The U.S. Isn’t Ready for the AI Jobs Shock
AI may disrupt millions of careers. But America’s job-training system is barely equipped to help workers adapt.
The uncomfortable truth: Washington already spends tens of billions of dollars on workforce training—and most of it doesn’t work very well.
A recent analysis of 56 randomized trials found that U.S. workforce programs increased long-term employment by just 1.7 percentage points and annual earnings by about $800. Large-scale government programs evaluated in the study produced no significant lasting results.
That matters because AI is about to make the need for effective retraining much bigger.
The answer isn’t to abandon job training. It’s to stop paying for programs that don’t deliver.
What works is already hiding in plain sight
Some sector-based training programs have produced dramatically better results. Their model is straightforward: Work directly with employers to identify the skills companies actually need, train people intensively, and measure whether graduates end up in better-paying jobs.
These programs also recognize something traditional workforce policy often misses: getting someone trained isn’t the same as getting them ahead.
Successful programs may provide transportation and child care, operate on a months-not-years timetable, selectively admit applicants and continually measure outcomes. The focus is employment, earnings and mobility—not simply enrollment or graduation.
Per Scholas offers one example. The nonprofit provides free IT training in 25 cities and has reported that participants earned 16% more than comparable peers over the decade after graduation. But it reaches only a few thousand students annually.
The challenge, then, isn’t simply scaling one organization.
It’s scaling the discipline that makes programs like it work.
Pay for results—not participation
America’s workforce system is fragmented across federal, state and local programs, many with overlapping missions and disconnected incentives.
That makes accountability difficult.
A review of states’ 2024 plans under the Workforce Innovation and Opportunity Act found that only 13 states explicitly included minimum wage-outcome criteria. Programs can therefore continue receiving public money even when participants remain in or near poverty after completing training.
That is a design problem.
If taxpayers are funding workforce programs, those programs should have clear performance standards: Do participants finish? Do they find jobs? Are those jobs connected to employer demand? And, most important, do their earnings actually improve?
Recent policy changes point in that direction. Expanded Pell Grant eligibility for short-term training now includes requirements involving completion, job placement, employer alignment and earnings. Similar accountability should extend across the workforce system.
The AI transition demands a different bargain
Nobody knows exactly how quickly AI will reshape employment—or which occupations will change most. But the uncertainty itself makes effective worker mobility more important.
America doesn't need another alphabet soup of training initiatives.
It needs a workforce system that can prove its programs are changing people’s economic trajectories.
The principle is simple:
If a program consistently produces better jobs and higher earnings, fund it. If it doesn’t, fix it—or stop funding it.
AI may transform the labor market. The least we can do is make sure the system designed to help workers navigate that transformation is capable of producing results.