SignalTech/AI
AI coding agents generate more code, but not more software
AI coding agents increase code output without increasing software delivery.
Context
What happened: A Harvard study of coding practices across hundreds of firms found AI coding agents raise code volume but not software output, because human code review absorbs the efficiency gains.
Key numbers: AI agent adoption led to a 30 percent rise in lines of code, 20 percent more commits, and 23 percent more pull requests.
Why it matters: The study authors report little evidence that firms increase software output or reduce employment using these tools, as code review lengthens, pull requests need more revisions, and reviewers leave more comments.
What changed
A Harvard study of coding practices across hundreds of firms found AI coding agents raise code volume but not software output, because human code review absorbs the efficiency gains.
Why it matters
The study authors report little evidence that firms increase software output or reduce employment using these tools, as code review lengthens, pull requests need more revisions, and reviewers leave more comments.