AI coding agents generate more code, but not more software

A new study reveals that while artificial intelligence coding tools help software developers write significantly more code, they do not increase overall software output. According to the report by Harvard researchers Fiona Chen and James Stratton, the introduction of AI coding agents leads to a 30 percent increase in total lines of code generated.

The study analyzed aggregated analytics data from Jellyfish encompassing 300 million work events across more than 700,000 employees at over 700 software firms from 2021 through March 2026. Despite the surge in raw code, the resolution rate for tracked software features did not change in a statistically significant way. The researchers found that efficiency gains during the coding phase are absorbed by a human review bottleneck.

The average review time for a pull request balloons by 49 percent after AI agents are introduced, with the share of pull requests requiring changes nearly doubling and comments increasing by 35 percent. Consequently, firms saw a 14 percent increase in workers performing code reviews, though overall employment levels remained steady.

While 95 percent of firms in the study have implemented AI coding agents, the data shows that human reviewers still handle the vast majority of review work.

The report