As a wave of companies blames workforce cuts on artificial intelligence, a new Anthropic report argues it is still too early to draw firm conclusions about AI’s impact on jobs. Released on March 5, 2026, and authored by economists Maxim Massenkoff and Peter McCrory, the report introduces a metric called “observed exposure” to measure not just which tasks AI could theoretically handle, but which are actually being automated in practice.
Measuring observed exposure
The observed-exposure metric draws on three data sources: whether a large language model such as Anthropic’s Claude can perform a task roughly twice as fast as a human, how Claude is actually used in work-related settings, and how job descriptions reflect AI use in specific roles. The central finding is a wide gap between capability and adoption. For computer and math workers, the report estimates that large language models are theoretically capable of handling about 94 percent of tasks, yet Claude currently covers only around 33 percent of those tasks in observed professional use.
At the occupation level, computer programmers show the highest observed exposure at roughly 74.5 percent, followed by customer service representatives (about 70 percent), data-entry keyers (about 67 percent), and medical-record specialists (about 67 percent). In short, actual AI use in the workplace is lower than many assume, even in highly exposed roles.
Exposure has not yet become unemployment
Critically, the report found no systematic increase in unemployment among workers in heavily exposed occupations since late 2022, including among groups concentrated in those roles, who tend to be older, more educated, better paid, and more likely to be women. It did, however, find suggestive evidence that hiring of younger workers has slowed in the most exposed fields. The authors frame this as a signal worth watching rather than proof of widespread displacement.
A nervous job market
The report lands amid an unsettled labor market in which several companies have tied layoffs to AI. On February 26, 2026, Block chief executive Jack Dorsey announced cuts of about 4,000 employees, nearly half the company’s workforce, attributing the move to AI-driven efficiency and predicting that most companies would make similar structural changes within a year. Other firms, including Oracle, Pinterest, Salesforce, and HP, have cited AI in connection with layoffs or restructuring plans.
Experts caution against reading too much into these announcements. “Given that we are not yet three and a half years into the AI era, it is difficult to trust any quantitative measures of the impact on labor,” said Michael Bennett, associate vice chancellor for data science and artificial intelligence strategy at the University of Illinois Chicago. Bennett added that some employers may be using the idea of AI displacement to justify cuts they would have made anyway, “while others are simultaneously seeing signs of real obsolescence in their workplaces and handing out pink slips.”
Concern is especially acute among programmers and engineers, as AI coding tools such as Anthropic’s Claude Code and OpenAI’s Codex grow more capable. Bennett described the observed-exposure metric as a useful way to identify the changes AI is making in the workplace, while calling for “more and more granular metrics” to track the shift over time.
Limitations and what to watch
Several caveats apply. The report measures exposure and usage, not causation, and explicitly notes that the AI era is too young for reliable labor-impact estimates. Its figures depend on usage data from one company’s model, which may not capture the full market, and “observed exposure” reflects current adoption that could change quickly as tools improve and diffuse. Layoff announcements that cite AI can also conflate genuine automation with cost-cutting or favorable public framing, making them an unreliable proxy for real displacement. The more durable takeaway is that capability currently outpaces adoption, and that early signals, such as slower hiring of younger workers, deserve close monitoring rather than firm conclusions. Related coverage on this site examines how AI is reshaping roles in data and analytics.
The full study is available from Anthropic.
