1. Home
2. News
3. Anthropic

Anthropic says junior roles fell from 27% to 9% of its technical hires

The company that sells the agents says its own engineers now produce five times the code, and it needs fewer of them starting out.

By Think Facility · October 17, 2026 · Anthropic AI policy Save

On October 16, 2026, Anthropic published a study of its own workforce. The number that matters is who it stopped hiring: junior roles were 27% of its technical hires in early 2025, and they're 9% now.

Four economists at the Anthropic Institute wrote it, Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory. They worked from internal records on hiring, AI use, output, calendars and Slack, plus 17 interviews.

Junior technical hires

from 27% to 9%

Junior technical staff

from 21% to 14%

Agent to employee hours

6 to 1 by August

Code commits per engineer

5x their velocity as 6 months ago

Company size

tripled in the last year

What changed in hiring

It started in January 2026. That's when the teams leaning hardest on Claude found they could hand it the code-writing and bug-hunting a junior engineer would once have done, and talk of needing fewer of them began. By April, engineering and research had stopped emphasizing lower-band hires. About a third of machine learning managers said that month they were hiring less because of AI (for software managers it was a fifth).

Anthropic hasn't stopped growing, to be fair. It still adds engineers in absolute terms (sales, finance and recruiting just grew faster). Clerical work barely grew at all. An executive assistant who built a scheduling agent told the researchers that "we've already moved past the point of Claude augmenting away the really entry-level admin type role."

Interviewers want different people now. Notes on a candidate's record with AI tools tripled for technical roles between 2024 and 2026, and quadrupled for everyone else. For engineers, the biggest gains were comments on autonomy and judgment. Domain expertise shrank.

The agents outnumber the people

In March 2026 agent work hours passed human hours across the company. That's everything, fleets included. Count only the agents individual employees run themselves and they still put in six times their handlers' hours by August.

Is the extra code worse? The authors checked. They compared an engineer's own commits with their agent's, in the same month and the same directory, and found no difference in how often lines were later deleted or reverted. Granted, they add that AI errors may be harder to spot, and I think that caveat deserves more weight than it gets.

So people are the bottleneck. Interactive Claude sessions sit idle about 80% of the time, waiting for a human. Reading everything their agents write would take even non-technical staff more than half a work week. Managing agents (writing specs for them, or setting one agent to check another's work) went from 10% of the average employee's sessions in fall 2025 to 22% in spring 2026, and the median employee's meeting time is up 20%.

What it means for people starting out

Anthropic is one company, and a bit of an odd one. Its staff use all day the model it sells, and the authors say many of its practices can't be copied across the economy. Still, reading the pattern forward, they conclude that AI diffusion "could risk worker displacement". They cite earlier research finding that AI-native startups and heavy adopters employ fewer entry-level people.

Graduates are already in a rough market. The New York Fed tracks recent college graduates aged 22 to 27, and put their unemployment rate at about 5.6% in the second quarter of 2026. Underemployment edged up to 42%. Neither the Fed page nor Anthropic's report pins those national figures on AI. What the report adds is narrower: one employer's own records, showing the junior rung thinning while the company tripled.

The authors do suggest what other employers could try. One is broad stock ownership, so staff share in what automation earns. Another is an early warning system Anthropic is building, which flags jobs at risk so people can move before the work goes. It's already moving them: staff with a year's tenure were 50% more likely to switch departments each month in 2026 than in 2025.

They're candid about where that stops. I'd point anyone in a policy job at this line:

Their answer is that employers can't fix it alone and that government needs policy on retraining and income support. They don't propose a specific bill. As of August, Anthropic's staff expected AI to be able to do 75% of their work by 2030. Other Americans said 28%.