What A.I. Is Actually Doing to the Economy

What It Covers

NYT economics correspondent Ben Casselman examines why AI's actual economic impact remains difficult to measure, using two contrasting 1990s case studies — the gradual Internet revolution and the rapid China trade shock — to frame two possible futures for AI-driven labor disruption across the U.S. economy.

Key Questions Answered

• •Data Infrastructure Gap: U.S. government economic data, including the monthly jobs report, does not track the tech industry as a standalone category — it is split across information, professional services, and manufacturing sectors. Policymakers and workers relying on official data to detect AI-driven job losses will miss early warning signals entirely until disruption is already widespread.
• •J-Curve Adoption Pattern: Economists use a J-curve framework to explain why transformative technologies initially reduce productivity before gains emerge. Companies and workers spend the early phase figuring out how to use the tool effectively. AI may currently sit in the downward scoop of that curve, meaning visible economic impact — positive or negative — is still ahead, not yet present.
• •CEO Incentive Distortion: When companies announce AI-driven layoffs, treat those claims with skepticism. Investors currently reward AI adoption narratives with higher stock prices and increased funding. CEOs who overhired during the post-pandemic boom are financially incentivized to attribute workforce reductions to AI productivity gains rather than to prior hiring mistakes or slowing business conditions.
• •Speed Determines Severity: The Internet revolution eliminated travel agents, typists, and bank tellers gradually over decades, allowing workers time to retrain and pivot. The China trade shock wiped out furniture and textile manufacturing in concentrated regions like Hickory, North Carolina within months, triggering lasting community collapse. AI's economic harm will scale directly with how fast displacement occurs.
• •Policy Readiness Gap: Economists recommend three immediate steps: improve real-time labor measurement tools, strengthen the existing unemployment insurance system — exposed as structurally fragile during the pandemic — and redesign trade adjustment assistance programs that failed displaced manufacturing workers in the 1990s. No comprehensive AI-specific labor policy currently exists at the federal or state level.