Agent autonomy

Every change an AI coding agent makes gets one of three levels. The riskier the change, the more trust its part of the code must have earned from earlier releases.

• Assisteda person approves each step
• Superviseda person approves the rollout
• Hands-offships on its own