Documentation-as-runtime for AI coding agents

The problem

AI coding agents hallucinate and drift when they lack a precise model of the system they're changing. Pointing an agent at a large codebase and trusting the result usually doesn't work — the specification gaps are where it goes wrong.

The approach

• ▸Structured the domain model, system invariants, ERD, and API contracts as machine-readable context, loaded at the start of every agent session.
• ▸Built a multi-agent governance layer: intent-driven directory hierarchies with trust levels, and AGENTS.md routing contracts that form deterministic context chains.
• ▸Cross-linked the documentation so agents navigate it autonomously and load only the context relevant to the task at hand.

The outcome

• ✓Agents reason about the platform's architecture on their own, instead of guessing.
• ✓The specification gaps that cause hallucinations are closed at the source.
• ✓Production-grade code ships from agent sessions, with humans reviewing rather than rewriting.

Stack