By Roman Aglaganov
Anthropic’s economics team has published a report on how AI has changed work inside the company itself. “How employment is changing at Anthropic”, released on 16 October, draws on staff records, surveys, calendar and Slack data, agent logs and interviews with 17 employees. The authors present it as an early look at where other workplaces may be heading.
Who gets hired
Fewer juniors: the share of junior staff in technical departments fell from 21% to 14% since early 2025, and junior technical hires from 27% to 9%. Junior job postings dropped sharply after December 2025.
Different skills: hiring notes mention candidates’ AI tool skills about three times as often for technical roles and four times for non-technical ones. Comments on autonomy and judgment grew most; mentions of domain expertise shrank.
How the work changed
Agents outwork people: agent working hours passed human hours in March 2026. By August, agents worked about six hours for every human hour in sessions staff directed.
More output: code commits per engineer are about five times their level six months earlier, with no rise in code later removed or reverted. Sales and contract work rose too.
More meetings: median meeting time is up 20% on a year ago. About 80% of interactive agent session time is spent waiting for a person, and reading all the agents’ output would take some teams over half a working week.
One engineer described managing “an army of Claudes, with every one asking for my attention.”
The caveats
The authors list them. Employees chose when to adopt agents, often on projects that were growing anyway. More code is not necessarily better code. Anthropic was growing fast, and it sells the models it studies, so its practices may not carry over to other employers. They also say internal fixes do nothing for the junior workers who are no longer being hired, and call for government policy on retraining and income support.
Why it matters
Most claims about AI and jobs are forecasts. This is a measurement, from the company with the most aggressive adoption and the best data. Its two clearest findings point in different directions: AI is replacing the entry-level work people used to learn on, and the work left for humans is more judgment, review and coordination, not less.
Disclosure: AI Pulse is written with AI tools, including Claude, which is made by Anthropic.
