Most debates about AI and jobs run on forecasts. This one runs on payroll data. On October 16 Anthropic published a study of its own workforce, written by economists Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory. It uses internal records on hiring, agent usage, code output, surveys, calendars and 17 employee interviews. The headline: at the company that builds Claude, the machines are now doing most of the hours.
Six agent hours for every human hour
Anthropic tracks how long its agents actually run. In March 2026, the combined hours of autonomous and interactive agents passed human work hours for the first time. Narrow it down to only the agents that individual employees direct, and by August those agents worked six times as many hours as the people steering them.
The method matters. Agent time counts runtime, including thinking and tool use, but drops any wait longer than 30 minutes. Human time is simply headcount multiplied by 40 hours a week. It is a rough comparison, but the trend line is not subtle.
The junior door is closing
The most quoted number is about who gets hired. The share of junior technical hires at Anthropic fell from 27% to 9%. The share of junior staff in technical departments slipped from 21% to 14% since early 2025. Postings for junior roles dropped especially fast after December 2025.
According to the report, managers started talking about lower demand for junior engineers in January 2026, once the teams leaning hardest on Claude found that writing code and finding bugs could be handed off. One person described a new model's boost for performance engineers as something that "basically overnight changed what we were looking for." By April, engineering and research teams had deemphasized hiring for lower band technical positions, and around a third of machine learning managers and a fifth of software managers said they were hiring less because of AI.
This is not a shrinking company. Anthropic tripled in size in the past year, and technical hiring still grew in absolute terms. Sales, finance and recruiting simply grew faster. Administrative roles, which normally scale with company size, barely did.
Output jumped, and not only in code
Code commits per engineer are now running at five times the pace of six months earlier. To check that this was not just the company's booming business, the researchers followed 192 engineers who adopted coding agents and compared them with 74 who had not yet. Output rose sharply after adoption. They also looked for "AI slop" and found no difference in how often agent-written code was later deleted or reverted compared with code the same person wrote by hand.
Non-technical staff changed too. As agents spread, lawyers processed more contracts and accounts payable moved more invoices, and sales reps closed deals with more clients after they adopted agents. Coding-agent sessions among non-technical employees went from 2 a month in August 2025 to 90 in May 2026. One sales rep built a custom dashboard in three to four hours, something they said was out of their wheelhouse before.
Now the humans are the bottleneck
All that speed has a cost: attention. Interactive Claude sessions spend about 80% of their time idle, waiting for a human to reply. For non-technical teams, simply reading everything their agents produce would take more than half the work week. One engineer described facing "an army of Claudes, with every one asking for my attention," and compared it to babysitting kids who treat an unopened juice box and a cut finger as equal emergencies.
So people are turning into managers of machines. "Meta-work," like writing specs for multi-agent runs, setting one agent to review another, and building monitoring dashboards, rose from 10% of employees' Claude sessions in fall 2025 to 22% in spring 2026. Written standing instructions for agents grew from a few hundred in July 2025 to 8,500 a year later. Meanwhile meetings went up, not down: the median employee spends 20% more time in them than a year ago.
Hiring criteria shifted with it. Interviewers now comment far more on candidates' creative use of AI tools, and on autonomy and judgment, while talk of narrow domain expertise shrank.
What the staff expect next
A companion report compared Anthropic employees with a Morning Consult survey of nearly 11,000 US adults. 91% of staff, versus 39% of the public, expect AI to make a Nobel-level discovery by 2030. More than half of staff expect AI to be able to do at least 80% of their current work by then; only 5% of Americans say the same about theirs. The authors argue the two groups share the same theory of how AI hits jobs. They just have very different experiences of what AI can already do.
Why this matters, and the caveats
Anthropic says it has set up an early warning system to flag jobs at risk of disruption and is prioritizing moving people into new internal roles. Employees with at least a year at the company were 50% more likely to switch departments each month in 2026 than in 2025. But the authors are blunt that none of this helps the people who are never hired in the first place, and they call for government policy on retraining and income support.
The biggest caveat is the one the report makes itself: Anthropic is an AI lab with fast revenue growth, deep AI adoption and widespread employee stock ownership. It is not a typical employer. Still, if the company closest to the technology is quietly cutting its junior pipeline while its agents clock six hours for every human one, it is a signal worth watching.
