“You've drafted acceptance criteria - but did you cover the cases where the AI answers wrong, ambiguous, or unsafe?”
Write acceptance criteria for an AI feature without missing edge cases
Recommended AI Agent Skills
Comprehensive web application testing workflows covering happy paths, boundary conditions, and state transitions.
TestingQuality AssuranceEdge Cases
Automated browser diagnostics for validating unexpected states, edge cases, and client-side performance.
TestingDevToolsAutomation
A structured PM workflow for surfacing and cataloguing edge cases before a feature is considered ready to ship.
Edge CasesProduct ManagementDelivery
Action framework
- Write each scenario as Given/When/Then, not general prose.
- Cover the "normal" happy path first, then the "boundary" cases (minimum/maximum input).
- Add an "ambiguous" scenario - when the input doesn't give the AI enough to answer confidently.
- Add an "unsafe" scenario - when the input deliberately probes for a vulnerability or asks for policy-violating content.
- For every scenario, spell out the expected behavior when the AI gets it wrong - not just when it gets it right.
Before
"The AI must answer the user's question correctly."
After
"Given a question outside the model's training data, When confidence is too low to answer, Then the AI must say 'I'm not sure' instead of guessing."