Your goal: take a failing AI pilot from "it is wrong" to a measured fix and a release you can trust, keep it safe, and decide what it may do alone.
How this works (about 25 minutes)
- Five small ideas: listen then size it, trace it and fix by cause, gate every release, the lock goes in code, and who acts alone. Zoe shows each one in her 3D deploy lab first, then you try it.
- Each idea has two quick checks: one you tap, one you answer aloud. Stuck? Say "I don't know" and Zoe gives a hint.
- After each idea you say it as a rule in your own words. Your rules stay on the board.
- A two-minute timed question at the end: Zoe holds the facts. Feedback follows.
Tips
- Say "explain again", "another example", "harder", "skip" or "go back" at any time.
- You can say your answer instead of tapping.
- No reading needed first. The module reading, Evals, Safety and Cost, goes deeper afterwards.