Anthropic, the AI company behind the Claude assistant, has published its own internal data showing that artificial intelligence is already reshaping its workforce: the share of junior technical employees fell from 21% to 14% in about a year, the junior share of technical hires collapsed from 27% to 9%, and by August 2026 its AI agents were logging six times as many work hours as the human staff who direct them.
What Happened
Illustration: engineers in a modern AI office supervising multiple glowing screens running automated AI agent sessions.
On October 16, 2026, Anthropic's in-house research unit published "How employment is changing at Anthropic," a report by its economists Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory that draws on internal employment records, AI-use telemetry, output metrics, employee surveys, calendars, hiring data and 17 employee interviews. The company says it may be a leading indicator of how AI will shift work organisation and labour demand across the wider economy — and the numbers are stark.
The junior rung is shrinking: since early 2025, junior-level employees in technical departments fell from 21% to 14% of staff, and the junior share of technical hires collapsed from 27% to 9%. Postings for junior roles fell sharply after December 2025. Around a third of machine-learning managers and a fifth of software-engineering managers told researchers they were "hiring less due to AI."
The agents have taken over the hours. Combined AI-agent work hours passed human work hours for the first time in March 2026, and by August, interactive agent hours outnumbered employees six to one. Code commits per engineer are about five times their level of six months earlier. Among 192 adopters versus 74 non-adopters, output rose sharply with no higher revert rate — though adoption was not random, and early adopters were already ramping faster.
The gains spread beyond engineering: lawyers processed more contracts, salespeople closed more deals. Non-technical staff went from two coding-agent sessions a month in August 2025 to 90 by May 2026. "Meta-work" — writing agent specifications and monitoring pipelines — grew from 10% to 22% of the average employee's interactive Claude sessions, and the median employee now spends 20% more time in meetings. Humans are doing less doing and more supervising.
Hiring managers describe a changed bar: mentions of AI-tool proficiency climbed from 5% to 15% of technical assessments and from 7% to 28% of non-technical ones, and interviews now emphasise "autonomy and judgment" over "domain expertise." One unnamed source put it bluntly: "I think we've already moved past the point of Claude augmenting away the really entry-level admin type role."
Background
Illustration: a row of empty junior desks in a tech office while a senior manager monitors a wall of AI agent dashboards.
Anthropic is an unusual company to publish this: it is the vendor whose agents it credits for the shift, it roughly tripled headcount over the past year, and it reports two to four times the revenue per employee that Meta and Google managed at a similar stage. Even a company in a hiring boom is hiring fewer beginners.
The report is explicit that it is a single-company self-portrait, not a census of the labour market, with the authors' own warnings baked in. But it joins growing evidence: independent surveys have found employers adopting AI are hiring more senior staff while entry-level openings thin out. Anthropic's figures are the first time a frontier lab has published its own data confirming the pattern from the inside.
A revealing detail: interactive agent sessions sit idle about four-fifths of the time, waiting for a person — one engineer described facing "an army of Claudes, with every one asking for my attention." Human attention, not model capability, is the binding constraint. Developer-blogger Simon Willison highlighted Anthropic's October 16 account of its Claude Code team, where Claude writes 90% of the codebase and merged pull requests doubled June–September while reverts fell 40%.
The expectations gap is telling: staff expect AI to do 75% of their work by 2030, versus 28% for other Americans. Meanwhile, long-tenure staff were 50% more likely to move across departments monthly in 2026 than in 2025, as the company prioritised internal redeployment over external junior hiring.
Why It Matters
This report matters because it is the dog that finally barked. For two years, economists have argued about whether AI would augment or replace workers, mostly from outside the labs. Now a frontier lab has published its own books — and the first rung of the technical career ladder is the one that got cut. If Anthropic is a leading indicator, the junior developer, entry-level analyst and junior paralegal everywhere are the workers to watch.
The 27%-to-9% figure needs careful handling: it is a share of hires, not a headcount — Anthropic hired more people overall but filled the seats with senior staff. That is exactly the pattern labour economists feared. The authors are frank about their remedies' limits: the company's early-warning system and internal redeployment do not "directly address the slowdown in junior hires documented in this report," they write, since "employment protections for incumbents will not benefit those not hired to begin with."
That admission points to the policy problem. If the market's response to AI productivity is to stop hiring beginners, the pipeline producing tomorrow's senior engineers breaks. The apprenticeship model — learn by doing routine work, graduate to hard problems — assumes the routine work is done by humans. Claude Code writing 90% of its own codebase suggests that assumption is expiring inside the labs first.
What Happens Next
Illustration: a young graduate at a career crossroads looking at a chart where an AI figure rises beside a shrinking career ladder.
The first test is whether the pattern travels. Independent researchers will now try to replicate the findings across industries: do law firms, banks and consultancies show the same junior-hire squeeze, and does the "no difference in revert rates" result hold under external audits?
The second test is political. Anthropic has now both pitched a draft AI safety bill to Congress and asked governments to step in on jobs — arguing AI's disruptions need public management on two fronts. The employment data gives worker advocates a new, company-sourced exhibit: not a warning from critics, but an admission from the industry itself.
The third is educational: universities and bootcamps still train students for entry-level jobs the data shows are shrinking. The surviving jobs look like coordination, supervision and judgment — the median employee now spends 20% more time in meetings — not keystrokes.
Watch the next round of lab disclosures: rivals will be under pressure to publish comparable data — or explain why they won't. The question hanging over the industry is the one Anthropic's own staff answered: if AI can do 75% of their work by 2030, what will the other 25% be, and who will be hired to do it?
Sponsored: explore today's featured sponsored offers
FAQ
Q: What did Anthropic's employment study find?
A: In "How employment is changing at Anthropic," published October 16, 2026, company researchers report the junior share of technical staff fell from 21% to 14% since early 2025 and the junior share of technical hires from 27% to 9%. By August, AI agents directed by employees logged six times as many work hours as their human handlers.
Q: Did the AI tools make Anthropic's engineers more productive?
A: Dramatically: code commits per engineer are about five times their level of six months earlier, and among 192 engineers who adopted coding agents versus 74 who did not, adopters' output rose sharply — with agent-written code no more likely to be reverted. The authors caution adoption was not random: early adopters were already ramping faster.
Q: Is Anthropic hiring fewer people overall?
A: No — headcount roughly tripled, and technical hiring still grew in absolute terms. The shift is in the mix: sales, finance and recruiting hiring grew faster, and within technical hiring the company strongly favoured seniors. About a third of ML managers and a fifth of SWE managers said they were hiring less because of AI.
Q: What is Anthropic doing about the junior-hire slowdown?
A: The company set up an early-warning system for at-risk roles, prioritised internal redeployment (long-tenure staff were 50% more likely to switch departments monthly in 2026 than in 2025) and points to broad share ownership. The authors admit these steps don't directly fix the junior-hire slowdown and call for government retraining and income-support policy.
Q: Does this mean AI will replace all entry-level tech jobs?
A: Not necessarily — this is one company's self-reported data, not a labour-market census. But it is the first time a frontier AI lab has published internal evidence that AI is shrinking entry-level technical hiring while agents take on more work, matching patterns other employers have begun reporting.
Sources
- Anthropic Institute (Oct 16, 2026) — "How employment is changing at Anthropic" (Wilmers, Catipon, Massenkoff and Peter McCrory: junior technical staff 21% to 14%, junior technical hires 27% to 9%, agent hours 6 to 1 by August, code velocity 5x, 192 vs 74 engineer comparison, meta-work and meeting-time figures, retraining and income-support call.
- AI Weekly (Oct 17, 2026) — junior technical hire share 27% to 9%; about a third of ML managers and a fifth of SWE managers hiring less due to AI; staff expect AI to do 75% of their work by 2030.
- The Inference (Oct 17, 2026) — agent work hours vs human hours; fivefold code velocity; revenue per employee 2–4x Meta and Google at a similar stage; attention as the binding constraint.
- Simon Willison, via AI News Brief (Oct 16, 2026) — Anthropic's Claude Code team: Claude writes about 90% of the codebase, merged pull requests doubled June to September, reverts fell 40%.
Comments
Post a Comment