MIT economist David Autor and co-author Neil Thompson have published a peer-reviewed framework arguing that artificial intelligence does not eliminate jobs so much as it separates workers who hold genuine expertise from those who rely on routine tasks that machines can now perform. Their paper, published in the Journal of the European Economic Association, arrives alongside federal data showing that while AI adoption among U.S. businesses is accelerating, very few companies report cutting staff because of it. The finding reframes the dominant fear of mass displacement into a sharper question: not who loses their job, but whose skills actually hold up under pressure.
Automation Splits the Workforce by Skill, Not by Sector
The core argument from Autor and Thompson centers on a distinction most coverage of AI and employment overlooks. When AI automates what they call “inexpert” or supporting tasks, it tends to raise wages for the skilled workers who remain while constraining the number of available positions. But when AI automates “expert” tasks, the effect reverses: wages fall for incumbents even as the door opens for more workers to enter those roles. A junior associate who once needed years of training to draft legal briefs, for instance, can now lean on AI tools to produce competent work faster, which compresses the premium that seasoned attorneys once commanded.
That two-track dynamic matters because it challenges the blanket narrative that AI either helps or hurts workers. The real outcome depends on which layer of a job gets automated. If AI handles the grunt work, experienced professionals become more valuable. If it handles the judgment calls, experienced professionals lose their pricing power. The framework suggests that mid-career workers in fields like finance, law, and software development face the most acute version of this tension, since their roles blend both types of tasks in ways that vary from firm to firm. In practice, that means two people with the same job title can experience automation very differently depending on whether their employer uses AI to support their expertise or to substitute for it.
Federal Data Shows Growth Without Mass Layoffs
Government numbers back the idea that AI is reshaping work without triggering the layoff waves that headlines often predict. The U.S. Census Bureau’s Business Trends and Outlook Survey, through its dedicated AI supplement, tracks how widespread AI use is among businesses, what types of tools firms deploy, and how adoption affects employment and organizational structure. The survey draws from a nationally representative sample and is designed as an ongoing measurement program, giving policymakers and researchers a near-real-time view of how AI diffuses across the economy instead of relying on occasional one-off studies.
Reporting on those survey results found that AI adoption among U.S. businesses climbed from 3.7% to 5.4% during the period studied, with projections reaching about 6.6% by early fall. The information technology sector led adoption, along with firms concentrated in Colorado and Washington, D.C., where knowledge-intensive industries cluster. Yet the same data showed that few firms reported layoffs tied directly to AI tools. Instead, respondents more often cited efficiency gains, process changes, or new product lines. That gap between rapid adoption and minimal job cuts is consistent with the Autor and Thompson framework: businesses are integrating AI into workflows, not swapping it in for headcount, and the technology is changing what workers do inside their roles rather than eliminating the roles themselves.
Job Postings Reveal Shifting Demand in Real Time
A separate line of research reinforces the pattern from a different angle. Economists Daron Acemoglu, David Autor, Jonathon Hazell, and Pascual Restrepo used online job vacancy data from Burning Glass to examine how AI changes labor demand signals as they appear in real time through job ads. Their National Bureau of Economic Research working paper finds that AI alters the composition of what employers ask for rather than simply reducing the volume of postings. Firms adjust skill requirements, rewrite job descriptions, and shift hiring criteria in ways that reflect new internal workflows, often emphasizing complementary capabilities like communication, problem-solving, or domain knowledge alongside familiarity with AI tools.
This matters because job postings are one of the earliest and most granular indicators of how employers actually respond to new technology. By the time official employment statistics register a change, the shift has usually been baked into hiring patterns for months. The Acemoglu team’s approach captures that leading edge, and what it shows is a labor market that is recalibrating around AI rather than contracting because of it. The picture that emerges from federal survey instruments and from private-sector job listings is broadly consistent: demand is not vanishing, but it is moving toward roles that can orchestrate, supervise, or meaningfully augment AI systems instead of competing with them task by task.
Why White-Collar Work Survives, With a Catch
Historical precedent offers some reassurance for office workers worried about obsolescence. White-collar work thrived through earlier waves of digital disruption because computers rarely replaced entire jobs in a single stroke. Instead, they automated specific activities inside broader roles, leaving enough of the job intact that workers adapted rather than disappeared. Spreadsheets did not kill accountants; they changed what competent accounting looked like. Legal databases did not eliminate paralegals; they shifted the balance from manual document retrieval to higher-order synthesis and case strategy.
The catch with generative AI is that it reaches further into the judgment layer than previous technologies did. A spreadsheet could organize numbers but could not write the memo interpreting them; a large language model can do both at a passable level. That does not mean white-collar jobs will vanish, but it does mean the bar for what counts as genuine expertise is rising. Workers who relied on process knowledge (the ability to follow established procedures accurately) face more competition from AI-assisted newcomers who can produce adequate work quickly. By contrast, workers who bring strategic thinking, deep client relationships, or domain-specific creativity retain an edge that the technology cannot yet replicate. In this sense, AI compresses the middle of the skill distribution while stretching the top, rewarding those who can define problems and integrate machine outputs into coherent decisions.
The Real Divide Is Visibility, Not Vulnerability
The most useful way to read the emerging evidence is not as a story about job loss but as a story about exposure. AI acts like a stress test for professional competence. In fields where performance was historically hard to observe (because clients could not easily judge legal writing quality, or managers could not see every line of code), automation makes relative ability more visible. Tools that standardize routine outputs narrow the gap between average and mediocre workers, while simultaneously amplifying the advantages of those who can use the same tools to produce exceptional results. In Autor and Thompson’s terms, the technology separates inexpert support tasks from expert judgment, and then shines a spotlight on who truly excels at the latter.
That visibility-driven divide also helps explain why the data show reconfiguration rather than collapse. The Census Bureau’s AI survey and the Burning Glass vacancy analysis both point to organizations redesigning roles so that fewer people handle more complex, higher-leverage work, supported by AI systems that take on lower-level tasks. For workers, the implication is not that every job is safe, but that vulnerability tracks with how easily their contributions can be standardized and monitored by software. For policymakers and educators, the challenge is to expand access to the kinds of training and career paths that build genuine expertise, so that more people can move into the segments of the labor market where AI functions as an amplifier rather than a replacement. In that world, the central question is less whether AI will take jobs and more whether workers will have the opportunity to develop skills that remain valuable when the machines arrive.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
