AI coding agents generate more code, but not more software
Summary
Ars Technica reports on a Harvard Jellyfish study showing AI coding agents significantly increase raw code output but do not boost overall software delivery due to longer code reviews and bottlenecks. The study finds a 30% rise in lines of code, 20% more commits, and 23% more pull requests after AI agent adoption, while issue resolution times increase and human reviewers remain central.