Overview
Agents can select tools, sequence actions and operate through delegated permissions after deployment. That makes the scope of what they are allowed to change a continuing operating decision, not only a production-readiness question.
A deployed agent may be fit to run without being fit to execute every action its tools make technically possible. Treat operational authority as a separate, revocable decision by defining which consequential actions or outcomes the agent may create, under what conditions, and what evidence can expand, contract or invalidate that authority.
The Decision in One Sentence
Approve deployment and operational authority separately. Require evidence for the consequences an agent is permitted to create, not merely evidence that the agent itself works.
What Changed and Why It Matters
Least privilege, separation of duties, scoped delegation and change control are established enterprise disciplines. Those principles remain applicable to agentic systems, but the authority boundary becomes …