AI agents at Anthropic now run for six hours for every hour its staff work, and the junior share of its technical hiring has fallen from 27% to 9%, according to a report the company published on 16 October 2026.
Code commits per engineer are running at about five times the rate of six months earlier.
The report, How employment is changing at Anthropic, was written by Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory of The Anthropic Institute.
It draws on the company's internal data on hiring, AI use, output, staff surveys, Slack messages and calendars, plus interviews with 17 employees.
Anthropic, the San Francisco AI company behind Claude, presents itself as a leading indicator. Its argument is that what is happening inside an AI lab now may show how other workplaces change later.
How much work are Anthropic's AI agents doing?
More than its people. Combined agent work hours passed human work hours for the first time in March 2026. By August, agents directed by individual employees were running for six times as many hours as the humans directing them.
The report counts agent time as runtime, including processing and tool use, and leaves out any wait longer than 30 minutes. Human time is headcount multiplied by a 40-hour week.
Non-technical staff have taken to agents too. Their coding-focused agent sessions rose from two a month in August 2025 to 90 a month in May 2026, and agents now appear in nearly half of the company's active public Slack channels.
Did output actually rise?
Yes, though the report is careful about cause and effect. Code commits per engineer reached about five times the rate of six months earlier.
Anthropic compared 192 engineers who adopted coding agents with 74 who did not. Code written by agents was no more likely than human code to be removed or reverted later.
Sales representatives who adopted agents closed more deals. The report notes that people with more sales opportunities may have been the ones who automated first.
One account executive estimated automation had replaced about 20% of their old tasks and let them talk to 30% more customers.
Output also rose elsewhere. Lawyers processed more contracts and the accounts payable team handled more invoices.
Why is Anthropic hiring fewer junior staff?
Because the work it needs people for has changed. Junior employees made up 21% of technical departments in early 2025 and 14% now. Among new technical hires, the fall was steeper, from 27% to 9%, with junior job postings dropping sharply after December 2025.
In April 2026, about a third of machine learning managers and a fifth of software engineering managers said they were hiring less because of AI. Technical departments fell from 61% to 56% of total employment over the past year.
What interviewers look for has shifted as well. Anthropic coded hiring scorecards for thousands of hires since 2024. Mentions of candidates' AI tool use roughly tripled for technical roles and quadrupled for non-technical roles.
For technical candidates, the biggest increases were in comments about autonomy and judgment, while domain expertise took up a smaller share. "Increasingly, hiring managers assess applicants based on their judgment and commitment," the report says.
One employee described how a new model changed hiring for performance engineers: "it basically overnight changed what we were looking for."
What is meta-work, and why is it growing?
Meta-work is the work of managing the agents: building tools for them, reviewing what they produce and watching automated pipelines. It rose from 10% of interactive Claude sessions in autumn 2025 to 22% in spring 2026.
People have become the bottleneck. About 80% of interactive session time is spent waiting for a human to respond, and the number of standing instructions written for agents grew from a few hundred to 8,500 in a year.
Agents have not cut meetings either. Median meeting time per employee rose 20% year on year. In an April 2026 survey, 20% of open-ended answers described a shift from doing the work to managing Claudes.
How far do Anthropic staff think AI will go?
Much further than the public. As of August 2026, Anthropic employees expected AI to be able to do 75% of their work by 2030. The report puts the figure for other Americans at 28%.
Staff also feel more productive. The share agreeing that work processes make them as productive as possible rose from 44% in late 2024 to 75% in April 2026, although the wording changed slightly between surveys.
The report suggests responses for other employers, including employee share ownership, an internal early warning system for jobs at risk, and moving people into new roles rather than replacing them. It says private action needs to be matched by government retraining and income support.
What does this mean for UK founders and graduates?
The squeeze on entry-level roles is already visible in the UK. Graduate vacancies on job site Adzuna fell to 8,383 in July 2026, down 45% from 15,397 a year earlier and the lowest since Adzuna began tracking them in 2016, The London Economic reported.
For startups, the Anthropic data points two ways. Small teams can now produce far more, which is why AI tools let solo founders do more than they could a year ago.
But the people being hired are expected to direct and check agents, not just complete tasks. Founders writing job specs in 2026 may get more from testing judgment and AI fluency than from asking for years of experience in a narrow skill.
The policy side is moving too. The debate over how AI is reshaping the UK jobs market has reached the point where former prime minister Rishi Sunak has called for National Insurance to be abolished.
The report admits its own limits. Anthropic tripled in size in the past year, staff chose whether to adopt agents, and an AI lab is not a typical employer. It also accepts that internal protections do nothing for people who are never hired, which is the group the junior hiring figures describe.
Bizztor covers AI and the UK startup economy every week. Follow Bizztor on LinkedIn for the latest.
