By Hamza Faisal

Anthropic has turned the microscope on itself, and the numbers are striking. On October 16, the company's Anthropic Institute published "How employment is changing at Anthropic," an internal study of how Claude agents have reshaped its own workforce — and the data suggests the first rung of the technical career ladder is getting much harder to reach.

According to the report, written by researchers Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory, the share of junior-level employees in Anthropic's technical departments has slipped from 21% to 14% since early 2025. The share of junior technical hires fell even faster — from 27% to 9% — with postings for junior roles dropping particularly sharply since December 2025.

That collapse happened even as Anthropic tripled in size. Technical hiring kept growing in absolute terms, but hiring in sales, finance and recruiting grew faster, and the technical departments' share of the overall workforce slipped from 61% to 56%. The company also says it takes in two to four times as much revenue per employee as Meta and Google did at a similar age.

Six agent hours for every human hour

The junior-hiring numbers are only half the story. Based on agent telemetry, the combined working hours of autonomous and interactive AI agents passed human work hours for the first time in March 2026. By August, the agents that individual employees directed were working six times as many hours as their human handlers.

Output rose with the machines. Code commits per engineer are about five times their level of six months earlier. To separate AI adoption from the company's own rapid growth, the researchers compared 192 engineers who adopted coding agents with 74 who had not — the adopters' output rose sharply, and agent-written code was no more likely to be removed or reverted later. Lawyers processed more contracts, accounts payable cleared more invoices, and salespeople closed more deals as agents spread.

The study is careful to caveat its own findings: agent adoption was not random, and more code does not necessarily mean better code.

Human attention is now the bottleneck

The same telemetry that shows agents' speed also shows their limits. Interactive agent sessions sit idle about four-fifths of the time, waiting for a person. Reading everything the agents produce would take more than half the work week, even for non-technical staff. One engineer described facing "an army of Claudes, with every one asking for my attention."

Work has rebuilt itself around supervision. Non-technical staff went from two coding-agent sessions a month in August 2025 to 90 by May 2026, and meta-work such as writing agent specifications and monitoring pipelines has roughly doubled its share of sessions. Hiring managers now screen for autonomy, judgment and a track record with AI tools over narrow domain expertise.

Staff with at least a year at the company were 50% more likely to move across departments each month in 2026 than in 2025, and the company has set up an early-warning system to track jobs at risk of disruption.

A leading indicator, or a special case?

Anthropic frames itself as a leading indicator of how AI will shift work organization and labor demand across the economy. In a survey, staff estimated AI could do 75% of their work by 2030 — compared with 28% in a survey of other Americans.

But the report's authors are frank about what the company cannot fix on its own. None of its internal steps "directly address the slowdown in junior hires documented in this report," they write, adding that "employment protections for incumbents will not benefit those not hired to begin with." They call for government policy on retraining and income support.

Whether this pattern holds outside an AI lab that builds the models it uses is the open question. For now, Anthropic's self-portrait is the clearest data point yet that agentic AI is not just changing what work looks like — it is changing who gets to start doing it.