# Agent operating model

Use this operating model to make agentic AI ownership explicit before systems gain runtime authority.

## Ownership lanes

| Lane | Accountable owner | Evidence required |
| --- | --- | --- |
| Business outcome | Process owner | Baseline, target, adoption signal |
| Product behavior | Product lead | User journey, approvals, escalation path |
| Platform reliability | Engineering lead | Logs, dashboards, rollback plan |
| Risk and compliance | Control owner | Policy mapping, access review, audit trail |
| Operations | Service owner | Support queue, runbook, incident owner |

## Operating rituals

- Weekly evaluation review for failures, regressions, and drift.
- Release gate for prompts, tools, models, retrieval indexes, and policy changes.
- Monthly authority review for data access, tool use, and human approval thresholds.
- Incident retrospective that turns findings into tests, controls, or runbook changes.

## Decision rule

No agent should operate without a named business owner, service owner, escalation owner, and control owner.