A frontier lab measures its own bottom rung
Anthropic’s study of its own workforce, published on 16 October by researchers at the Anthropic Institute, found that the junior share of its technical hires fell from 27% to 9%, a drop of two-thirds. The authors suggest that what happens inside the company may show early how AI will reshape hiring and labour demand elsewhere. Our news report covers the study’s main findings. This piece asks a narrower question: if Anthropic is an early signal, what should UK graduate employers and job seekers take from it, and how much of the same pattern is already visible in British data?
Strategic Insight: The report concedes that the measures Anthropic has taken to protect its staff do nothing for people it never hired. Its authors write that “employment protections for incumbents will not benefit those not hired to begin with”. Graduates looking for a first job are, by definition, on that side of the line.
What the Anthropic data does and does not show
The report dates the shift precisely. Talk of weaker demand for junior technical staff surfaced in Anthropic’s core technical teams in January 2026, as the teams leaning hardest on Claude found they could hand it more of the code-writing and bug-hunting. By April, the company’s research and engineering teams were putting less weight on hiring for lower-band technical roles. In a survey that month, around a third of machine learning managers and a fifth of software engineering managers said AI meant they were hiring fewer people.
Shares, not headcount
The headline figures are proportions, and that matters. Anthropic “has tripled in size in the last year”, and the report says technical hiring has kept growing in absolute terms. What fell is the junior share of it, along with technical departments’ share of total employment, which slipped from 61% to 56% as recruitment into sales, finance and recruiting teams grew faster. The study describes a company changing whom it hires as it expands. It is not evidence of falling technical employment.
The figures that matter for the UK question sit side by side below.
| Signal | Figure | Source | What it suggests |
|---|---|---|---|
| Junior share of Anthropic’s technical hires | 27% to 9% | Anthropic | Junior intake shrank while the company tripled in size |
| Agent working hours relative to the staff directing them (August 2026) | 6 to 1 | Anthropic | Implementation work is moving to agents that employees supervise |
| Fall in UK online job adverts, most vs least AI-exposed third of occupations | 15% vs 6% | Bank of England staff blog | UK hiring has weakened most where AI exposure is highest |
| UK employers that have cut entry-level roles in the past year | 36% | Work Foundation | Entry-level hiring is shrinking across a third of UK employers, not only at AI firms |
| Graduate hiring change in 2026, ISE survey | Up 4% | Institute of Student Employers | ISE’s chief executive says AI is not, so far, why employers cut intake |
Strategic Reality: The UK numbers point in different directions because they measure different things. The Bank of England figures cover online adverts across 26 occupational groups, the Work Foundation polled 1,001 businesses of every size about entry-level roles, and the ISE speaks for some of the UK’s leading graduate employers.
Why judgement-first hiring creates a pipeline problem
Just as relevant for graduates is what Anthropic now looks for in anyone it hires. The authors had an automated system classify interviewers’ scorecards for thousands of people hired since 2024. In 2024, a candidate’s record with AI tools came up in around 5% of technical assessments and 7% of non-technical ones. By 2026 such mentions were three times as frequent in technical assessments and four times as frequent in non-technical ones.
For technical candidates, the biggest gains were in comments on autonomy and judgement, whilst domain expertise took up a smaller share of the assessment. For non-technical roles, interviewers wrote more about orientation towards customers and the mission, and less about execution. One manager put the logic plainly: “I expect models to continue to be jagged, so for my team I want them to be fungible across domains.”
Critical Context: The report ties the move away from junior hiring to newer models. One staff member described the uplift newer models gave performance engineers as something that “basically overnight changed what we were looking for”.
Where does judgement come from?
The report notes that in some roles, employees “are increasingly removed from the object-level work that originally built their skills”. Across both technical and non-technical jobs, it lists the parts being automated: writing code, finding bugs, logging progress and drafting emails. Put that beside the scorecard data and a tension appears. In our reading, employers increasingly want new staff to arrive with judgement, whilst the routine work that once justified hiring juniors is the work being handed to agents.
A Bank of England staff blog published in August frames the same risk for the UK. Citing research by Friebel and colleagues, it describes a workforce pyramid resting on big intakes of juniors that may be turning into a diamond shape, with an empty base and experienced people in the middle. Working from home may have started that shift by making on-the-job training more expensive, the post argues, and AI strengthens the incentive to make it permanent, because senior staff can now absorb tasks that were once the reason to hire juniors.
The good-employer trap
Anthropic’s response to disruption is to watch for it and move people. It is building an early warning system that tracks role-level demand and productivity against AI capabilities, and it prioritises internal redeployment. Staff with at least a year’s service were half as likely again to change department in any given month in 2026 as in 2025.
The report is candid that this carries a cost. It notes that commitments to redeploy existing staff could make employers “hesitant to take a risk on hiring new employees”.
⚠️ Warning: Redeployment protects incumbents. A UK employer that adopts Anthropic’s model to look after its current workforce may, on the report’s own reasoning, become more cautious about taking on graduates. The report itself treats this as a trade-off, not a free win.
Who in the UK is already feeling it
The Bank Underground post by Haley Schlicht of the Bank’s Data and Statistics Division, which we covered in August, combined several published AI exposure measures and rebuilt them on UK employment weights. Online adverts fell 15% for the most exposed third of occupations, 10% for the middle third and 6% for the least exposed. Customer service adverts fell 23% and administrative adverts 22%. Vacancies have fallen sharply in finance (where highly exposed occupations make up 84% of jobs), ICT (81%) and professional services (63%).
Administrative work is one place where the UK data and Anthropic’s account line up. Anthropic describes itself as clerical-light despite rapid growth, and one of its executive assistants told the researchers: “I think we’ve already moved past the point of Claude augmenting away the really entry-level admin type role.”
The Work Foundation survey, run by Survation across 1,001 businesses between 7 and 20 May, found 36% had reduced entry-level roles over the past year, according to Reuters. Small businesses reported cuts at 24%, against 48% of medium-sized and 46% of large employers. The share saying AI and automation had cut the jobs on offer was 24% among small firms, against 58% and 60% among larger ones. Reuters also reported that Adzuna counted 49% fewer July job ads asking for graduates than a year earlier.
Where it is not showing yet
The Institute of Student Employers tells a different story. Its 2026 survey, reported by The Guardian, found graduate hiring rose 4% in 2026, following an 8% fall in 2025, and employers expect it to grow by another 7% over the coming year. Underneath that average, 40% of respondents reduced their intake and 45% increased it. Chief executive Stephen Isherwood was clear about the cause of the cuts: “They [companies] are not managing down graduate recruitment because of AI. It’s because of economic uncertainty.” He placed the AI concern “two or three years down the line”.
| Group | What is changing | What is still unknown |
|---|---|---|
| Graduates seeking technical roles | At Anthropic, a far smaller junior share of technical hires, assessed more on autonomy and judgement | Whether UK technology employers follow; ISE’s chief executive says AI is not driving cuts |
| Administrative and customer service roles | The steepest UK advert falls (22% and 23%); Anthropic reports needing few clerical roles | How much of the UK fall is AI rather than cost and cyclical pressure |
| Larger UK employers | More likely than small firms to have cut entry-level roles, and far more likely to cite AI | Why ISE respondents’ graduate hiring still rose 4% this year |
What this means for graduate assessment
Anthropic’s interviewers increasingly reward candidates with a record of inventive AI use. UK graduate recruiters count unauthorised AI use in interviews as misconduct. The ISE’s survey found 92% of employers suspected or confirmed misconduct in selection, and 66% had seen at least one candidate misuse AI in an interview, as our October report set out. The two are not opposites: one rewards a track record of using AI well, the other penalises unauthorised help in the room. If the frontier pattern spreads, the question for graduate employers is less whether candidates used AI than whether the assessment can tell judgement exercised through AI from answers supplied by it. Our coverage of how the Big Four are reworking graduate training follows one sector’s response.
What remains unresolved
Nobody owns the missing first rung
The report says plainly that private employer action has to come with forward-looking government policy on retraining and income support. In the UK, the scale of the problem is already on the record: according to Reuters, over a million 16 to 24-year-olds were not in education, employment or training between January and March 2026, a level last seen in late 2013, and a May government report raised the risk of a “lost generation”. Redeployment schemes inside firms cannot reach people who never got through the door.
Attribution is not settled
The Bank of England post concludes that pandemic over-hiring, the economic cycle, labour costs, remote working and AI arrived so close together that “confident attribution remains premature”. Isherwood blames the economy. Even Anthropic’s own productivity comparison carries a caveat from its authors: engineers chose when to adopt agents, so the estimates could be biased. The direction of travel is clearer than the cause.
One company is not a market
The authors themselves warn that many parts of the company’s employment practice could not be copied economy-wide. It is three times the size it was a year ago, its revenue per employee is two to four times what Meta and Google achieved during their own rapid growth, and widespread stock ownership gives staff a direct stake in its success. A company growing that fast can cut its junior share while still adding technical jobs. A UK firm with flat headcount that makes the same shift would be hiring fewer juniors in absolute terms.
Reality Check: The UK evidence supports a narrower claim than the headlines. Entry-level hiring is under pressure, the pressure is heaviest in the occupations most exposed to AI, and larger employers are more likely to name AI as a cause. Whether AI is the main driver, rather than one of several, is not established by any of the UK sources examined here.
What to watch
Three checkable signals will show whether Anthropic’s experience is a preview or an outlier.
- The next ISE annual survey. Isherwood put AI’s effect on graduate hiring two or three years away. If next year’s survey shows employers citing AI as a reason for cutting intake, the frontier timeline is arriving in the UK faster than ISE’s chief executive expects.
- The Bank of England’s three series. The Bank Underground post names them: whether job numbers fall after vacancies, whether entry-level hiring keeps weakening, and whether measured AI exposure turns into actual adoption. All three moving together would make the AI explanation much harder to dismiss.
- Official youth NEET estimates. The count of 16 to 24-year-olds outside education, employment or training passed a million in the first quarter. A continued rise alongside stable graduate scheme numbers would suggest the pressure is landing below graduate level, in the administrative and customer service roles the vacancy data already flag.
Anthropic’s authors argue that a company this far ahead in adoption could see these changes first. The useful reading for UK employers is less about headcount than about criteria: at Anthropic, hiring has moved towards judgement and away from the execution and narrow expertise that AI tools absorb. That shift showed up in interview scorecards while technical hiring was still growing.
Strategic Insight: The report’s own authors admit that none of Anthropic’s measures “directly address the slowdown in junior hires” they document. For UK employers, the question worth asking now is who in their organisation will be developing the judgement they intend to hire for in five years’ time.
Source: How employment is changing at Anthropic by Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory (The Anthropic Institute, 16 October 2026). UK data from Canaries in the column: AI exposure and the UK’s hiring slowdown by Haley Schlicht (Bank Underground, 6 August 2026); the Work Foundation survey as reported by Reuters, filed by David Milliken (via Zawya, August 2026); and the Institute of Student Employers’ 2026 survey as reported by The Guardian (14 October 2026).
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will be watching whether next year’s graduate recruitment surveys begin to show the shift Anthropic has measured inside its own walls. Read more analysis at Insights, or get in touch.