TLDR Tech / News
AI coding assistants shift bottlenecks to code review, Harvard study finds
A Harvard working paper analyzing 718 companies found that engineers utilizing AI coding agents produced approximately 30 percent more code, but overall output did not see a significant increase, while code review time rose by 49 percent. Researchers Fiona Chen and James Stratton observed that the accelerated code generation overwhelmed the available reviewers, leading to nearly double the proportion of pull requests requiring formal modifications. These results suggest that the implementation of AI tools redirects bottlenecks rather than enhancing total productivity.
Read the full story