TechAI coding agents produce more code, but not more software

A study covering more than 700 software development firms finds that human code review bottlenecks AI coding tools. While coding itself became more efficient, reviews took longer and code was more likely to need revisions, with little evidence of increased software output or reduced employment.

artificial intelligencecoding toolscode review

AArs Technica★★☆☆☆2026-10-09 19:43Original