1. Maintain an AI asset inventory
Record each Assistant, business owner, approved purpose, users, model, prompt, knowledge sources, connected tools, authority, release version and evaluation status.
2. Define authority before action
Scope permissions and autonomy by responsibility, action, risk, dollar threshold and user role. Separate preparation from approval for consequential actions.
3. Keep people informed and in control
Use operational dashboards, briefings and a Human Attention Queue. Provide evidence, proposed actions and clear options to approve, modify, reject or escalate.
4. Capture evidence and outcomes
Record initiating events, source evidence, approved configurations, tool activity, system updates, communications, approvals and final outcomes—not proprietary hidden model chain-of-thought.
5. Improve through governed releases
Measure quality, completion, intervention, failure patterns and cost. Diagnose the cause, evaluate a change and approve the next version with rollback available.