AI coding agents generate more code, but not more software
Study finds coding efficiency gains get "absorbed" by human review "bottleneck."
Recent research conducted by Harvard University researchers Fiona Chen and James Stratton reveals that while AI coding assistants and agents can generate vast amounts of functional code efficiently, they do not necessarily lead to increased software output or reduced employment in firms. The study analyzed data from over 700 software development companies and found that human code review acts as a significant bottleneck, limiting the overall efficiency gains from AI coding tools.
The effort required to review the AI-generated code is substantial, and this increased review time results in longer pull requests, more revisions, and more comments from reviewers, which in turn can lead to decreased productivity. The increased coding phase efficiency is ultimately absorbed by downstream production constraints, making AI coding tools less effective in boosting software output or decreasing employment.
Written by urgent.news from Ars Technica's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
