AI Made Every Test Pass. The Code Was Still Wrong.
Summary
The article demonstrates how AI-generated tests can green-light a bug by validating the implementation rather than the intended behavior. It describes using real contracts to stress-test a Solidity converter and advocates AST-level ground-truth comparison to ensure semantic correctness, highlighting the need for a known-good reference when using AI for testing.