AI Doesn’t Need to Take Every Job to Break the Labor Market

Artificial intelligence does not need to eliminate 300 million jobs to radically reshape the economy. It only needs to make enough workers easier to replace, enough entry-level positions unnecessary and enough companies confident that they can produce more with fewer people. That shift can weaken wages and bargaining power even if headline unemployment never approaches Depression-era levels.

The most dramatic estimates are often misunderstood. Goldman Sachs has estimated that roughly 300 million full-time jobs globally are exposed to some degree of automation from generative AI, while the International Monetary Fund has estimated that about 40% of global employment could be affected by AI. Exposure does not mean elimination, however, because many jobs will be augmented rather than replaced, and the IMF estimates that in advanced economies roughly half of the occupations exposed to AI could experience productivity gains instead of outright displacement.

That distinction matters, but it should not be used as reassurance that everything will work out automatically. A job does not need to disappear completely for the worker occupying it to lose economic power. If one attorney using AI can perform work that previously required an attorney and two junior associates, employment can fall even though the legal profession remains very much alive.

AI Is Different Because It Is Coming for White-Collar Work

Previous waves of automation were often associated with factories, warehouses and other repetitive physical work. Artificial intelligence is unusual because many of the tasks it performs are cognitive: writing, coding, summarizing documents, analyzing information, creating images and communicating with customers.

The IMF estimates that about 60% of jobs in advanced economies are exposed to AI in some way, substantially more than in lower-income countries. The reason is straightforward: wealthy economies employ far more people in professional, managerial, administrative and technical occupations built around information processing. Those were precisely the jobs many people were told would provide protection from automation.

AI therefore challenges one of the central assumptions behind the modern education system. For decades, the answer to automation was essentially to become more educated, acquire cognitive skills and move into knowledge work. Now the technology is improving fastest at many of the tasks performed by the educated workers who followed that advice.

That does not mean engineers, lawyers or accountants are about to disappear. It means firms may need fewer people to produce the same output, particularly at junior levels where employees historically performed research, drafting, basic analysis and other tasks now increasingly assisted by AI.

Companies Do Not Need to Announce “AI Layoffs” for Automation to Reduce Employment

The impact of AI can appear gradually rather than through dramatic mass firings. A business employing 500 people may introduce AI tools, allow natural attrition to reduce headcount to 450 and simply stop replacing every person who leaves.

From the outside, nothing looks like a technological revolution. There is no factory shutting down and no CEO announcing that robots replaced an entire department. Yet 50 jobs have effectively disappeared because the remaining employees can handle more work.

This pattern is particularly important for entry-level employment. Goldman Sachs reported in March that AI’s impact is already visible in technology, knowledge and creative occupations and expects the effects to become much more significant over the next decade. The IMF has similarly warned that entry-level and middle-skilled positions may be among the areas squeezed as companies redesign jobs around AI.

That could create a long-term career-development problem. Senior employees do not appear spontaneously; they usually begin as juniors and learn through years of increasingly complex work. If AI removes a large share of beginner tasks, companies may eventually discover that they have also removed the training ground that produced their future experts.

Productivity Gains Are Great for Companies Until Someone Asks Who Gets Them

The economic case for AI is extremely strong if the technology increases productivity. A company that can serve twice as many customers with the same number of employees can lower costs, increase profits, reduce prices or invest in expansion.

The critical question is how those gains are distributed. If workers become twice as productive and their wages rise substantially, AI can raise living standards broadly. If the company instead reduces staffing while directing most of the additional profits toward shareholders, the technology can make the economy richer while making many workers less economically secure.

Federal Reserve Governor Michael Barr warned in July that AI could either expand opportunity or deepen existing inequalities. He noted that the top tenth of U.S. households own roughly 59% of household wealth, while the bottom half own less than 3%, meaning increases in the value of businesses and financial assets disproportionately benefit households already holding substantial capital.

This is the distribution problem at the center of the AI revolution. Machines may increase the economic pie while simultaneously changing who has the strongest claim on each new slice.

The Owner of the AI May Gain More Than the Worker Using It

Imagine two companies introducing identical AI systems. At the first company, workers use the technology to become more productive and receive higher wages because their skills become more valuable. At the second, management uses the same technology to reduce staffing by 20% while keeping salaries largely unchanged.

The technology is identical, but the economic outcome is completely different. AI itself does not determine whether workers benefit; ownership, competition, labor markets and corporate decision-making determine where the productivity dividend flows.

Recent IMF research reinforces that concern. Using global AI usage data, researchers found that productivity gains from AI currently tilt toward higher-paid occupations in nearly every country studied, although the degree varies substantially. Another IMF study notes that high-income workers are also more likely to own the risky financial assets that could appreciate if AI increases corporate profitability, giving them two potential channels for benefiting from the technology.

That creates the possibility of a compounding advantage. Wealthier workers can benefit from AI at work while simultaneously benefiting as investors in the companies profiting from automation.

Human Labor Can Become Less Valuable Without Becoming Worthless

Economic discussions about automation often fall into an unnecessary binary. Either AI replaces people completely, or humans remain indispensable and there is nothing to worry about.

Labor markets do not work that way. A technology that reduces demand for a type of worker can lower wages or slow hiring long before the occupation disappears.

Consider customer service. If an AI system handles 70% of routine inquiries and humans address only the complicated cases, the company may still employ customer-service representatives. It simply needs fewer of them, and the remaining jobs may become more demanding because easy calls have disappeared.

The same mechanism can operate in software engineering, legal research, marketing and graphic design. An employee may remain useful while the number of employees required to produce a particular amount of work declines significantly.

Entry-Level Workers Could Bear the Most Dangerous Cost

The most immediate AI inequality may not occur between highly educated professionals and blue-collar workers. It may appear between established workers and young people trying to enter professional careers.

Senior employees possess judgment, relationships, institutional knowledge and experience that current AI systems cannot easily replicate. Junior employees often spend much more of their time drafting basic documents, conducting research, creating first versions of presentations, writing routine code or performing other tasks where AI is becoming increasingly capable.

That makes the economics of entry-level hiring less attractive. Why hire six graduates if three experienced workers equipped with AI can produce comparable output?

The long-term consequences could be severe even if corporate profits improve. A generation that struggles to obtain its first meaningful job cannot build the experience required for the more sophisticated positions companies still need later.

A Shrinking Middle Class Would Change What Companies Sell

If AI increases wealth concentration, businesses will adapt to the customers who still have substantial money to spend. That process is already visible in markets where luxury spending remains strong even while many households complain about affordability.

This does not necessarily mean society literally divides into billionaires and everyone else. The number of affluent households can grow even while inequality increases, because a growing economy can simultaneously produce more millionaires and leave median households feeling financially squeezed.

The danger is a bifurcated consumer economy. Companies can increasingly focus either on wealthy consumers willing to pay high prices for premium experiences or on mass-market customers demanding the cheapest possible product.

The middle can become harder to serve profitably. That pattern already exists in industries ranging from airlines to fashion, where premium products and ultra-low-cost offerings can outperform traditional middle-market options.

The “Whale Economy” Shows How This Can Work

Video games provide an unusually clear example of a market that does not require every customer to spend equally. Free-to-play games can attract millions of users while deriving a disproportionate share of revenue from a relatively small group of heavy spenders often called whales.

The economic lesson extends far beyond gaming. A company does not necessarily need a large prosperous middle class if a smaller group of wealthy customers generates enough spending to support the business.

Luxury automobiles, private aviation, elite travel, premium healthcare and high-end financial services already operate on versions of this model. If wealth becomes more concentrated, companies have an incentive to devote more innovation and resources toward serving customers who possess the greatest purchasing power.

That can create an unsettling feedback loop. Automation reduces labor costs, profits flow disproportionately toward capital owners, wealth becomes more concentrated and businesses increasingly design products for the people who captured those gains.

UBI Is Not a Ridiculous Response, but It Is Not a Free Solution

Universal basic income inevitably enters discussions about mass automation because it offers a straightforward answer to declining labor demand. If machines produce more goods and services while fewer people are required to work, governments could redistribute part of the resulting wealth through unconditional cash payments.

Critics are correct that unconditional income can affect work behavior. A large U.S. randomized experiment involving 1,000 low-income participants receiving $1,000 per month for three years found that labor-force participation fell by 4.2 percentage points and that participants worked roughly one to two fewer hours per week, while non-transfer income declined by about $1,900 annually relative to the control group.

Another experiment in Compton, California found no reduction in labor supply among people already working full time but found a larger decline among participants who had been part-time workers before the program. These results complicate both ideological extremes because unconditional cash neither caused everybody to stop working nor had zero effect on labor participation.

The bigger question is what happens if employment opportunities themselves shrink because of automation. Reducing work slightly when plenty of jobs exist is different from receiving income support because the economy genuinely needs fewer workers.

The UBI Debate Could Look Very Different in an AI Economy

Today’s UBI experiments occur in an economy where paid employment remains the primary method for distributing purchasing power. That makes it reasonable to worry that unconditional payments can discourage some work at the margin.

A heavily automated economy creates a different problem. If businesses can produce enormous quantities of goods with dramatically less labor, there must still be customers capable of purchasing those goods.

Companies cannot indefinitely replace workers with machines while expecting former workers to remain equally powerful consumers. At some point, the system has to distribute purchasing power through wages, capital ownership, government transfers or some combination of all three.

This is the economic contradiction that receives too little attention in utopian automation forecasts. Production can become almost infinitely efficient, but consumer economies still require consumers with money.

Ownership May Become More Important Than Employment

If AI raises the productivity of capital faster than the value of human labor, owning productive assets becomes increasingly important. Someone whose entire economic life depends on a paycheck is more exposed to automation than someone who also owns stocks, businesses, real estate and other assets.

This is not a new concept, but AI could accelerate it. The Federal Reserve’s July discussion of AI and inequality emphasized that wealth is already heavily concentrated and that compounding investment returns can cause those who own appreciating assets to pull progressively farther ahead.

That suggests one response to AI cannot simply be retraining workers endlessly. Skill development remains essential, but households may also need broader participation in the ownership of the technologies and companies generating productivity gains.

Retirement accounts, employee stock ownership, broad-market investment and other mechanisms that expand capital ownership therefore become more than personal-finance advice. They become part of the larger question of how an automated economy distributes its wealth.

AI Could Still Create Millions of Jobs We Cannot Yet Imagine

There is an important reason not to assume permanent mass unemployment. Previous technological revolutions destroyed occupations while creating entirely new industries that earlier generations could not have predicted.

AI itself is already generating demand for data centers, power infrastructure, semiconductor manufacturing, cybersecurity, model evaluation and many other supporting activities. Goldman Sachs specifically expects AI investment to create employment tied to the massive physical infrastructure required to support the technology.

Productivity growth can also lower costs and generate entirely new forms of demand. If businesses become more efficient, households can spend savings elsewhere, creating employment in industries that may have little direct connection to AI.

The uncertainty is timing. New industries do not necessarily appear in the same places, at the same wages or quickly enough to employ everyone displaced from existing occupations.

The Transition Is More Dangerous Than the Destination

Economists often point to history and note that technological progress has ultimately generated more wealth and new forms of employment. That observation can be correct while still ignoring enormous human costs during periods of transition.

A 52-year-old accountant displaced by AI cannot necessarily become a data-center electrical engineer because economists predict the economy will eventually create enough new jobs. Geography, education, age, family responsibilities and wages all limit how easily workers can move between occupations.

The same problem applies to entire communities. If AI-driven job losses become concentrated in cities dominated by administrative, customer-service or professional employment, the local economic effect can resemble previous industrial declines even if national GDP continues to increase.

Economic output therefore is not enough to judge whether the transition is successful. The distribution and speed of change matter just as much.

AI Could Make Society Richer and Workers Poorer at the Same Time

This is the central paradox policymakers need to confront. Artificial intelligence can increase national productivity, corporate profits and total wealth while reducing the economic value of certain forms of human labor.

There is no law requiring productivity growth to translate automatically into proportional wage growth. If labor becomes easier to substitute while ownership of AI systems remains concentrated, a growing share of economic gains can flow toward shareholders and highly skilled workers who complement the technology.

The IMF has repeatedly warned that AI could deepen inequality even while boosting global growth. It estimates that around 40% of jobs globally and roughly 60% in advanced economies will be affected, with outcomes ranging from improved productivity to lower labor demand and job disappearance.

That does not justify stopping technological development. It does justify abandoning the assumption that whatever maximizes corporate efficiency will automatically maximize social welfare.

The Biggest AI Risk Is Not That Nobody Works

The most extreme version of the automation debate imagines billions of people permanently unemployed while machines perform everything. That future may never arrive, and focusing exclusively on it can distract from a much more plausible economic disruption.

AI can leave plenty of jobs while making good jobs harder to obtain. It can reduce entry-level hiring, suppress wage growth in exposed occupations and create enormous productivity gains that flow disproportionately to people who already own valuable assets.

That society can still have low unemployment. People can move into lower-paying service jobs, gig work, care work and occupations where human presence remains economically valuable even as high-paying knowledge work becomes more concentrated.

The unemployment rate might therefore look surprisingly normal while the underlying distribution of economic power deteriorates.

The Answer Cannot Simply Be “Learn to Use AI”

Workers should absolutely learn how to use AI. The IMF’s labor-market research shows that new skills are increasingly rewarded and that job postings requiring emerging skills tend to command wage premiums.

But “learn AI” cannot be the entire national strategy because everyone cannot gain a permanent advantage from using the same widely available tool. Once AI becomes standard software, proficiency may become a baseline requirement in the same way email, spreadsheets and internet research eventually became ordinary job skills.

The early adopters can earn a premium, but competition gradually eliminates some of that advantage. The larger question remains how many people companies need after everyone becomes AI-enabled.

Training can help workers compete for the jobs that remain. It cannot guarantee that the total number of desirable jobs grows as quickly as productivity.

We Should Want the Productivity Revolution Without Pretending Distribution Will Fix Itself

AI has the potential to produce enormous benefits. It can eliminate tedious work, accelerate scientific research, improve healthcare, reduce administrative costs and allow small businesses to perform tasks that once required entire departments.

Rejecting those gains because existing institutions distribute wealth imperfectly would be a mistake. But allowing the technology to evolve while refusing to address the distributional consequences would be another.

The economic objective should be making AI increase human prosperity rather than simply corporate output. That could involve stronger worker training, broader capital ownership, wage supplements, portable benefits, redesigned tax systems and eventually some form of guaranteed income if labor demand weakens substantially.

None of those policies has a perfect answer today. What is becoming increasingly clear is that hoping every displaced worker simply finds a better job is not a serious strategy.

AI does not have to make humans obsolete to transform capitalism. It only has to make human labor less scarce while making ownership of productive technology more valuable. If that happens, the defining economic divide of the next generation may no longer be between people with good jobs and bad jobs, but between those who own the machines and those who still have to compete with them.