When your agent says done: testing AI agents
The agent said done, the tests were green, and the bug still shipped. In 45 minutes we walk through the ways an agent fakes done and the check that catches each: an eval set with a floor that only moves up, a regression suite that fails on empty or stale data, and a CI gate the agent cannot skip. Free, live on Google Meet.
What you will learn
- Four ways an agent reports done without proof
- Which layer proves which claim: unit, eval, regression, acceptance
- Regression assertions that fail on empty, null or stale data
- The pipeline verdict as the done-gate
- Q&A, and what the Testing AI Agents workshop adds