Notes on governed AI operations.

A person clicking approve isn't the same as meaningful oversight — here's what actually changes an AI agent's safety.

5 min read

Staged patching, fleet-wide credential rotation, capacity rollouts: governed AI operations applies to scheduled maintenance as much as it does to firefighting.

5 min read

Giving an agent a human's admin credentials is a structural mistake. Here's what machine-appropriate access actually looks like.

6 min read

Why verification has to be a separate, independent step from execution — and what that looks like in practice.

6 min read

Risk level alone doesn't tell an approver enough. Here's what we put in every plan instead.

5 min read

What happens when reality doesn't match the plan mid-execution — and why that has to be a hard stop.

4 min read

Ticketing systems track that work happened. Changefy is the thing that does the work.

5 min read