What ICM actually is
Folder structure as agent architecture. That is the 1 line version. Here is the version you can use.
The idea
Instead of building a system to coordinate your AI, you arrange folders. Numbered ones. Each folder holds 1 job and the context for that job. The agent reads the folder it is standing in, and that is what tells it what to do.
No framework. No orchestration tool. No code holding it together. The folders are the structure, and plain text files are how everything talks.
Why it works: an AI does not browse your files looking for what matters. It opens what you point it at. So the arrangement of your folders is not housekeeping. It is the instruction.
What it looks like
01_research/ what you learned, and about who 02_writing/ how you write, and what you sell 03_publishing/ where it goes, and the checks before it does
The numbers are the order the work happens in. Research comes before writing, so research is 01. When you need a 4th thing, you add 04 and nothing else has to move.
Why the old way stopped working
Most note systems die the same way. You set one up, capture everything for a few weeks, and then the pile gets messy. Not untidy: unfindable. You start searching instead of opening, search gives you a screen full of results, and you read none of them. Then you stop trusting it, so you stop opening it.
That happens because everything touched everything. 3 folders with 3 jobs cannot pile up the same way. Storage gets fuller. A brain gets bigger.
Who made this
ICM was created by Jake van Clief. I did not invent it. I use it, I teach it, and I build inside it. If you want the source material rather than my explanation of it:
Where to start
Do not read more about it. Make 3 folders and point your agent at 1 of them.