Experts Question Anthropic’s AI Work Report

Anthropic said in a report published on Thursday that it is hiring fewer junior technical workers and that human attention has become a “bottleneck” as the company relies more on Claude AI agents. The report says that from May 2026, agents logged more work hours than human employees. Employees spend about 22% of their time managing Claude agents, more than double the amount a year earlier, based on Anthropic’s analysis of chat summaries.

Anthropic said it produces more code per person, closes more sales per representative and processes more contracts through what it calls “deep adoption of AI.” It said outsourcing some coding and sales tasks to agents raised productivity within a few months. Since March 2025, when Claude Code had been released, entry-level technical hires fell from nearly 30% to 9%, while the share of headcount fell from about 21% to 14%. Anthropic attributed those changes to AI use and described workers as a “thin layer” atop a growing group of agents.

The report suggests wider AI adoption could make software engineering jobs harder to secure and make it more difficult for entry-level tech workers to gain skills and experience. It presents single-person startups run by founders directing groups of agents as one possible future. However, several researchers questioned whether the company’s findings predict changes across other businesses. They said the report lacks independent outside review and does not show how Anthropic employees feel about their work or use the models. Anthropic did not respond to The Guardian’s request for comment.

Morgan Frank of Case Western Reserve University said a paper under peer review found job markets exposed to AI were deteriorating before ChatGPT launched in November 2022 and Claude followed a few months later. State unemployment claims began rising at the start of 2022, the paper says. Frank said ChatGPT’s launch did not correspond to a major increase in job losses or worsening outcomes, and cited pandemic-era hiring cuts and “AI-washing,” when employers blame AI for layoffs they might have made anyway, as possible factors.

Yale social scientist Julián Posada said the report came from a company seeking investors ahead of a possible stock market debut as soon as November. Dylan Baker of the Distributed AI Research Institute called the shift toward agents and fewer entry-level hires “shortsighted” and disputed the report’s description of humans as the bottleneck in software work. The report proposes employee stock ownership as one possible way to make AI adoption benefit workers.

The report appeared one day after Anthropic pledged 5% of its shares to the “public good,” without giving details. A draft S-1 filing showed an $80bn net loss in the first half of 2026, and Anthropic received a junk credit rating in June. Investors now project a $1.5tn valuation, down from earlier estimates of $2tn or more, amid slowing revenue growth and concerns about cloud partner CoreWeave’s debt and falling stock. The Guardian also noted that a majority of Americans are wary of AI, particularly over job losses, citing Pew Research Center.

Anthropic said in a report published on Thursday that it is hiring fewer junior technical workers and that human attention has become a “bottleneck” as the company relies more on Claude AI agents. The report says that from May 2026, agents logged more work hours than human employees. Employees spend about 22% of their time managing Claude agents, more than double the amount a year earlier, based on Anthropic’s analysis of chat summaries.

Anthropic said it produces more code per person, closes more sales per representative and processes more contracts through what it calls “deep adoption of AI.” It said outsourcing some coding and sales tasks to agents raised productivity within a few months. Since March 2025, when Claude Code had been released, entry-level technical hires fell from nearly 30% to 9%, while the share of headcount fell from about 21% to 14%. Anthropic attributed those changes to AI use and described workers as a “thin layer” atop a growing group of agents.

The report suggests wider AI adoption could make software engineering jobs harder to secure and make it more difficult for entry-level tech workers to gain skills and experience. It presents single-person startups run by founders directing groups of agents as one possible future. However, several researchers questioned whether the company’s findings predict changes across other businesses. They said the report lacks independent outside review and does not show how Anthropic employees feel about their work or use the models. Anthropic did not respond to The Guardian’s request for comment.

Morgan Frank of Case Western Reserve University said a paper under peer review found job markets exposed to AI were deteriorating before ChatGPT launched in November 2022 and Claude followed a few months later. State unemployment claims began rising at the start of 2022, the paper says. Frank said ChatGPT’s launch did not correspond to a major increase in job losses or worsening outcomes, and cited pandemic-era hiring cuts and “AI-washing,” when employers blame AI for layoffs they might have made anyway, as possible factors.

Yale social scientist Julián Posada said the report came from a company seeking investors ahead of a possible stock market debut as soon as November. Dylan Baker of the Distributed AI Research Institute called the shift toward agents and fewer entry-level hires “shortsighted” and disputed the report’s description of humans as the bottleneck in software work. The report proposes employee stock ownership as one possible way to make AI adoption benefit workers.

The report appeared one day after Anthropic pledged 5% of its shares to the “public good,” without giving details. A draft S-1 filing showed an $80bn net loss in the first half of 2026, and Anthropic received a junk credit rating in June. Investors now project a $1.5tn valuation, down from earlier estimates of $2tn or more, amid slowing revenue growth and concerns about cloud partner CoreWeave’s debt and falling stock. The Guardian also noted that a majority of Americans are wary of AI, particularly over job losses, citing Pew Research Center.