Economists at some of the world’s leading research institutions are raising alarms about the speed at which artificial intelligence is threatening American jobs. New analyses from the International Monetary Fund, the Brookings Institution, and the Federal Reserve suggest that tens of millions of U.S. workers face significant disruption to their daily tasks, with the sharpest effects landing not on factory floors but in offices, clinics, and classrooms. The warnings arrive as workplace AI adoption accelerates and policymakers struggle to keep pace.
How Many Jobs Face Real Disruption
The scale of potential displacement is difficult to overstate. In advanced economies like the United States, about 60% of jobs may be impacted by AI, according to the International Monetary Fund. The IMF’s analysis splits that exposure into two broad categories: roughly half of those affected positions could see productivity gains from AI tools, while the other half face reduced labor demand as machines take over core tasks. That split means the difference between a worker who uses AI to do their job faster and a worker whose job simply disappears.
Brookings Institution researchers put a finer point on the U.S. picture, finding that more than 30% of American workers could see at least 50% of their occupation’s tasks disrupted by current generative AI capabilities. That finding challenges a common assumption that automation primarily threatens low-wage, repetitive work. Instead, generative AI targets cognitive and nonroutine tasks, meaning mid- to higher-paid professions such as legal analysts, financial advisors, and software developers sit squarely in the crosshairs. For many of these workers, disruption will arrive not as a single layoff notice but as a steady erosion of responsibilities as software takes over research, drafting, and analysis.
White-Collar Workers Bear the Heaviest Burden
A Pew Research Center analysis using O*NET work activities found that in 2022, 19% of U.S. workers held jobs in the top quartile for AI exposure, based on the importance of high-exposure activities to their daily roles. These positions cluster among higher-educated, white-collar professionals in fields such as management, finance, and technical services. Pew’s demographic breakdown also documents uneven exposure across education levels, gender, and race and ethnicity, suggesting that the economic fallout will not be distributed equally across the workforce and that some already advantaged groups may paradoxically face the most direct technological pressure.
This pattern inverts decades of automation history. Between 1980 and 2016, automation in the United States primarily displaced routine manual and clerical workers, according to NBER Working Paper 32536. That study found that technologies such as industrial robots and office software contributed materially to increases in between-group inequality, widening the earnings gap between workers with and without college degrees. AI-driven disruption, by contrast, is climbing the income ladder. Professionals who once considered their analytical skills a shield against automation now find those same skills are exactly what large language models replicate most efficiently, raising the prospect that high-paying occupations could become more polarized between a small number of AI-augmented stars and a larger group of displaced specialists.
Adoption Is Accelerating Faster Than Policy
The Federal Reserve has documented rapid growth in workplace AI adoption through its review of available time-series data, though the board’s researchers caution that estimates vary widely depending on survey design. Firm-level surveys, worker-level surveys, and differences in question wording all produce different adoption figures, and some measures capture experimental pilots while others track only fully deployed systems. That measurement gap matters because policymakers cannot design effective retraining programs or safety nets without reliable data on how quickly AI is entering workplaces and which sectors are moving fastest.
The disconnect between adoption speed and policy response creates a window of vulnerability for workers. Even if aggregate economic gains from AI eventually materialize, the transition period could be brutal. Evidence on past automation waves documents a mechanism called “rent dissipation,” in which the productivity gains from new technologies are offset by downward pressure on wages for displaced workers and intensified competition for remaining roles. In plain terms, the economy can grow while individual workers get poorer, a dynamic that played out with industrial robots and could repeat with generative AI on a much larger scale if institutions such as unemployment insurance, job placement services, and community colleges are not upgraded in parallel.
Adaptive Capacity Separates Winners from Losers
Not every worker in a high-exposure job faces the same risk. Brookings researchers measuring adaptive capacity found that of the 37.1 million U.S. workers in the top quartile of occupational AI exposure, 26.5 million have above-median adaptive capacity, calculated using Lightcast data on skills, education, and local job openings. That means roughly 10.6 million highly exposed workers lack the transferable skills, educational credentials, or geographic mobility to pivot into new roles if their current positions erode. These workers face the steepest welfare costs from displacement, not only because they are vulnerable to job loss but because they have fewer realistic pathways into new occupations that benefit from AI rather than compete with it.
The gap between those who can adapt and those who cannot is where policy intervention matters most. Workers in smaller metro areas or rural communities, for instance, often have fewer alternative employers and less access to retraining programs. A financial analyst in New York with strong data skills may transition to an AI-adjacent role with relative ease, leveraging tools to expand their productivity. A medical coder in a mid-sized city with fewer employers and narrower skill sets faces a far harder path. Targeted training subsidies, relocation assistance, and investments in broadband and remote-work infrastructure could help narrow that divide, but only if they are informed by granular labor-market data rather than national averages.
Modest Macro Gains May Not Offset Worker Pain
One of the most striking counterpoints to AI optimism comes from economists who have examined the likely macroeconomic payoff. NBER Working Paper 32487 assesses broad claims about AI’s economic potential and concludes that headline-grabbing projections of explosive growth are far from guaranteed. While AI can raise productivity in specific tasks and industries, the paper argues that bottlenecks in complementary investments, such as organizational change, worker training, and new capital, could limit the speed and scale of aggregate gains. In this view, AI looks less like an instant general-purpose revolution, and more like a gradual, uneven diffusion that may leave many workers bearing costs long before the benefits show up in national statistics.
This tension between micro-level disruption and macro-level modesty complicates the politics of AI policy. If national output grows only slowly while visible layoffs hit white-collar workers, public support for experimentation could erode. Policymakers therefore face a dual challenge: cushioning the immediate blow for affected workers while also making the long-term investments needed to unlock AI’s productive potential. That includes funding community colleges and apprenticeship programs, modernizing labor-market data systems, and coordinating with employers on standards for responsible deployment so that productivity gains translate into better jobs rather than a one-sided transfer of income from labor to capital.
Why Better Data Will Shape the Response
Across these studies, a common thread is the need for more precise information on who is at risk and how they are faring over time. Much of what economists know about employment, wages, and job transitions comes from long-running federal surveys such as the Current Population Survey, which underpins the official U.S. unemployment rate. While invaluable, these instruments were not designed with AI-specific exposure in mind, and they often lack the detailed task-level information needed to distinguish between jobs that are AI augmented and those that are AI substituted. Integrating new questions on software use, automation tools, and remote-work arrangements into existing surveys would give policymakers a clearer view of how AI is reshaping work in real time.
Linking richer survey data with administrative records and private-sector sources could also sharpen the picture. For example, combining CPS employment histories with firm-level information on AI deployments and job postings would allow researchers to track whether workers in high-exposure occupations are actually moving into new roles or simply dropping out of the labor force. That kind of evidence would help answer the central policy question raised by the IMF, Brookings, the Federal Reserve, and NBER researchers: not whether AI will change the labor market, it already is, but whether the institutions that govern work can change quickly enough to ensure that the technology’s gains are broadly shared rather than concentrated among a narrow slice of firms and highly adaptable professionals.
More From The Daily Overview
*This article was researched with the help of AI, with human editors creating the final content.
Grant Mercer covers market dynamics, business trends, and the economic forces driving growth across industries. His analysis connects macro movements with real-world implications for investors, entrepreneurs, and professionals. Through his work at The Daily Overview, Grant helps readers understand how markets function and where opportunities may emerge.
