AI coding agents break your code one in four times, new data shows

AI coding agents, when they contribute code, create 'merge conflicts' in over a quarter of their changes. This means software development teams will spend more time fixing these errors than previously understood.

Everyone assumed AI coding agents would just make software development faster. This paper shows that integrating AI-written code comes with a significant hidden cost: a quarter of the time, the AI's changes will clash with existing code. This means companies deploying AI agents must now factor in substantial human effort to resolve these conflicts, slowing down the very process AI was supposed to accelerate.