The predictions have been loud and consistent. AI was going to hollow out white-collar work, kill coding jobs, and fundamentally rupture the relationship between labor and capital. The technology, we were told, would not merely change work — it would end large portions of it.
The record, so far, does not cooperate.
Peter McCrory, head of economics at Anthropic, has published an essay confronting what he calls the central mystery of this moment: why hasn't AI increased unemployment? McCrory is not a skeptic of AI's power — he runs the research team at one of the most consequential AI companies in the world, and he studies, in his own words, how AI is reshaping work 'including my own.' His team digs into how people and businesses actually use AI, then sets that against broader economic statistics.
What he finds is a genuine puzzle. One in five American companies now uses AI in at least one part of their business. In the information sector, that share climbs to two in five. Quality-adjusted AI output — a measure of how much more capable the technology became, not merely how widely it was deployed — grew over 2,000 percent in both 2024 and 2025. These are not marginal numbers. By any reasonable standard, effects on the labor market should already be visible.
They are not, or not yet. The American job market, McCrory writes, is 'remarkably stable.'
He offers one observation that anchors the analysis: 'Despite the incredible advance of AI and its rapid adoption throughout our economy, there is not one job for which all associated tasks are systematically handled by AI.' That sentence deserves a second read. Not one job — not a category, not a sector, not a skill set — has been fully automated out of existence. The technology is transforming tasks within jobs. It is not, at least not yet, eliminating the jobs themselves.
McCrory speaks from direct experience. As an economist, he has watched Claude and successive frontier models automate time-consuming work he once did himself — estimating statistical models, solving systems of mathematical equations, building visualizations. 'While I'm still directing my own research,' he writes, 'the work is increasingly being done by AI.' The direction remains human. The execution is shifting.
The honest question he poses — is today's stability the calm before the storm, or are the apocalyptic predictions simply wrong? — does not yet have a clean answer. The data he cites establishes the puzzle without resolving it.
---
Say it plainly: the AI-kills-jobs narrative has been more useful as a political and cultural instrument than as an economic forecast. It has fed regulatory appetite, justified expanded bureaucratic oversight of the technology sector, and given progressive institutionalists a lever to slow an industry they did not build and do not control.
The actual evidence, assembled by someone whose job is to find it, points somewhere more interesting and less catastrophic. AI is changing what work looks like from the inside — compressing tasks, shifting the ratio of human direction to human execution — without, so far, producing the mass displacement that was supposed to demand a policy response. Free markets, it turns out, are absorbing a 2,000 percent capability shock with considerably more resilience than the doomsayers projected. That is not a reason for complacency. It is a reason to follow the evidence rather than the press release.
