I got my first client a fellow dad form pre-school who runs a 20 man Industry Climbing company. They are very efficiently run processes and one Saas solution acting as their operations system. This created a tension for me to picture what the first folder could even be to help them out. Then I saw a door of creating a folder system holding the context what the owner needs to do each day and for the week. Basically a secratary, so he doesn't have to spend so much time on the computer. I will also be working on listing their capabilities so that they can make more strategic plans of how to recombine their capabilities in new packages or find other capabilities that are missing for their business development and growth management. While also holding the recursive learning loop which was always a part of the agile movement at the core of their values.

🎁 $1,000 IN EDUBAWARE CREDITS 🎁 One winner takes it. 📋 THE CHALLENGE Build a folder-based AI translator that takes one kind of work and turns it into another kind of work. Same shape in, same shape out, every time. Not a summarizer. Not a writer. A converter with a contract. This week's deliverable is one translator folder that someone could drop into a Claude project, feed it the input it expects, and get back the output it promises. Every time. Without surprises. 🎯 PICK YOUR CONVERSION The conversion is yours. Pick one you do by hand right now and hate. A few sparks to get you thinking: - 📞 Sales call transcript → CRM notes in your team's exact fields - 📝 Long-form essay → LinkedIn carousel, slide by slide - 🎙️ Meeting recording → product requirements doc - 🔬 Research paper → investor one-pager - 🐛 Bug report thread → Jira ticket with repro steps - 📧 Customer email → support ticket with severity and category - 📊 Spreadsheet export → weekly status update - 📖 Interview transcript → case study draft - 🧾 Receipt photos described in text → expense report line items - 📋 Discovery call notes → SOW first draft The more locked-down the output, the better. "Turns notes into a doc" is not a contract. "Turns discovery call notes into a five-section SOW where section 3 is always scope exclusions" is. 🔥 THE ANGLE THIS WEEK Last comp was The Auditor. Every finding cited a provision so a reader could open the standard and check. The comp before that was The Cartographer. Every card cited a file and a line so a reader could open the source and check. The Translator is the same discipline, one more time. Every line in the output traces to a line in the input. ↔️ A translator has three properties that a summarizer does not: 1. The output has a fixed shape. Same fields, same order, same format, regardless of what the input looked like. If the input was messy, the output is still clean. If the input was short, the output still has every field, marked empty where there was nothing to fill it.

This is a follow-up to my previous post about using ICM for AI-driven machine learning research. In one week of ICM-assisted research, I've moved further than I managed in roughly six months of my master's thesis. This is what that looks like in practice: ICM/ ├── README.md ├── skills/ ├── agents/ │ ├── literature-intake.md │ ├── research-development.md │ ├── code-development.md │ ├── results-evidence.md │ ├── research-argument.md │ ├── thesis argument.md │ ├── self-review.md │ └── ... ├── research/ │ ├── context/ │ ├── literature/ │ ├── development/ │ ├── results/ │ └── reports/ ├── code/ │ ├── src/ │ ├── tests/ │ └── runs/ ├── thesis/ └── _system/ ├── rules/ ├── templates/ └── schema/ I can queue several research goals across different chats and let each one keep moving. The agents are not all doing the same job with different names. Each one has a bounded responsibility, its own context and a clear handoff to the next part of the work. - The literature-intake agent turns papers into usable research context. It extracts the claims, methods, datasets, assumptions and limitations that matter for my problem. It helps answer: what has already been tried, what can actually be reused and what still needs to be tested? - The research-development agent turns vague ideas into explicit questions, hypotheses and experiments. It forces the research to become testable before implementation begins. Instead of “try an LSTM”, the goal becomes something like: under these conditions, does recursive probabilistic prediction outperform a defined baseline? - The code-development agent owns the implementation. It builds the data pipeline, model interfaces and experiment code while respecting the assumptions defined by the research question. Its job is not to decide whether the research is meaningful. Its job is to make the proposed experiment executable and reproducible. - The testing agent checks whether the implementation behaves as intended. It verifies data shapes, transformations, edge cases, saved artifacts and the parts of the pipeline that can be checked mechanically. A passing test gives me confidence in the software. It does not give me scientific confidence in the conclusion.

Every.to, one of my favorite newsletters about all things AI over the last 3 years is starting a new series called Show me your folder. It is interesting have a read. https://every.to/context-window/show-us-your-folders?utm_source=email&utm_medium=post&utm_campaign=4482&utm_content=link_2&utm_term=free_2026-09-16&ev_email_id=post_4482_post_9feb1b2bba72&ev_link_id=post_4482_post_9feb1b2bba72_link_2&ev_audience=free&ev_post_id=4482&ev_email_type=post&ev_send_date=2026-09-16

I found this thread yesterday about a project from Stanford University to change how we interact with research. There is a lrage "problem" in society now that the best research lives the live as a stale pdf behind paywalls from major universities. So Standford researchers created Paper2Agent in order to enable the reader to extract the models and knowledge out research and put it into Github and build an MCP server that allows them to call it and interact with it. This is very similar to what we are doing right now with ICM. I found this really exciting. https://github.com/jmiao24/Paper2Agent