Anthropic cut junior tech hires from 27% to 9% as Claude took on the code

Anthropic, the company behind the Claude chatbot, now hires far fewer beginners for its technical teams. Junior roles made up 27% of its technical hires at the start of the period the company studied, and that share has since dropped to 9%, according to a report Anthropic published about its own staff. The company that sells AI coding agents to everyone else has become the first large employer to show, from its own payroll, what happens to entry-level jobs once those agents do the work.

For anyone hoping to start a career in software, that is the figure to watch. Anthropic’s teams found that writing code and hunting bugs, the tasks that normally train new engineers, could increasingly be handed to Claude, and engineering and research teams stopped prioritising hires for lower-level technical posts. Around a third of machine-learning managers and a fifth of software-engineering managers told an internal survey they were hiring less because of AI. Across technical departments, the share of junior employees fell from 21% to 14%.

The agents, meanwhile, are working long shifts. Counting only the sessions that employees start and steer themselves, Claude logged six times as many working hours as the people supervising it. Code commits per engineer now arrive five times faster than six months before, and the report says lawyers process more contracts and salespeople close more deals. Staff with no code commits to their name went from 2 coding-agent sessions a month to 90, and one sales representative built a custom dashboard in three or four hours that they said was out of their wheelhouse.

All that output has to be read by someone, and the report names human attention as the new bottleneck. Interactive Claude sessions spend around 80% of their time idle, waiting for a person to answer. Simply reading everything Claude writes in those sessions would take more than half the work week, even for employees in non-technical jobs. One engineer described facing “an army of Claudes, with every one asking for my attention,” and compared it to babysitting, where a request to open a juice box is “voiced with the same level of concern as the one who has sliced their finger and is bleeding.” Written standing instructions for agents grew from a few hundred to 8,500 in a year, and the share of sessions spent on this kind of meta-work, from writing specs for agent runs to monitoring pipelines, rose from 10% to 22%.

One thing AI did not shrink is the meeting. Calendar data shows the median employee spends 20% more time in meetings than a year earlier, and engineers without management duties spend 16% more. Employees repeatedly described face-to-face coordination as indispensable, and the report notes that a break from managing the agents “typically looks like jumping into a meeting.” TechCrunch led its coverage of the report with the meetings figure.

What Anthropic looks for in new hires has changed too. Hiring managers now write more about a candidate’s autonomy and judgment and less about domain expertise, and comments on how well applicants use AI tools have tripled for technical roles and quadrupled for non-technical ones. Anthropic’s own employees expect AI to be able to do 75% of their work by 2030, against 28% for other Americans surveyed.

The numbers deserve caution. The report was written by Anthropic’s own researchers about a company whose business is selling these agents, and no outside group has checked the data. Anthropic tripled in size over the past year and pays staff partly in stock, which makes it a poor stand-in for a typical employer. The authors concede that engineers who moved to fully autonomous agents were already on a steep upward path before they did, which flatters the productivity gain, and their test for sloppy AI code, which found no rise in deleted or reverted lines, rests on a handful of proxies. The company’s main answer to disruption, an early-warning system that moves existing staff into new roles, does nothing for people who were never hired. The report says so itself: “Employment protections for incumbents will not benefit those not hired to begin with.”

The study, published by the Anthropic Institute on October 16, was written by economists Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory. It draws on internal hiring, calendar, Slack and agent data from early 2025 to this summer, plus 17 employee interviews. Postings for junior roles have fallen especially sharply since December 2025, and engineering and research teams deprioritised lower-level hiring by April 2026. Anthropic says it will keep running its early-warning system on roles at risk and wants governments to pair it with retraining and income support. For the juniors it no longer hires, the company’s own remedy is a request to someone else.

Tags: Anthropic, AI agents, Claude, Entry-level jobs, Future of work, Anthropic Institute