Harvard Study Finds AI Coding Agents Lift Code 30% but Fail to Raise Software Output
Updated
Updated · Ars Technica · Oct 9
Harvard Study Finds AI Coding Agents Lift Code 30% but Fail to Raise Software Output
3 articles · Updated · Ars Technica · Oct 9
Summary
Across 700-plus software firms, AI coding agents raised lines of code 30%, commits 20% and pull requests 23%, yet issue and feature completion rates showed no statistically significant gain.
Human review absorbed those coding gains: pull-request review times lengthened, revisions became more likely, and reviewers left more comments, creating a downstream bottleneck.
Harvard researchers based the study on 300 million work events covering more than 700,000 employees from 2021 through March 2026, using measured AI adoption and GitHub activity to track rollout timing.
The findings suggest AI coding tools are speeding code generation more than end-to-end software delivery, with little evidence firms are increasing output or cutting engineering jobs.