A Quinnipiac poll in March found that 70% of Americans think that artificial intelligence will lead to fewer job opportunities
for human beings, up from 56% a year ago. Thirty percent say they’re worried for their own jobs. And why not? Warnings of a coming labour market apocalypse feature prominently in the remarks of AI leaders.

Dario Amodei, CEO of Anthropic, says that up to half of all entry-level white collar jobs will dissolve within the next five years. Mustafa Suleyman, CEO of Microsoft AI, believes that most white-collar work will “be fully automated by an AI within the next 12 to 18 months”. OpenAI released a policy paper calling for a 32-hour workweek so that AI creates mass leisure rather than mass joblessness. Outside my window at The New York Times, there is a large billboard for an AI company I’ve never heard of that reads: “Stop Hiring Humans.”

Thanks.

If you believe the story the AI labs are telling, it’s hard to see what stands between us and mass unemployment. AI has been designed to cheaply mimic what human beings can do on a computer, but never needs to sleep, never tries to form a union and often outperforms real people on real tasks; of course companies will want to replace human beings with this human-being-replacement machine. Maybe they already are. Tech companies like Block, Meta, Oracle and Microsoft are laying off or buying out workers and naming AI as the reason.

But it’s worth being cautious. These tech companies might be unwinding a hiring binge and telling the stock market the tale likeliest to excite or appease investors. AI leaders might understand neural nets better than they understand labour markets – or they might have bought too deeply into their own marketing materials.

For one thing, the macrodata isn’t matching the anecdata: the [US] unemployment rate was 4.3% in March; in March of 2020, it was 4.4%. Average hourly earnings are stable. Claude Code is a marvel, yet demand for software engineers is booming. Maybe mass layoffs are coming. But maybe not.

Economists, I’ve found, are quite sceptical that mass joblessness is on the horizon. In “What Will Be Scarce?,” Alex Imas, an economist at the University of Chicago, tries to clarify the mistake most AI discourse, in his view, makes. “The answer to any question about the future economics of advanced AI begins with identifying what becomes scarce,” Imas writes.

For most of human history, calories were scarce. Our energy went into finding or growing food. Agriculture steadily made food more plentiful and goods became scarce. Then goods were scarce; hand-me-down clothes were common and tools were expensive. Innovations in technology and manufacturing made goods cheaper. Then, technical knowledge became scarce: doctors, lawyers and software engineers are paid high salaries because of the rarity of what they know. The fear is that AI will make knowledge plentiful; that it will turn the fruits of learning into a commodity as surely as manufacturing turned clothing into a commodity and industrial agriculture made strawberries commonplace.

But something is always scarce. People are looking at the economy as it exists and asking which tasks AI can do; they should be asking which jobs people won’t want AI doing, or which services AI will make us want more of.

Here is a poetic finding from econometrics: as the rich get richer, they want more from other humans, not less. They “shift their spending toward goods and services where the human element, the experience or the social meaning matters more”, Imas writes. They seek out clothing with a story, food with a provenance, doctors who make house calls, therapists who make them feel seen, tutors who know their children and personal trainers who work around their injuries. This, Imas says, is “the relational sector” of the economy, and it will explode. Instead of so many human beings working with computers, they will work with other human beings.

The more automation there is, the more people value a human’s touch. Take coffee. It was once laborious to make espresso at home. Now, Nespresso machines are everywhere. Has that led to Starbucks closing and neighbourhood coffee spots dropping their prices? Of course not. There are more baristas than ever. There are more coffee shops than ever. Coffee as a commodity led to more demand for coffee as an experience. “The fact that the good is scarce is exactly what gives it meaning,” Imas writes.

Imas’ story suggests a place where human labour might move amid mass automation: toward more human roles. But it’s also possible that human labour won’t need to move that much at all.

In 1979, VisiCalc, the first electronic spreadsheet, was released for the Apple II. It could do in minutes what previously took teams of accountants days. There were predictions of mass unemployment for bookkeepers. Instead, the number of accountants quadrupled over the next 40 years. “The spreadsheet didn’t replace the accountant,” Eldar Maksymov writes. “It unleashed latent demand for financial intelligence that had been there all along, waiting for costs to fall far enough to be satisfied.”

Maksymov, an accounting professor at Arizona State University, is describing the “Jevons Paradox”, so named for William Stanley Jevons, a British economist. Jevons, writing in 1865, was interested in Britain’s use of coal. Not long before, James Watt had invented a new-and-improved steam engine, which generated more than twice as much power from a given amount of coal. But instead of cutting Britain’s use of coal, demand for coal tripled. Cheap coal didn’t lead to less coal being used; it led to coal being used for more things than anyone had previously thought possible.

This, Maksymov thinks, is what AI is likely to do even in the industries most exposed to its disruption. He thinks that, in part, because it happened before. “In every major occupational group that adopted computers heavily, employment grew faster than in groups that did not,” he writes. “Computers eliminated specific tasks within jobs – but the resulting cost reductions created so much new demand that the occupations expanded overall.”

Computers can do much that humans once did, but they didn’t put humans out of work. The ability to do more made people realise there was more to do.

This is fairly common. When I started my podcast 10 years ago, I was its only researcher; now I have an extraordinary team of people who help me prepare episodes. Has that made my job easier? Not in the least. I spend far more time researching and prepping because they bring me so much more to absorb and consider, and I choose to do more challenging episodes because I am confident I can do them.

Every enthusiastic AI adopter I know is working harder than ever because there is more they can do. Whether they are working smarter is arguable. Studies differ on whether AI is making people more productive or simply giving them (and their bosses) the illusion of productivity. Slowly reading a difficult book is far better than rapidly absorbing summaries of 12 books; struggling through a first draft will lead you to more new ideas than editing five AI-written drafts. It’s my belief that the feeling of efficiency should be mistrusted. The people I know who have small armies of AI agents working on their behalf certainly feel more productive, but I have not noticed their work improving. In some cases, it has clearly declined.

But I digress.

Ethan Mollick, a professor at the Wharton School, once told me about a test he used for AI: is it better than your best available human? The question, as he saw it, wasn’t whether AI was better than the best editor or coder or therapist or researcher or dermatologist or travel agent; it was whether it was better than the best such person you could call in your moment of need.

Over the past year, I have watched the AIs I use become better than my best available person quite often. I have an amazing editor, but he needs to sleep and work with other writers; I have a wonderful therapist, but I see her once a week, if that; I have access to good doctors, but it takes work to see them. Perhaps I had hit the event horizon I had been warned of, and AI would begin replacing the humans in my life.

But the opposite happened. The better the AI got, the more I had to discuss with the humans in my life. The AI thought my symptoms were concerning, so I made an appointment with my doctor (allergies, it turned out); it had a good insight on a personal challenge, and that opened a new conversation with my therapist; it allowed me to validate a research idea, and that opened up a new question to explore with my editor; AI has made it possible to caption videos easily, and now I work with more video editors. The better my AI has gotten, the more I’ve wanted from the human beings around me – and from myself.

While I don’t believe full automation of the economy, or even mass unemployment, is likely, I don’t totally discount the possibility. AI is a different kind of technology than what has come before: Perhaps its flexibility and conversational nature will make it a substitute when previous tools have proved to be complements. What’s likelier, though, is that AI doesn’t take all, or most, of the jobs. But it does take some. And that, strangely, is the possibility we’re least prepared for.

A world where AI displaces 8 million workers might be harder to handle than a world where it displaces 80 million workers. A mass unemployment event would force a wholesale restructuring of our economy. Covid offers an example: the shock was so total that we didn’t revert to our normal habit of blaming workers for their misfortunes; instead, we created an unprecedented architecture of income support that made lockdown a period of unexpected prosperity for millions of workers.

We are crueller when the displacement is more limited. The best estimates of job loss from competition with China put it around 2 million jobs. That’s small in the context of the entire US economy, where roughly 5 million people are hired each month and about 5 million people leave or lose their jobs each month. But it was devastating for particular communities, and we did very, very little to help them.

If everyone with a marketing degree or a trucking job found themselves jobless, we might act; if unemployment among marketers or truckers merely triples, we will do what we’ve done in the past: suggest it’s their fault, give them a few months of unemployment insurance or some retraining options that don’t work and then mostly ignore their plight.

Then there’s the reality that even as AI makes relational skills more valuable, it may make those skills rarer. Young people have gone from spending about 12 hours a week with friends in 2003 to about five hours a week in 2024. The number of high school seniors who report going on a date fell from 80% in 2000 to 46% in 2024. About a quarter of Gen Zers report that they haven’t had sex in the past year. AI could be a handmaiden of this social dissolution, offering a digital simulacrum of friendships and relationships without opening people to the beauty and agony of real relationships, in which we learn to relate to other people who are truly other people, and whose desires are not simply extensions of our own.

If Imas is right – and I think he is – our ability to relate, sensitively and deeply, to other human beings will be a central and valuable skill. That, I fear, is the exact skill we are breaking down in the young.

When I am feeling optimistic about the world AI might make possible, I imagine a world in which we are richer than we are today and are encouraged to live more fundamentally human lives, doing more fundamentally human things. When I am feeling pessimistic, I imagine something like that same world, but the wealth will be hoarded and we will value a depth of human connection that we no longer know how to provide.

This article originally appeared in The New York Times.

Written by: Ezra Klein

© 2026 THE NEW YORK TIMES