Interpretable Context Methodology (ICM)

ICM replaces complex multi-agent frameworks with a filesystem-based architecture that uses numbered folders and markdown files to orchestrate sequential AI workflows.

Interpretable Context Methodology (ICM) is a systems-first approach to AI orchestration that treats folder structures as agentic architecture. Developed by Jake Van Clief, the methodology eliminates the need for heavy coordination code by decomposing workflows into numbered stages (e.g., 01_research, 02_drafting). Each stage operates under a strict contract defined in a CONTEXT.md file, which specifies inputs, processes, and outputs across a five-layer context hierarchy. By using the filesystem as the state machine and plain text as the universal interface, ICM allows a single LLM to execute complex, multi-step pipelines with high transparency and manual review gates at every handoff.

https://arxiv.org/abs/2603.16021