The AI ​​skills gap is here, AI companies say, and power users are leading the way.

A recent study by Anthropic shows that while AI is rapidly changing the way work is done, it is not meaningfully eliminating jobs. At least not yet. But behind the “still healthy” labor market, there are early signs that it is having an uneven impact, particularly on younger workers just entering the labor market, said Peter McCrory, head of economics at Anthropic.

In an interview on the sidelines of the Axios AI Summit in Washington, D.C., McCrory said the company’s latest economic impact report found little evidence of widespread job displacement so far.

“There is no real difference in unemployment rates” between workers who use Claude for “the most core tasks of their jobs in an automated manner,” such as technical writers, data entry clerks, and software engineers, and workers in occupations less exposed to AI that require “physical interaction with the real world and dexterity.”

But that could change quickly as AI adoption spreads across industries. If Anthropic CEO Dario Amodei is to be believed, AI could eliminate half of all entry-level white-collar jobs and raise unemployment to 20% in the next five years.

“Displacement effects can materialize very quickly, so we need to build a monitoring framework to understand them before they materialize and ideally identify appropriate policy responses,” McCrory told TechCrunch.

Staying ahead of these trends is why it’s important to track AI growth, adoption, and diffusion, he said.

McCrory said that in theory, an AI model like Claude could do almost anything a computer can do. In reality, most users only scratch the surface of these features.

He said Anthropic looked into what tasks AI was particularly good at, what tasks were already being automated, and what roles were tied to real-world workplace use cases.

Anthropic’s fifth economic impact report, released Tuesday, found a growing skills gap between early Claude adopters and newcomers, even in places where there isn’t yet much migration.

Early adopters are more likely to get much more value from their models by using them for work-related tasks rather than casual or one-off purposes, and by using them in more sophisticated ways, such as as “thinking partners” for iteration and feedback.

McCrory said the findings show that AI is becoming a technology that rewards those who already know how to use it, and that employees who can effectively integrate AI into their work will increasingly have an advantage.

The benefits are not evenly distributed geographically. The report also found that “Claude is used more intensively in high-income countries, in places with a large population of knowledge workers in the United States, and for a relatively small number of specialized tasks and occupations.”

That said, despite the promise of equalizing AI, adoption may already be leaning towards the wealthy, and these benefits could be amplified as power users move further ahead.