Labor Matters: My Assessment of AI's Impact on the US Economy and Labor Market
This is an executive summary of a long piece over at my Substack: https://gadlevanon.substack.com/p/my-assessment-of-ais-impact-on-the
Summary
Media and social media are full of opinions about AI's impact on the US economy and labor market. My view is that most of the debate is badly underestimating how much AI has already changed both.
The evidence, as I read it, points to a much larger impact than the conventional view allows. This matters because a consensus that says "not yet" is a consensus that prepares for nothing, at exactly the moment preparation is still possible.
Proving that any particular economic development was caused by AI is extraordinarily difficult. The economy offers no clean experiments, and AI is arriving alongside many other changes. What follows is therefore not definitive proof. It is my reading of the evidence.
1. Growth is stronger than it looks, and AI is the biggest reason.
I put underlying GDP growth right now at about 3 percent at a time when the population is barely growing, implying unusually strong growth in GDP per capita. The main driver is a surge in business investment, much of it AI-related. The usual measure of AI investment counts only the visible part, chips, data centers and power; the plants and grid that have to be built to build them are not labeled AI in anyone's accounts, so the measured contribution is a floor. When actual spending beats expectations three years running, the problem is the expectations.
The impact goes beyond direct investment. The AI-driven surge in technology stocks has created a large wealth effect, sustaining solid consumer spending even though job growth outside health care has been close to zero over the past two years. Between the investment boom and the wealth effect, AI has been a major reason the economy kept growing nicely despite tariffs, federal spending cuts, the war with Iran, and unusually high uncertainty.
2. Productivity has stepped up, and the gains are concentrated where technology is changing work fastest.
Productivity has grown about 2.5 percent a year over the three years through mid-2026, more than twice its pace in the 2010s. That is too large, and has lasted too long, to be noise.
The gains are also highly concentrated. They are strongest in finance, insurance, information, and professional and business services, what I call FIIPB, where much of America's office work is done and AI use is high. Productivity in those industries has grown about 4 percent a year. Growth is also elevated in advanced manufacturing and retail, two sectors being reshaped rapidly by technology. Across much of the rest of the economy, productivity growth has been close to zero.
That pattern doesn't prove AI caused the acceleration, but it is one of the main reasons I think AI and related technologies are already contributing a lot. If this were a broad cyclical rebound, I'd expect the gains to be spread more evenly. Instead they're concentrated in the parts of the economy where technology is changing work fastest.
3. The missing jobs aren't a weak economy. They're productivity.
Job growth has been weak for two years. Payrolls have grown by about 1.3 million over the 24 months through August, roughly 55,000 a month, and outside health care and social assistance the economy has added essentially no jobs. Most people read that as weakness. I read it as productivity.
Weak hiring is usually taken as a sign that demand is weak. That's hard to square with output. Output has kept growing while employment and hours have barely moved, so firms are producing considerably more without adding many workers. Whatever its cause, that is higher productivity.
The clearest example is FIIPB. For 36 straight months, from September 2023 through August 2026, employment in these industries was lower than a year earlier even as employment in the rest of the economy kept rising. Every other run like that since 1990 came during or right after a recession. This one didn't. These industries had been major engines of US job growth for decades. Something changed around 2023.
At the same time, from the end of 2022 through the end of 2025, real output in FIIPB rose about 13 percent while hours worked fell about 2 percent. More output with fewer hours is productivity.
The harder question is what caused it. I think AI and related technologies are an important part of the answer, but not the whole answer. So far, the clearest employment effects from actual AI deployment appear in clerical and other routine jobs, where tasks are easy to codify and carry less judgment or responsibility. Within professional occupations, existing evidence points to larger effects at the entry level, where more of the work is research, drafting, coding and analysis, tasks today's systems can increasingly do.
There is also an anticipatory effect on hiring. Firms don't staff for today's technology; they hire for the work they expect to need over the next several years. If employers expect AI to absorb a growing share of certain tasks, they get more cautious about adding headcount before deployment is complete.
4. Young graduates are the exception in a low-unemployment labor market.
Since late 2022, unemployment among bachelor's degree holders aged 22 to 27 has risen about 1.5 percentage points, three times the increase for young workers without a degree and about twice the increase for older graduates. Young-graduate unemployment is now unusually high for an expansion, and their pay advantage is shrinking.
Broad economic weakness can't explain a pattern this specific. Nor can the post-2022 tech layoffs: the deterioration has spread well beyond tech and has now lasted three years past that correction. I think AI is an important mechanism. Much of the work that used to go to junior professionals is work AI can now do, and firms expecting it to do more are hiring fewer of them.
A longer-running trend is making the problem worse. The share of each cohort earning a bachelor's degree has risen for decades, so more graduates are competing for a slow-growing pool of entry-level professional jobs. That glut didn't cause the recent deterioration, but it means a hit to demand lands harder than it would have a generation ago.
5. The studies that find nothing are using the wrong variable.
Studies finding no measurable AI effect on the labor market go viral and fuel the belief that AI hasn't yet had an impact. Most rely on AI exposure scores, find little relationship with employment, and conclude that AI is not doing much. I largely discount those conclusions because exposure, even if measured correctly, is not the same thing as displacement.
Exposure scores do not tell us whether employers actually adopt AI, whether it substitutes for labor, or whether demand expands enough to absorb the extra output. Using exposure on its own as the sole proxy for AI's labor-market impact is a non-starter. A null result from that test is not strong evidence that the true effect is small.
6. The big job losses haven't happened yet. They'll come in the next recession.
Firms have barely started. The gap between what the best models can do and what most firms actually use them for is enormous, and I expect it to keep growing. Closing that gap will be one of the most important forces shaping the US economy and labor market in the coming years and decades.
Conditions so far have been unusually forgiving. There has been no recession since ChatGPT's release. Growth has been solid, corporate profits are at a record share of GDP, and the layoff rate is about as low as it gets. AI is also creating jobs. Thousands of AI companies are founded every year, concentrated in tech hubs like the Bay Area, and existing firms are adding AI roles at a faster clip still. I estimate that 1.5 to 2 percent of knowledge workers now hold jobs that would not exist without AI. And that count excludes the physical buildout: the data centers, power plants, transmission lines, and chip fabs behind the AI boom are employing electricians, construction crews, engineers, and technicians in significant numbers.
That combination has produced a wave of articles declaring the AI job apocalypse dead. I think they're premature. Firms make their biggest workforce cuts in recessions, when they have both the opportunity and the pressure to restructure. The manufacturing experience is a useful warning: the impact of automation looked much smaller in 2000 than it did by 2010, after two recessions had forced firms to reorganize around the technology. By the next downturn, employers will have had years to learn where AI can substitute for labor. I expect large cuts in office jobs when it comes, and I don't expect many of those jobs to come back. That young college graduates are struggling even in today's favorable conditions is what worries me most.
The prevailing view that AI has not yet affected the labor market much is not a harmless analytical mistake. It tells policymakers, colleges, and young workers: no rush. If I am right, the next recession will make these effects much more visible. By then, employers will have had years to learn where AI can substitute for labor, and many adjustments that could have been made gradually will become much harder. The time to prepare for that labor market is before it arrives.