Peter McCrory spent 18 months reviewing Bureau of Labor Statistics data, occupation-level unemployment figures, and Anthropic’s own internal research on how workers actually use Claude. On July 24, he published what he found. It wasn’t what his boss had been predicting.
McCrory is Anthropic’s head of economics. In a lengthy essay on X (the former Twitter), he laid out the data and concluded the U.S. labor market has not yet taken a visible hit from AI.
His CEO, Dario Amodei, has spent the past year repeatedly warning that a white-collar jobs crisis is coming fast and coming hard, Fortune reported. The data McCrory found tell a different story.
What Peter McCrory found when he looked at U.S. labor market data
McCrory started with the basics. The U.S. unemployment rate sat at 4.2% in June. The Federal Reserve considers that full employment. Job openings roughly matched the number of unemployed workers. Prime-age employment was near multi-decade highs.
He also ran a more specific test. McCrory looked at unemployment rates among workers whose jobs have the highest concentration of tasks that Claude is used to automate. He compared those workers to people in roles with less AI exposure. He found no relative deterioration in the more-exposed group.
“I don’t expect unemployment to be noticeably higher a year from now — at least not because of AI,” he wrote.
McCrory traces that finding to what he calls AI’s “stubbornly jagged” capability profile, a term borrowed from Wharton professor Ethan Mollick.
No occupation in the Labor Department’s taxonomy has all of its tasks handled by Claude. When McCrory looked at how people actually use Claude at work, the pattern was workers bringing it into their process to iterate and refine, not handing entire tasks over to it.
Anthropic CEO Dario Amodei made very different prediction on AI jobs impact
Amodei has not been quiet about where he thinks this is heading. In May 2025, he told Axios AI could eliminate half of all entry-level white-collar jobs and push unemployment to somewhere between 10% and 20% within one to five years. He said companies and policymakers were sugarcoating the risk and needed to stop.
In January 2026, he published an essay calling AI a “general labor substitute.” He said it would push work from lower-skill roles up toward upper ones, potentially leaving workers without jobs or stuck on very low wages for good.
By June 2026, he was calling for universal basic income and wage insurance. He said significant job loss might be “an intrinsic property of the technology.”
McCrory’s data don’t prove that wrong. What they show is that the crisis scenario Amodei has described hasn’t arrived yet, at least not in the aggregate labor statistics. Both men point to the same vulnerable group: early-career workers in AI-exposed roles. The disagreement is over how bad it will get, and how fast.
Eric/Getty Images
Where early warning signs already appear in U.S. employment
McCrory isn’t saying everything is fine. Hiring has softened for young workers in roles with high AI exposure over the past year. Stanford researchers studying the same trend have called those workers “canaries in the coal mine.”
The Bureau of Labor Statistics projects slower employment growth through 2034 for technical writers, data entry workers, and customer support roles. Those are exactly the categories McCrory’s analysis flags as most exposed to AI automation.
McCrory also points to a growing gap between workers who use AI as a core part of how they work and those who don’t. Power users are getting more productive. Everyone else is mostly staying the same. That shows up first in hiring and wages, before it ever reaches unemployment data.
What a delayed AI labor shock means for U.S. economy
McCrory’s data raises a specific question. If AI is lifting productivity among a subset of workers without causing broad job losses yet, where are the economic gains going?
Companies with AI-fluent workers are producing more without hiring more. That runs straight to the earnings line. But that gain is sitting inside a relatively small group of companies and workers. The broader consumer economy isn’t seeing the same lift.
Most of the productivity gain is concentrated among high-skill, high-income workers. If that stays true, the income gap between AI-fluent workers and everyone else keeps widening.
Spending by lower and middle-income households tends to be more consumption-driven, so a widening wage gap at the bottom eventually shows up in slower consumer spending growth, which is an economic problem that compounds over time.
McCrory’s essay points to one more practical implication. If the job disruption is real but still building, companies and policymakers have more time to respond than Amodei’s timeline suggests.
Retraining, education, and safety net adjustments are all easier to build before unemployment rises than after. That window is open right now. Nothing in McCrory’s data says it stays open forever.
Related: Mark Cuban has strong words on AI companies and job losses
