An AI agent still needs an owner.

Define authority before autonomy\nList what the system can read, draft, change, and send. These are different permissions. A system that can draft a reply should not automatically be allowed to send it.\n\n## Make evidence visible\nA recommendation should be traceable to its input. Keep records of sources, decisions, and approvals. Never present generated examples as actual customer activity.\n\n## Plan for failure\nDecide what happens when an API is unavailable, a confidence check fails, or a customer asks for a human.