AI is making developers write more code. It is not making them ship more software.
Harvard researchers Fiona Chen and James Stratton studied coding teams using AI tools and found lines of code, commits, and pull requests all rose 20 to 30 percent. Yet the amount of software actually completed showed no statistically significant increase. The reason: the time to review and merge each pull request rose from roughly 7 days to 10.5 days, reviewer comments per pull request climbed sharply, and the share of employees doing code review rose 14 percent. The paper also found that AI-generated code needed substantially more corrections before it was ready to merge.
This flips the usual AI-coding pitch. The limiting factor was never how fast code gets typed. It is how fast a human can read that code, question it, and decide whether it is safe to ship, and that job has not gotten any easier just because a machine wrote the first draft.
Companies chasing productivity gains by pointing AI at the keyboard might get further pointing some of that budget at reviewers instead.