Artificial intelligence is likely to increase productivity across large parts of the U.S. economy, but the same technologies that allow companies to produce more with fewer workers also create a significant labor-market risk. The central concern is not that AI will eliminate all employment. It is that it may reduce demand for labor across a wide range of occupations faster than workers can move into new jobs, creating prolonged periods of displacement, lower earnings, weaker consumer demand and greater economic uncertainty.
The early evidence suggests that the first effects may appear through reduced hiring rather than large, highly visible layoffs. Companies adopting generative AI can automate portions of administrative, analytical and customer-facing work without eliminating an entire occupation. A department that previously required 20 employees may still exist but operate with 14 or 15. Over time, this type of incremental reduction can materially affect employment even when no profession disappears outright.
Research from Stanford's Digital Economy Lab has found weaker employment outcomes among younger workers in occupations with high exposure to generative AI. Workers ages 22 to 25 in highly exposed occupations have experienced weaker employment growth than similarly aged workers in less-exposed fields. The finding is important because entry-level positions are often the easiest for companies to reduce when AI is capable of performing basic research, drafting, coding, document review, customer support or administrative tasks.
The occupations most vulnerable in the near term are those in which a substantial share of work involves processing information, applying routine rules, generating standardized content or communicating through digital systems. These include customer-service representatives, administrative assistants, data-entry workers, bookkeepers, billing and payroll clerks, insurance claims processors, loan-processing staff, recruiters, paralegals, transcriptionists, translators, research assistants, copywriters, marketing support personnel, junior financial analysts and some entry-level software-development roles.
The risk is not necessarily that these occupations disappear. It is that they require fewer people.
If an accounting firm can handle the same volume of work with 25 percent fewer employees because each accountant is more productive with AI, the profession continues to exist while employment contracts. The same logic applies to law, banking, insurance, consulting, advertising, media, corporate communications and software development. The economic effect depends less on whether AI can fully replace an individual worker than on whether it enables one worker to perform tasks previously distributed among several employees.
At scale, that distinction becomes important. A company that reduces a department from 100 employees to 75 has not eliminated the occupation, but it has removed 25 jobs. If similar reductions occur across thousands of employers, the cumulative effect can become substantial without producing a single moment that looks like a traditional employment crisis.
McKinsey Global Institute has estimated that approximately 11 million U.S. workers in declining occupations may need to transition into different jobs over the next decade, with a range of roughly 6 million to 16 million depending on the pace of automation and changes in labor demand. These figures should not be interpreted as a forecast of unemployment. Many workers will retire, retrain or move directly into other occupations. The significance of the estimate is the scale of potential labor reallocation.
Large-scale reallocation is difficult because workers are not interchangeable across industries. A mid-career insurance employee cannot automatically move into cybersecurity. An administrative worker cannot immediately enter nursing or skilled construction. Differences in education, geography, age, family obligations and wage expectations slow the adjustment process. Even workers who eventually find new jobs may experience long periods of reduced income or may accept positions that pay substantially less than the jobs they lost.
That distinction is economically important. A worker who moves from an $80,000 job to a $50,000 job is counted as employed, but the household still experiences a substantial decline in purchasing power. At scale, those income losses can weaken consumer demand even without a major increase in the official unemployment rate.
This creates a broader macroeconomic risk. Consumer spending accounts for a large share of U.S. economic activity. Households that lose income typically delay major purchases, reduce discretionary spending and save less. Lower household spending then affects businesses that may have little direct exposure to AI. A reduction in spending on automobiles, travel, restaurants, home improvement and retail can weaken employment in those sectors, creating secondary effects beyond the occupations initially disrupted by automation.
The resulting feedback loop is relatively straightforward. AI raises productivity and reduces labor demand in some industries. Affected workers lose jobs, suffer wage reductions or become more cautious about spending. Lower household spending reduces revenue elsewhere in the economy. Businesses facing weaker demand then slow hiring or reduce employment themselves.
This mechanism is one reason the employment effects of AI could matter far beyond the technology sector.
The International Monetary Fund has estimated that roughly 60 percent of jobs in advanced economies are exposed to artificial intelligence. In many of those occupations, AI is expected to complement workers and increase productivity. In others, however, it may replace enough tasks to reduce labor demand, hiring or wages. The outcome will differ substantially across occupations, firms and industries.
The distributional effects may be particularly important. Productivity gains do not automatically translate into broadly shared income gains. If companies retain most of the benefits through higher profits while workers bear most of the losses through reduced employment or wages, national output can rise even as household economic security deteriorates.
This is a central issue in the AI employment debate. GDP, corporate profits and productivity could all increase while a significant number of workers experience lower incomes and more unstable careers.
Younger workers may face a distinct problem because many of the tasks AI performs well are tasks traditionally assigned to employees at the beginning of their careers. Junior lawyers conduct document review and basic research. Junior accountants perform routine financial work. Entry-level programmers write and test relatively simple code. Marketing assistants prepare drafts, conduct research and produce reports.
If AI reduces the need for workers performing these tasks, companies may hire fewer junior employees while continuing to retain experienced personnel. That would weaken the traditional pipeline through which workers gain experience and advance into senior positions.
The longer-term consequence could extend beyond employment and into the development of expertise itself.
Experts in most fields are not created through education alone. They develop through years of practical work: handling routine assignments, making mistakes, observing experienced colleagues, learning how unusual cases differ from standard ones and gradually taking responsibility for more difficult decisions. Much of that process begins with lower-level work.
If AI removes a large portion of those early-career tasks, fewer workers may receive the practical experience needed to become highly skilled professionals later.
This could create a significant pipeline problem in fields such as law, accounting, engineering, medicine, finance, journalism and software development. Companies may become more efficient in the short term by automating junior work while simultaneously reducing the number of people being trained through real-world experience. Over time, the result could be a smaller pool of genuinely experienced professionals, particularly in occupations where good judgment depends on years of exposure to unusual problems, failures and complex decisions.
In that sense, AI could begin to hollow out the apprenticeship system that has traditionally produced experts.
An organization might save money by assigning routine analysis to software instead of junior employees. Years later, however, it may discover that there are fewer mid-career professionals with the experience necessary to supervise complex work. The short-term labor savings could therefore create a long-term shortage of human judgment.
That issue is especially important in professions where AI itself will still require oversight. If fewer people are allowed to perform the work at an early stage of their careers, there may eventually be fewer experienced people capable of determining when the AI is wrong.
The effect could become self-reinforcing. The more organizations rely on AI because experienced workers are scarce, the fewer opportunities younger workers receive to develop those same skills.
There are strong countervailing forces. New technologies have historically created new occupations and industries, and AI is already increasing demand for data scientists, machine-learning engineers, cybersecurity specialists, semiconductor workers, electricians and employees involved in data-center construction and energy infrastructure. The Bureau of Labor Statistics continues to project overall employment growth in the United States.
The unresolved question is whether new employment opportunities will emerge quickly enough, in sufficient numbers and with sufficiently similar skill requirements to absorb displaced workers.
That is the principal source of economic uncertainty.
The most plausible near-term risk is therefore not mass unemployment on the scale of the Great Depression. It is a more gradual deterioration in employment quality and job availability across selected occupations, particularly among younger and middle-skilled workers. This could include slower hiring, fewer entry-level positions, more frequent career changes, weaker wage growth and downward occupational mobility.
Such a transition would be difficult to detect using unemployment statistics alone. A worker who accepts a lower-paying job remains employed. A graduate who never receives an entry-level offer does not appear as a laid-off worker. A company that replaces departing employees with software rather than new hires does not report a major reduction in force.
For this reason, the most important labor-market indicators may eventually include not only unemployment, but hiring rates, wage growth, entry-level job creation, occupational mobility and the share of income flowing to labor rather than capital.
Artificial intelligence is likely to make the U.S. economy more productive. The unresolved issue is whether the labor market can absorb the resulting displacement without a prolonged period of weaker household income, reduced consumer demand and growing economic instability.
The risk is not that AI eliminates all work. It is that it reduces the number of workers required across enough occupations, and does so quickly enough, that job creation, retraining and professional development cannot keep pace.
If that happens, the United States could face an unusual economic environment: higher productivity and strong corporate earnings alongside weaker employment security, lower household purchasing power and fewer pathways for younger workers to develop the expertise needed to replace today's senior professionals.
The long-term question is therefore larger than how many jobs AI will eliminate. It is whether the economy can continue producing both enough work and enough experienced people to sustain the institutions, professions and consumer demand on which the broader system depends.
