A study by Harvard researchers Fiona Chen and James Stratton analyzed 300 million work events from over 700 software firms between 2021 and March 2026. The data, sourced from Jellyfish, tracked the impact of AI coding assistants and agents on engineering output.
The findings show that AI coding agents increase total lines of code by 30 percent, total commits by 20 percent, and pull requests by 23 percent. However, these gains do not lead to a statistically significant increase in the resolution rate of software features tracked in Jira.
The bottleneck lies in the code review process, which takes 49 percent longer after the introduction of AI agents. Pull requests now require more revisions, and the number of comments per request has risen by 35 percent. Consequently, 14 percent more workers are now involved in code reviews.
Despite 80 percent of firms using some form of AI-assisted code review by March 2026, AI agents generated only 23.3 percent of review comments and 10.8 percent of pull requests. Researchers found no evidence that AI tools have led to significant changes in total employment levels at these companies.
