Anthropic published a study of its own workforce. It says junior staff fell from 27% to 9% of its technical hires. Junior staff also slipped from 21% to 14% of technical departments since early 2025. Teams that lean on Claude found that writing code and finding bugs could increasingly be handed to it. Code commits per engineer are now five times the pace of six months ago. Technical hiring still grew in absolute terms.
The authors call the link between junior hiring and AI use descriptive, so cause is not proven. By August, agents run by individual employees worked six times as many hours as those employees. In interactive sessions, agents spent about 80% of the time waiting for a human. To test quality, the authors compared agent-written and human-written code from the same employee, month and directory. Lines later removed or reverted did not differ. They add that AI errors may be harder to spot.
For buyers, the study names human attention as the bottleneck, and Anthropic's hiring managers now stress autonomy and judgment. A removed-or-reverted check may miss errors that are hard to see. Ask who reviews agent-written code in your teams, and what you measure besides reverts.