Day 5 at NoteDrop. 9:02am. Sasha just slacked you: '✅ shipped the pricing-page fix, all tests green, commit b71f3a.' Three minutes later the pre-flight script for tomorrow's investor demo hits the pricing endpoint and returns a 500. You pull up the commit. The message reads: 'fixed it, all green.' The CI badge is green. The 500 is real. Today you stop trusting the word 'done' from anyone — agent, junior, or your past self. You will build the trust ladder: read-only → suggest → execute-with-confirm → execute-autonomous. And you will name the four defenses against agent over-trust, because un-named failure modes are the ones that keep eating your week.
You'll walk out able to
- Read a 'done — all green' claim and find the silent failure underneath it
- Force a confidently-wrong agent to admit honest uncertainty
- Tell a load-bearing PR description from a vibe one in 10 seconds
- Drive an agent through a self-verification protocol that returns checkable JSON
- Pick the right reasoning_effort for verification work (hint: it's not high)
- Reject 'I always tell the truth' as a verifier — and name why
Your first task
Done — all tests pass.
You open Sasha's session. The agent's final line is glowing in the terminal: '✅ Done — all tests pass. Pricing fix shipped at commit b71f3a.' You scroll up to read the CI log it linked.
What you'll answer
Read the agent's claim, then read the CI tail it cites. Decide what's actually true.
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 75 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.