Updated

A new study found that while AI coding agents are generating more code than ever, the efficiency gains are not translating into more finished software, because productivity improvements are being absorbed by a human-review bottleneck. The findings add nuance to claims about AI's transformative impact on software engineering.