ICM
ICM stands for Interpretable Context Methodology. It's the folder structure I use to give an AI agent everything it needs to work inside a business without getting confused or making things up.
Most businesses that try to use AI end up dumping everything into one chat and hoping for the best. That works for a question. It falls apart for real, ongoing work.
ICM is the alternative: organize a business, a project, or a body of knowledge into a folder structure that an AI agent can walk on its own, the same way a new hire would learn where things live and what matters.
The deployments below are real examples: pitch cards for what an ICM-organized AI agent looks like in different industries, and a few of the systems I've actually built this way.
ICM is a framework developed by Jake Van Clief, not something I invented. eduba.io
What a build actually consists of
The harness
A coding agent, Claude Code, Codex, or something similar, that actually walks the folders and does the work. Not a chatbot. Something that reads, writes, and acts inside the files.
The folders
The ICM structure itself, organized so the agent always knows where something lives and what it means, the same way a new hire would learn the business.
The connections
Wherever the business's real information already sits: email, SharePoint, Drive, a CRM, whatever's actually in use. These matter just as much as the harness and the folders, but they vary by implementation, every business keeps its information somewhere different.
Deployments