Anthropic says junior technical hires fell as agent work grew
Illustration made with AI for ai notisAnthropic's own report, published on 16 October 2026, says the junior share of its technical hires fell from 27% to 9% while agents took on more of the work.
- The Anthropic Institute report, published 16 October 2026, says the junior share of Anthropic's technical employees fell from 21% to 14% since early 2025.
- The report says the junior share of technical hires fell from 27% to 9%, and that by August interactive agent hours were about six times employee hours.
- The authors compared 192 engineers who adopted coding agents with 74 who did not, and say adoption was not random, so their estimates may be biased.
The report, "How employment is changing at Anthropic", is by Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory. It says the junior share of Anthropic's technical employees fell from 21% to 14% since early 2025, and the junior share of technical hires fell from 27% to 9%. Agent work passed human work hours in March 2026, and by August interactive agent hours were about six times employee hours.
The authors report that code commits per engineer rose to about five times their level six months earlier. They compared 192 engineers who adopted coding agents with 74 who did not, and note that adoption was not random, so the estimates may be biased. Roughly 80% of interactive Claude session time is idle waiting for human input, and the share of sessions spent on work that builds tools and monitors agents rose from 10% to 22%.
The data are Anthropic's own: internal data covering employment, AI use, output metrics, employee surveys, internal communication patterns, employee calendar data and hiring decision records, plus 17 employee interviews. The figures describe one company, not the labour market.
What it means for you Our view
This is one company's account of itself, written from its own internal records, so it describes Anthropic and not the labour market. The authors themselves warn that adoption of coding agents was not random, which limits what the comparison can show. Still, it gives a concrete picture of what an agent-heavy team looks like, including a rise in sessions spent on building tools and monitoring agents from 10% to 22%. If you manage engineers, check how your own junior hiring and review workload have changed over the past year before you set targets based on these ratios, and ask who reviews agent output on your team.