AI-rewritten: This is a summary of an article from Ars Technica, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

Harvard researchers Fiona Chen and James Stratton analyzed data from over 700 software firms between 2021 and March 2026 using Jellyfish analytics. Their study found that while AI coding agents significantly increase the volume of code written, they do not necessarily lead to more finished software or reduced employment. Specifically, introducing these tools resulted in a 30 percent increase in total lines of code generated, a 20 percent rise in commits, and a 23 percent increase in pull requests. However, the resolution rate for major software features tracked by tools like Jira did not change significantly after AI adoption.

The primary reason for this lack of increased output is that the efficiency gained during coding is offset by longer review times. On average, the time required to review code before merging it into a project increased by 49 percent after AI agents were introduced. Consequently, the number of pull requests requiring revisions nearly doubled, and the average number of comments per request rose by 35 percent. In response to this added workload, the share of workers performing code reviews increased by 14 percent, but there was no significant change in total active employment across the firms studied.

Although many companies now use AI for reviewing code, human reviewers remain responsible for the vast majority of the work. By March 2026, while 80 percent of measured firms used some form of AI code review, AI agents were only responsible for 23.3 percent of all review comments and 10.8 percent of all pull requests. The researchers noted that while AI tools are becoming more common, teams are still learning how to best deploy them. Ultimately, the study suggests that the time and effort spent reviewing AI-generated code currently counteract the speed benefits of faster initial coding.