Anthropic’s internal employment study found that junior workers’ share of technical hiring fell from 27 per cent to 9 per cent while technical hiring continued to grow in absolute terms. Researchers also tracked code output and adoption of coding agents. The findings describe one rapidly expanding company, with non-random tool adoption and indirect code-quality measures limiting what can be inferred about AI’s effects on employment.
The AI developer Anthropic published an internal study of its changing workforce on October 16. Its own employment records show junior workers’ share of technical hires falling from 27 per cent to 9 per cent. Their share of employees in technical departments fell from 21 per cent to 14 per cent. Technical hiring continued to grow in absolute terms, while sales, finance and recruiting expanded faster. This is evidence from one company’s study.[1]
The researchers examined employment, tool use, output, communications, calendars and hiring assessments, alongside 17 employee interviews. They followed 192 engineers who adopted coding agents and compared them with 74 engineers who remained non-adopters throughout the study. Within-person comparisons showed higher code output after adoption. Tool adoption was not random: preparation for large projects could influence the results, and autonomous-agent adopters already had strongly rising output beforehand.[1]
Code-quality checks compared human and agent contributions by the same employee, month and directory, using later deletions, bug-fix deletions, reverts and test-skipping comments as proxies. Those measures cover only some dimensions of quality. Waiting periods exceeding 30 minutes were removed from agent-time calculations. The estimate for people used a weekly total of 40 hours. By August interactive agent hours were six times estimated employee hours. Anthropic says it has introduced early-warning metrics for changing roles and prioritizes internal redeployment.[1]
