AI coding agents generate more code, but not more software
A new study finds that code review time rose 49 percent after companies introduced AI coding agents. Gains in writing code are offset by more human work checking it.
Software companies that have adopted AI coding agents are seeing code review take markedly longer. That is the finding of a study by researchers Chen and Stratton, which draws on data from the engineering analytics firm Jellyfish and ends in March 2026. The study concludes that the amount of code grows, but the amount of finished software does not grow to the same degree.
Longer reviews
The average time between a pull request (a proposed code change) being submitted and merged into the codebase rose 49 percent after AI agents were introduced. More detailed data points the same way: the share of pull requests with changes requested nearly doubled, and the number of comments per pull request rose 35 percent.
The researchers also found a 14 percent increase in the share of workers performing code reviews. They checked total active workers using Jellyfish and LinkedIn data and concluded that they could not attribute significant employment changes to AI.
AI offers little help with review
In theory AI could also help with reviewing, but so far the effect has been marginal. By March 2026, 80 percent of the measured firms used some form of AI code review. Even so, AI agents produced only 23.3 percent of all review comments and were involved in 10.8 percent of all pull requests, so humans still do most of this work.
Will it change?
AI coding agents are a relatively new technology, and their output has improved considerably since the study's data cutoff. Some 95 percent of the firms in the study have now adopted the agents, but many are probably still learning when and how to use them best. As engineering teams gain experience, the trade-off between coding time and review time may improve. For now, the increase in coding speed is offset by a similar increase in human review time and effort.
