Runbook: an AI agent misbehaves in production

Runbook: an AI agent misbehaves in production A workflow diagram generated by Archify. 01 / Detect 02 / On-call 03 / Mitigate EX / Learn Triage Stop the harm Close Alert fires · errors · cost · evals · Detect › Triage Alert fires errors · cost · evals Acknowledge · open this runbook · On-call › Triage Acknowledge open this runbook Size the impact · one tenant or all? · On-call › Triage Size the impact one tenant or all? Escalate · page the owner · On-call › Stop the harm Escalate page the owner Verify recovery · metrics back to normal · On-call › Close Verify recovery metrics back to normal Pause or limit · kill switch · rate cap · Mitigate › Stop the harm · undo noted Pause or limit kill switch · rate cap undo noted Roll back · prompt · model · config · Mitigate › Stop the harm Roll back prompt · model · config Postmortem · blameless · actions · Learn › Close Postmortem blameless · actions cause unclear confirmed still failing recovered Legend Agent logic Policy Tool action Context / trace Cloud service External system