Rundown This guide teaches a workflow for turning selfies into professional-looking headshots with Google’s Nano Banana Pro. You will learn how to avoid the “uncanny” AI-generated look by using a clean source photo and prompting the model to edit the image instead of generating a face from scratch. Expect to test multiple photos and iterations before finding the best result. #### Who This Is Useful For - Jobseekers who want to upgrade their headshot - Anyone who needs a new LinkedIn photo - Teams that want to unify company profiles without hiring an expensive photographer #### What You Will Build You will create a polished, reusable professional headshot from your own photo. - A clean source photo taken with good lighting and composition - An AI-enhanced professional headshot - A reusable prompt for future headshot edits ![Generated Image February 02, 2026 - 3_55PM](https://tru-images.b-cdn.net/guide-assets/69/698117e2ee667d93fdd4ff73/generated-image-february-02-2026-3-55pm.jpeg) #### What You Need - Google Workspace account - Access to Google AI Studio and Nano Banana Pro - An API key with Gemini and Nano Banana Pro enabled, if needed - A webcam or phone camera - A neutral background and direct or natural lighting #### Going Further - After creating your headshot, use Gemini to generate additional prompt concepts. Paste in the provided prompt and ask for four new headshot concepts using the same principles, making clear that the prompts should edit photos rather than generate new images from scratch. ### Step 1: Take a clean source photo Start with a strong source photo. Bad lighting can cause AI to regenerate your face from scratch, which is what creates the “uncanny” look. The goal is to give Nano Banana Pro a clean image to edit rather than replace. - Background: Use a neutral background, similar to a passport photo. - Lighting: Use direct lighting. If you do not have a light, try your phone’s flashlight, a white background on a second monitor, or natural light near a window. Avoid direct sunshine. - Camera: Use your webcam if possible. If you use your phone, use the front camera with a timer to avoid awkward selfie angles. - Composition: Keep your upper chest visible. Position the camera lens between shoulder height and the top of your head. If needed, prop up your laptop or phone to reach the right height. ![Gemini_Generated_Image_44bwi644bwi644bw](https://tru-images.b-cdn.net/guide-assets/69/69811c6cabd37ea46302a532/gemini-generated-image-44bwi644bwi644bw.png) In short: pretend you are taking a passport photo and use your laptop’s webcam. AI can upscale it. Snap a couple of photos so you have options to test. > Pro tip: Try rotating your torso 45 degrees while looking into the camera. This was the best pose for Nano Banana Pro besides straight on. ### Step 2: Open Nano Banana Pro in Google AI Studio Go to aistudio.google.com and click Playground. It should show that you are using Nano Banana Pro. ![ai_studio](https://tru-images.b-cdn.net/guide-assets/69/69811c8fce5fbc656db7f39d/ai-studio.png) > Pro tip: Make sure you click the key and add an API key that has Gemini and Nano Banana Pro enabled. ### Step 3: Enhance your photo with the editing prompt To avoid the deepfake look, tell Nano Banana Pro to edit your photo instead of generating a new image from scratch. Upload one of your photos and send this prompt: ```text Generate: Post-processing enhancement, professional color grading, balanced studio lighting, remove noise and grain, sharpen focus, upscale to 8k, skin texture refinement, subtle dodging and burning, clear and crisp details, maintain exact original facial structure and clothing, photorealistic ``` Try the prompt with a few of the photos you took. The source workflow tested 20+ prompts before getting a few good results, so expect to iterate. When you get one you like, increase the quality to 4K under “Resolution,” then prompt the AI to generate the same exact photo again. > Pro tip: Start a new chat if you start getting repetitive or low-quality results. ### Pro tip > Start a new chat if you start getting repetitive or low-quality results. ### Guide: Write Viral YouTube Scripts With This Free Tool > Learn how to use NotebookLM to analyze YouTube videos, extract the structure behind successful content, and turn that structure into reusable blueprints for your own scripts. - Canonical page: https://app.therundown.ai/guides/write-viral-youtube-scripts-with-this-free-tool - Machine-readable page: https://app.therundown.ai/guides/write-viral-youtube-scripts-with-this-free-tool.md - Difficulty: beginner - Featured tool or company: Google - Categories: Content & Creative, Marketing & Growth - Instructors: Billy Howell - Published: 2026-02-19 - Updated: 2026-07-30 ### Overview #### The Rundown Learn how to use NotebookLM to analyze YouTube videos, extract the structure behind successful content, and turn that structure into reusable blueprints for your own scripts. You will produce a smart notebook that can help draft outlines, scripts, titles, and video angles from a reference video and your brand context. #### Who This Is Useful For - Product managers who want tighter product announcement videos - Video interviewers who need help structuring their interviews - Anyone who records YouTube content #### What You Will Build ![image-2070](https://tru-images.b-cdn.net/guide-assets/69/6997a0751c1f0d8814ad19ff/image-2070.png) You will build a smart NotebookLM notebook that reverse-engineers YouTube videos and creates reusable blueprints for drafting video scripts. - A source set with a reference YouTube video and your brand context - A saved blueprint based on a video structure you like - A script outline, draft script, and video title ideas generated from that blueprint #### What You Need - Google Workspace or Gmail account - A video idea or two - A long-form YouTube video with a structure you like - Optional: your website link or PDFs for extra brand context #### Going Further - After creating your first blueprint, keep building a library of reusable video structures for different angles or use cases, such as product announcements. You can also analyze your own videos, collect strong hooks in a Google Doc, and add that document as a Drive source so your notebook improves over time. ### Step 1: Add your video and brand context Find a long-form YouTube video with a structure you like. Go to notebooklm.google.com, create a new notebook, choose Add sources, select Websites, and paste the YouTube link. Create a note with a brief summary of your brand, your content vision, and your audience. Click the three dots, then choose Add as source. To give NotebookLM more context, you can also add your website link or upload PDFs as sources. ### Step 2: Reverse-engineer the video into a blueprint Make sure the YouTube video and your brand statement are selected. Paste this prompt into the chat: ```text Reverse engineer the structure of this video. Turn it into a blueprint that outlines the structure so I can use it for [my brand] content. Take note of the different scenes and possible angles for a video structured like this. ``` Click Save to note in the chat. ![image-2071](https://tru-images.b-cdn.net/guide-assets/69/6997a0911c1f0d8814ad2286/image-2071.png) Rename the note “Blueprint 1,” then click the three dots and choose Convert to source. You can also click the three dots on a note to export it as a Google Doc or Sheet. ### Step 3: Turn the blueprint into a script Now you have a modular blueprint you can use to write scripts on any topic. De-select the YouTube video in the sources pane so that only your brand statement and Blueprint 1 are selected. Send a prompt like this: ```text Use blueprint 1 to create an outline, a script and four video titles for my brand on the topic of [your topic]. ``` Export the script to Google Docs and edit it. Over time, build a library of blueprints you can use to try different video angles. ![image-2072](https://tru-images.b-cdn.net/guide-assets/69/6997a0aa1c1f0d8814ad2b3c/image-2072.png) Product announcements are a strong example use case. You can create a bespoke notebook for product announcements, turn several product announcement videos you like into blueprints, and test a few of those blueprints the next time your company needs to script one. ### Step 4 Going further: Build a running hook library You can also have NotebookLM analyze your own videos and keep a running list of feedback. Link to one of your videos and try a prompt like this: ```text Watch this video and give feedback on engagement. Highlight specific areas of high engagement that could work as hooks for short-form videos. ``` Copy those hooks into a Google Doc titled “Running List of Hooks.” Then choose Add sources, select Drive, and add that document. Your notebook will get smarter as you analyze more videos without cluttering your sources. ### Pro tip > To give NotebookLM more context, add your website link or upload PDFs as sources. ## Courses ### Course: AI Automations (New & Improved) > Build reliable AI automations with n8n, Claude, and structured outputs to classify data, orchestrate multi-step workflows, debug integrations, and ship practical systems that save time. - Canonical page: https://app.therundown.ai/courses/ai-automations-course - Machine-readable page: https://app.therundown.ai/courses/ai-automations-course.md - Level: Intermediate - Categories: Automation & Agents - Instructors: Nate Grahek - Duration: 3 hours 42 minutes - Lessons: 3 - Chapters: 20 - Published: 2026-05-04 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **Pick High-ROI Use Cases**: Choose narrow, repeatable automation projects with accessible data, clear quality standards, and realistic business value. - **Build Agentic Workflows**: Build N8N workflows with triggers, data sources, AI steps, loops, and output updates that move information reliably from start to finish. - **Collaborate with Claude**: Use Claude to plan automations, generate workflow JSON, write prompts, create schemas, and adapt templates to your tools. - **Prompt for Reliable Output**: Design focused AI nodes with stable system prompts, dynamic inputs, and structured JSON outputs for dependable downstream processing. - **Debug with Confidence**: Test and debug workflows by inspecting node outputs, pinning data, using execution history, and reconnecting credentials, fields, and destinations. - **Optimize and Deploy**: Improve automations by evaluating output quality, comparing models, estimating API costs, and publishing workflows when they are ready to run. ### Tools covered - n8n - Claude - Gemini - Google Sheets - Notion - WordPress ### Course: AI Automations Made Simple with Google Workspace Studio > Build practical Google Workspace Studio automations that triage email, draft grounded replies, generate recurring research updates, and keep human review at the center of reliable AI workflows. - Canonical page: https://app.therundown.ai/courses/google-workspace-studio - Machine-readable page: https://app.therundown.ai/courses/google-workspace-studio.md - Level: Beginner - Categories: Automation & Agents, Productivity & Knowledge Work - Instructors: Nate Grahek - Duration: 49 minutes - Lessons: 1 - Chapters: 6 - Published: 2025-12-13 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Spot Automation Opportunities**: Identify repetitive triage, drafting, summarizing, extraction, and research tasks that are safe to delegate while keeping final decisions in your hands. - **Automate Inbox Triage**: Build Gmail automations that use triggers, filters, AI decisions, labels, archiving, and starring to turn your inbox into a cleaner action list. - **Draft Safer AI Replies**: Create grounded reply workflows that extract questions, draft responses from approved documents, and prepare emails for human review. - **Connect Workflow Steps**: Use variables to pass email content, decisions, extracted details, drafts, dates, and research outputs between workflow steps. - **Design Practical Agents**: Design narrow, reliable AI agent tasks for decisions, summaries, labels, drafts, and extractions without over-automating judgment. - **Schedule Research Updates**: Create scheduled research workflows that generate formatted updates and append them to Google Docs for reusable context over time. ### Tools covered - Google Workspace Studio - Google Workspace - Gmail - Gemini - Google Docs ### Course: AI Essentials for Leaders & Managers > Build practical AI adoption habits that help leaders improve productivity, guide teams safely, design low-risk workflows, and create clear policies for responsible use. - Canonical page: https://app.therundown.ai/courses/ai-essentials-for-leaders-and-managers - Machine-readable page: https://app.therundown.ai/courses/ai-essentials-for-leaders-and-managers.md - Level: Beginner - Categories: Business Operations & Leadership - Instructors: Nate Grahek - Duration: 1 hour 10 minutes - Lessons: 1 - Chapters: 6 - Published: 2025-11-16 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Lead Practical AI Adoption**: Set AI priorities that emphasize safe ROI, team trust, practical experimentation, and repeatable business value. - **Engineer Better Context**: Give AI the company, role, market, goals, constraints, examples, and data it needs to produce useful work. - **Build Reusable AI Workspaces**: Create reusable AI projects for leadership tasks such as one-on-one preparation, performance reviews, sales enablement, and RFP support. - **Design AI Workflows**: Break complex processes into smaller AI tasks, workflow steps, and narrow agent decisions that are easier to test and improve. - **Govern AI Responsibly**: Define clear AI usage expectations, reduce shadow AI risk, and evaluate higher-risk workflows with real examples and accuracy checks. - **Prepare for AI Agents**: Document repeatable processes as SOPs that support delegation, automation planning, and future agent readiness. ### Course: AI Foundations (New & Improved) > Master the essential skills for working effectively with AI tools like ChatGPT through practical, hands-on techniques that go beyond basic prompting. - Canonical page: https://app.therundown.ai/courses/ai-foundations-course - Machine-readable page: https://app.therundown.ai/courses/ai-foundations-course.md - Level: Beginner - Categories: AI Foundations & Prompting - Instructors: Nate Grahek - Duration: 2 hours 52 minutes - Lessons: 3 - Chapters: 22 - Published: 2026-04-28 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **Choose the Right AI Approach**: Separate chat, automation, projects, research, and custom tools so each AI use case gets the right level of planning and effort. - **Engineer Better Context**: Provide roles, goals, constraints, examples, and business context that help AI produce specific, useful work instead of generic output. - **Guide AI Conversations**: Use interview-style prompting, fresh starts, and handoff summaries to guide AI through complex work without losing quality. - **Build Reusable AI Projects**: Create reusable projects with instructions, knowledge files, templates, and examples for repeated business tasks you can confidently evaluate. - **Run Practical AI Research**: Scope deep research tasks, synthesize findings, and turn useful insights into decision-ready reports or project knowledge files. - **Create AI-Powered Artifacts**: Design lightweight dashboards, calculators, and interactive artifacts that reveal patterns, test scenarios, and support better decisions. ### Tools covered - ChatGPT - Claude - Gemini ### Course: AI Foundations for Marketers: 2025 Edition > Build practical AI systems for marketing that improve output quality with better context, reusable project libraries, reliable workflows, and focused prototypes. - Canonical page: https://app.therundown.ai/courses/ai-foundations-for-marketers-2025-edition - Machine-readable page: https://app.therundown.ai/courses/ai-foundations-for-marketers-2025-edition.md - Level: Beginner - Categories: Marketing & Growth, AI Foundations & Prompting, Automation & Agents - Instructors: Nate Grahek - Duration: 1 hour 19 minutes - Lessons: 1 - Chapters: 6 - Published: 2025-11-09 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Improve AI Results**: Diagnose generic or inaccurate AI outputs by improving context, scope, examples, audience details, and success criteria. - **Reuse Marketing Context**: Build reusable context libraries and projects that preserve company, customer, campaign, and process knowledge for repeated marketing work. - **Prompt with Control**: Structure AI work sessions that ask clarifying questions, develop ideas progressively, and avoid overloaded conversations. - **Use AI Strategically**: Apply AI to research, synthesis, critique, drafting, and customer insight work where it can reliably improve marketing speed and quality. - **Automate Repeatable Work**: Design narrow AI workflows for lead enrichment, proposal support, competitor monitoring, CRM cleanup, and other repeatable operations. - **Prototype Marketing Tools**: Prototype dashboards, calculators, lead magnets, and internal tool mockups that make marketing decisions and software ideas easier to evaluate. ### Course: AI Growth Systems > Build tool-connected AI growth workflows that monitor competitors, surface Reddit opportunities, draft review-ready responses, and deliver structured reports with human oversight. - Canonical page: https://app.therundown.ai/courses/ai-growth-systems - Machine-readable page: https://app.therundown.ai/courses/ai-growth-systems.md - Level: Intermediate - Categories: Marketing & Growth, Automation & Agents, Competitive Intelligence, Social Media Marketing - Instructors: Pietro Montaldo - Duration: 2 hours 30 minutes - Lessons: 3 - Chapters: 24 - Published: 2026-05-11 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **Spot Agent Opportunities**: Map repetitive growth, sales, support, finance, and operations work that AI can handle while humans keep strategy and trust-building. - **Design Agent Workflows**: Design agents with clear triggers, AI steps, tool actions, branching logic, and human review points. - **Choose the Right Tools**: Choose between chatbots, copilots, agents, bought tools, and custom builds based on control, complexity, cost, and customization needs. - **Prompt Reliable Outputs**: Create structured prompts with goals, outputs, guardrails, context, model choices, and formats that downstream tools can use. - **Track Competitors**: Build competitor monitoring systems that research public channels, synthesize strategy insights, and deliver reports or spreadsheet-ready findings. - **Find Reddit Opportunities**: Monitor Reddit communities for relevant discussions, rank opportunities, and draft credible comments for human review before posting. ### Tools covered - Relay App - Gmail - Google Sheets - Perplexity - Claude - Reddit ### Course: An Intro to ChatGPT Sites for Beginners > A beginner-friendly workshop on using ChatGPT Sites to find patterns in your data, publish a report your team can open with a link, and schedule it to update automatically. - Canonical page: https://app.therundown.ai/courses/an-intro-to-chatgpt-sites-for-beginners - Machine-readable page: https://app.therundown.ai/courses/an-intro-to-chatgpt-sites-for-beginners.md - Level: Beginner - Categories: AI Websites, Web Design, Dashboards, Data Analysis - Instructors: Nate Grahek - Duration: 1 hour 23 minutes - Lessons: 1 - Chapters: 8 - Published: 2026-08-01 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Turn a spreadsheet or CSV into a live, shareable report that updates itself. ### What you will learn - **Plan Site Scope**: Define the audience, access level, purpose, assets, and interactivity before creating a ChatGPT Site. - **Launch Landing Pages**: Create a focused page with a clear offer, call to action, waitlist flow, and interactive elements. - **Refine with Feedback**: Use annotations, screenshots, and batched instructions to edit layouts, copy, colors, and components efficiently. - **Connect Forms & Domains**: Set up dependable lead capture and attach a custom domain while checking each AI-guided step. - **Build Decision Dashboards**: Turn spreadsheets, CSVs, or connected business data into KPIs, filters, charts, and scenario calculators. - **Automate Updates**: Schedule dashboard refreshes, test runs immediately, and adjust cadence for freshness, accuracy, and token use. ### Tools covered - ChatGPT - Claude - Chrome ### Course: Automated Deep Research: Track Competitors While You Sleep > Build an n8n-powered competitor monitoring system that uses Perplexity research, Notion archives, change detection, and stakeholder alerts to surface meaningful market shifts automatically. - Canonical page: https://app.therundown.ai/courses/automated-deep-research-track-competitors-while-you-sleep - Machine-readable page: https://app.therundown.ai/courses/automated-deep-research-track-competitors-while-you-sleep.md - Level: Intermediate - Categories: Data, Research & Analysis, Marketing & Growth, Automation & Agents - Instructors: Nate Grahek - Duration: 54 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-09-27 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Choose the Right Automation**: Identify repetitive, research-heavy competitor monitoring tasks that are good candidates for reliable AI automation. - **Map the Workflow**: Design a structured competitor research workflow with clear inputs, outputs, guardrails, and exception-based alerts. - **Build in n8n**: Configure n8n to run repeated competitor research using company variables, API inputs, and AI-assisted debugging. - **Create Usable Reports**: Turn messy API responses into clean fields and readable Notion reports with citations, links, and skim-friendly structure. - **Detect Competitive Changes**: Compare current and previous competitor reports to detect meaningful changes in launches, pricing, positioning, and market activity. - **Alert the Right People**: Send concise stakeholder notifications that summarize what changed, prioritize importance, and link back to the full source report. ### Tools covered - n8n - Perplexity - Notion - ChatGPT - Claude - Gemini ### Course: Become an AI-Native Leader > Build an AI-native operating model that ties automation to revenue, reduces decision bottlenecks, and turns team knowledge into reusable workflows and shared context. - Canonical page: https://app.therundown.ai/courses/become-an-ai-native-leader - Machine-readable page: https://app.therundown.ai/courses/become-an-ai-native-leader.md - Level: Intermediate - Categories: Business Operations & Leadership, AI Strategy, AI Workflows - Instructors: Lauren Vriens - Duration: 2 hours 23 minutes - Lessons: 3 - Chapters: 19 - Published: 2026-06-13 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **Measure AI by Business Impact**: Tie AI initiatives to revenue, retention, margin, cash flow, or throughput instead of generic productivity gains. - **Prioritize Revenue Workflows**: Find the bottlenecks, handoffs, and translation loss that slow customer value, then apply AI where it improves flow. - **Build Shared AI Context**: Create reusable markdown files, instructions, and shared context systems that keep people and AI working from the same source of truth. - **Productize Repeatable Work**: Design repeatable AI jobs, skills, and agents around clear outcomes, constraints, inputs, tools, and success criteria. - **Lead AI-Native Teams**: Set operating principles, review rhythms, and adoption rituals that make AI useful without flooding leaders with low-value output. - **Improve AI Output Quality**: Use AI to surface assumptions, ask better questions, organize knowledge, and turn repeated corrections into stronger workflows. ### Tools covered - ChatGPT - Google Drive - SharePoint - Notion - Slack - Microsoft Teams ### Course: Build a Marketing Creative Studio in Your Pocket > A 3-session course where you design, produce, and refine an AI marketing workflow you can run on your own. - Canonical page: https://app.therundown.ai/courses/ai-for-marketing-with-melanie-moeller - Machine-readable page: https://app.therundown.ai/courses/ai-for-marketing-with-melanie-moeller.md - Level: Beginner - Categories: Marketing & Growth, Content & Creative - Instructors: Melanie Moeller - Duration: 2 hours 44 minutes - Lessons: 3 - Chapters: 17 - Published: 2026-07-09 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Build a Marketing Creative Studio in Your Pocket ### What you will learn - **Build campaign intelligence**: Use deep research to map audience needs, trust signals, competitors, trends, and opportunities before creating assets. - **Create a messaging system**: Turn research into a campaign brief, message hierarchy, proof points, tone, visual direction, and reusable guidance summary. - **Produce visual campaign assets**: Generate images, build a storyboard, animate shots, and assemble a coherent short-form campaign in CapCut. - **Add voice and music**: Create timed voiceover and music that support the campaign message, pacing, and emotional arc. - **Test and improve creative**: Use CTA diagnosis, synthetic personas, brief scoring, and performance data to drive the next iteration. ### Tools covered - ChatGPT - Gemini - Midjourney - Hailuo - Veo - Sora-class tools - ElevenLabs - Suno - CapCut ### Course: Build AI agent teams for Sales & Marketing with Zapier Agents > Build focused Zapier Agents that enrich and qualify leads, route prospects, draft follow-ups, and streamline marketing content workflows across CRM, docs, and team communication tools. - Canonical page: https://app.therundown.ai/courses/build-ai-agent-teams-for-sales-and-marketing-with-zapier-agents - Machine-readable page: https://app.therundown.ai/courses/build-ai-agent-teams-for-sales-and-marketing-with-zapier-agents.md - Level: Beginner - Categories: Automation & Agents, Sales, Consulting & Customer Success, Marketing & Growth - Instructors: Grace Lu, Anika Gupta - Duration: 55 minutes - Lessons: 1 - Chapters: 6 - Published: 2025-10-17 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Map Agent Workflows**: Design focused AI teammates around repeatable sales and marketing workflows with clear goals, inputs, tools, and handoffs. - **Configure Zapier Agents**: Build Zapier agents from templates or prompts, then configure triggers, app authorizations, and tools for reliable automation. - **Enrich CRM Leads**: Automate CRM research by enriching new contacts with company and prospect details while controlling when records are updated. - **Qualify and Route Prospects**: Connect agents so enriched lead data can be scored against qualification criteria and routed to the right sales channel. - **Create Content Assets**: Generate marketing content, save assets to shared workspaces, and notify collaborators through tools like Google Docs, Drive, and Slack. - **Control Reviews and Sharing**: Add human review steps, draft social posts, and share agents with the right permissions for safer team collaboration. ### Tools covered - Zapier Agents - HubSpot - Slack - Google Docs - Google Drive - LinkedIn ### Course: Build an AI-Powered Operating System for Your 1 on 1s > Build a simple AI-powered system that helps you run more effective 1:1s and improve team performance. - Canonical page: https://app.therundown.ai/courses/build-an-ai-powered-operating-system-for-your-1-1s - Machine-readable page: https://app.therundown.ai/courses/build-an-ai-powered-operating-system-for-your-1-1s.md - Instructors: Adam Birnbaum - Duration: 57 minutes - Lessons: 1 - Chapters: 7 - Published: 2026-09-03 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **AI Workspace Setup**: Create a dedicated AI workspace with instructions, Markdown context files, a profile, a running log, and a commitment tracker for each direct report. - **Context-Driven Prep**: Use accumulated history to surface open commitments, unresolved issues, recurring themes, risks, and high-value topics before each conversation. - **Structured Agendas**: Build agendas that move routine updates aside and focus live time on check-ins, decisions, roadblocks, support needs, and commitments. - **Better Questions**: Ask coaching questions that uncover challenges, clarify priorities, build resourcefulness, and deepen trust. - **Follow-Through Tracking**: Turn meeting notes into reviewed log updates and accurate commitment records that keep future conversations grounded. - **Responsible AI Use**: Apply human review, privacy safeguards, and company data-handling policies before relying on AI-generated management support. ### Course: Build and run your ad creative strategy with Claude > Build the personas, write the concepts, test them, and scale the winner, with Claude as the strategist, analyst, and buyer you don't have. - Canonical page: https://app.therundown.ai/courses/build-and-run-your-ad-creative-strategy-with-claude - Machine-readable page: https://app.therundown.ai/courses/build-and-run-your-ad-creative-strategy-with-claude.md - Instructors: Rishi Lalwani - Duration: 1 hour 19 minutes - Lessons: 1 - Chapters: 8 - Published: 2026-09-11 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **Claude Project Setup**: Create a reusable Claude workspace with brand context, avatar research, project instructions, and Meta Ads data access. - **Avatar Research**: Use real market language from reviews, forums, and social sources to define a specific buyer with clear desires, objections, and motivations. - **Ad Inspiration**: Analyze competitor and adjacent-market ads to identify transferable hooks, formats, offers, and creative patterns without copying. - **Concept Adaptation**: Turn proven ad structures into brand-specific concepts by combining persona, angle, offer, proof, and format. - **Performance Analysis**: Name and group ads by creative variables so Claude can surface winning combinations and diagnose CPM, thumbstop, CTR, and conversion issues. - **Creative Testing Decisions**: Build a focused testing plan that matches budget, target CPA, localization needs, targeting approach, and production tools to the next best action. ### Course: Build Professional PowerPoint Decks with Claude Skills > Build repeatable PowerPoint deck workflows with Claude Skills by turning approved templates, structured context, and fresh inputs into polished recurring presentations faster. - Canonical page: https://app.therundown.ai/courses/build-professional-powerpoint-decks-with-claude-skills - Machine-readable page: https://app.therundown.ai/courses/build-professional-powerpoint-decks-with-claude-skills.md - Level: Beginner - Categories: Productivity & Knowledge Work, Content & Creative - Instructors: Nate Grahek - Duration: 1 hour 12 minutes - Lessons: 1 - Chapters: 9 - Published: 2025-10-31 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Choose the Right Decks**: Identify recurring presentation workflows where Claude Skills can save meaningful time without replacing your design judgment. - **Plan Before PowerPoint**: Structure presentation work into planning, synthesis, approval, and file-generation steps so changes happen while they are easiest to review. - **Manage Claude Context**: Use Projects, current inputs, and progressive disclosure to give Claude enough context without overwhelming the workflow. - **Build Reliable Templates**: Prepare PowerPoint templates with fixed layouts, clear placeholders, alt text instructions, and practical character limits. - **Create Deck Skills**: Create and test a custom Claude Skill that fills approved slide structures with updated business content. - **Improve and Scale**: Spot common automation issues, keep stable design elements fixed, and scale from a few slide types to larger recurring decks. ### Tools covered - Claude - PowerPoint ### Course: Build Reliable AI Assistants: Your RAG System Guide > Build reliable RAG assistants that ground answers in searchable documents, metadata, vector retrieval, and verifiable sources using practical Supabase and n8n workflows. - Canonical page: https://app.therundown.ai/courses/build-reliable-ai-assistants - Machine-readable page: https://app.therundown.ai/courses/build-reliable-ai-assistants.md - Level: Intermediate - Categories: Coding & App Development, Automation & Agents, Data, Research & Analysis - Instructors: Nate Grahek - Duration: 47 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-09-19 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Grounded AI Answers**: Use RAG to ground AI answers in current, specific source material instead of relying on model memory or guesswork. - **RAG Use Case Validation**: Validate a use case with a small file-based prototype before investing in a full custom RAG system. - **Knowledge Base Design**: Design chunks, metadata fields, and source references that make retrieved information accurate, traceable, and useful. - **Search Strategy**: Choose between keyword search, semantic search, and structured queries based on whether users need exact matches, meaning-based retrieval, or calculations. - **Vector Database Setup**: Build a Supabase vector database that stores text chunks, embeddings, and metadata for reliable retrieval. - **Reliable Chat Workflows**: Configure retrieval, re-ranking, and prompting so assistants answer consistently with evidence-backed source details. ### Tools covered - Supabase - n8n - Cohere ### Course: Context Engineering: The Core Skill Worth Learning Even As Models Improve > Build reliable AI workflows by supplying stronger context, structuring inputs with Markdown and JSON, managing conversation limits, and turning vague prompts into repeatable collaboration patterns. - Canonical page: https://app.therundown.ai/courses/context-engineering-the-core-skill-worth-learning-even-as-models-improve - Machine-readable page: https://app.therundown.ai/courses/context-engineering-the-core-skill-worth-learning-even-as-models-improve.md - Level: Intermediate - Categories: AI Foundations & Prompting, Automation & Agents - Instructors: Nate Grahek - Duration: 1 hour 4 minutes - Lessons: 1 - Chapters: 8 - Published: 2025-10-25 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Diagnose Bad Outputs**: Diagnose weak AI outputs by identifying missing context, poor task fit, vague goals, or overloaded conversations. - **Provide Better Context**: Improve results with examples, business details, constraints, success criteria, and trusted source material before rewriting prompts. - **Control the Workflow**: Guide AI through interview-style workflows, approved steps, and numbered options to keep collaboration focused and efficient. - **Manage Conversation Limits**: Manage context limits by chunking complex work, summarizing progress, starting fresh chats, and reusing saved project materials. - **Structure AI Work**: Structure inputs and outputs with Markdown or JSON so AI can follow priorities and produce predictable results. - **Match Tools to Tasks**: Choose the right setup for deep research, recurring projects, RAG-style knowledge bases, and focused coding plans. ### Tools covered - LinkedIn ### Course: Create Viral Short-form Videos with AI > Discover the system behind a channel that posts viral talking-head short-form videos every week: educational clips starring an AI clone of you, so you can publish consistently without filming every video yourself. - Canonical page: https://app.therundown.ai/courses/create-viral-short-form-videos-with-ai - Machine-readable page: https://app.therundown.ai/courses/create-viral-short-form-videos-with-ai.md - Level: Beginner - Categories: Marketing & Growth, Content & Creative - Instructors: Nic Conley - Duration: 36 minutes - Lessons: 1 - Chapters: 6 - Published: 2026-07-13 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Build a repeatable short-form video system starring your AI clone. ### What you will learn - **AI Video Workflow**: Build a repeatable workflow for turning strong topics into AI-assisted short-form videos without daily filming. - **Viral Topic Selection**: Find proven viral ideas in your niche and reshape them with emotional angles that drive retention and reaction. - **Short-Form Scripting**: Train AI writing tools on high-performing examples to create hooks, pacing, tone, and scripts suited to your target platform. - **Voice, Avatar, and B-Roll**: Create expressive voiceovers, avatar segments, and supporting visuals that make AI-led videos feel dynamic and watchable. - **Fast-Paced Editing**: Combine avatar clips, real footage, AI visuals, cuts, and pacing into polished videos built for short-form attention spans. - **Publishing and Monetization**: Choose monetization paths, posting cadence, and hiring processes that support consistent output and business results. ### Tools covered - HeyGen - Fish Audio - AI avatars - Short-form video ### Course: Editing Video with Claude > Build repeatable Claude workflows that turn source footage, websites, and presentation assets into polished shorts, promos, and slide decks with clear planning, review gates, and reusable skills. - Canonical page: https://app.therundown.ai/courses/editing-video-with-claude - Machine-readable page: https://app.therundown.ai/courses/editing-video-with-claude.md - Level: Intermediate - Categories: Content & Creative, Video Editing, Automation & Agents - Instructors: Nate Grahek - Duration: 1 hour 1 minutes - Lessons: 1 - Chapters: 8 - Published: 2026-07-01 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **AI Operator Workflow**: Move from context to plan, draft, approval, and final build so Claude stays aligned with your creative judgment. - **Style References**: Use templates, screenshots, brand rules, file paths, and high-quality references to communicate the exact style and constraints. - **Short-Form Repurposing**: Define crops, placement, captions, length, and test renders to turn long-form source material into vertical shorts. - **Promos and Decks**: Turn websites, notes, and assets into animated promos or HTML slide decks by approving the structure before design work begins. - **Preview and Calibrate**: Use local preview tools, low-quality renders, and custom controls to fine-tune movement, layout, cropping, and timing. - **Reusable Skills**: Package proven workflows with rules, assets, commands, and pitfalls so repeat projects are faster and more consistent. ### Tools covered - Claude - ChatGPT - Gemini - Codex - Claude Code - Co-Work ### Course: Essential ChatGPT Tips: Latest Features & Practical Use Cases > Build better ChatGPT results with practical techniques for prompts, projects, custom GPTs, personalization, Canvas editing, reusable context files, and lightweight dashboards. - Canonical page: https://app.therundown.ai/courses/essential-chatgpt-tips-latest-features-and-practical-use-cases - Machine-readable page: https://app.therundown.ai/courses/essential-chatgpt-tips-latest-features-and-practical-use-cases.md - Level: Beginner - Categories: AI Foundations & Prompting, Productivity & Knowledge Work - Instructors: Nate Grahek - Duration: 1 hour 4 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-08-30 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Control Better Outputs**: Create clearer prompts by defining goals, context, examples, constraints, and success criteria before asking for an output. - **Build Reusable Assistants**: Use projects, custom GPTs, source files, and instructions to make recurring personal or team work more consistent. - **Structure AI Workflows**: Choose structured workflows over high-autonomy agents when reliability, risk control, and repeatable quality matter. - **Raise the Quality Bar**: Improve drafts, decisions, and analyses with rubrics, feedback loops, role-play, critique, and targeted revision. - **Personalize Your Workspace**: Configure personalization, memory, Canvas, and file context so ChatGPT adapts to your role without cluttering future work. - **Prototype Practical Tools**: Prototype focused tools, dashboards, and interactive reports from small datasets, screenshots, and plain-language requirements. ### Tools covered - ChatGPT ### Course: Finally Getting AI to Do Real Work > Learn why AI outputs still fail in everyday work, then apply durable habits for context, prompting, projects, and tool-connected workflows that make models useful for real documents, decks, research, and operating work. - Canonical page: https://app.therundown.ai/courses/finally-getting-ai-to-do-real-work - Machine-readable page: https://app.therundown.ai/courses/finally-getting-ai-to-do-real-work.md - Level: Beginner - Categories: Productivity & Knowledge Work, AI Foundations & Prompting, Automation & Agents - Instructors: Nate Grahek - Duration: 1 hour 11 minutes - Lessons: 1 - Chapters: 8 - Published: 2026-05-18 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview A practical workshop on turning AI from a generic chat box into a reliable working partner. ### What you will learn - **Diagnose weak outputs**: Spot whether bad AI work is coming from missing context, bad scoping, hallucination, sycophancy, or context rot. - **Package reusable context**: Build hot-start documents and project instructions so new chats begin with the right background. - **Split complex work**: Break research, outlining, writing, and production into separate focused AI passes. - **Prompt before prompting**: Use AI to interview you, write stronger briefs, and generate better research or system prompts before the real task. - **Work as the orchestrator**: Decide what belongs with you, what belongs with the model, and what should connect back into your everyday tools. ### Tools covered - ChatGPT - Claude - Gemini - NotebookLM - Google Workspace - Microsoft 365 - Notion - Asana ### Course: Get Consultant-Grade Strategy From the AI You Already Pay For > Build consultant-grade AI advisory workflows that use structured context, contrarian analysis, bias checks, and source-grounded critique to make higher-stakes business decisions with more rigor. - Canonical page: https://app.therundown.ai/courses/get-consultant-grade-strategy-from-the-ai-you-already-pay-for - Machine-readable page: https://app.therundown.ai/courses/get-consultant-grade-strategy-from-the-ai-you-already-pay-for.md - Level: Intermediate - Categories: Business Strategy, AI Productivity, Research & Analysis, Consulting - Instructors: Nate Grahek - Duration: 1 hour 26 minutes - Lessons: 1 - Chapters: 8 - Published: 2026-06-26 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **AI Advisory Boards**: Build multi-perspective AI advisory prompts that compare options, surface tradeoffs, and challenge your default assumptions. - **Context Engineering**: Organize reusable research, documents, examples, and numbers so AI starts with the context needed for useful strategy work. - **Bias and Risk Checks**: Apply WRAP, steelman prompts, and pre-mortems to widen options, test assumptions, and prepare for decisions being wrong. - **Failure Mode Control**: Prevent cold starts, context rot, hallucinations, and sycophancy with clarifying questions, source grounding, outlines, and timely chat resets. - **Project Instructions**: Create project instructions that define the role, scope, workflow, tone, output requirements, and source-use rules for repeated work. - **Quality-Gated Outputs**: Use examples, templates, outline approvals, and draft reviews to move from raw context to higher-quality AI-assisted deliverables. ### Tools covered - Fable - Opus - X‏ ### Course: Getting Hands-On with ChatGPT Agents > Build practical ChatGPT agent workflows for research, web actions, file creation, recurring tasks, and connected-account work while applying clear prompts, supervision, and security guardrails. - Canonical page: https://app.therundown.ai/courses/getting-hands-on-with-chatgpt-agents - Machine-readable page: https://app.therundown.ai/courses/getting-hands-on-with-chatgpt-agents.md - Level: Beginner - Categories: Automation & Agents, Productivity & Knowledge Work, Business Operations & Leadership - Instructors: Dr. Alvaro Cintas - Duration: 1 hour 7 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-07-25 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Choose the Right Agent Tasks**: Identify when an agent is the right choice for multi-step research, web navigation, file creation, form filling, or account-based work. - **Configure Agent Tools**: Set up browsing, code execution, connectors, memory, and scheduled tasks in ways that match your workflow and privacy needs. - **Prompt for Reliable Workflows**: Write focused prompts with clear goals, context, constraints, output formats, and guardrails for better autonomous results. - **Stay in Control**: Supervise logins, confirmations, CAPTCHAs, purchases, and other sensitive steps without giving up control of important decisions. - **Build Everyday Automations**: Create practical agent workflows for research reports, meeting prep, email drafts, competitive analysis, lead research, and recurring tasks. - **Evaluate Agent Platforms**: Compare ChatGPT agents with alternatives such as Perplexity, Manus, and Genspark to choose the best tool for the job. ### Tools covered - ChatGPT - Perplexity - Manus - Genspark - LinkedIn - X (Twitter) ### Course: Getting Knowledge Work Done with ChatGPT Work and GPT-5.6 > Nate Grahek demonstrates how to use ChatGPT Work and GPT-5.6 to turn source files, research, and an existing brand template into a polished, editable PowerPoint presentation while showing where human review still matters. - Canonical page: https://app.therundown.ai/courses/getting-knowledge-work-done-with-chatgpt-work-and-gpt-5-6 - Machine-readable page: https://app.therundown.ai/courses/getting-knowledge-work-done-with-chatgpt-work-and-gpt-5-6.md - Categories: Productivity & Knowledge Work, Content & Creative, Data, Research & Analysis - Instructors: Nate Grahek - Duration: 1 hour 17 minutes - Lessons: 1 - Chapters: 6 - Published: 2026-07-18 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Cut through the launch noise and see what the new ChatGPT experience actually changes for everyday professional work. ### What you will learn - **Plan AI Deliverables**: Use a gather, brainstorm, draft, and build workflow to turn messy inputs into polished presentations, reports, sites, and other work products. - **Curate Project Context**: Organize source files, templates, data, and references so ChatGPT can inspect, summarize, and work from the right material. - **Build Editable Decks**: Create on-brand PowerPoint drafts from structured outlines, reusable placeholders, source data, and design references. - **Create Interactive Tools**: Turn spreadsheets, surveys, and analysis outputs into HTML dashboards, calculators, and internal strategy pages. - **Automate Repeatable Work**: Package proven workflows into custom skills, scheduled tasks, and browser-based automations while keeping judgment and review under your control. - **Evaluate Model Quality**: Benchmark new AI models against real tasks, choose effort levels wisely, and challenge outputs before committing to final artifacts. ### Course: Getting Started With Claude Cowork > Build reliable Claude Cowork workflows for proposals, decks, documents, and agent-assisted knowledge work by structuring context, checkpoints, reusable instructions, editable outputs, and safe human review. - Canonical page: https://app.therundown.ai/courses/getting-started-with-claude-co-work - Machine-readable page: https://app.therundown.ai/courses/getting-started-with-claude-co-work.md - Level: Beginner - Categories: Productivity & Knowledge Work, Automation & Agents - Instructors: Nate Grahek - Duration: 1 hour 23 minutes - Lessons: 1 - Chapters: 8 - Published: 2026-05-29 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **AI Collaboration Mindset**: Keep ownership of context, judgment, taste, and final quality while using Claude as a structured collaborator. - **Controlled Kickoff Prompts**: Write kickoff prompts that define the role, require one question at a time, and pause for approval before drafting. - **Gather, Draft, Build**: Move from source gathering and brainstorming to plain-text drafts before generating editable documents, decks, or spreadsheets. - **Project Folders & Templates**: Organize inputs, standards, examples, and reusable templates so Claude can produce work that matches your business context. - **Reusable Instructions**: Codify proven workflows into project instructions and custom skills that enforce validation, preferences, and approval gates. - **Safe AI Workflows**: Spot prompt injection risks, scope access carefully, and review AI-drafted outputs before they reach people or systems. ### Tools covered - Claude - PowerPoint - Microsoft Word - Excel - Slack ### Course: Getting Started with Google Gemini CLI Installation: Usage and Real-world Applications > Install and use Gemini CLI to generate, edit, document, and organize project files from the terminal with practical command-line workflows. - Canonical page: https://app.therundown.ai/courses/getting-started-with-google-gemini-cli-installation-usage-and-real-world-applications - Machine-readable page: https://app.therundown.ai/courses/getting-started-with-google-gemini-cli-installation-usage-and-real-world-applications.md - Level: Beginner - Categories: Coding & App Development - Instructors: Dr. Alvaro Cintas - Duration: 51 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-07-11 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Install Gemini CLI**: Install the CLI with Node.js and npm, then authenticate with a Google account or API key. - **Navigate the Terminal Workflow**: Start Gemini from the right project folder and use natural-language prompts, slash commands, and shell commands effectively. - **Manage Project Context**: Add specific files to context so Gemini can inspect, explain, and modify code without scanning unnecessary project files. - **Edit Code with AI**: Generate, edit, debug, and improve project files directly from terminal prompts with approval-based changes. - **Use Built-in Tools**: Use built-in tools to read files, search text, fetch web content, run commands, and reason across larger codebases. - **Automate Practical Tasks**: Create README files, organize folders, analyze images, rename assets, and support GitHub-related command workflows. ### Tools covered - Gemini CLI - Node.js - npm - Google AI Studio - VS Code - GitHub ### Course: How to run OpenAI’s open weight models locally on your computer > Run open-weight reasoning models on your own computer with Ollama and LM Studio while balancing hardware limits, privacy, model size, and hosted testing options. - Canonical page: https://app.therundown.ai/courses/how-to-run-openais-open-weight-models-locally-on-your-computer - Machine-readable page: https://app.therundown.ai/courses/how-to-run-openais-open-weight-models-locally-on-your-computer.md - Level: Beginner - Categories: AI Foundations & Prompting, Coding & App Development - Instructors: Dr. Alvaro Cintas - Duration: 1 hour 2 minutes - Lessons: 1 - Chapters: 10 - Published: 2025-08-08 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Pick the Right Model**: Choose between GPT-OSS, hosted OpenAI models, and other open models based on privacy, offline access, cost, customization, and model quality. - **Evaluate Hardware Fit**: Assess whether GPT-OSS 20B or 120B fits your machine by comparing memory, parameter size, architecture, benchmarks, and practical performance needs. - **Run Models with Ollama**: Install Ollama and run GPT-OSS locally from the terminal, with smaller model alternatives ready when your computer cannot handle larger models. - **Use LM Studio Locally**: Use LM Studio to download local models, chat through a visual interface, and query local documents with retrieval-augmented generation. - **Protect Private Data**: Decide when fine-tuning, retrieval, or local execution is appropriate for sensitive legal, healthcare, finance, research, coding, or internal business workflows. - **Experiment Safely**: Test hosted demos safely for quick experimentation while keeping sensitive prompts and files reserved for local workflows. ### Tools covered - Ollama - LM Studio - ChatGPT - Grok ### Course: How to supercharge your Bolt app with Supabase and Stripe > Build a Bolt currency app with Supabase authentication, database-backed live exchange rates, secure edge functions, and Stripe subscription checkout for paid access. - Canonical page: https://app.therundown.ai/courses/how-to-supercharge-your-bolt-app-with-supabase-and-stripe - Machine-readable page: https://app.therundown.ai/courses/how-to-supercharge-your-bolt-app-with-supabase-and-stripe.md - Level: Beginner - Categories: Coding & App Development - Instructors: Tomek Sułkowski - Duration: 1 hour 1 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-07-09 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Prototype a Bolt App**: Turn a focused product prompt into a working Bolt interface and refine it before adding backend complexity. - **Add User Authentication**: Connect Supabase authentication, test sign-up and login flows, and verify users in the dashboard. - **Store Data in Supabase**: Move changing currency data into Supabase, review migrations, apply security policies, and confirm live database reads. - **Fetch Live Rates Securely**: Use Supabase edge functions and secrets to fetch external API data securely and update database values on a schedule. - **Monetize with Stripe**: Create a Stripe subscription flow, track subscription data in Supabase, and unlock app features based on active status. - **Build for Production**: Choose when to hardcode, store, export, or synchronize app logic so dynamic data, secrets, and payment flows stay reliable. ### Tools covered - Bolt - Supabase - Stripe - GitHub ### Course: Intro to AI Evals > A beginner-friendly course on building a first AI evaluation with a spreadsheet, Claude Code, known outcomes, and a practical accuracy-and-cost scorecard. - Canonical page: https://app.therundown.ai/courses/intro-to-ai-evals - Machine-readable page: https://app.therundown.ai/courses/intro-to-ai-evals.md - Instructors: Nate Grahek - Duration: 1 hour 17 minutes - Lessons: 1 - Chapters: 6 - Published: 2026-08-31 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Learn how to measure AI quality, compare models, and trust workflows before you scale them. ### What you will learn - **Find Practical AI Wins**: Identify narrow, repeatable AI workflow opportunities where reliability, cost, speed, and accuracy can be measured. - **Design Reliable Evals**: Build workflow evals using real inputs, expected outputs, prompts, models, and answer keys from historical human decisions. - **Choose Automation Targets**: Select automation candidates with clear categories, meaningful volume, business value, and low-risk human-in-the-loop deployment paths. - **Measure Model Tradeoffs**: Compare prompts and models on the same task to balance output quality against latency and operating cost. - **Boost Eval Accuracy**: Improve accuracy with targeted business rules, examples, context, and decomposed workflow steps. - **Preserve Workflow Knowledge**: Capture tacit human expertise and manage AI context so workflow knowledge stays usable across experiments. ### Course: Intro to Vibe Coding > Anyone can build real software now. Learn how, without the spiral of broken code, wasted tokens, and builds that go off the rails. - Canonical page: https://app.therundown.ai/courses/intro-to-vibe-coding - Machine-readable page: https://app.therundown.ai/courses/intro-to-vibe-coding.md - Level: Beginner - Categories: Coding & App Development - Instructors: Nate Grahek - Duration: 4 hours 18 minutes - Lessons: 3 - Chapters: 22 - Published: 2026-04-28 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **Plan AI Builds**: Turn vague app ideas into focused PRDs, MVP scopes, and step-by-step build plans before generating code. - **Prompt Like a CTO**: Use one-question-at-a-time interviews, multiple-choice prompts, and clear context to guide AI like a technical co-founder. - **Prototype Before Coding**: Create mockups, compare UI directions, and validate user flows with sample data before connecting real APIs or accounts. - **Pick the Right Tools**: Choose AI coding tools, models, subscriptions, and workflows based on project complexity, budget, control, and token cost. - **Ship Safely**: Debug and deploy projects using local previews, browser inspection, console errors, environment variables, and production hosting workflows. - **Handle Production Risk**: Make practical decisions about authentication, payments, databases, pricing limits, API keys, and security before inviting real users. ### Tools covered - Replit - Lovable - Vercel - Bolt - Cursor - Claude Code ### Course: Introduction to Agentic Development > Build an agentic development workflow for creating, debugging, and shipping web apps with Warp while staying in control of code changes, API integration, and project context. - Canonical page: https://app.therundown.ai/courses/introduction-to-agentic-development - Machine-readable page: https://app.therundown.ai/courses/introduction-to-agentic-development.md - Level: Beginner - Categories: Coding & App Development, Automation & Agents - Instructors: Zach Lloyd - Duration: 55 minutes - Lessons: 1 - Chapters: 5 - Published: 2025-09-09 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Agentic Workflow**: Use a prompt-driven workflow to plan, steer, verify, and preserve code changes while staying in control of the implementation. - **Guided Development**: Guide an agent through project setup, feature development, terminal commands, and code inspection in small, verifiable steps. - **Code Ownership**: Review plans, diffs, commits, saved prompts, and project context to keep agent-generated work understandable and reusable. - **Practical Debugging**: Debug failures by sharing errors, validating assumptions, checking server inputs, and iterating until the app works end to end. - **API Integration**: Integrate external APIs safely by providing current documentation, configuring secrets properly, and testing the full request flow. - **Tool Selection**: Choose appropriate agentic tools, project types, languages, context sources, and deployment paths for real-world development work. ### Tools covered - Warp - Git - Docker - Figma ### Course: Land Your First AI Consulting Gig > Learn a practical, market-led sequence for turning AI building skills into paid consulting work: master one tool, read real Upwork demand, win early contracts, and turn proof into inbound opportunities. - Canonical page: https://app.therundown.ai/courses/land-your-first-ai-consulting-gig - Machine-readable page: https://app.therundown.ai/courses/land-your-first-ai-consulting-gig.md - Instructors: Aniket Panjwani - Duration: 58 minutes - Lessons: 1 - Chapters: 6 - Published: 2026-08-22 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview You can build with AI. Here's how to get someone to pay you for it, starting with one real Upwork listing. ### What you will learn - **Frame Client Value**: Spot real business problems where AI can speed up delivery without distracting from the outcome a client is paying for. - **Learn by Building**: Build practical AI skills through small projects that stretch your ability and map directly to paid work. - **Find Paid Demand**: Use Upwork job posts to identify warm demand, beginner-friendly opportunities, and the technical skills worth practicing next. - **Screen Jobs Wisely**: Evaluate freelance opportunities by checking scope clarity, client signals, technical difficulty, and your ability to prove you can deliver. - **Win Early Trust**: Create concise proposals and profile assets that show proof, relevance, and trustworthiness even with limited client history. - **Grow From Scraping**: Turn simple scraping tasks into a path toward stronger projects involving authentication, dashboards, APIs, CRMs, and web apps. ### Course: Launch Your OpenClaw from Scratch > Learn how to plan, launch, and refine your OpenClaw agent. - Canonical page: https://app.therundown.ai/courses/launch-your-openclaw-from-scratch - Machine-readable page: https://app.therundown.ai/courses/launch-your-openclaw-from-scratch.md - Level: Beginner - Categories: Automation & Agents - Instructors: Michael Shimeles - Duration: 2 hours 23 minutes - Lessons: 1 - Chapters: 11 - Published: 2026-04-28 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### What you will learn - **Understand Agent Basics**: Explain how LLMs use tokens, embeddings, context, and prediction so you can design clearer prompts and workflows. - **Prompt with Precision**: Write concise instructions that define the task, constraints, process, and definition of done for reliable agent behavior. - **Set Up OpenClaw**: Deploy OpenClaw, connect model access and communication, and configure the core files that shape identity, tools, memory, and routines. - **Connect Practical Tools**: Connect services such as Notion, Trello, Google Calendar, search, and email, then validate each tool with real actions. - **Automate Workflows**: Build skills, subagents, and scheduled automations that preserve context and handle repeatable work without blocking your main agent. - **Refine and Scale**: Improve OpenClaw over time with recursive skill refinement, reverse prompting, memory, and lean context management. ### Tools covered - OpenClaw - Composio - Notion - Trello - Google Calendar - Telegram ### Course: Let ChatGPT 5.6 Build Your Next Presentation > A beginner-friendly, step-by-step workshop on using ChatGPT to gather content, draft a strong structure, and build a finished PowerPoint presentation your team can still open and edit. - Canonical page: https://app.therundown.ai/courses/let-chatgpt-5-6-build-your-next-presentation - Machine-readable page: https://app.therundown.ai/courses/let-chatgpt-5-6-build-your-next-presentation.md - Level: Beginner - Categories: AI Productivity, Presentations, Business Communication - Instructors: Nate Grahek - Duration: 1 hour 12 minutes - Lessons: 1 - Chapters: 8 - Published: 2026-07-25 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Build a polished, editable PowerPoint deck from blank page to finished file. ### What you will learn - **Choose the Right Deck**: Identify recurring presentations where AI can save preparation time while meeting real business and collaboration needs. - **Prepare AI-Ready Inputs**: Assemble templates, brand rules, context, and clean data formats that give AI enough information to produce accurate work. - **Plan Before Building**: Use clarifying questions, outline options, objections, and previews to align the story before generating a final file. - **Create Editable PowerPoints**: Specify native charts, tables, placeholders, and layouts so teammates can revise the deck in PowerPoint. - **Design Reusable Templates**: Curate large slide libraries into a small set of practical layouts with clear usage instructions. - **Automate Repeatable Decks**: Turn proven prompts, guardrails, and fixed templates into reliable workflows for recurring presentation updates. ### Tools covered - ChatGPT - PowerPoint - Gemini ### Course: Mastering AI Agents for Customer Service with Fin > Build and optimize Fin-powered customer service agents with reliable knowledge, controlled escalation, performance metrics, and safe rollout practices. - Canonical page: https://app.therundown.ai/courses/mastering-ai-agents-for-customer-service-with-fin - Machine-readable page: https://app.therundown.ai/courses/mastering-ai-agents-for-customer-service-with-fin.md - Level: Beginner - Categories: Sales, Consulting & Customer Success, Automation & Agents, Business Operations & Leadership - Instructors: Brian Byrne - Duration: 1 hour 1 minutes - Lessons: 1 - Chapters: 8 - Published: 2025-08-13 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Plan a Safe Rollout**: Choose focused use cases, success criteria, rollout scope, and ownership before turning on an AI agent. - **Train With Control**: Connect reliable knowledge sources and write clear guidance that controls tone, behavior, escalation, and edge cases. - **Design Support Workflows**: Build multi-step flows that troubleshoot issues, use customer context, collect details, and hand off with the right information. - **Test Before Launch**: Validate responses with targeted testing, scoped audiences, and channel-specific checks before customers rely on them. - **Measure Performance**: Track resolution, handoffs, abandonments, topic trends, and customer experience signals to find improvement opportunities. - **Optimize Continuously**: Prioritize updates to content, data access, integrations, and automation based on evidence from real support conversations. ### Tools covered - Fin - Intercom - Stripe ### Course: Mastering Prompting: Techniques for Better AI Responses > Build clearer, more reliable prompts for AI tasks by choosing the right model, structuring requests with context and constraints, and applying techniques like zero-shot, few-shot, role prompting, prompt chaining, and reasoning-aware prompt. - Canonical page: https://app.therundown.ai/courses/mastering-prompting-techniques-for-better-ai-responses - Machine-readable page: https://app.therundown.ai/courses/mastering-prompting-techniques-for-better-ai-responses.md - Level: Beginner - Categories: AI Foundations & Prompting, Productivity & Knowledge Work - Instructors: Dr. Alvaro Cintas - Duration: 1 hour 11 minutes - Lessons: 1 - Chapters: 6 - Published: 2025-08-01 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Choose the Right Model**: Select the right model type for simple requests, creative work, coding, planning, logic, and other complex tasks. - **Apply Prompting Techniques**: Use zero-shot, few-shot, role, chain, tree-of-thought, and step-by-step patterns to match the task at hand. - **Structure Strong Prompts**: Create prompts with clear tasks, context, input data, constraints, tone, and output formats for more reliable responses. - **Prompt for Real Use Cases**: Adapt prompt patterns for coding, debugging, content creation, research, marketing, education, and style matching. - **Improve AI Outputs**: Refine weak prompts with clearer instructions, examples, delimiters, formatting requirements, and prompt improvement tools. - **Work Across AI Tools**: Manage long conversations, model costs, provider differences, and visual or audio prompting quirks more effectively. ### Tools covered - PowerPoint ### Course: Nano Banana Pro for Slide Decks > Create consistent, editable presentation visuals with Nano Banana Pro by using stronger prompts, reusable style systems, accuracy checks, and progressive disclosure for clearer audience communication. - Canonical page: https://app.therundown.ai/courses/nbp-for-slide-decks - Machine-readable page: https://app.therundown.ai/courses/nbp-for-slide-decks.md - Level: Beginner - Categories: Content & Creative, Productivity & Knowledge Work - Instructors: Nate Grahek - Duration: 48 minutes - Lessons: 1 - Chapters: 6 - Published: 2025-12-06 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Purposeful AI Visuals**: Create illustrations, diagrams, icons, and conceptual graphics that help audiences grasp ideas faster. - **Stronger Visual Prompts**: Write clear prompts with audience, message, style, and constraints to get more relevant slide assets. - **Accuracy Control**: Verify labels, numbers, claims, and embedded text so generated visuals stay accurate and presentation-ready. - **Consistent Deck Style**: Build reusable style references and visual rules that keep deck imagery consistent across slides. - **Research-Led Ideation**: Use source material to brainstorm slide structure and visual concepts before generating polished graphics. - **Progressive Disclosure**: Reveal complex visuals in stages with cropping, masking, and simplified graphics that guide audience attention. ### Tools covered - Nano Banana Pro - PowerPoint - Keynote ### Course: No Code Product Building: From Idea to Product-Market Fit with AI > Go from idea to working product without a technical team. Marily Nika teaches the product development process, shows how AI replaces work you would otherwise hire for, and gives you a practical path to a working prototype with real user feedback. - Canonical page: https://app.therundown.ai/courses/no-code-product-building-from-idea-to-product-market-fit-with-ai - Machine-readable page: https://app.therundown.ai/courses/no-code-product-building-from-idea-to-product-market-fit-with-ai.md - Level: Beginner - Categories: No-Code, Product Management, AI Product Development, Prototyping - Instructors: Marily Nika - Duration: 1 hour 38 minutes - Lessons: 2 - Chapters: 13 - Published: 2026-08-10 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Take an idea from notes-app concept to prototype and real user feedback with AI. ### What you will learn - **Validate Real Demand**: Use AI-assisted research and real user conversations to identify pain points, willingness to pay, and product signals before building. - **Define Focused Requirements**: Turn validated insights into a prioritized PRD that clarifies users, use cases, success metrics, and the smallest valuable feature set. - **Review With AI Agents**: Set up product, customer, engineering, and risk reviewers to pressure-test assumptions, scope, privacy, feasibility, and overpromising. - **Prototype From Evidence**: Generate and refine UI concepts from product requirements, then move selected screens into the right no-code or AI builder. - **Separate Prototype From MVP**: Use clickable prototypes to gather evidence while defining MVP tests around desirability, feasibility, business viability, and the riskiest assumption. - **Measure Product Quality**: Create rubrics and evaluation loops that score AI outputs for relevance, groundedness, completeness, safety, and hallucination risk. ### Tools covered - Perplexity - Reddit - Gemini - ChatGPT - Claude - Stitch ### Course: NotebookLM for Work > Use Google NotebookLM to turn trusted sources into grounded research syntheses, reusable context, and business-ready outputs for sales prep, reports, and AI project workflows. - Canonical page: https://app.therundown.ai/courses/notebooklm-for-work - Machine-readable page: https://app.therundown.ai/courses/notebooklm-for-work.md - Level: Beginner - Categories: Data, Research & Analysis, Automation & Agents - Instructors: Nate Grahek - Duration: 44 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-12-20 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Build Research Notebooks**: Turn files, websites, videos, and internal materials into a focused knowledge base for business research and execution. - **Engineer Better Context**: Use stored sources, examples, and constraints to reduce generic AI output and support repeatable work. - **Prepare for Meetings**: Research companies, identify risks and opportunities, and prepare sharper talking points before sales calls or meetings. - **Create Reusable Outputs**: Synthesize expert content from multiple sources into structured guides, reports, tables, and reusable reference material. - **Manage Source Quality**: Find, review, refine, and remove sources so each notebook stays accurate, relevant, and grounded in trusted material. - **Connect AI Workflows**: Move distilled insights into Gemini, gems, projects, docs, or sheets to support drafting, planning, and production workflows. ### Tools covered - NotebookLM - Gemini - Claude - Google Docs - Google Drive - YouTube ### Course: Template-First AI Workflows: From Messy Data to Perfect Documentation > Build template-first AI workflows that turn messy inputs into polished documents, decks, summaries, and content with repeatable automation and human review. - Canonical page: https://app.therundown.ai/courses/template-first-ai-workflows-from-messy-data-to-perfect-documentation - Machine-readable page: https://app.therundown.ai/courses/template-first-ai-workflows-from-messy-data-to-perfect-documentation.md - Level: Beginner - Categories: Automation & Agents, Productivity & Knowledge Work, Business Operations & Leadership - Instructors: Nate Grahek - Duration: 1 hour - Lessons: 1 - Chapters: 9 - Published: 2025-09-05 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Find Workflow Opportunities**: Spot repeatable business tasks where AI can turn messy inputs into useful, structured outputs. - **Design From Templates**: Use finished examples and templates to define the format, quality, and style AI should reproduce. - **Control AI Outputs**: Write clearer prompts with constraints, placeholders, and examples that reduce ambiguity and output drift. - **Prioritize Automation**: Decide which tasks are ready to automate based on business value, repeatability, available examples, and review needs. - **Build Modular Workflows**: Break complex work into focused workflow steps that summarize, draft, structure, parse, and populate documents. - **Keep Quality Human-Led**: Review, polish, and simplify automated outputs so useful drafts can become publishable deliverables. ### Tools covered - n8n - Zapier - Make - Google Slides - Google Docs - LinkedIn ### Course: The AI Foundation Workshop: Your New Operating System for the Future of Work > Build a practical AI operating system for repeatable work by using context-rich prompts, reusable projects, structured files, and low-risk workflows that turn messy tasks into clearer drafts, critiques, and handoffs. - Canonical page: https://app.therundown.ai/courses/the-ai-foundation-workshop-your-new-operating-system-for-the-future-of-work - Machine-readable page: https://app.therundown.ai/courses/the-ai-foundation-workshop-your-new-operating-system-for-the-future-of-work.md - Level: Beginner - Categories: AI Foundations & Prompting, Automation & Agents - Instructors: Nate Grahek - Duration: 1 hour 3 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-08-22 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Prompt with Context**: Turn vague requests into focused AI work by defining roles, context, criteria, examples, and constraints. - **Clarify Before Drafting**: Use AI as a thinking partner to clarify goals, ask better questions, and create stronger first drafts. - **Manage AI Effectively**: Set expectations, provide feedback, choose the right model or mode, and restart conversations when context gets overloaded. - **Build Reusable AI Projects**: Create reusable projects or custom GPTs with curated files, instructions, examples, and frameworks for repeatable work. - **Structure AI Knowledge**: Use markdown to structure instructions, knowledge files, summaries, and handoffs that AI can follow reliably. - **Orchestrate AI Workflows**: Choose between chat, projects, workflows, and software based on risk, repeatability, accuracy needs, and scale. ### Tools covered - LinkedIn - GPTs ### Course: The Human-First Agentic Content Workflow Workshop > Build human-in-the-loop n8n workflows that turn existing team knowledge into polished content assets while preserving brand judgment, approvals, and scalable publishing paths. - Canonical page: https://app.therundown.ai/courses/the-human-first-agentic-content-workflow-workshop - Machine-readable page: https://app.therundown.ai/courses/the-human-first-agentic-content-workflow-workshop.md - Level: Beginner - Categories: Automation & Agents, Content & Creative, Business Operations & Leadership - Instructors: Nate Grahek - Duration: 1 hour 2 minutes - Lessons: 1 - Chapters: 7 - Published: 2025-11-21 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Find Automation Opportunities**: Spot repetitive, high-value bottlenecks where AI can extend existing human effort instead of generating work from scratch. - **Build Agentic Workflows**: Design N8N workflows that move source content through analysis, strategy, drafting, formatting, review, and publishing preparation. - **Repurpose Source Content**: Turn human-made videos, podcasts, trainings, or process recordings into useful drafts, SOPs, and channel-ready content with stronger context. - **Co-Create With AI**: Use prompts, JSON, code nodes, HTTP requests, and API documentation to troubleshoot and expand automations with AI assistance. - **Keep Humans in the Loop**: Add approval checkpoints where taste, accuracy, brand quality, or publishing decisions need human judgment before automation continues. - **Scale for Production**: Scale workflows with triggers, reusable stored outputs, and flexible destinations like Notion, WordPress, Slack, email, or project tools. ### Tools covered - n8n - Gemini - YouTube - Notion - WordPress - Slack ### Course: The State of AI Presentation Tools in 2026 > Build reliable AI-assisted presentation workflows that turn messy ideas and research into editable, on-brand PowerPoint decks using Claude, templates, and structured assembly. - Canonical page: https://app.therundown.ai/courses/ai-presentations-with-claude-skill - Machine-readable page: https://app.therundown.ai/courses/ai-presentations-with-claude-skill.md - Level: Beginner - Categories: Productivity & Knowledge Work, Content & Creative - Instructors: Nate Grahek - Duration: 1 hour 21 minutes - Lessons: 1 - Chapters: 6 - Published: 2026-04-03 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Orchestrate Presentation Workflows**: Break complex deck work into reliable AI-assisted steps for research, structure, messaging, drafting, and final assembly. - **Choose the Right Tools**: Assess presentation tools by export quality, editability, model transparency, limits, and fit for mission-critical PowerPoint work. - **Draft From Spoken Ideas**: Capture rough ideas through dictation and guide AI with focused questions before moving into substantial drafting. - **Summarize Research Into Slides**: Convert dense research, transcripts, and source material into skimmable slide artifacts that support faster review and decision-making. - **Automate Template Decks**: Build reusable PowerPoint templates with placeholders, schemas, brand guidance, and approved JSON content for consistent deck production. - **Extend With Focused Automation**: Use AI extensions and advanced automation for focused tasks such as feedback, edits, imagery, and troubleshooting without losing workflow control. ### Tools covered - PowerPoint - Claude - NotebookLM - YouTube ### Course: Vibe Coding in Cursor for Non-Developers > Build practical custom software ideas with Cursor by planning clear requirements, generating prototypes, preparing data, managing context, and debugging confidently without traditional coding experience. - Canonical page: https://app.therundown.ai/courses/vibe-coding-in-cursor-for-non-developers - Machine-readable page: https://app.therundown.ai/courses/vibe-coding-in-cursor-for-non-developers.md - Level: Beginner - Categories: Coding & App Development - Instructors: Nate Grahek - Duration: 52 minutes - Lessons: 1 - Chapters: 9 - Published: 2025-10-04 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Workshop recording ### What you will learn - **Plan Buildable Apps**: Turn vague software ideas into focused, buildable concepts with clear goals, constraints, and scope. - **Prepare Strong Context**: Create structured PRDs, sample data, and visual references that give AI coding tools the context they need. - **Prototype Before Coding**: Use free AI builders to explore mock-ups, compare directions, and choose a practical starting point before coding. - **Manage Cursor Effectively**: Guide Cursor with the right files, modes, checkpoints, and small requests to make steady progress without context overload. - **Fix Problems Systematically**: Debug issues by collecting screenshots, browser errors, relevant files, and clear diagnosis prompts before making changes. - **Speak the Language**: Use practical software terminology to communicate clearly about front ends, back ends, assets, APIs, and documentation. ### Tools covered - Cursor - Claude - WhisperFlow - Monologue - Superwhisper - YouTube ### Course: Your AI Reset Workshop > A beginner-friendly reset that makes the core AI concepts clear, explains inconsistent results, and gives you a practical system for prompting, reusable context, and repeatable work. - Canonical page: https://app.therundown.ai/courses/your-ai-reset-workshop - Machine-readable page: https://app.therundown.ai/courses/your-ai-reset-workshop.md - Instructors: Nate Grahek - Duration: 1 hour 15 minutes - Lessons: 1 - Chapters: 5 - Published: 2026-08-22 - Access: See https://app.therundown.ai/plans for current membership and enrollment terms. ### Overview Learn the fundamentals, give AI the right context, and build repeatable workflows in ChatGPT and Claude. ### What you will learn - **Diagnose AI Failures**: Identify why an AI output failed and fix the underlying issue with better context, instructions, model choice, or task design. - **Use AI Foundations**: Use models, tokens, context windows, reasoning, tools, and agents in practical terms to make better workflow decisions. - **Build Context Docs**: Create reusable personal, role, company, and project context so AI can produce more relevant work from the start. - **Create Better Projects**: Set up AI projects with durable instructions, knowledge files, templates, and examples for repeatable knowledge-work outputs. - **Manage Chat Quality**: Reduce context rot by starting fresh conversations while preserving important knowledge in reusable project environments. - **Run AI Workflows**: Apply a staged workflow that gathers context, shapes direction early, invites pushback, and generates final outputs only when the plan is sound.