Study Finds Half of AI-Generated Code Passing Automated Tests Would Be Rejected by Human Developers

Summary

A new study has found that approximately 50% of AI-generated code that passes standard industry automated testing benchmarks would be rejected during human code review by professional developers. The finding highlights a significant gap between automated quality metrics and the standards real engineering teams apply in practice, including considerations around readability, maintainability, architectural consistency, and adherence to team conventions that automated tests typically do not capture.