How AI Is Restructuring the American Labor Market in Ways the Unemployment Rate Doesn’t Capture

As of January, the unemployment rate—that one figure that cable anchors and economists love to toss around—was 4.3%, which on paper seems almost uninteresting. steady. Depending on who you ask, it might be a little soft. However, if you spend any time speaking with recruiters in Manhattan, recent graduates seeking their first paralegal or marketing position, or mid-career copywriters who used to charge $80 per hour, the image becomes less dull. The headline number isn’t designed to see what’s rearranging beneath the surface.

Walking through office districts in cities like Austin and Atlanta gives the impression that although the buildings are still full, the desks inside them now have a different meaning. Three new hires used to fill a junior analyst position annually, but now there is only one. A creative agency that used to have a bullpen of associate writers now has a stack of model subscriptions and one editor. In a dramatic wave, no one was fired. No one applied for unemployment benefits. The jobs simply ceased to be posted.

That’s what the unemployment rate actually fails to account for. It gauges those who actively seek employment but are unsuccessful. The gradual disappearance of ladder rungs is not measured by it. The research team at Goldman has been fairly transparent about this, predicting a roughly 0.6 percentage point increase in unemployment if the adoption of AI takes ten years, and cautioning that the impact becomes more severe if the transition front-loads. Although Anthropic’s own labor-market report, which was published earlier this year, did not find a systematic increase in unemployment among highly exposed occupations, it did note something more subtle and perhaps more significant: hiring of younger workers in those occupations has slowed. There is no slamming of the door. It simply won’t open.

In the meantime, places that no one romanticizes are experiencing a true boom in labor demand. contractors for HVAC systems. Line workers and electricians. Since 2022, more than 200,000 construction-related jobs have been added as a result of the data center build-out, which includes all those windowless gray boxes being built outside of Phoenix, Columbus, and rural Virginia. It’s the type of work where the employee perspires without being asked. It also pays well. The peculiar irony of the AI boom is that its physical presence is generating precisely the type of blue-collar demand that economists have consistently claimed has disappeared forever.

Using a synthetic differences-in-differences approach, Yale’s Budget Lab found no statistically significant labor shocks caused by AI in its May report. The researchers approached it with caution, if not outright caution. They pointed out that basic comparisons were misleading because exposed and unexposed occupations were structurally different in the first place. That prudence seems appropriate. Because the dashboard hasn’t been constructed yet, it’s possible that we’re looking at the first few frames of something massive and interpreting it as nothing.

Observing this unfold, I find it unsettling that the most exposed individuals resemble those the U.S. economy spent forty years advising to become more knowledgeable, educated, and credentialed. administrative and office personnel. Advertisers. Young programmers. Paralegals. They are more likely to be female, to hold a four-year degree, and to have been assured of their safety. We can determine when they have stopped working by looking at the unemployment rate. It won’t let us know when they were initially let go. Additionally, the restructuring will be nearly complete by the time it does.