The AI Context Loop

In one line: Docs are context for the AI, not just humans — better docs produce more aligned output, which produces more docs, compounding over time.

The most significant benefit of living documentation is the compounding knowledge flywheel, not human communication.

The loop operates as follows:

1. Documentation provides context — the AI reads CLAUDE.md, memory, and linked docs at session start, building a mental model of the architecture, constraints, and design philosophy.
2. Context produces alignment — with the why, the AI generates code that fits existing patterns on the first attempt instead of after several corrections.
3. Implementation generates documentation — the commit pattern (Section 11.4) ensures every significant change produces an updated page, diagram, or ADR.
4. Documentation enriches context — the new docs are available to future sessions, so each implementation benefits from all prior ones.

Each cycle adds context depth. Over a long project with this pattern active, the documentation corpus gives the AI a level of understanding that would take a new human developer weeks to acquire by reading code alone. The docs are the primary mechanism by which the AI's effectiveness compounds.

The practical implication: documentation quality directly affects velocity. An hour writing a clear page is not overhead — it is an investment in every future session that reads it. This inverts the usual cost-benefit: instead of a cost that pays off only when a human reads it (rarely), it pays off every time the AI starts a session (frequently).

---