Hard rules
Guardrails that make the safe choice the easy default — and flag risky operations before they happen.
What this layer is
The fourth layer is governance. Rules that cannot be talked around, written in language the AI cannot misinterpret, scoped to the operations where small mistakes have large consequences.
Most of the rules are short. Most of them are about not doing things. Together they create a system that is much harder to break than it would otherwise be — and a system that is honest with the engineer when it does not know what to do next.
These rules are what make autonomy safe. The Epic Loop runs at a graded level of independence — from human-gated review all the way to auto-merge — but a hard denylist means protected or client-owned code is never merged automatically, whatever the autonomy setting. Every decision the controller and the maker take is deterministic, logged and auditable, so autonomy never means opacity.
Failure mode it prevents
Well-intentioned changes to the parts of the system that least tolerate them.
A focused walk through the methodology.
A working session covering all five layers and how they would map to the systems you actually have to modernise.