AI coding tools boost code output, but software development remains unchanged

A recent study reveals that while AI coding agents significantly increase code generation, they do not lead to a measurable rise in software output or reduced employment. Harvard researchers Fiona Chen and James Stratton analyzed data from over 700 software firms, covering 300 million work events from 2021 to 2026. The study found that human code review remains a major bottleneck, with the review process growing longer and requiring more revisions. As a result, firms do not see a net gain in efficiency. The data, sourced from Jellyfish, highlights that while AI tools improve coding speed, downstream constraints in the development process offset these gains. The findings suggest that AI coding assistance, while valuable, does not fundamentally alter the software development lifecycle.