Anthropic's AI Workforce Experiment is Already Squeezing Out Junior Jobs as Agents Work Six Times More Than Staff

Anthropic's AI agents are now working six times as many hours as the employees directing them, while the share of junior workers among new technical hires has plunged from 27% to 9% — offering an early glimpse of how AI could reshape white-collar hiring, productivity and the structure of work.

The findings come from Anthropic's new report, which draws on internal employment, AI-use, output, survey, communication, calendar and hiring data, alongside interviews with 17 employees. The company says its own workplace could act as a "leading indicator" of how AI changes work across the wider economy.

Hiring Fewer Junior Technical Workers

The biggest employment shift is happening at the entry level. Anthropic said the share of junior employees within its technical departments dropped from 21% in early 2025 to 14%, while the proportion of new technical hires who were junior fell much faster, from 27% to 9%. Job postings for junior roles also declined sharply from December 2025.

The shift began as teams increasingly found that Claude could handle work previously assigned to junior technical staff, including writing code and finding bugs. By April, engineering and research teams had reduced emphasis on lower-band technical positions, while around a third of ML managers and a fifth of software-engineering managers reported hiring less because of AI.

"The thing that raised eyebrows was the uplift for performance engineers in particular and it basically overnight changed what we were looking for," Anthropic quoted one employee as saying.

At the same time, technical departments accounted for 56% of Anthropic's workforce, down from 61% a year earlier. Technical hiring continued to rise in absolute numbers, but sales, finance and recruiting grew faster.

AI is also allowing non-technical employees to perform work that once required engineers. One sales representative built a customized dashboard and pipeline using AI in three to four hours, despite saying such a project had previously been outside their expertise.

Coding-agent sessions among non-technical employees jumped from two per month in August 2025 to 90 in May 2026.

AI Agents Now Outwork the Humans Directing Them

Anthropic's agent data shows how deeply AI has penetrated its operations. In March 2026, autonomous and interactive agent work hours surpassed human work hours for the first time. By August, agents directed by individual employees were working six times as many hours as their human handlers.

The agents are being used for large-scale software engineering, testing, infrastructure maintenance and code translation, while personal assistants handle calendars, messages, scheduling and meeting preparation.

The productivity gains extend beyond coding. Anthropic said engineers' code commits rose to five times their velocity six months earlier, while lawyers processed more contracts, accounts-payable workers handled more invoices and salespeople closed more deals.

A study following 192 engineers who adopted coding agents and 74 who did not found that the amount of code added rose sharply after adoption. Anthropic cautioned that more code does not automatically mean better code or greater business impact, but said it found no difference in certain measures of code subsequently removed or reverted.

New Scarce Resource is Human Attention

The AI-driven productivity surge has created an unexpected constraint -- humans cannot keep up with the work their agents generate.

Anthropic said interactive Claude sessions spend around 80% of their time waiting for human input, meaning the agents themselves are often ready to continue while employees become the limiting factor.

One employee described the situation as having "an army of Claudes, with every one asking for my attention." Another said: "things we expect in a week are things that would take three months elsewhere."

The company found that even non-technical employees generate enough AI output to require more than half a working week to read everything. As a result, workers are increasingly doing what Anthropic calls "meta-work", building tools to monitor agents, writing instructions, creating dashboards and reviewing one agent's work with another.

Meta-work rose from 10% of employees' average interactive Claude sessions in fall 2025 to 22% in spring 2026. Meanwhile, the number of written standing instructions for agents grew from a few hundred in July 2025 to 8,500 a year later.

The change is also reshaping what Anthropic wants from new hires. Hiring managers increasingly emphasize autonomy and judgment rather than narrow technical expertise and implementation skills.

Mentions of candidates' AI-tool experience appeared in about 5% of technical assessments and 7% of non-technical assessments in 2024; by 2026, those rates had tripled for technical roles and quadrupled for non-technical roles.

The second Anthropic study shows why its employees may see this transformation differently from the wider public. On average, Anthropic staff expect AI to perform 75% of their current work by 2030, compared with 28% among U.S. workers. Ninety-one percent of Anthropic staff expect AI to make a Nobel-level discovery by 2030, versus 39% of U.S. adults.

Yet Anthropic employees do not expect automation to mean the company stops hiring. Ninety-five percent expect Anthropic's headcount to grow over the next 12 months, with the median employee expecting 66% growth. Anthropic stresses these are employees' personal expectations, not company plans.

The broader warning in the research is therefore not simply that AI replaces workers. It is that the composition of work may change before total employment does, with fewer entry-level technical jobs, broader responsibilities for remaining employees and a growing premium on people who can direct, evaluate and govern fleets of AI agents.

As Anthropic's researchers put it, "Employees increasingly see their work as managing teams of Claudes," while human attention and judgment remain the bottlenecks the system struggles to overcome.