Unwritten Process
- IDENTIFIED
- OCTOBER 13, 2026
A team's engineering judgment lives in heads, calls, and onboarding rather than in the repository. Humans absorbed it by proximity; AI coding agents inherit only what is written down, so every unwritten rule is bypassed on every commit.
The Problem
Teams blame agents for taking shortcuts, but the agent configuration is full of API maps and naming conventions and contains not one line about how the team reasons. The process was never written down. For twenty years that was a hidden tax paid slowly through onboarding and repeated mistakes. Agents turned it into a multiplier: whatever is not written down, they will not know, and they will execute at machine speed whatever is.
Root Causes
Process lives in people, not in the repository: The rules that came out of years of incidents exist only in the heads of the people who learned them, so no machine and no new hire can inherit them without years of the same tuition.
Proximity stopped scaling: Unwritten wisdom degraded gracefully when teams learned it by sitting near each other. Distributed teams and autonomous agents both break that assumption, and both pay for it in repeated failures.
Postmortems end at the memo: Lessons are documented for humans who may or may not read them, never placed in the artifact the executing system reads on every commit, so each incident teaches the team but not the agents.