AI tools are moving fast. ICM is a way to build your AI work as folders and markdown — portable across the next agent harness, readable to humans, runnable without writing code.
AI tools are changing every few months. The framework you wired your work into last quarter might be the one teams are ripping out this quarter. The model that just shipped does what your custom plugin used to do. The ground keeps shifting.
ICM — Interpreted Context Methodology — is a way to organize AI work that doesn't depend on any particular framework or model. You write your workflow as a folder of plain-English markdown files. An AI agent reads the folder, follows the structure, and runs the work. When the next harness ships, your folder still works.
This series is five parts: why frameworks keep breaking, how the layered approach works, where ICM fits in the broader stack, what real workspaces look like, and where the methodology is heading next.
Read in order, or jump to any part.
AI orchestration is in flux. Don't build your house on sand.
More context doesn't always make AI smarter. Here's the layered approach ICM uses to keep agents lean.
Chat, skills, ICM, or framework: the principle that helps you decide which fits.
Two real workspaces, walked end to end: what they look like, how they run, and the open convention that keeps them portable.
What ICM doesn't yet handle, and the direction worth moving toward: durability, observability, restartability.
A workspace is just a folder. The question is which agent reads it.
Mac/Windows desktop app. Point it at a folder, it reads your CLAUDE.md, runs the workflow, and asks for approval at checkpoints. Designed for non-developers — no terminal, no code.
MIT-licensed open-source alternative if you'd rather not be locked into Anthropic. Runs locally. Note: I haven't personally tested it but the no-lock-in posture is right.
An ICM workspace can include an AGENTS.md alongside its CLAUDE.md so it also runs on Cursor, Codex CLI, Windsurf, GitHub Copilot, and Devin. Those are developer tools — but the point is portability. When a new non-tech harness ships, your folder is ready.
A note on the rest: ChatGPT Workspace Agents and Microsoft Copilot are powerful but connect to cloud apps rather than local folders — they're a different shape of tool.
Two skills to bootstrap your first workspace, free to download.
icm-workspace-builder Interview-driven scaffolding for a brand-new workspace. Walks you through eight questions about your workflow, then generates the folder structure with a CLAUDE.md and per-stage CONTEXT.md files.
skill-to-icm-converterFor when you have an existing Claude skill that's grown to 3+ steps and is starting to tangle. Reads the skill, identifies the implicit step boundaries, and proposes a workspace structure that preserves what the skill already does.
To use either: drop the .md file into your Cowork skills folder, then ask Cowork "build an ICM workspace for [topic]" or "convert this skill to an ICM workspace."
Interpreted Context Methodology was created by Jake Van Clief (RinDig on GitHub), who developed and documented the approach across his repository, a 2026 paper, and a 19-video YouTube series. This series synthesizes Jake's methodology for a non-technical audience and extends it with a forward-looking chapter on durable execution.
Where to find Jake's original work:
For the full attribution and what's mine vs. what's Jake's, see the About this series page.