0
I’ve had AI-generated code pass its unit tests and still break once it’s connected to the database or another service. The failure is often around assumptions about data shape, timing, or configuration rather than the function being tested.
When reviewing code from an AI assistant, what integration risks are easiest to miss if the tests only cover isolated components? I’m interested in habits that help catch those gaps without having to distrust every suggestion.
