NYT: AI Is Reshaping the Economy, but the Data Can't Keep Up

Ben Casselman's piece in The New York Times sets up the frustration cleanly: pretty much everyone agrees AI has the potential to reshape the economy in the coming decades, but no one is sure what effect the technology is having right now. Generative AI has taken less than four years to go from a novelty useful mostly for writing limericks to a tool adopted by the world's largest corporations, and the statistical apparatus we use to measure the economy has not caught up.

The reason the debate feels unresolved is that different data sources point in opposite directions. Some measures suggest AI is already destroying tens of thousands of jobs and is showing up in unusually high unemployment among new graduates. Other measures suggest companies adopting AI most aggressively are hiring more people, not fewer. Casselman leans on newer datasets to make the point. The Yale Budget Lab's monthly occupational churn measure, designed as an early warning system for shifts in the job mix, has so far found no clear connection between AI usage and changes in employment or unemployment. Ramp and Revelio Labs data go further in the other direction, showing that heavy AI investors have grown employment faster than laggards, entry-level hiring included.

Why any of this matters if you are not a labor economist: the political and corporate decisions being made right now, from workforce planning to regulation, are being made on top of numbers that were built for a slower-moving economy. Government statistics are backward looking by design and better at broad trends than at specific sectors or regions, and Casselman notes that the federal statistical system is under real strain from falling survey response rates and funding pressure. That is a bad combination when the technology you are trying to track is being adopted this quickly.

The caveat the piece is careful about, and worth keeping in mind, is that the Ramp and Revelio data skew toward tech-savvy firms, so the 'AI adopters hire more' finding is not automatically representative of the whole economy. What the reporting does not resolve is which specific occupations are actually shrinking, or how much of the current confusion is a productivity J-curve, meaning firms are still in the experimentation phase and the gains are yet to show up in output figures.

The useful takeaway is a posture rather than a forecast. Treat any confident number about AI's current effect on jobs, in either direction, as provisional, and pay more attention to the small set of alternative datasets trying to close the gap than to the headline unemployment print. The story of AI's economic impact is being written in data we do not yet know how to read.