Anthropic’s own workforce shows AI’s first casualty is the junior hire, and human attention is the new bottleneck

Anthropic has published a report on its own workforce, and its most striking finding concerns who gets hired. According to the company, junior employees fell from 21% to 14% of its technical departments between early 2025 and now. Over the same period, junior hires fell from 27% to 9% of technical hiring. That means fewer than one in ten new technical hires is now at entry level, down from more than one in four.

The report includes other changes too. Technical teams shrank from 61% to 56% of the total workforce. Anthropic credits two things for that: business functions growing as the company scales, and coding tools that let non-technical staff build software themselves. In an April 2026 internal survey, 27% of employees said Claude had made them more productive. Another 20% said their job had shifted ‘from actually doing work to managing Claudes.’ One engineer compared it to babysitting, where every child demands equal attention. Calendar data shows meeting time per employee up 20% from a year ago, and up 16% for technical contributors. The company attributes this to the need for ‘human trust and accountability.’

Anthropic presents itself as a leading indicator. It says its workplace trends may show ‘how AI will shift work organization, hiring, and labor demand across the economy,’ because it ‘has thoroughly molded its workflows around AI.’ That framing deserves caution. Anthropic is an AI lab selling AI tools, and its staff work under what one employee called an ‘eternal pressure’ to ‘use Claude for everything.’ Few companies will adopt AI this intensely or in quite the same way. The productivity claims also need a careful reading. The company cites more code per person, more sales closed per rep and more contracts processed. It also admits that ‘more code does not necessarily mean better code or business impact.’ It says only that it has found no sign that code quality is declining. That is not the same as evidence that business results have improved.

The point that most deserves attention is the one Anthropic makes almost in passing. The report says ‘human attention and judgement’ are becoming the productivity bottleneck, as Claude produces a ‘deluge’ of text that someone has to read and organize. This matches a pattern we have tracked before: a study of 100,000 engineers found agentic coding raised lines of code, commits and pull requests but did not significantly change the number of tasks completed. AI speeds up output. The constraint moves to review, integration and decision-making, which are still human jobs.

Our reading: this is the clearest first-party evidence yet for our employment thesis that AI changes which work matters rather than simply eliminating jobs. It also points to a real short-term problem we will not soften. If the entry rung of the career ladder disappears, the industry has to find a new way to produce senior engineers, because judgment has traditionally been learned by doing the junior work that agents now absorb. The rise in meetings is the hopeful part. When execution gets cheap, coordination, accountability and face-to-face trust become more valuable. Over the long run that points to work centered on judgment and relationships, which fits our view that abundance frees people for higher-value work. Getting there will require companies, universities and governments to redesign how people gain experience, and to do it soon.