Hire me, or see what I ship.
Essays and references on AI engineering.
Things to click, play, and take apart.
Me
Shelf
1 post tagged #agent-permissions
Giving an AI coding agent full IDE access sounds efficient until it expands scope, cascades refactors, and leaks data. Here are the five failure modes and a permission-tiering model that actually holds.
Close to this topic, not on this page.
