AI coding agents create merge conflicts at 3x the rate of human developers

Researchers analyzed 142,000 pull requests written by AI coding agents on GitHub and found that 27.67% produced merge conflicts — integration problems that require manual resolution. This means AI-generated code is significantly harder to integrate into existing projects, forcing human developers to spend more time untangling what the AI wrote before merging it in.

AI coding agents are now writing real code in production repositories, but nobody has measured how well that code actually integrates with human-written code. This dataset reveals a concrete friction point: AI agents generate code that collides with existing work frequently enough to be a real cost. Development teams that adopt AI coding agents aren't just getting faster code generation — they're also getting a hidden tax of integration work that doesn't exist (or exists at much lower rates) in human-only workflows.