A Google Team Measured Half of My Argument, and Left the Other Half Open

A recent Google paper demonstrated that spec-driven test generation, where an AI agent first writes a contract by reading existing code, significantly improves bug detection. The author argues this result only validates half of their own hypothesis, as the contract is still derived from the implementation, meaning tests and code can't truly disagree. A proposed methodology called Independent Spec-Driven Development (ISDD) suggests separating the specification, giving the coding agent only the requirements while withholding the acceptance criteria from which tests are generated. This separation ensures the test suite serves as an independent check, and the author's open-source tool is designed to explore whether this method catches more defects.