Ryan “RyMac” McKinney · USMC · 21 August 2026 · 4 minute read
Interpretable Context Methodology is 3 serious words for a filing cabinet. You put plain text files in ordinary folders, arranged so an AI can open one, read where it is, and know what to do next. That is the idea. Everything else is detail.
Interpretable. You can open it and read it. There is no database, no hidden setting, no dashboard that knows something you do not. If your AI does something strange, you open the file and see the sentence that told it to.
Context. Everything the AI can see at the moment it does a job. Not what you typed. What is in front of it. That distinction is the entire ballgame, and I wrote it out longer in context engineering vs prompt engineering.
Methodology. A repeatable order. Same 2 files, same shape, every project, so you stop reinventing it and the machine stops guessing.
Jake Van Clief and David McDermott wrote the paper. Jake calls the setup the ICM Folder Structure and teaches it himself over at Clief Notes. I did not invent any of this. I pay for his group every month, which is why I answered the accusation in is Jake Van Clief legit instead of staying quiet.
If you want the research version, read the paper. If you want it without the research language, I sat down and did that in the ICM paper in plain English.
Most people arrive here after buying something else. A multi agent course. A memory tool. Something with a canvas and boxes and arrows between them. You may hear that middle layer called orchestration, which simply means the part that decides who does what next.
All of that exists to move information from where it is to where the AI can see it. A folder already does that. It has done it since 1984. You are not missing software. You are missing an arrangement.
I said the same thing on camera the day I stopped building agents, and the video is here if you would rather watch than read.
2 files and some folders. That is not a simplification for the article. That is the actual thing.
~/business/ ├── CLAUDE.md <- the map. read first, always ├── Context.md <- the desk. what we are doing today └── ops/ ├── clients/ ├── offers/ ├── sops/ └── archive/
The map does no work. It says one thing: before you touch anything, go read the desk file. The desk file says what is happening right now. The folders hold the material. Every folder is named for what a person would call it, and I walked through every one in the ICM folder structure.
You can buy a tool that remembers things for you. When it remembers wrong, you cannot see why, because the remembering happens somewhere you do not have access to. I wrote about that trap in AI agent memory.
With a file, you open it. The reason is a sentence. You change the sentence.
That is not a small convenience. It is the difference between running your business and renting a black box that runs it for you.
Somebody running a business that now has an AI in it, who is tired of explaining the business again every single morning.
You do not need to be technical. I am not. I am a Marine who could not write a line of code and still cannot, and I explained how that works in ICM for people who cannot write code.
About 20 minutes to set up. About a week of real work before you trust it.
You can build the empty version yourself tonight. A folder, 2 text files, 4 folders underneath. Nothing is stopping you and I would rather you did that than paid somebody $5,000 for it.
What I send instead is a project already underway, so you learn it on real work rather than on a pretend folder. If you want to see it running before you touch anything, I let it audit itself live on camera across 2 businesses with no script.
The kit is 6 files. It takes 20 minutes. A start page. A map. A desk file. The folder skeleton. A setup sheet. A list of what to cancel. No card. No call. No course to buy.
Send me the folder system