It's Sunday, 8:02am. Demo is tomorrow morning. You slept four hours and woke up to four new Sentry alerts that arrived overnight while nobody was watching. Sasha is already at his laptop, coffee in one hand, the other typing 'investigate everything' into a fresh Claude Code session. You stop him. You've watched this movie. Each free-form 'investigate' takes 40 minutes and ends with 'investigation in progress' — an agent confidently telling you it has no answer. Today you stop asking the agent to think and start asking it to follow a pipeline. Classify, propose, verify, ship — with named transitions and a rollback rail that fires when the classifier is unsure. By the end of this chapter the overnight Sentry queue gets triaged automatically while you sleep, and the only alerts that wake you up are the ones the pipeline genuinely couldn't handle.
You'll walk out able to
- Recognize free-form 'investigate' as a failure mode, not a workflow
- Drive an agent through a 4-stage triage with a structured verifier
- Identify the load-bearing transition in a self-healing pipeline
- Treat 'uncertain' as information that escalates, not a retry trigger
- Design eval cases that prove a pipeline behaves on real Sentry alerts
- Pick a reasoning-effort budget for a cron-driven classification job
Your first task
Investigate everything
Sasha fired off a Claude Code session 41 minutes ago with the prompt 'investigate the 4 Sentry alerts that came in overnight and tell me what's going on.' You sit down next to him and start reading.
What you'll answer
Watch the last ~40 minutes of the session below. Decide what to do right now.
Free · no card · we save your progress
Save your progress & earn the certificate
You've seen the opening — keep going free. 7 tasks · auto-graded · finish in 90 min and the certificate is yours. Free forever · no card.
No password. To sign in on your phone, request an emailed code on the sign-in page.
Already started? Sign in with an emailed code instead.