← Automation AI Engineer

Chapter 1 · 60 min · free

Inherit the crisis

It's Monday, 9:14am. Four days ago NoteDrop's founder Mira pushed a commit, closed her laptop, and disappeared. She was the only engineer who knew the codebase. Sentry has 23 open alerts. The CEO has a Series-A demo in 72 hours — fail it and the round dies. You were hired this morning as the new technical lead. Your only help is Sasha, a junior dev who loves Claude Code but trusts it the way a tourist trusts a Vegas dealer. In the next hour you will learn to look at an agent loop and know — in under a minute — whether it's working or burning your runway.

You'll walk out able to

  1. Diagnose a Claude Code session as healthy, dying, or dead within 60 seconds
  2. Size a sprint to match the available failure signal
  3. Write a TDD prompt that gives the agent a verifier and you a kill-switch
  4. Catch scope-bleed in an agent diff before it ships
  5. Pick the right parallelism strategy for a backlog of tickets

Task 1 of 7 · no sign-up

Is this loop alive?

Sasha kicked off a Claude Code session 23 minutes ago to fix the login bug, then went to lunch. The CEO just walked over: 'Is it working?' You attach to the tmux pane and start reading.

Watch the last ~7 minutes of the session below. Decide what to do right now.

session replayt=0s / 1380s
0000suserFix the login bug — users on Safari can't sign in after the magic link redirect. See Sentry SAFARI-AUTH-2031.

Watch the session, then decide:

Already started? Sign in with an emailed code instead.