Machine-Written Pull Requests Contain 70% More BugsCodeRabbit analyzed 470 GitHub pull requests and found AI-generated code introduces more defects than human-written code across logic, security, maintainability and performance categories, with severity spiking higher as well. The use of AI code generation has expanded across the industry.
Most of the bugs that make it to production are where the code seems totally reasonable, but someone misunderstood how systems interact, or what really happens in a corner case. AI is great at making reasonable looking code. It’s terrible at… understanding.
Me today: And make sure the tests pass. AI: Great, the test cases are all updated! Me: You didn’t actually run then. AI: Perfect! You’re absolutely right, I will run the tests now.


