Share your best AI workflow. We could show it to 2M+ people.

I work with long talking-head videos for my online course and Instagram. Turning them into vertical shorts used to take hours per video: finding the best moments, cropping to 9:16, adding captions, or paying for clipping subscriptions. I built a video-clipping agent as a Claude skill that runs entirely on free, open-source tools. Step-by-step: 1. I drop a long video—either talking-head footage or an already-produced 16:9 video—into a folder and invoke my "clipper" skill in Claude. 2. Claude transcribes the video locally with faster-whisper using word-level timestamps, so no API key is needed. ElevenLabs Scribe is an optional upgrade for tricky audio. Before transcription, the workflow cleans the audio with declipping and normalization because clipped microphone peaks can make Whisper skip words or hallucinate. 3. The skill scores every candidate moment against a rubric: hook strength (30 points), whether it can stand alone (25), emotional charge (20), rhythm (15), and ending (10). Only segments scoring 70 or higher survive, so I do not publish a weak clip simply because it exists. 4. Claude cuts the selected moments with word-level precision and reframes them to 9:16 in one of two modes: face-tracking crop for raw footage, or "pad" mode, which centers the full 16:9 video over a blurred background. The latter lets me repurpose already-edited videos without chopping off centered graphics or captions. 5. ffmpeg masters everything to a fixed specification: 1080x1920, 24fps, H.264 CRF18, and audio normalized to -14 LUFS. The files are ready for Instagram, TikTok, and Shorts. 6. I review the scored shortlist and publish the winners. What used to take hours—or require a monthly subscription—now takes minutes and costs zero. The scoring rubric also makes the agent explain why each clip deserves to exist instead of simply cutting wherever the waveform looks loud.

I’ve found it useful to use Claude while reading nonfiction books, such as *Factfulness* by Rosling. I ask Claude for a synopsis of the book and save the result as a PDF on my computer for future reference. I also ask Claude to quiz me about the book, either with multiple-choice questions or in short-answer format, to test whether I’ve fully understood what I’ve read. In an ongoing conversation, Claude can also relate the current book to other books I’ve recently read and provide additional synopses. Step-by-step: 1. I tell Claude which nonfiction book I’m reading, such as *Factfulness* by Rosling. 2. I ask Claude to create a synopsis of the book. 3. I save the synopsis as a PDF on my computer for future reference. 4. I ask Claude to quiz me using either multiple-choice or short-answer questions. 5. I continue the conversation so Claude can relate the book to other books I’ve recently read and provide further synopses.

I own a rental property where a water leak has occurred roughly every year or two. The leak typically runs for weeks before the water utility detects usage above its threshold. Because billing cycles last three months, the utility may notify me weeks or months after the problem begins. By then, the leak can have produced a bill more than $1,000 higher than the usual $100–$300 amount. The utility offers a one-time, per-account waiver for accidental leak overages. After using that waiver the first time, later incidents are entirely out of pocket. The utility also cannot notify me sooner than when usage exceeds 25,000 gallons during a billing cycle, which moves the account into a quadruple-rate tier for the rest of that cycle. I repeatedly asked whether they could provide an immediate alert when a user-set or company-set daily usage threshold was exceeded, but they said they had no system or solution for it. I tried checking my usage manually every day, but after weeks or months of normal readings, it was easy to become complacent or forget. After receiving another $1,300-plus bill, I asked ChatGPT whether I could automate the process of logging into my utility account, checking usage daily, and emailing me about the prior day’s usage or an overage. ChatGPT suggested several options, including paid AI-agent tools and a free script running on my own hardware. I wanted a completely free, cloud-based solution that would not require my computer to stay on, so I compared the paid options, including Google Spark, with a GitHub-based system. GitHub apparently includes 2,000 minutes of script runtime per month, while my system was estimated to use about 100 minutes. I spent part of a day asking ChatGPT questions, copy-pasting code into GitHub, and refining it. I now have a cloud-based system that logs into my water utility account, checks daily usage, emails me when my daily or seven-day-average thresholds are exceeded, and adds each day’s usage to an Excel spreadsheet for ongoing history. I have verified that it works, and it is set up to keep running and sending alerts without ongoing cost. I had never coded before. The system uses Python, GitHub Actions/YAML, Playwright, pandas, openpyxl, Excel, and Gmail for email alerts. Excel and Gmail were the only tools in that list I had used previously. If I want to change an alert threshold or another setting, I can log into GitHub and ask ChatGPT for the relevant code adjustment. I also added a second rental property in the same city that uses the same water utility. Replicating the process for that property required only a small amount of additional code and took almost no time. I now have a perpetually self-updating, cloud-based water-usage database with daily email alerts for both rental properties, at zero ongoing cost.

QuickBooks stores customer data across multiple windows, including notes, deposits, invoices, payments, aging, and email. To see everything, we previously had to keep several windows open while constantly searching, opening, and closing screens. Communicating with customers beyond QuickBooks’ standard letters was also difficult and time-consuming. I used Claude to write a program that brings this information together in one view. It creates an aging summary of all past-due accounts, with sortable columns for customer name, aging category, and total. I also added columns for the last payment and payment date. Clicking anywhere on a customer’s row opens a full drill-down of that customer’s data on one page. The accounting person can view invoices filtered by date or open invoices, see payments and the invoices they were applied to, review aging detail, run reports without leaving the screen, send emails, record collection notes, and print invoices. The software supports simultaneous access for multiple authorized users. The data stays up to date and can be refreshed at any time. This saves us hours of time compared with working through QBO. My next step is to have Claude automate reminder notices at 30 and 60 days, based on user-selected settings, as well as 90-day collection letters. These notices will be customizable while still allowing automation. Step-by-step: 1. I identified the customer information that was spread across QuickBooks windows, including notes, deposits, invoices, payments, aging, and email. 2. I used Claude to write a program that creates an aging summary of all past-due accounts. 3. I added sortable columns for customer name, aging category, total, last payment, and payment date. 4. I enabled users to click a customer row and open a full drill-down of that customer’s data on one page. 5. I included access to date-selectable invoices, open invoices, payments and their applied invoices, aging detail, reports, emails, collection notes, and invoice printing. 6. I set up the software so multiple authorized users can access the data simultaneously and refresh it whenever needed. 7. I plan to add customizable, automated reminder notices at 30 and 60 days and collection letters at 90 days.

Every RSVP tool I found wanted a monthly subscription to collect a few hundred yes-or-no answers, and none would return my data in a format I could use. So I built my own with Claude Code. The RSVP system is small: a form deployed on Vercel and embedded into our wedding website. We also emailed the direct link because plenty of guests might never visit the site. The form asks for more than a yes-or-no response. It collects meal choices, dietary restrictions, and plus-ones. Every submission adds a row to a Google Sheet through the Sheets API. That sheet ended up mattering more than the app. Months later, when it was time to create the seating chart, every guest was already represented in clean, structured rows that I could give to a completely different tool. For seating, I opened the same sheet in Claude and wrote out the rules in plain English: who had to sit together, who could not sit anywhere near each other, which relatives needed a quiet corner, and which friends could anchor a table of people who did not know anyone else. Claude produced a first draft that respected those constraints, along with an interactive board where I could drag people between tables. Step-by-step: 1. Tell Claude Code which fields you want on the RSVP form and have it build the front end and back end. Have Claude deploy it to Vercel. 2. Set up a Google Cloud service account, enable the Google Sheets API, and share the sheet with the service account’s email address. Configure each submission to append a row. 3. Embed the form on your wedding website and email the direct link separately. Some people will never open the website. 4. Ask for everything you will eventually need, not just attendance. Your caterer and venue will want meal counts and dietary restrictions, and going back to collect that information later is miserable. 5. When it is time to seat people, point Claude at the same sheet and write your rules in plain language: who must sit together, who cannot sit together, who needs a quiet table, and who can anchor a table of strangers. 6. Have Claude generate a first draft based on those rules, then move people yourself. This works anywhere you have a public form collecting data and a private tool that needs to reason about it. Put a spreadsheet between them instead of building one app that tries to do both. The form stays simple, and whatever you build six months later can read from the same place.

I used to create a skeptical VC investor persona through prompting whenever I ran ideas and actual business plans through AI. Recently, after working on a business plan for weeks, I used Claude Cowork with Opus 5/High to evaluate the same concept. Instead of prompting for critique and redesign each time, I built a Skill upfront. The concept is the same, but the Skill makes the execution much more effective. It reviews ideas or detailed business plans, researches first, and critiques only after that. It also analyzes and suggests wedges and new moats and, when instructed, generates detailed business plans. It works well as a one-shot analysis, but it is most effective when I push back and challenge it further. I’m sharing the Claude Skill for free under the MIT license: https://github.com/zszendro/vc-teardown Step-by-step: 1. I used to prompt AI to role-play a skeptical VC investor when reviewing ideas and business plans. 2. After working on a business plan for several weeks, I ran the same concept through Claude Cowork with Opus 5/High. 3. Instead of prompting separately for critique and redesign, I built a Skill upfront. 4. I designed the Skill to research first and critique only afterward. 5. I used it to review ideas or detailed business plans, analyze and suggest wedges and new moats, and generate detailed business plans when instructed. 6. I continued the analysis by pushing back and challenging the Skill beyond its initial one-shot response. 7. I shared the Claude Skill for free under the MIT license at https://github.com/zszendro/vc-teardown

I was tired of opening Spotlight every time I wanted to switch apps. The search results would move around, I would occasionally open the wrong app, and those few wasted seconds kept adding up throughout the day. Instead of memorizing a collection of unrelated hotkeys, I used ChatGPT Work to design and build a personalized shortcut system in Apple Shortcuts. The shortcuts can open individual apps, jump directly into specific Safari or Chrome profiles, or launch an entire work mode with one keystroke. For example, a recording shortcut could open your microphone, camera, recording software, and notes. An analytics shortcut could open all the dashboards you check each week. Step-by-step: 1. I opened the ChatGPT desktop app, started a new chat, and switched to Work mode. 2. I asked ChatGPT to plan shortcuts around my real workflow: 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. 3. I told it which apps, browser profiles, URLs, and work modes I use most. 4. I narrowed the list before making any changes by keeping the highest-frequency shortcuts and avoiding macOS or app conflicts. 5. In ChatGPT, I went to Settings → Computer Use and turned on Any App. 6. I gave ChatGPT the approved list and clearly limited its scope: Set these up in Apple Shortcuts. Do not change anything outside this list. Test each shortcut. 7. I opened All Shortcuts in Apple Shortcuts and ran each one manually. I used Quick Actions to add or change its keyboard shortcut. I like Control + Option because it is less likely to conflict with existing Mac commands. 8. I tested every shortcut while working in another app, checking that it opened the correct app, account, browser profile, or collection of tools before building more. 9. Once everything worked, I asked ChatGPT to create my reference: Write me a one-page cheat sheet of all the shortcuts you set up, where they live, and what’s left to do. The result is a personal shortcut system built around how I actually work, plus a one-page cheat sheet so I don’t have to memorize everything immediately. For the complete walkthrough, exact prompts, screenshots, and setup instructions, follow my full Rundown University guide: Use ChatGPT to Build a Custom Mac Shortcut System. Tools used: ChatGPT Work, Apple Shortcuts Industry: Cross-industry Tags: #chatgptwork #appleshortcuts #macautomation #productivity #workflowautomation

AI coding tools can build quickly, but they can also build the wrong thing quickly. Starting implementation from a vague feature request leaves important decisions about scope, architecture, edge cases, success criteria, and expected behavior to be made implicitly during coding. I created a spec-generator agent that sits between an idea and implementation. I give it a feature request, product vision, or rough description of what I want to build. It investigates the existing project, identifies missing decisions and constraints, researches external dependencies when necessary, and turns the request into a detailed specification that another AI agent can implement without having to guess what I meant. The finished specification becomes the source of truth for the rest of the development workflow. Step-by-step: 1. I give the spec-generator the feature or product idea I want to build, along with any existing requirements, vision documents, or constraints. 2. I have it inspect the existing project before proposing a solution. It needs to understand the current architecture, conventions, capabilities, and relevant prior decisions rather than designing the feature in isolation. 3. I have it identify ambiguities and missing decisions, including questions about users, behavior, scope, dependencies, edge cases, data requirements, integrations, and what is explicitly out of scope. 4. I have it research external technologies, APIs, libraries, or platform capabilities when the design depends on facts that cannot be determined from the repository alone. 5. I have it translate the idea into a layered specification: first the product purpose and desired outcomes, then the technical architecture, and finally the detailed implementation requirements. 6. I have it define measurable success criteria and acceptance tests so that “done” means something concrete rather than simply “the code was written.” 7. I have it persist the finished specification in the project so developers or coding agents can treat it as the source of truth during implementation. 8. I pass the specification through a separate review or validation step before coding begins, resolving gaps or contradictions in the spec rather than discovering them halfway through implementation. Instead of asking an AI coding agent to interpret a rough idea while it writes code, I separate figuring out what should be built from building it.

I produce cinematic AI video. What ruins a sequence is almost never image quality; it is drift. The face, jacket, and light change between shots, so what should read as one scene becomes a gallery of near misses. Most people fight this by regenerating until something matches. I built the opposite: a set of chained Claude skills that locks a character once and never lets the model improvise again. The lesson that started it cost me 72 credits in one unusable shot. I fed a finished POV frame as an image reference to “set the world” for a motorcycle sequence. The model treated it as a strong identity reference, cloned the entire cockpit from that frame, and ignored the bike I was describing in the prompt. Image references are read as identity, not atmosphere. That misunderstanding is why most people’s characters mutate. Step-by-step: 1. I lock the character before generating anything. I fill nine layers in order: identity and facial structure, skin realism, wardrobe and materials, emotional energy, environment, lighting scheme, camera language, vertical 9:16 framing, and a fixed block of negative instructions. A generic prompt returns a generic face. This is direction, not a portrait. 2. I shoot a character sheet: a neutral, multi-view image of the locked character on white. That sheet—not a pretty hero frame—is what I reuse as a reference from then on. 3. I expand coverage instead of reinventing the scene. One locked scene becomes a full shot list with hero angles, macro inserts, and establishing shots. The subject, wardrobe, weather, and lighting never change between prompts. Only the camera, focal length, framing, and distance change. Every prompt restates the entire world from scratch because the generator has no memory between images. Consistency comes from redundancy, not from the model remembering. 4. I time the story before animating. I create a 3x3 board with nine beats over fifteen seconds, with each panel showing its timecode, shot title, and action. The rule is strict: the character, wardrobe, world, and light stay identical across all nine panels; only the story advances. The emotional scale rises toward panel nine. 5. I generate with reference discipline. I pass the object or vehicle in a clean exterior shot as the only image reference, then describe the world, framing, and interior in text, repeating that description word for word across shots. I chain consecutive shots by feeding the last frame of one as the start image of the next. The joins are invisible. 6. I review the board against the renders, keep the shots that hold, and regenerate only the ones that drift. The result is that I stopped paying the regeneration tax. Shots are budgeted, not gambled, and a fifteen-second sequence holds together because it was blocked like a real scene—with coverage and continuity—before a single frame existed. The nine-layer character build and the coverage framework are adapted from two AI video courses plus twenty years of production. I learned the reference discipline in step five the expensive way.

A friend who had just started a pet-sitting and dog-walking business asked me to build a booking app. I used Google AI Studio to design the initial prototype. The process was straightforward, and I had a basic working system running within two hours. AI Studio created a Firebase database to store the details and was also effective at designing frontend changes. It published the app for me, and the resulting UI was intuitive. After demonstrating the app, I identified many additional features that needed to be added. This went beyond AI Studio’s capabilities, so I exported the code from AI Studio and started using Claude. I asked Claude to analyze the code and suggest the required changes. Claude identified critical security flaws in the database. I fixed them manually at first, but then realized that Claude could access the Firebase database and fix issues automatically. I continued prompting Claude with additional feature requests, and it built them. There were errors along the way, so I needed to be familiar with Google Chrome’s developer tools to copy the errors and ask Claude how to fix them. I used MailJS for email templates and Resend for email transport. I stored the app in GitHub and allowed Claude to access the repository so it could commit changes automatically. I ran the app locally with npm during development and then hosted it on Vercel. Eventually, I purchased an inexpensive domain name, and the app is currently hosted at Names. One major problem was that when Google AI Studio created the Firebase database, the permissions were locked, preventing me from making administrative changes. I solved this by recreating the database and asking Claude to write the SQL query to set it up. The permissions were still tricky, and I had to continue asking Claude to correct them. Google’s service permissions can be difficult to understand, and finding the correct settings was not always easy. Firebase was also complex to navigate, and getting the permissions configured correctly took time. Once I allowed Claude to connect to the app and its background services through the Google plugin, development became much faster. However, you need to trust the tool carefully and always work on a copy of the live app. Overall, I think app development with Google AI Studio and Claude is impressive. You can create professional apps quickly. I come from an IT support background, though, and I think people who are new to IT may find it difficult to troubleshoot errors without a basic understanding of networking and systems administration. I also built a litter-tracking app using ChatGPT, and it was equally effective. I eventually started using Codex and Claude Code, but I think standard Claude and ChatGPT are more intuitive for nontechnical users. Step-by-step: 1. I used Google AI Studio to create an initial booking-app prototype for a pet-sitting and dog-walking business. 2. I used the Firebase database created by AI Studio to store the app’s details and used AI Studio to design frontend changes. 3. I published the initial app with AI Studio and demonstrated it to identify additional features. 4. I exported the code from AI Studio and asked Claude to analyze it and suggest changes. 5. I addressed the critical database security flaws identified by Claude, first manually and later by allowing Claude to access the Firebase database. 6. I prompted Claude to build additional features and used Google Chrome’s developer tools to copy errors and ask Claude for fixes. 7. I recreated the Firebase database when AI Studio’s locked permissions prevented administrative changes, then asked Claude to write the SQL query to set it up. 8. I used MailJS for email templates and Resend for email transport. 9. I stored the app in GitHub and allowed Claude to commit changes automatically. 10. I ran the app locally with npm during development, hosted it on Vercel, and later purchased an inexpensive domain name that is currently hosted at Names. 11. I connected Claude to the app and background services through the Google plugin, while continuing to work on a copy of the live app.

I enjoy researching football accumulators, but I did not want a workflow that simply asks AI for “the best bets.” I wanted a repeatable process that starts with evidence, makes uncertainty visible, and is allowed to say “pass.” I started building it on 4 August 2026. The result is a private Football Lab covering the Premier League, Championship, and League One. It is for personal research and entertainment only—not a public tips service, income claim, or automated betting system. The workflow pulls public football data into a private, traceable store, then cleans, reconciles, and blends it before analysis: - football-data.co.uk: 7,420 normal-context matches from 2021–22 to 2025–26, including results, basic statistics, referees, and historical odds. I excluded COVID-affected 2020–21. - Fixture Download: An initial 2026–27 schedule snapshot containing 1,484 fixtures across the three divisions. - Premier League public match feed: A five-season layer covering referees, cards, event timing, added time, and 469 penalty kicks split into scored, saved, and missed. - Official Premier League Transfer Watch and BBC Sport: A source ledger for squad movement. - Official EFL appointment pages: Timestamped Championship and League One weekend checks. Schedules, appointments, and transfers retain their source and capture time instead of becoming untraceable web snippets. Step-by-step: 1. I validate fixture identity, duplicates, dates, missing fields, and team-name mismatches before modelling. I then reconcile the sources into a common club and fixture record. A tidy report built on a broken fixture list is still wrong. 2. I build separate Elo and Poisson baselines that turn historical team performance and home advantage into expected goals and home/draw/away probabilities. The divisions remain separate, so Championship form is not quietly treated as Premier League form. Each fixture is predicted before its result updates the model, preventing hindsight from creeping in. I backtested the baseline against 1,484 completed 2025–26 fixtures to establish an honest benchmark rather than claim a magic model. 3. I add context that the baseline cannot see alone. When an official referee appointment is confirmed, I timestamp it and match it to the fixture. The Lab can then show competition-specific cards, dismissals, and—where Premier League evidence exists—penalty-kick and added-time patterns. The question is not whether a referee picks a winner, but whether the match environment looks more volatile or the sample is too thin to support a useful conclusion. Unknown or changed appointments remain neutral. 4. I maintain a private append-only ledger of source-backed squad movement and label every club as established, promoted, relegated, or limited history. A signing does not automatically improve a probability, and an old-division record is not treated as identical new-division form. Until those effects earn a tested role in the model, they widen uncertainty or rule out a fragile fixture. 5. I begin with the complete fixture board rather than a short list of favourites. For every game, I combine baseline probabilities, expected goals, team context, confirmed squad changes, referee environment where evidence exists, and unresolved live checks. I write a plain-English match story explaining what the baseline sees, what could make the fixture fragile, and whether the sensible outcome is candidate, watch, or pass. 6. I preserve the full board in a private Weekend Sheet, along with the model read and reason, uncertainty flags, and the small number of research candidates that survive the checks. There can be up to seven candidates, but seven is never a quota: three strong games means three, and none means pass. 7. Before results, I record each run’s data cutoff, model version, and referee-status snapshot. After the round, a separate debrief compares the original probabilities and swerves with what happened, checks whether the flags caught fragile fixtures, and identifies one bounded improvement. The Lab remains in private paper-run mode while a recurring live results source, appointment capture, and market checks earn their own evidence gates. AI helps turn public-source data into an inspectable research workflow. It makes uncertainty visible and treats “pass” as just as valid as a confident call.

Ten days ago, I posted about keeping a `HANDOFF.md` so an AI film project doesn't lose the decisions it has already made. This is the other half, and it turned out to be the expensive one: the decisions that haven't been made yet. Credit where it belongs: Tony Ojeda posted a spec-generator agent that sits between an idea and implementation. The rule I borrowed is his: the agent inspects what already exists before proposing anything. I applied it to film production instead of code. I pointed an agent at my production documents for a short film I hadn't started generating and asked it to identify what was still undecided, what each item affected, and what would break if it were decided late. It came back with ten open decisions. One of them was worth the whole exercise. My environment description is locked verbatim across all 40 shots so the world stays identical. The film is called *The Thaw*. Whether the ice visibly melts during the film is written down nowhere. If I decide that in week three, all 40 keyframes get regenerated at once. I would have found out around shot 12. The counterintuitive part is who writes the list: not me. Asking the person who already has the whole film in their head what's missing gets you very little, because they have all of it—and that's exactly why they can't see the hole. The list has to come from something that only knows what's written down. Step-by-step: 1. I put the open questions in section 0 of the handoff, above everything else, so it's the first thing read and the first thing emptied. 2. I have the agent write that list, not me. It reads every existing document for the project and returns only what it cannot know from them. 3. I have it return three things for each item: what's undecided, which shots it affects, and what breaks if it's decided late. The third column sets the priority. 4. I have it sort by what costs the most to change afterward, not by what's easiest to answer. 5. I keep one hard rule: while an open question affects a shot, that shot doesn't get generated. The question gets decided, or deferred in writing with the cost of being wrong stated. 6. Every answered item leaves section 0 through one of two doors: into the closed canon or into the rejected list. Nothing is simply deleted. 7. I ask specifically about the things that go missing every time: the rule of the world; the physical scale of anything impossible, such as whether the character can touch or climb it; any object appearing in more than one shot; screen direction; how it ends; who speaks and in what voice; and the delivery format. 8. When it finds nothing real, it says so. A list padded to look thorough is worse than an empty one. The version I'd used for a year ran at the end of a session and recorded what got decided. Running it at the start, focused on what hasn't been decided, is the same document pointed the other way—and it's the direction that saves money.

As an AI consultant, I need to stay ahead of a landscape that evolves daily by reading dozens of industry articles each week. Manual tech watch quickly becomes a major time sink and pulls me away from client work. I built a specialized AI agent for challenge-focused summarization. Instead of producing generic, passive summaries, it answers two questions for every article: What specific problem, friction, or limitation does the article highlight? What practical solutions, tools, or actionable steps does it provide? The agent strips away marketing hype, introductory fluff, and generic definitions. It identifies the author’s core pain point and produces a quick-read summary focused on actionable technical or business solutions. This turns 15-minute reads into 30-second, high-value digests. By focusing on problem-solving rather than passive reading, the workflow helps me identify practical tools and frameworks I can apply directly to client projects and improve my consulting work. Step-by-step: 1. I provide the agent with a raw article link or the article text. 2. The agent filters out marketing hype, introductory fluff, and generic definitions. 3. It identifies the core pain point, friction, or limitation discussed by the author. 4. It extracts the concrete answers, including practical solutions, tools, and actionable steps. 5. It synthesizes the findings into a quick-read format focused on actionable technical or business solutions. 6. I use the digest to identify tools and frameworks that may apply directly to client projects.

My girlfriend has a professional camera and lens for photographing birds. Because the camera captures an enormous number of images per second, sorting through them and keeping only the good ones after a day at the wetlands took a very long time. She showed me the two tools she was using to tag, crop, view, and zoom into photos, inspect focal points, and view the color histogram. I decided to vibe-code a C# tool using Cursor, Grok, and Claude Opus. The tool reads the EXIF and MakerNotes metadata in each picture and lets her scroll through all the photos in fullscreen with the mouse. It displays the histogram in the top-right corner and useful information—aperture, shutter speed, focal length in millimeters, and ISO—in the top-left, with color coding from red to green. It also shows the Nikon focus zone. Clicking the left mouse button zooms to 75%, allowing us to pan, while clicking the right mouse button zooms to 100%. Once we decide visually on the best crop and judge the photo’s potential, pressing a keyboard key instantly crops the image at the current view, saves a copy, tags the photo, and adds her copyright in the bottom-right corner. This makes the process much faster because she can view, tag, crop, and add copyright at the same time. It took about two hours of working with her to iterate on the tool and find the right balance of features and ease of use. Step-by-step: 1. I observed the two tools she was using to tag, crop, view, and inspect bird photographs. 2. I used Cursor, Grok, and Claude Opus to vibe-code a C# application for her workflow. 3. I made the tool read each photo’s EXIF and MakerNotes metadata. 4. I added fullscreen mouse scrolling, a histogram in the top-right, and camera information in the top-left, including aperture, shutter speed, focal length, and ISO. 5. I added red-to-green color coding and a display of the Nikon focus zone. 6. I set up left-click zoom to 75% with panning and right-click zoom to 100%. 7. I added a keyboard shortcut that crops the image at the current view, saves a copy, tags the photo, and adds her copyright in the bottom-right corner. 8. I iterated on the tool with her for about two hours until the features and ease of use were balanced for her needs.

I set up an AI assistant to run my entire writing business on autopilot. Every morning, it pulls RSS feeds from more than 30 AI and writing sources, deduplicates them against the previous day’s digest, curates the top items, and sends me a single Telegram message with numbered, linked items before I wake up. At 11 a.m. each day, it generates an original writing craft post. The topic comes from a rotation pool of more than 15 categories, and the assistant avoids anything used in the last 30 days. It also creates accompanying artwork in a rotating fine-art style, then cross-posts the content to Facebook, X, and my blog, including the featured-image upload to WordPress. Each week, it compiles and sends an email newsletter to my subscriber list through Brevo. It pulls from a curated candidates file that I approve before the newsletter goes out. Behind the scenes, the assistant manages a fleet of five servers, including servers for my wife, daughter, and two business colleagues. It handles daily backups, monitors costs across providers, and reminds me when context windows are becoming expensive. The key insight wasn’t the automation; it was the partnership model. My assistant has a persona file (`SOUL.md`) that defines how it communicates, a memory file (`MEMORY.md`) with everything it needs to know about my life and business, and a playbook of behavioral rules built from real mistakes over time. It pushes back on bad ideas, flags risks before executing, and has genuine opinions about craft and content. That shift—from “tool you talk to” to “colleague who has your back”—is what I wrote my book about. *Harnessing the Machine* is the field guide I wish I’d had when I started. It isn’t a tutorial, because the technology changes weekly; it’s a guide to building a working relationship with something that remembers yesterday. The tech stack is OpenClaw, GLM-5.2 as the primary model, DeepSeek V4 Pro as the fallback, and AWS Lightsail. The total monthly cost is under $30. The real cost was calibrating the assistant: teaching it what I care about, what “good” looks like, and when to ask versus when to act. That’s the part most people skip, and it’s why most “AI automation” posts feel like demos rather than relationships. Tools used: OpenClaw, GLM-5.2, DeepSeek, Telegram, WordPress, Brevo Step-by-step: 1. I configured OpenClaw with a persona file (`SOUL.md`), a memory file (`MEMORY.md`), and a behavioral playbook built from real mistakes. 2. I connected it to RSS feeds from more than 30 AI and writing sources and had it deduplicate, curate, and send a numbered Telegram digest each morning. 3. I created a rotation pool of more than 15 writing categories and instructed it to avoid topics used in the previous 30 days. 4. I scheduled it to generate a daily writing craft post, create artwork in a rotating fine-art style, and cross-post the result to Facebook, X, and my WordPress blog with a featured image. 5. I set up a weekly Brevo newsletter that pulls from a curated candidates file I approve before sending. 6. I connected the assistant to five servers, including servers for my wife, daughter, and two business colleagues, and had it manage daily backups, provider costs, and expensive context windows. 7. I configured GLM-5.2 as the primary model, DeepSeek V4 Pro as the fallback, and AWS Lightsail as the hosting environment. 8. I calibrated the assistant by teaching it my standards, what “good” looks like, and when to ask for approval versus acting on its own.

It starts when someone clicks “Find My Best AI Opportunity” on my website. Instead of going straight to a booking page, they enter a short AI chat. The assistant asks about their work, business, main pain point, AI experience, urgency, name, and email. The workflow runs in n8n. Once the chat has enough information, it creates an AI Readiness Summary, saves the lead in Notion, sends me an internal brief, and emails the visitor their summary with a link to book a 30-minute call through Cal.com. The result is a better-qualified call: the visitor gets useful value first, and I have the context I need before we meet. Step-by-step: 1. A visitor clicks “Find My Best AI Opportunity” on my website. 2. The visitor completes a short AI chat about their work, business, main pain point, AI experience, urgency, name, and email. 3. n8n uses the collected information to create an AI Readiness Summary. 4. The workflow saves the lead in Notion and sends me an internal brief. 5. The visitor receives their summary by email, along with a link to book a 30-minute call through Cal.com. 6. I review the context before the call, making it better qualified.

I built an autonomous, end-to-end AI Sales Development Representative (SDR) engine entirely in n8n. On a scheduled trigger, the agent calculates targeting parameters, reads long-term CRM memory to avoid duplicate outreach, searches for and qualifies prospective leads, scrapes company websites for buying signals, drafts tailored outreach emails, writes structured relational data to PostgreSQL, and reports execution summaries through Telegram—with zero manual intervention. The system is currently deployed in production for a B2B agricultural export business, generating qualified international wholesale leads on a recurring schedule. Most AI automations rely on simple linear scripts or break down when handling complex agentic tool workflows. This system addresses three common failure points: - High API costs: Re-sending large system prompts and tool schemas on every agent iteration drains tokens. - Context blindness: Agents without memory of previous contacts can send duplicate outreach. - Database crashes: Agents may hallucinate ENUM values or fail to insert nested one-to-many arrays into relational tables. The workflow uses a Cloudflare-proxied Claude Sonnet 4.6 model with prompt caching, persistent CRM memory reads, and a fault-tolerant parallel database-write architecture. The stack includes n8n as the orchestrator; Claude Sonnet 4.6 through a Cloudflare Worker proxy as the LLM core with ephemeral prompt caching; PostgreSQL for CRM contacts, intelligence, and outreach tables with custom ENUMs; SerpAPI for prospect discovery; Firecrawl for website content extraction; and Telegram for execution reporting. The workflow exposes these tools to the n8n agent: - `read_relationship_memory`: Read-only SQL access to historical contact and outreach data, preventing duplicate prospecting. - `Lead_Finder`: Searches for and identifies target prospects by country and sector. - `Scrape_Website_Content`: Extracts website content, buyer-intent signals, and objections from discovered domains. - `write_relationship_memory`: Writes leads, intelligence facts, and drafted emails to Postgres in one resilient call. Step-by-step: 1. A Schedule Trigger feeds a JavaScript “Country Calculator” node that resolves the day’s targeting parameters—region and industry focus—using ISO week rotation. This cycles outreach across markets automatically. 2. The AI Agent connects to an OpenAI Chat Model node whose Base URL points to a custom Cloudflare Worker. The worker translates OpenAI-formatted requests into Anthropic’s Messages API, enabling Claude Sonnet 4.6 while injecting ephemeral cache-control headers into the system prompt and tool definitions to reduce repeat-token costs. 3. Before researching, the agent calls `read_relationship_memory` to check relationship status and outreach history, preventing duplicate contact attempts. 4. `Lead_Finder` searches target sectors in the day’s region and returns seven filtered candidates. `Scrape_Website_Content` then visits each domain, extracts clean page text, and surfaces offerings, value propositions, and likely objections. 5. The workflow writes nested one-to-many data—multiple facts and one outreach log per contact—without item duplication or ENUM crashes. The tool schema requires a strict JSON array with exact ENUM string choices spelled out in the description. 6. A sub-workflow triggered by “When Executed by Another Workflow” splits the array, then flattens nested `contact.*` fields to root keys using JavaScript. 7. An upsert query, `ON CONFLICT (email) DO UPDATE`, writes the contact, increments `email_count` for repeats, and returns `contact_id`. 8. A “Re-attach Context” node merges `contact_id` back with the original intelligence array and outreach payload because n8n strips extra data through single-row database nodes. 9. Two parallel branches run: one inserts the outreach log with `ON CONFLICT DO NOTHING`, while the other splits and inserts each intelligence fact with defensive ENUM sanitization. This eliminates crashes and duplicate rows during retries. 10. The agent’s final output triggers a Telegram message summarizing the discovered leads, extracted facts, and drafted emails, sent directly to the operator’s phone. The result is a production-grade, self-healing AI outbound pipeline running with zero manual intervention. It maintains CRM data integrity, avoids duplicate outreach, and uses prompt caching to keep LLM costs low at scale.

I recently built a month-end close workflow in Awish for a client at a finance company. The problem was not creating the final report. The real bottleneck was collecting data from different systems, checking what was missing, reconciling totals, chasing exceptions, and getting the report to the right people. I built the entire process in Awish by describing what I wanted. Step-by-step: 1. I opened the Awish chat and wrote: “At every month-end close, collect journal entries, invoices, vendor bills, and financial records from NetSuite together with reporting workbooks from Excel and SharePoint. Check submission completeness, reconcile totals across sources, identify missing data or unusual variances, prepare the management-reporting workbook, send unresolved exceptions to Finance in Microsoft Teams for approval, and once approved export the final report to PDF, store it in SharePoint, and distribute it through Outlook.” 2. Awish understood the request, planned the workflow, and selected NetSuite, Excel, SharePoint, Microsoft Teams, and Outlook for the required steps. 3. I reviewed the plan, connected the client’s accounts, and approved the automation. 4. At month-end, Awish pulls the required financial data and reporting files, checks whether anything is missing, reconciles totals, and flags unusual variances. 5. It updates the management-reporting workbook and sends only the unresolved exceptions to the Finance team in Microsoft Teams. 6. Once Finance approves the exceptions, Awish finalizes the report, exports it to PDF, stores it in SharePoint, and sends it to the authorized recipients through Outlook. The useful part is that Finance no longer has to spend most of the close manually collecting and checking information before making a decision. The repetitive reconciliation work is handled automatically, while the team retains control over unexplained exceptions and the final report. Trigger → Analyze → Approval → Action Month-end close → Reconciliation \u0026 variance checks → Finance approval → Final report \u0026 distribution

I turned an open-source XML lookup fix into a debugging walkthrough that other developers can run. I used Codex, Python, and GitHub to document a bug where a document style containing quotation marks could be saved but then fail during retrieval because its name was interpolated into an XPath expression. Step-by-step: 1. I gave Codex the original patch and inspected the affected code, tracing the stored value through the public API, the wrapper, and the XML library. 2. I extracted the smallest standalone reproduction with `lxml`. It included a name containing both single and double quotes and kept the failing lookup visible. 3. I replaced interpolation with a bound variable: `styles.xpath("style[@name=$name]", name=name)`. I checked that the wrapper forwarded variables while preserving its namespace mapping. 4. I ran both versions. The interpolated version raised `XPathEvalError` in my reproduction. Binding the value matched five exact names, while an absent name returned no match. 5. I published the explanation, executable example, and upstream patch together. I inspected the rendered article and links and disclosed AI assistance. A prompt to reuse: "Reduce this lookup failure to a runnable example. Keep the failing case, show the fix, and check ordinary text, both quote types, custom namespaces and a missing value. Report which checks actually ran." The original fix had already been merged, so this workflow documents it. Broader service behavior needs separate tests. I used my existing Codex and local Python setup and did not measure time savings. I am Hồ Khắc Huy, a freelance software engineer. This is my independent open-source work. The linked article includes the runnable example, upstream contribution, and my contact details: https://github.com/builtbyhuy/builtbyhuy/blob/main/notes/2026-09-06-xpath-variables.md