Incidents & recovery
Free. No email required.
What the organization does after its first agent-caused failure.
- Anyone on the team can stop the line the moment they spot a defect, regardless of role or seniority. That's the andon-cord rule, borrowed straight from manufacturing: the authority to stop something is not the same as the authority to approve it. Stopping belongs to whoever's closest to the defect, not whoever holds the title.
- Verified means someone other than the author checked the work against a standard and left evidence of what they checked. Not “it looks right.” Not the agent's own account of what it did. Trust but verify.
- Two ladders exist here and I don't let them blur into one. The blast-radius tier is about prevention: how much checking a piece of work needs before it ships. Incident severity, SEV1 through SEV4, is about response: how hard to react once something's already gotten through. A high-tier rollout can still throw a low-severity incident, and a low-tier one can still surprise you.
- When something does go wrong, the order matters:
- Stop the line.
- Cut off whatever the agent could still reach.
- Capture the full chain of what happened before anyone reruns or cleans anything.
- Preserve that record.
- Only then remediate — because remediation can destroy the evidence you'd need to actually understand what happened, except when the harm is still active and time-critical.
- No incident review closes without answering “what should have caught this before a person had to.” That includes reviews an agent drafted. A human still checks for blame creeping back onto a person before it circulates.
What's Behind This Card
No essay behind this one, three templates instead: the Agent Rollout Pre-Flight Checklist before a rollout, Definition of Verified for sign-off, and the Agent Incident Runbook for when something breaks.
This is one of six cards today. The field guide maps seventeen practice areas, and I'm growing this set as each one earns its own depth.
This is a fit question, not a recipe. Where your team actually sits against this list, and what that's costing, is what the Engineering Ways of Working Diagnostic is built to find.