Lesson 2: Keeping the agent healthy once it is live
23 min read·13 Oct 2026
Module 16 · Lesson 2 · About 55 minutes · Needs: Module 12 (kill switch, audit trail), Module 13 (the eval gate), Module 14 (run records and alert rules in returns_desk/tracing.py)The situation
It is Tuesday 23 June 2026, week 11. At 09:00 a prompt change, v2, ships without the eval gate. By mid-morning Priya is asking why her team is drowning in hand-offs, and Dev rolls back at 10:25.
Module 14's lab wrote an alert rule that would have caught this, but nobody was on the other end. This lesson replays that morning with a dashboard, alerts that reach a person, and a runbook. Then Sam runs a drill where the agent does damage, and Marta asks how long transcripts are kept. All BrightCart data is simulated.