Do you know how many agents are in production and who owns each? Choose an option No inventory We know it by memory Partial list, no owners List with owners Live registry with an owner per agent
Does each agent have a defined expected behaviour and SLA? Choose an option No, they act freely Informal expectation Described for some Defined for most SLA and limits per agent
Can you see what they do live (logs, alerts, cost)? Choose an option Blind Only when it breaks Logs, no alerts Logs and alerts Live observability and cost
Do you evaluate quality continuously, not just at launch? Choose an option Not evaluated Only at launch Sporadic manual review Periodic sampling Continuous automated evals
Do you control permissions, data and actions each agent can run? Choose an option Broad access, no control Inherited unreviewed permissions Some manual limits Role-based permissions Least privilege and audit
Can you reconstruct why an agent decided what it decided? Choose an option No Only the outcome Partial logs Trace of most decisions Full, reproducible trace
Is there a process to change or stop an agent without breaking production? Choose an option No, hot changes By hand and fearful Some version control Reviewed deploys Versioned changes with rollback