Before the public began to worry in recent weeks that AI would destroy humanity altogether, we mostly worried it would take our jobs. For the past several years, we’ve heard a stream of corporate executives and economists warning about the looming AI employment apocalypse. The media seized on tech layoffs and a spike in the youth unemployment rate in 2025 as the first signs of the great job disappearance.
But in reality, economists are seeing few signs of AI disruption in the labor market. Youth unemployment fell to 7.1% in August 2026, down from an average of 8.3% in 2025. Employment in computer and mathematical occupations such as software developers and information security specialists reached an all-time high in August, according to government surveys. And overall private employment keeps climbing month after month, buoyed in part by data center spending, which is pushing up demand and wages for occupations such as electricians.
The Bureau of Labor Statistics (BLS) recently released its assessment of which occupations had “very high,” “high,” “moderate,” and “low” exposure to AI. The BLS warns that “[a]n exposure category is not a forecast of employment growth or decline.”
Indeed, many occupations that have “very high” or “high” exposure to AI grew over the past year, while other occupations with “low” exposure to AI are shrinking. For example, in the first eight months of 2026 compared to the first eight months of 2025, employment is up by 1.3% in computer and mathematical occupations (which include computer programmers, software developers, data scientists, and mathematicians), all of which uniformly have “very high” exposure to AI. Similarly, employment is up by 1.7% in architectural and engineering occupations, all with high or very high exposure to AI. Employment is up by 1.8% in life, physical, and social science occupations (which include biologists, physicists, historians, and economists), almost all of which have high or very high exposure to AI.
Meanwhile, employment in many hands-on and high-touch occupations, which were supposed to fare well in the AI age, has declined. Employment in health care support occupations (which include home health aides, nursing assistants, dental aides, and pharmacy aides) with low to moderate exposure to AI is down 2.7% over the last year, perhaps reflecting the Trump Administration’s unceasing war on foreign-born workers. Similarly, employment of personal care and service workers (which include barbers, hair stylists, manicurists, and personal trainers), with low to moderate exposure to AI, is down by 1.6%. Protective service occupations (which include firefighters, police officers, security guards, and lifeguards) saw a 4.5% decrease. And installation, maintenance, and repair occupations (which include auto mechanics, HVAC technicians, and linemen) saw a 2.1% decrease.
Health care support deserves particular attention. The general consensus has been that health care occupations would continue to grow as the U.S. population ages, and the sector even became the only reliable source of job growth as the labor market cooled in 2025. The latest employment projections from the BLS call for the private health care and social assistance sector to add 2.2 million jobs by 2035.
When considering the decline in employment in health care support occupations, there are factors beyond AI to consider: health care support workers themselves are aging and retiring; there are fewer younger workers to replace them; immigration has fallen under the Trump Administration, removing a key source of labor (foreign-born workers account for an outsized share of health care workers); and there has been consolidation across the health care sector.
We are not arguing that these trends will necessarily persist. Policymakers need to realize that there is a wide range of potential forecasts of the AI impact on the labor market, and we are far from settling which one is right. The increases in employment levels for information-driven occupations and decreases for high-touch occupations could reverse next month or next year. It is noteworthy that many analyses have also found no evidence of AI disruption on workers, though.
The Budget Lab finds no clear evidence of labor market disruption associated with AI, no connection between measures of AI usage and changes in employment or unemployment, and no change in the occupational mix (or the percentage-point shift in the occupational composition of major industry groups such as construction and information) that clearly aligns with the introduction of AI. Put another way, they find no evidence of AI disruption at the broadest level or in underlying shifts within industries. Researchers from the Federal Reserve Bank of Atlanta and the Federal Reserve Bank of Richmond also find “little evidence of near-term aggregate employment declines due to AI” in a survey of nearly 750 corporate executives. (The researchers do, however, find evidence of compositional reallocation of labor within and across firms, with clerical roles declining and skilled technical roles increasing.)
There are a few important caveats to be sure. First, a growing body of analysis suggests that AI is not destroying jobs but rather shifting the tasks that workers perform. The Burning Glass Institute finds that “AI isn’t simply replacing jobs or enhancing them. It’s doing both—within the same roles, at the same time.” Existing labor market data reveals a lot about industries and occupations, but it doesn’t tell us much about tasks. It is difficult to capture shifts in the tasks that workers perform. Secondly, as Jed Kolko, Senior Fellow at the Peterson Institute for International Economics, points out, the prevalence of LLMs is so recent that any lasting economic impact would likely take years to show up in economic data. Next, existing research has focused on labor demand, or the willingness of employers to hire workers, but not on how AI might affect labor supply, or the availability of people to work. And finally, analyses have only evaluated existing occupations and have not analyzed potential job growth associated with AI. This is partially by definition, but AI could spur job creation or create entirely new occupations.
Moving forward, it is also important to remember that forecasts can identify broad trends about industries and the direction of those trends, but they are less accurate in identifying the magnitude of trends or details about specific occupations. Forecasts often rely on historical trends and are less likely to account for unforeseen economic shocks or rapid technological developments.
Analyses of the impact of AI on workers have important implications for workforce development, as policymakers seek strategies to retrain and upskill workers who may be dislocated by AI and as future workers decide between college and non-college career pathways. This is especially true at a time when there is renewed interest in the skilled trades, particularly among Gen Z, and declining perceptions about the value of a college degree. The Progressive Policy Institute will continue to track the impact of AI on workers.
Michael Mandel is vice president and chief economist at the Progressive Policy Institute. Michael Pearson is director of workforce development policy at the Progressive Policy Institute.
