Anthropic says only 9% of its new engineers are beginners, down from 27%

Nine percent. That is now the share of beginners among the engineers and researchers hired by Anthropic, down from 27% before. The figure comes not from a union or a research firm, but from the company itself, in a report published on October 16 on how employment is changing inside it.

The document is signed by four Anthropic economists: Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory. It describes what is happening to jobs at the maker of Claude since its employees started working with AI agents, programs that can chain tasks together on their own. One caveat is needed: Anthropic sells these very agents, and has every interest in showing that they transform work. Its internal data cannot be verified from the outside.

Fewer juniors, but not fewer engineers

The figure measures a share, not headcount. Technical hiring has continued to grow in absolute terms. The share of juniors on technical teams fell from 21% to 14% since early 2025, and the share of technical departments in total employment fell from 61% to 56% in a year. According to the report, entry-level job postings dropped sharply after December 2025. In an internal survey in April, about a third of machine learning team leads and a fifth of software development managers said they were hiring less because of AI.

Six hours of agent work for every human hour

The report also quantifies the role agents have taken on. In August 2026, agents run by employees reportedly logged six times more hours of work than humans. The calculation assumes 40 hours per week per employee, and counts the agents' compute and tool-use time. The number of code contributions per engineer reportedly increased roughly fivefold in six months.

One detail is less flattering: Claude sessions spend about 80% of their time waiting for a human response. Anthropic's employees, for their part, expect AI to be able to do 75% of their work by 2030, versus 28% for other Americans surveyed.

What the report does not prove

The authors themselves list several limitations. Employees do not adopt tools at random. Employees who adopted autonomous agents were already producing more and more before adopting them. And more code does not mean better code. Much of the material is descriptive: internal data, 17 interviews and automated analysis of agent logs. The reception was partly wary. On the developer platform daily.dev, which compiles Hacker News discussions, skepticism dominated, notably toward a report published by the company itself and judged unverifiable.

Anthropic is not an ordinary company, either. Its revenue reportedly reached $16.8 billion in three months, according to the press. According to the report, its revenue per employee is two to four times that of Meta or Google at the same age. The authors acknowledge that many of their practices cannot be transposed elsewhere.

Who will pay for the transition?

The report admits that Anthropic's internal measures, such as redeploying employees or employee share ownership, will not prevent job losses and do not address the slowdown in junior hiring. Its authors call on governments to fund retraining and income support. On October 18, the company announced a public-interest fund endowed with 5% of its shares, subject to shareholder approval. The fund could notably finance responses to labor market instability. Anthropic presents this mechanism as an alternative to the growing stock bonuses that tech companies often grant their executives.

One question remains that the report does not address: if the most advanced companies hire fewer beginners, where will tomorrow's experts be trained?