AI coding agents generate more code, but not more software

A Harvard study of Jellyfish data covering 300 million work events at 700+ firms found AI coding agents raise lines of code by 30%, commits by 20% and pull requests by 23%, yet Jira-tracked issue and feature resolution does not change significantly. Review time before merge grows 49%, the share of PRs with requested changes nearly doubles and comments per PR rise 35%.

• AI agents add 30% more lines of code, 20% more commits, 23% more PRs
• PR review time before merge grows 49%; comments per PR up 35%
• Share of pull requests with requested changes nearly doubles
• Jira-tracked issue and feature resolution shows no significant change