AI Coding Agents Produce More Code, but Firms Ship No More Software

A study of more than 700 software firms found that AI coding agents increased code production, while tracked software output did not rise significantly. Longer reviews and more requested changes absorbed much of the coding time saved.

Slashdot has reported:

AI coding agents increased the amount of code produced at more than 700 software firms, but the study found no statistically significant rise in the software features those companies completed. The research, which examined actual coding practices, points to human code review as a bottleneck that absorbs efficiency gains from AI-assisted coding.

After firms introduced AI coding agents, total lines of code generated rose by 30 per cent on average, total commits increased by 20 per cent and pull requests grew by 23 per cent. The researchers found no statistically significant change in the resolution rate for Issues and Epics tracked by tools such as Jira. They also found no shift in the size or complexity of those tracked issues after AI tools were introduced.

The review process took longer after the agents arrived. The average time from submitting a pull request to merging it into a codebase increased by 49 per cent. The share of pull requests requiring changes nearly doubled, while comments per pull request rose by 35 per cent, according to the researchers.

The study found that the share of workers performing code reviews increased by 14 per cent after AI agents were introduced. The researchers wrote that they “cannot attribute significant employment changes to AI” after examining total active workers listed in Jellyfish data and cross-referencing those firms with LinkedIn data.

AI tools were also used for some code reviews, but their contribution remained limited in the study. By March 2026, 80 per cent of the measured firms used some form of AI code review. AI agents produced 23.3 per cent of review comments and were involved in 10.8 per cent of pull requests, leaving humans responsible for most of the review work.

The researchers said any efficiency gained during coding was “absorbed by downstream constraints in the production process.” They found that “the code review process significantly increases in length, pull requests are more likely to require revisions, and reviewers leave more comments.” The findings point to a gap between producing code and completing software work: more code and submissions did not translate into a measurable increase in resolved features.

The study’s results were reported by Ars Technica and relayed by Slashdot. It covered more than 700 software firms and examined changes following the introduction of AI coding agents, including output, reviews and employment. While AI-assisted tools increased coding activity, the researchers found “little evidence that firms increase software output or reduce employment” through their use.