do you trust tests written by the same model that wrote the code

A developer encountered issues with AI-generated tests, where the tests asserted the code's output rather than the specified requirements, leading to a "tautology with a green checkmark." The developer now hand-writes assertions for business logic, while the AI handles routine coverage like input validation and edge cases. They are curious about how other developers balance AI and human involvement in test writing, specifically regarding trust in AI-generated tests and assertions.

This report uniquely highlights a specific failure mode of AI-generated tests, where the AI validates its own implementation rather than the intended specification, unlike general discussions on AI in testing.