The invention of electricity made menial jobs such as lamplighters, elevator operators, and knockers (the human equivalent of modern alarm clocks) irrelevant. Computers made data entry clerks, switchboard operators, and file clerks obsolete.
Anthropic, the artificial intelligence (AI) company that emerged in 2026 as an existential threat to a market worth billions, is back with every stunning new feature in its Cloud model, warning that outdated AI tools could displace vast swaths of jobs. The AI giant was founded by former OpenAI employees who were obsessed with both AI safety and AI advancement. The gap between these two numbers is both reassuring and worrying, depending on your field of work.
In a report titled “The Impact of Artificial Intelligence on the Labor Market: New Measures and Early Evidence,” authors Maxim Massenoff and Peter McCrory found that actual AI adoption is only a small part of the actual capabilities of AI tools.
In theory, AI can cover most tasks in fields such as business and finance, management, computer science, mathematics, law, and office management. However, in most areas, actual adoption rates (which the researchers measured using work-related usage data from Anthropic’s AI model Claude) are only a fraction of theoretical capabilities.
Business leaders have been heeding warnings about artificial intelligence replacing white-collar jobs for months. Anthropic CEO Dario Amodei said last year that the technology could disrupt half of all entry-level white-collar jobs. Mustafa Suleyman, Microsoft’s director of artificial intelligence, made a similar prediction, estimating that most professional jobs will be replaced within a year to 18 months.
The researchers attribute this lag to existing legal constraints and technical barriers, such as model limitations, the need for additional software tools and the need for humans to still vet AI work. But that’s only temporary, they expect.
The study introduces what it calls “observational exposure”—a new metric that compares theoretical AI capabilities to real-world usage data derived directly from Cloud’s interactions in professional settings. The obvious finding is this: AI has only scratched the surface of its technological capabilities. When the gap does close, the workers most at risk are older, more highly educated, and higher-income workers.
The workers who bear the brunt of this situation are not who most people imagine. Compared with the group with the least exposure, those with the most exposure to AI are 16 percentage points more likely to be women, earn 47% more on average and are almost four times more likely to have a graduate degree. That’s lawyers, financial analysts, software developers, not warehouse workers. Computer programmers, customer service representatives and data entry workers are among the most vulnerable occupations.
But even those occupations most exposed to AI’s capabilities have yet to fully experience a workplace reckoning. The researchers gave an example of what they believe is a fully exposed task that doctors typically perform: authorizing a pharmacy to refill a medication. Artificial intelligence can certainly automate this task, but they note that they have not observed Cloud performing it, although it could theoretically be done with large language models.
The results are stunning. For computer and mathematics workers, large language models can theoretically handle 94% of their tasks. However, Cloud currently completes only 33% of these tasks in observed professional use. The same gap exists between office and administrative roles – 90% theoretical ability, only a small percentage is actually used.
The “red zone,” as the researchers describe it, describes actual uses of AI and is dwarfed by the possible “blue zone.” As capabilities increase and adoption deepens, red will gradually fill in with blue, the researchers wrote. On the other hand, 30% of workers have zero exposure to AI — cooks, mechanics, bartenders, dishwashers — jobs that require a physical presence that an LL.M. cannot replicate.
Peter Walker, head of insights at Carta, extrapolated the blue and red findings into bar graphs. “A universal truth: most radar charts should just be bar charts,” he wrote on X. “Love your stuff, Anthropic!”
The paper identifies a scenario that everyone in the knowledge economy should consider: the “Great Recession of White-Collar Workers,” noting that the U.S. unemployment rate doubled from 5 percent to 10 percent during the financial crisis of 2007 to 2009. The researchers note that their framework clearly detects a doubling of the share of occupations in the quartile of occupations most affected by AI – from 3% to 6%. This hasn’t happened yet, but it’s definitely possible.
If you think this is an AI company talking about their book, it’s a distinct possibility in many scenarios, well beyond the viral doomsday articles from the likes of Matt Schumer and Citrini Research. In a speech last month, Federal Reserve Governor Michael S. Barr laid out what he saw as three possibilities for adopting artificial intelligence.
The U.S. Bureau of Labor Statistics released a dismal jobs report on Friday. Employers cut 92,000 jobs in February, and the unemployment rate rose to 4.4%. Some companies have recently announced mass layoffs due to AI. Jack Dorsey’s Block laid off nearly half its staff last month, citing artificial intelligence. “We’re already seeing that the smart tools we’re creating and using, combined with smaller, flatter teams, are enabling a new way of working that fundamentally changes what it means to build and run a company,” Dorsey wrote in a post on
However, the problem, at least for younger workers, is not layoffs but a hiring slowdown in AI-exposed fields, with employment in the post-ChatGPT era down 14% compared to occupations exposed in 2022, the study found. However, the researchers noted that these findings were barely statistically significant. Research shows that so far there has been no systematic rise in unemployment. Citadel Securities isn’t known for publishing market research, but it was moved by a viral doomsday article to note that hiring of software engineers has actually increased in recent months.
Still, human researchers say the slight decline could signal a new reality for employment in the age of artificial intelligence, as it echoes other research on the state of the job market for younger workers. A similar study found that among workers aged 22 to 25, employment in jobs exposed to artificial intelligence fell by 16%.
For some young workers, this means avoiding the labor market altogether. “Young workers who are not employed may remain in their current jobs, take a different job, or return to school,” the researchers said.
This story originally appeared on Fortune.com
