Anthropic’s head of economics, Peter McCrory, just delivered a data-driven rebuttal to the loudest doomsday story in AI. In a lengthy X essay, McCrory writes that Anthropic “don’t see significant impact of AI on the U.S. labor market,” at least not yet. That claim lands in direct tension with Anthropic CEO Dario Amodei’s repeated warnings that a white-collar labor shock is coming, possibly severe enough to justify major policy responses.

McCrory backs up the “not yet” part with current labor conditions and with comparisons inside the labor market itself. He points to an unemployment rate of 4.2% in June, a level the Federal Reserve associates with full employment. He also notes that job openings roughly match the number of unemployed workers, and prime-age employment sits near multi-decade highs. Most pointedly, McCrory argues updated analysis using more recent Bureau of Labor Statistics data shows no relative deterioration in unemployment for workers whose jobs contain a large share of tasks that Claude is used to automate, compared with workers in less-exposed roles. “I don't expect unemployment to be noticeably higher a year from now-at least not because of AI,” he wrote. In other words: if AI had already started wholesale substituting out entry-level white-collar workers at the pace Amodei has previously forecast, you would expect the unemployment patterns to show it. McCrory says they do not.

So how did Anthropic end up with two competing timelines from two senior voices? Start with Amodei’s trajectory. The prediction that kicked off the current fight is rooted in Amodei’s earlier public stances on labor impacts. In May 2025, he told Axios that AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10% to 20% within one to five years, urging companies and policymakers to stop “sugarcoating” the risk. He doubled down in a January 2026 essay, “The Adolescence of Technology,” warning that AI functions as a “general labor substitute for humans,” displacing work from lower tiers of skill to the upper, and potentially creating a lasting underclass of unemployed or very-low-wage workers.

But by this May, Amodei moderated his framing somewhat. McCrory’s essay points to a shift toward a “multiplier” logic, drawing on the Jevons paradox recently repopularized by Apollo Global Management’s Torsten Slok. The paraphrased explanation McCrory’s piece cites is essentially this: “If you automate 90% of the job, then everyone does the 10% of the job,” and the “10% kind of expands to be 100% of what people do,” translating into roughly “10-times their productivity.” That multiplier story is closer to an augmentation model than a pure substitution model. Yet Amodei did not settle there. In June, he re-escalated again, arguing significant, enduring job loss might be “an intrinsic property of the technology” itself and calling for government responses including wage insurance and universal basic income.

McCrory’s analysis challenges the substitution-style “general labor substitute” reading, especially on the magnitude and timeline. The core of his technical argument is about how AI capabilities actually land on real jobs, not just on paper. He describes AI’s “stubbornly jagged” capability profile, borrowing the term made famous by Wharton’s Ethan Mollick. In Anthropic’s framing, no job in the Labor Department’s O*NET taxonomy has all of its tasks handled by Claude. Complex work still depends on human oversight to direct systems and catch their errors. In that world, AI is less like a single lever that removes a job and more like a patchwork tool that changes task mix, workflow, and who needs what kind of judgment.

McCrory also points to evidence that Claude usage correlates with users acting as “thought partners” rather than replacements, and that people with more domain expertise succeed more often and recover better when the AI stumbles. That is, in his telling, the opposite of a scenario where AI simply substitutes for human labor. And that is exactly why his findings matter for executives watching their own labor-risk narratives: McCrory’s occupation-level unemployment results would be harder to reconcile with a wholesale replacement story. If entry-level consultants, lawyers, and financial analysts were being systematically substituted out at Amodei’s earlier scale, then unemployment would likely show divergence across occupations with higher exposure to Claude-enabled tasks. McCrory says it doesn’t.

There is, however, an important wrinkle that keeps this from turning into a clean victory lap for either side. McCrory concedes that hiring has already softened for young workers in AI-exposed roles, which is the population that a slow, aggregate-level “bigger pie” effect might not protect. He also notes Bureau of Labor Statistics projections for slower growth through 2034 for occupations like technical writers, data entry workers, and customer support reps. So the “no big unemployment spike” conclusion does not mean “no labor pain.” It suggests the disruption may be uneven, landing first on early-career cohorts and specific job families, rather than showing up as an immediate, economy-wide unemployment shock.

Then there is the elephant both men acknowledge in different ways: uncertainty about the future. McCrory concedes that if AI begins automating innovation itself through recursive self-improvement, standard economic models allow for a scenario resembling the “singularity” that Amodei fears. McCrory’s view is simply that it would not arrive on the near-term timeline or scale Amodei has publicly forecast. That leaves Anthropic publicly straddling two positions: a data scientist in-house saying the disruption has not arrived in aggregate labor statistics, while a chief executive insisting the disruption is coming fast enough that it could justify major policy measures.

For boards, investors, and founders, the strategic stakes are obvious even if you never care about unemployment rates for their own sake. When a category shift is being debated in public, the mismatch between narrative and data can change regulation, hiring strategy, and capital allocation. McCrory’s essay implies executives should separate “catastrophe already happening” from “uneven transition already underway,” because those lead to very different decisions. In a labor market, the timeline matters, the exposed groups matter, and the mechanism matters. If your planning only assumes one big unemployment cliff, you might miss the more likely early signal: hiring softness in entry-level roles, task reshuffling, and productivity gains that do not distribute evenly.