Workflow Hub

I built this workflow for product teams that have research but no clear path from customer evidence to what to build next. It works one discovery stage at a time, keeps evidence separate from inference, and does not move forward until a person approves each stage. 1. Define the outcome you are trying to reach. 2. Upload your research or answer the guided questions. 3. GPT-5.4 Mini extracts evidence-backed facts, insights, assumptions, and quantified customer opportunities. I use this prompt for the extraction step: "Guide one discovery stage at a time, separate evidence from inference, preserve supported numbers, never invent findings, and require human approval." 4. Review the extraction and approve it. Nothing continues without this approval. 5. After approval, the workflow scores the opportunities, proposes solutions, and designs lean experiments. 6. Approve the results, then it generates the final Opportunity Solution Tree and a decision report. The app is built with OpenAI Sites, the Responses API, D1, and R2. What it produces is a traceable map from customer evidence to outcome, solutions, and experiments, with the numbers from the source material kept intact so the team can check every claim against what customers actually said. https://product-discovery-master.vspirola76.chatgpt.site/

I had a ChatGPT personalisation setup that worked well until product changes made it less reliable. I kept re-explaining how I work and what good alignment feels like, so I rebuilt it as a plugin. I think of it as a backpack ChatGPT carries: a skill is the operating guide, reference files are the notes and maps, and connected tools are the gear. The plugin loads personal context and collaboration preferences, not a persona. How it runs: 1. In normal mode, the plugin quietly loads my current alignment and context bundle, so ordinary chats stay ordinary. 2. I trigger it with words I already use when talking to ChatGPT. Nicknames like "Goldfish," "Fish," "Blub," or "Blubsie" act as recognition signals, so I don't have to type setup commands. 3. Repair mode runs only when I explicitly ask for re-alignment. It reads the current bundle, identifies what went wrong, corrects course, and can propose a small update. 4. If the context fails to load, ChatGPT is set to say: "Error: Blub not loading. Fish is non-active." It is silly on purpose, but I immediately know the problem is missing context and not a personality shift. 5. Specialised skills stay separate, such as image generation, writing, research, or archiving. I combine them with the context layer when needed. It saves me from re-explaining my working style at the start of every session, and it makes missing context visible instead of silent.

I trade gold futures (GC and MGC) during the New York session. I wrote this workflow to keep my process consistent before I scale to multiple accounts. It is for traders who want one market, one playbook, one risk framework and one journal. I analyze in TradingView and execute in Tradovate. 1. From 60 to 45 minutes before the session, I review the economic calendar, USD, Treasury yields and gold movement. 2. From 45 to 15 minutes before, I analyze D1/H4 and H1/M15 and mark key liquidity levels: previous-day high and low, Asia and London highs and lows, and major support and resistance. 3. I set my bias (bullish, bearish or neutral) and my daily risk limits. 4. At 9:30 AM ET I observe the US cash-market open. 5. From 9:35 to 11:00 AM ET I execute only confirmed setups, either Setup A (liquidity sweep reversal) or Setup B (trend continuation). I skip trades when structure is unclear, price is sideways, there is no entry confirmation, risk cannot be calculated, or trading restrictions apply. 6. I size each position with this formula, always rounded down: Contracts = Dollar risk budget ÷ (Stop ticks × Tick value + cost allowance). I risk 0.25% of my defined risk budget or less per trade. 7. After the session, I journal every trade, review P&L and evaluate performance. My risk limits: I stop after two planned losses or the firm's lower limit, take a maximum of 3 attempts per session, hold one gold trade idea at a time, and aim for at least 2:1 planned R:R, subject to testing. After any rule violation I stop and review. I never widen stops or increase size to recover losses. What it produces: a repeatable routine, a journal of 30 to 50 trades before I expand, and weekly measures of net profit after fees, win rate, expectancy, maximum drawdown and rule adherence.

Ask one AI a hard question and you get a confident answer. I wanted a careful one, so I built a panel. Models from OpenAI, Anthropic, Google, xAI and DeepSeek research the same question, argue it out over several rounds, and check each other's sources and sums. A sixth model, Kimi, takes no side and writes the record. I use it where one confident answer is a risk, like checking a scary AI headline against the public record or stress-testing a plan before anyone spends money on it. They are allowed to make their own code changes, so they have the ability to evolve. 1. I write a brief: the question, the sources to read first, and the shape of answer I want. 2. Each model searches with its own company's search engine, so they don't all find the same five links. 3. Each writes a private first answer without seeing the others, so nobody anchors on the first confident voice. 4. Open rounds: they challenge each other by name, and a second model checks every sum. A majority can close the debate; one holdout can delay it twice. 5. Kimi records what they agree on. Disagreements go in their own section, never in the conclusion. Software then opens every link and flags any that fail or never came from the research. 6. Each model gets one objection to the record, appended word for word, and a plain-English summary goes on top. 7. They propose new rules for themselves. I approve or reject each generation and rate whether the next debates got better. A debate costs a few US dollars and about half an hour, and every one is published, mistakes included: https://quietlyincharge.ai/?utm_source=rundown&utm_campaign=workflow-hub

Last year my daughter went to camp and wanted daily updates on her favorite TV show, which ran every night. The camp only accepts letters, and I did not want to handwrite them every night, so I built a workflow to send them for me. It works for anyone who needs to send physical mail on a regular schedule. Through a website, I used Make.com to build the workflow: 1. It starts at a specific time each day. 2. It asks Claude for a recap of the previous night's episode. 3. The recap is fed into the Make.com workflow. 4. Make.com connects to Lob through an API. 5. I submitted through Lob to create the letter. Every late morning I see the new letter go out, and my daughter gets the episode recap a few days later. Recently I used a very similar workflow to build websites that send letters for people who want to cancel gym memberships or send landlords an email to end a rental. Both are situations where only physical letters are accepted. Connecting these tools took a little time, but it opened up a lot of possibilities, as long as someone still needs a physical letter.

I used Claude to help me create my nonprofit's website, nomorecopays.org, in a day and a half. It walked me through the full process of setting up a nonprofit, helped with my legal paperwork, and guided me to the other tools I would need to run and grow the organization. This is for founders who are starting a nonprofit and want one place to plan the steps. 1. I asked Claude to walk me through the entire process of creating a nonprofit. 2. I had Claude create my website. 3. I had Claude help me edit my mission statement and review my attorney's drafts of the bylaws and other legal documents before I submitted them to incorporate. 4. I asked Claude to guide me on the other companies I would need to function and grow, including Namecheap, Google for Nonprofits, Google Workspace, Substack, and Zeffy. 5. I had Claude design my business cards so they share the look of my website. 6. I had Claude review the links and references for my Substack articles. Result: my nonprofit's website was created in a day and a half, with Claude's help across the legal, business card, and newsletter steps.

I'm a psychotherapist in New Zealand. My handouts were static and needed regular updating, so I wanted simple, slightly interactive websites that clients could engage with, without anything massive. I have no coding experience beyond programming a Commodore 64 in BASIC, so I asked Claude to build them. The sites needed to be static, and any data a client enters needed to stay private. 1. I gave Claude a simple design prompt for each tool, starting with the sites I needed. 2. First was an IFS (internal family systems) site, laid out as a house, where a client can map out their internal world and connect the key elements to understand the whole: https://ifs.padma.nz 3. Next was a second house offering a walk through basic tools for emotional regulation. It has no data entry, but includes downloadable information: https://kind.padma.nz 4. Then came a life narrative site using leaves on a tree. Events are logged on a branch with a common theme (for example, school), and photos and music relevant to a life event can be attached: https://leaves.padma.nz 5. Finally, because I use Jung's Active Imagination in my work as outlined by Robert Johnson, I built a site that gives instructions and maps out a pathway for this approach: https://clearing.padma.nz What I end up with is four simple, static sites that replace my out-of-date handouts, with the aim of keeping any client-entered data private.

I run a photography business and built a three-layer AI workflow to help with daily operations, business improvement, and watching for outside changes that might affect us. It's for small business owners who don't want to monitor every source or handle every issue themselves. The layers are an Operations Core (dashboard, estimates, schedules, reminders, project records, attribution, and other business state), a Work orchestrator with specialist AI agents, and a signal and triage system that watches platform alerts, vendor updates, local-search changes, newsletters, calendars, and marketing sources. Here is how a signal moves through it: 1. The triage layer picks up a signal from one of the sources it watches. 2. Work decides whether the signal actually matters and routes it to the right specialist. 3. The specialist reviews it. I use separate specialists for WordPress, Google Business Profile and local search, marketing and STR signals, AI systems, research, and technical implementation. The WordPress and Google Business Profile specialists are also connected directly to those systems through MCP/API connectors, so they can compare the incoming signal against the current state of the actual site or profile before deciding whether it is relevant or whether any action is needed. 4. If the issue needs deeper evidence, the research specialist investigates before anything moves forward. 5. Work reconciles the findings into a recommendation. 6. I approve or decline the recommendation. The system brings me in only when a real decision is needed. 7. The change is implemented. 8. The Operations Core is updated with the new business state. The persistent Business Desk files hold current priorities, rules, operating guidance, and the source of truth for business state, so each step works from the same context. The main benefit for me is that I don't have to personally track every source or jump into every issue.

I’m Steven Alber, founder of ARENOMICS Institute (https://arenomics.institute), an independent research initiative exploring how autonomous AI agents can operate within accountable economic and institutional systems. One of our central research questions is simple: Who gives an AI agent the authority to act? I use ChatGPT and Claude as research and development tools to investigate this question through a structured workflow. 1. I identify an institutional problem, such as how an autonomous agent receives permission to make decisions or execute actions on behalf of a person or organization. 2. I use ChatGPT to examine relevant research, technical standards, and existing approaches to agent identity, authorization, delegation, and governance. 3. I develop conceptual frameworks that distinguish an agent’s technical identity from the authority it receives and the specific mandate under which it operates. 4. I use Claude to help structure the research into technical explanations, institutional models, educational materials, and website content. 5. I compare the models’ outputs, challenge assumptions, examine weaknesses, and refine the proposed architecture before publication. This research direction is part of what I call Mandate Engineering: investigating how human or organizational intentions can be translated into explicit, limited, and accountable authority for autonomous agents. The institute is still developing. My goal is not simply to produce more AI-generated content, but to explore practical and scientifically credible approaches to governing increasingly autonomous intelligence.

AI shopping agents don't browse the way people do, and most online stores aren't built for them. I wanted to know where an agent would get stuck on my store, so I used Storewalk. It scans a store and shows exactly where an AI agent would stop: whether it can find your products, compare them, complete a purchase, or handle what comes after. This is for anyone running an online store who wants to find those gaps before shoppers' agents run into them. 1. Run a Storewalk scan on your store. 2. Review the checks it runs, which show where an agent would stop. 3. Look at your Ready score, out of 100. 4. Work through the fix list. Each fix names who should act and how much effort it takes. 5. Compare your score against up to two competitors. 6. Book the free consultation that comes with every full report at https://storewalk.ai/ What you get: a Ready score out of 100, a fix list with owners and effort levels for each item, and a side-by-side view against up to two competitors.

As an agency, we have to report the work we do for our clients and how efficient it is. Building these presentations by hand is very time consuming and boring, and there is little value in making them. At Le Trefle à 5 Feuilles we decided to automate this recurring task. Before, it took about 3 hours to gather every stats report, make the presentation available to the client, and interpret the stats and recommend further actions. I keep a human in the loop, because someone has to read the report before it is sent. Since we work in Google Workspace, I built the workflow with several Gemini Gems: 1. Google Apps Script watches one folder on Google Drive and detects when a new statistics report is posted. 2. The first Gem decodes the report and inserts the variables into it. 3. The second Gem interprets and sums up all the statistics. 4. The third Gem merges those statistics into the template the workflow calculated. 5. I review the finished presentation before sending it to the client. It now takes only 3 minutes to produce the report, so the consultant can focus on the real value: their analysis of the campaign.

The problem: I’m on the customer side, not the engineering side. When something breaks, I usually find out when a customer writes in, and by then a bunch of people have already hit it. We had an error chart in our analytics tool, but I don't often remember to check that dashboard every day, and a basic “alert me over X errors” setup would go off constantly for things that aren’t real problems. What I built: A Claude scheduled task that checks our error chart every morning, filters out the noise, and tells the whole team in Slack only when something’s actually wrong. I set it up just by describing what I wanted in plain English. How it works: Runs every morning before the team starts and looks at the full previous day. Pulls errors by type straight from our analytics tool. Claude connects to it directly, so it sees the same data as the dashboard. Ignores the noise. I gave Claude the rules on which error types are real problems and which aren’t, and it only counts the ones that matter. Checks a threshold. If the real errors go over the number I picked, it posts. Otherwise nothing happens. Posts one clear alert in a shared Slack channel with the total, which errors spiked, and a link to the chart. Never repeats itself. It checks the channel first so the same day never gets flagged twice. Why it’s worth it: I hear about issues before customers have to tell me. Everyone sees the same alert at the same time, so there’s no relaying messages or chasing people down. The right person just picks it up. Quiet days stay quiet, so when an alert shows up, people pay attention. I tested it against the past couple of weeks. It flagged every day something really broke and stayed silent on a busy day that was all harmless errors.

Subfox Industrial builds custom paperless shop-floor systems for specialist manufacturers that need accuracy, traceability and proof without the paperwork. Instead of a broad software rollout, we start with one workstation and prove it works before we scale. This is for manufacturers who want that kind of control over their floor. Here is how we approach it: 1. We map the real task, the checks and the decisions people make at that station. 2. We configure the workstation around that process. 3. We prove it on the floor. 4. Once it works, we scale to other workstations. Our workstations cover cutting, CNC machining, molding, quality control, maintenance and other process steps. They capture measurements and evidence as the work is done. Where it earns its place, AI adds extra checks and anomaly flags, and the customer keeps control of their data. An IRIS-certified rail parts factory runs on Subfox. The result is that measurements and evidence are captured as the work happens, not reconstructed afterward.

I use this when I review code for pull requests. Claude breaks the code into relevant sections and presents them in execution order instead of the alphabetical file order GitHub shows, which makes it easier to reason through the logic. 1. Have Claude break the PR into relevant sections, ordered by how the code executes. 2. Have Claude export the walkthrough as a markdown version of the PR diff report. 3. Paste the markdown into ChatGPT. 4. Enter ChatGPT voice mode and talk through the report with it. 5. As I read each section of the code, ask ChatGPT for a summary of that section. 6. Ask questions about the code, the logic, or the story's acceptance criteria to get a deeper understanding of what I'm reviewing. The result is a PR review that follows the code's actual flow, with a spoken walkthrough and section summaries I can check against the code as I go.

I've been working with ChatGPT on a personal archive project called Atlas. We developed a specific way of analyzing old conversations, discovering creative ideas, reviewing them one by one, and saving the approved results. The workflow worked well, but every new session meant bringing back the instructions and explaining the process again. So I asked ChatGPT's Create Plugin tool to package our process into a private plugin called Atlas Analysis. Here is how it runs now: 1. I open Atlas Analysis and upload my current project ZIP, the original conversation, and an earlier AI (Gemini) analysis. 2. ChatGPT examines the material and identifies potential discoveries. 3. We review each candidate together. I approve, revise, merge, or reject it. 4. Approved discoveries receive their permanent archive IDs, and the session produces an export for my existing local project. The plugin is a first version and I'm still testing and refining it. My favorite part is that I can package a way of working we already developed together, instead of designing every session around a tool's default behavior. I didn't write any plugin code myself. I explained what I wanted, provided my project materials, and ChatGPT built the first version. The result is a reusable tool that saves me from re-explaining the same process each time.

I used Claude Code to build a scenario-testing financial model for ANT, based on Anthropic's S-1 filing. It is for anyone who wants to see how different assumptions change the projections, rather than reading a static spreadsheet. 1. I pointed Claude Code at Anthropic's S-1 filing and had it extract the historical financials. 2. From that data, it built a complete set of income statements, balance sheets and cash flow statements. 3. I then had Claude Code build an HTML front-end for the model. 4. On the front-end, I added sliders to control the key variables, along with graphical displays that show the projections under each scenario. 5. I deployed the system on a website and wrote up the process on my Substack: https://onabetternote.substack.com/p/build-your-own-ant-financial-model The result is a three-statement model with an interactive interface where I can dial in different scenarios and see the projections change.

I'm Steven Alber, an independent researcher building Quantum Civilization (https://quantumcivilization.org). It's a research platform exploring how quantum computing, AI, autonomous agents, and emerging technologies could transform the future of human civilization. I use ChatGPT and Claude to do the work, and this workflow is for anyone who wants to run a research project with AI models and no traditional team. 1. I identify important developments in quantum computing, artificial intelligence, autonomous agents, and emerging technologies. 2. I use ChatGPT to research scientific publications, analyze the evidence, challenge my assumptions, and explore possible implications for civilization. 3. I develop research questions and conceptual frameworks that connect quantum technologies with AI, economics, and the future of society. 4. I use Claude to help turn the research into structured publications, website code, and interactive digital experiences. 5. I compare the outputs from both models, review the results, correct weaknesses, and publish the work on Quantum Civilization. The goal is to explore whether one independent researcher, working with advanced AI models, can develop a credible international research platform. The process is experimental and still evolving. I direct the research and make the final decisions.

I connected ChatGPT to Metricool to get more personalized analytics on how our social media is performing. This workflow is for anyone who manages social channels and wants to see which posts work best on each platform before writing the next batch. Step-by-step: 1. I connected ChatGPT to my Metricool account. 2. I used ChatGPT to analyze social media performance, including which posts were performing best on each platform. 3. I used those results to develop better-performing posts for each platform. 4. Once ChatGPT was trained on my style and content calendar, I automated it to create and schedule posts at optimal times.

I'm an immigration attorney, and I built this workflow so our team can fill in the template emails we use at each stage of a case without copying a template or hunting for client details. We call up a template by number or case stage, answer a few questions, and Gemini drafts the email in Gmail, ready to send to the client. Step-by-step: 1. I uploaded all of our template emails into a Gemini notebook project inside our Enterprise Gemini account. 2. I connected Gemini to Gmail. 3. I instructed the Gemini project to ask for the details that need to be filled into the templates. 4. The user calls up a template by number or its corresponding case stage. 5. Gemini asks for missing details, such as the client’s name or other information, and tells the user what needs to be attached. 6. Once the user provides all the necessary details, Gemini drafts the email and usually finds the recipient’s email address on its own. 7. Gemini prepares the draft in Gmail for the user to send to the client. The result is a finished client email draft generated from a template number and the missing details, so we don't have to fill in the template by hand.

My basement radon fan was pulling outside air down an outdoor stairwell floor drain, which weakened the suction under the slab. I didn't know what part would fix it, so I worked through the problem with my AI assistant, Grok Bot. At first, I asked for "a drain plug that lets water through but keeps air out," and it suggested general waterless trap seals. Once I explained that the drain runs under the slab to the radon fan, it changed its answer to a part made for radon systems: a retrofit floor drain seal with a weighted valve that opens for water and stays shut against the fan's suction. I sent photos of the drain with a tape measure in the shot. Grok Bot asked about the pipe size, the bowl width, and how far the pipe drops straight down, then confirmed that the seal would fit my 2-inch pipe. It compared three suppliers and helped me complete the checkout. We passed control of the screen back and forth to fill in the forms, and it showed me the full total before anything was charged. Apple Pay didn't work from its computer, so we switched to PayPal. I signed in and approved the payment myself. The part cost about $30, or under $50 with shipping and tax, and it arrived within a week. Afterward, I had Grok Bot turn the whole exchange into a slide deck containing the photos, its questions, my prompts, and the checkout back-and-forth. I used that deck to train my facilities leadership team. My point was that you don't need to be a tech person to do this: take a picture, explain the problem the way you would to a coworker, answer the assistant's questions, and stay in charge of anything that costs money. Step-by-step: 1. Describe the problem in plain language, including why it matters—in this case, the radon fan pulling air through the drain. 2. Send photos with a tape measure in the shot for sizing. 3. Answer the assistant's follow-up questions instead of guessing at part numbers. 4. Ask it to compare suppliers with links and prices. 5. Let it fill in the checkout, but review the total and approve the payment yourself. 6. Have it turn the conversation into a slide deck to train others. My first description was too vague and produced a generic answer, so I learned to provide the full context up front. Payment methods that depend on your phone, such as Apple Pay, may not work when the assistant is checking out on its own computer. Since this drain is outdoors, check the valve after hard freezes and leaf fall.