3th Floor AI · CAST 2026
BUG HUNT
AI-Driven Exploratory Testing
A live AI agent reads an API spec, hunts down defects,
and files reproducible bug reports while you watch.
What you'll see
Test Charter: the agent's plan, generated live from the API spec
Execution Log: every HTTP request the agent fires, color-coded pass / fail
Defect Cards: filed when a bug is confirmed across 3 independent runs
Two builds to hunt
v1.0 (buggy): 5 planted defects: float precision, pagination, unicode crash, missing auth, empty checkout
v2.0 (fixed + regression): 3 bugs claimed fixed. A new one snuck in. Find it.
Rules of the hunt
The 3× rule: a candidate must reproduce 3 for 3 before it earns a card. Flaky is not filed.
MOCK / LIVE: MOCK replays a recorded hunt for pacing. LIVE turns the agent loose for real.
The hunter: 3thfloor-tester, a 3B model running on this Mac. No cloud. No safety net.
click anywhere or press any key