AI-Written Tests Passed, But Mutation Testing Exposed 2 Undetected Code Faults

A developer used an AI agent to write both a code patch and its unit tests for a bounded queue implementation, and the test suite initially passed without errors. However, when mutation testing was applied by seeding four deliberate faults into the code one at a time, two of those faults went undetected by the tests. The root cause was that the AI-generated tests inherited the same flawed assumptions present in the AI-generated implementation, making the suite blind rather than truly correct. Mutation testing works by introducing small, targeted faults and checking whether the test suite catches them, providing a measurable signal beyond a simple pass or fail result. The author now runs mutation testing as a mandatory gate before merging any agent-written patch, arguing that a green test suite is only a claim until mutation testing turns it into a measurement.