🤖AI Coding Agents Boost Code Volume, Not Completed Work
AI coding tools add more code, but less actual work
TL;DR
A Harvard study finds AI coding agents increase code volume but not completed work. Agents boost lines of code by 30%, but don't improve issue resolution. Developers see more PRs and longer review times.
Harvard researchers found AI coding agents increase code volume but not completed work. Agents boost lines of code by 30%, commits by 20%, and pull requests by 23%. However, these tools don't improve the number of Jira issues resolved per worker. Developers see a 49% increase in PR review time and a 53% rise in comments per PR. This means more code, longer reviews, and no real productivity gains. If you're using AI coding tools, expect more code and more reviews, but no significant boost in actual work output.
Key Points
Harvard study of 718 firms found AI coding agents increase lines of code by 30%.
Commits increased by 20% and pull requests by 23% with AI coding agents.
AI coding assistants did not increase the number of Jira issues resolved per worker.
PR review time increased by 49%, comments per PR rose by 53%, and changes doubled.
Jellyfish analytics platform gathered data from 700,000 workers across 700+ firms.
Why It Matters
If you're using AI coding agents, expect more code and longer reviews but no significant boost in actual work output. The study shows that while lines of code and PRs increase, the number of resolved issues remains unchanged. This affects teams relying on AI to boost productivity, as the tools may add more work without delivering better results.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,569 builders reading daily.