Anthropic 内部研究：AI 智能体工时达员工六倍，初级技术岗位占比降至 14%

Anthropic is hiring fewer and fewer junior employees in technical roles, while the hours logged by its AI agents keep climbing.

According to an internal study published by the AI company, the share of junior positions in technical departments dropped from 21 percent to 14 percent since early 2025. New hires took a bigger hit. The junior share there fell from 27 percent to just 9 percent. Open junior roles declined especially sharply starting in December 2025.

Anthropic points to the growing capability of its AI agents as the reason. By August 2026, those agents were logging six times as many work hours as human employees. Tasks like writing code and finding bugs are increasingly handed off to Claude. Junior hiring declined the most in roles where Claude handles the most complex work.

Productivity rose across every department, according to Anthropic. Engineers shipped 5.1 times as many code commits, data scientists ran 3.1 times as many queries, and the legal team processed 3.1 times as many contracts. Sales closed 1.7 times as many deals, recruiters hired 1.4 times as many people, and accounting handled 1.4 times as many invoices.

The study itself admits that Anthropic is a special case, though. It's an extremely AI-heavy software company, and the results don't map easily onto the broader economy. Rapid business growth and voluntary adoption of AI tools could also be skewing the numbers. So far, the wider economy hasn't shown measurable productivity gains from AI .

Fewer junior roles now means fewer experts later

Researcher Nolan Lovett of the NATO Special Operations University recently described a "tragedy of the cognitive commons" in a research paper. When individual companies replace entry-level positions with AI, they benefit from the efficiency gains while the costs get spread across everyone who needs experienced talent later. Anthropic fits that pattern.

Skilled professionals only develop through years of practice in the very junior roles that are now disappearing. Lovett argues the full impact won't show up until 2030 to 2045 because today's available experts were trained 5 to 20 years ago. Software engineering, financial analysis, and legal research are the most exposed fields.

Lovett's prediction only holds if humans are still doing this kind of work in the future, though. That's something Anthropic itself probably doesn't expect: the company has suggested in the past that programmers and call center workers would eventually need to move into jobs like electrician or nurse.