When it comes to AI taking jobs, optimists like Jensen Huang point to the history of technology and note that humans have always found something to do. Others say that history is not a reliable guide, since technology that can do certain jobs better than humans is different from an all-purpose technology that is superior to us in every way.
I believe that even the optimists on job loss do not go far enough. Regardless of how AI progresses, there will always be work. This is true even in the extreme case where machines can do literally everything humans can, cheaper and better. This is the world that Bill Gates fears:
So there’s no doubt that to date, A.I. has created more jobs than it’s destroyed. The demand for the skill sets, even just to build the data centers, is very, very high. We have a reasonably low unemployment rate.
The superiority of these systems is subject to a threshold where you have to believe that it’s incredibly reliable. If you’re going to have your telesales or telesupport capability be A.I.-driven, you want to make sure that its accuracy is better than humans…
We will, in the next few years, cross over those thresholds for accounting, legal work, telesales, telesupport — where there’s service 24 hours a day, in every language, with infinite trivia capability and no urgency.
I imagine people starting with a mental checklist of jobs, and then crossing them off as computers can do them more affordably than humans. Attorney, accountant, nurse. Check, check, check. Before, we checked off things like farming. Now, technology comes for the brain work, mops up the last remnants of physical work, and at that point there will be nothing left. Sure, the pessimists acknowledge, in the past there was always something else for humans to do, but now we are reaching beyond the limit of this principle, at which point the members of our species will be completely superfluous.
There are two versions of this argument. One fundamentally misunderstands what a job is and is pure lump of labor fallacy. A more sophisticated version acknowledges we will get productivity gains but nonetheless rests on a broken understanding of political economy in which all the gains will go to the top and the human desire for both government redistribution and charity will be abolished in circumstances where we would least expect that to happen.
Luddites of previous generations were wrong not because of the nature of technological development, but because they lacked a sufficient appreciation of human nature and fundamental truths about the economy.
To say “this time is different” reflects someone not fully internalizing the lessons of history. As I’ve written before,
In the US, the unemployment rate is about 4%. In India, the government estimates an official unemployment rate of 3.1% in 2025. What a remarkable coincidence that India has four times as many people as the US, and somehow around four times as many jobs available. The coincidences don’t stop there. In the first US census in 1790, the country had about 4 million people, including slaves. Today, we have 342 million. Yet each society seems to end up with about the same number of jobs, in proportion to the population…
So “jobs” is not the reason some countries and societies are wealthier than others. Why, then, do people in the US enjoy much higher living standards than people in India, who are better off economically than people at the American founding, who didn’t have antibiotics, cars, indoor plumbing, air conditioning, and an endless number of modern conveniences we now take for granted?
The answer is technology and productivity. In a sense, you can say richer societies have “higher paying jobs.”
To talk about Jevons Paradox, in which new technology leads to more jobs in a specific field, misses the point. Jevons Paradox is sometimes true, and sometimes false. When it comes to the overall economy, it’s a red herring.
A job isn’t usually something that fills a societal “need.” It is a social construct. We don’t actually “need” accountants, psychiatrists, professional basketball players, airline pilots, etc. We can ban all of these jobs and it wouldn’t kill anyone.
We can define a job as “something that gives humans an excuse to transfer ownership rights over green pieces of paper between one another.” To say there will be no jobs is to say there is nothing a human being can do for you, which is in effect saying that all desires will be met, including those related to status – this latter part being impossible since status is zero sum.
With Enough Productivity, We Can Literally Just Play Games All Day
Ruxandra recently wrote an article in which she accepted the jobspocalypse view for the sake of argument while saying she doesn’t necessarily buy it, but I would object that this isn’t one of those things you can accept for the sake of argument because doing so requires such a radically unrealistic view of the human condition that it changes too much about any hypothetical for it to be useful. If I said “imagine if humans didn’t have families” without specifying anything else about how the world changes, it would be a bad thought experiment, because humans without families would be a different species, so you can’t usefully speculate about what the economy or social norms would look like in such a situation.
It’s the same with jobs. To say “imagine a world without work” is to say “imagine a world where humans have all their needs met and don’t care about status.” If nothing else, wealth will always rank human beings, and so people will try to gain as much wealth as possible. They’ll think about what they can do to get it, and whatever actions they take in that process society will call a “job.” For more sophisticated people, education, moral values, and refinement are considered more important than money for achieving status, and so we will have podcasters, columnists, writers of non-fiction books, etc. That class is already post-material in a sense, and we will always continue trying to impress each other. Making busywork for others can itself be a source of status – as the nonprofit and education sectors largely are – and here the possibilities are endless.
If, as imagined under the worst-case scenario, a small number of individuals capture all the gains from superintelligence, they’ll be able to pay enormous sums for whatever human-made products and services they prefer. Even if they spend only a small fraction of their wealth this way, a tiny portion of an economy this large is a great deal of money. The humans they pay will have preferences of their own for things only humans can provide, and so on down the chain.
Remember that machines doing “all the jobs” means the jobs get done, so aggregate consumption is far higher than today. In a world where gains are widely distributed, living standards are high and you have to pay well to get talented people to provide the services you want. Think the Baumol effect on steroids. If they’re highly concentrated, the same result might require an income floor, so that people whose labor has little market value can still live well and don’t have to take whatever the rich offer. That floor doesn’t require a new political outlook as sometimes suggested; keeping welfare spending constant as a share of GDP would be more than enough. Even a modest portion of the winners’ income going to charity would accomplish the same thing.
This is why I find Luke Drago’s essay “The Intelligence Curse” so unconvincing. It assumes that private and government redistribution depend on labor having leverage over capital. But the largest transfers in the American welfare state go to retirees, who have no labor leverage at all. The union membership rate has fallen by half since 1983, while social spending has risen as a share of GDP. Redistribution in democracies rests on votes, norms, and status competition, not on elites needing workers. Political scientists have generally found weak connections between economic self-interest and political attitudes.
Resource-rich states like the Gulf monarchies spend lavishly on their citizens despite not needing their labor. The Drago view requires assuming that those who capture the gains from AI will be driven almost entirely by narrow self-interest, with no role for moral norms or the status that comes from being seen as generous or benevolent. That is as unrealistic as the Marxist belief that people can be socialized to work only for the collective good.
The idea that all mechanisms of redistribution completely disappear in a world with superintelligence is part of the denial of human nature that is required for one to fear the jobspocalypse scenario.
Wealthier societies increasingly replace unpleasant work with amusing pastimes and hobbies. Take chess. All the way back in 1997, Deep Blue defeated Garry Kasparov, the reigning world champion, in a six-game match. By a decade later, computers were clearly better at it than every human, and the gap has grown since. Yet there are probably more people making a living off the game than ever before.
The same thing has happened with poker, where the machines passed humans around 2019. But between 2016 and 2026, the number of entries for the World Series of Poker Main Event went from 6,737 to 9,208, which we might take as a (very) rough proxy for the change in the number of professional poker players.
In chess and poker, we care about the relative ranking of humans. There are more people that can devote their time to playing chess and poker than in centuries past because we are a more productive society, even though machines are superior to people in these narrow domains. There’s nothing stopping competitive gaming, artwork, and other hobbies from becoming a much larger share of the economy.
Right now, there are around 450 players in the NBA with an average salary of $12 million a year. The reason that our society, unlike modern Congo or Ancient Greece, can afford to pay people so much to play an ultimately meaningless game is that we are wealthy. We also have the WNBA, where players earn an average of around $600,000 a year, despite the superstars of that league not being as good as the top male high school athletes. We chose to have a professional basketball league for women, which we are able to do because we are rich enough that a lot of people make a lot of money doing things that there isn’t even that much of a demand for.
There’s no theoretical reason why in a world with superintelligence the number of professional athletes in society can’t increase by multiple orders of magnitude, as long as there is some minimal level of redistribution. We can have basketball leagues for men under six feet tall living in Chicago, elderly Latinas, etc. Why would anyone watch any of this? I don’t know. Why do people watch high school basketball or the WNBA? Everyone does not simply prioritize seeing the best athletes over all else. Some people watch women’s basketball to make a political statement. High school and college sports depend on fanbases that care due to their personal connections to educational institutions. The wealthier society gets, the more we can indulge these kinds of preferences.
All an individual needs to do in order to have a job is to find some way to amuse or entertain fellow humans, or be part of a competitive enterprise or community that others care about. It’s unlikely that the entire labor force ends up being basketball players. Some will play chess or Dungeons and Dragons. Others will do hobby-like or hobby-adjacent work like podcasting, playing video games, or streaming.
In 1970, the average NBA salary was $40,000. Adjusting for inflation, that would be about $342,000 today, which is less than what WNBA players currently earn. NBA salaries have increased by around 35 times in real terms since 1970 not primarily because players have gotten better in the last several decades, even though they most certainly have. Rather, GDP has grown, which means that every form of work that still exists is much more highly compensated. Sometimes, as in farming, this involves a productivity increase, but this doesn’t really apply in professional sports, which involve zero-sum contests audiences watch for pleasure – though it’s possible that players being better athletes makes watching a game more enjoyable.
It might sound ridiculous to assume that everyone can practice some kind of hobby as a job. But from a historical perspective, many things we consider real jobs today would look like hobbies to a previous generation. Imagine telling an eighteenth-century English miner coughing up a lung that one day a woman will be able to live much better than he does by handing out cups of coffee and spending her hours chatting with customers and co-workers while running a cafe. A lot of small businesses are already a kind of hybrid between a traditional job and a hobby. Those who supply toilet paper to gas station bathrooms probably aren’t doing the thing that they are most passionate about, but those who own and work in yoga studios, gyms, and marijuana shops are doing what interests them. As we get richer, we have the luxury of increasingly blending work and pleasure. I think most women who open yoga studios could theoretically do something that makes more money, but they live in a society that is productive enough that there are a lot of ways for them to meet their basic needs and they can afford to spend their days doing something that they generally enjoy.
This analysis of course applies to people who want something that we would call a job. For many, after the singularity working won’t be necessary and leisure will be too enjoyable, and they may lack ambition or a desire to play the wealth status game. But this would just be a continuation of historical trends. Everything will be a hobby, and in situations where money changes hands, we’ll call a hobby a “job.” Everything else we’ll call leisure. There won’t be people who want to work but can’t, because if nothing else a society that is so productive that it can automate all human labor will have enough resources to subsidize an endless amount of useless activity.
This would just mean staying on the same current trajectory. We spend more time in school and live longer as retirees than previous generations due to economic growth. Keynes’ dream of productivity gains turning into less work has come true. He was only wrong in assuming that the work week would be shortened, when in reality we simply extended the periods of life during which humans don’t engage in the labor force. We’re already in a post-industrial world where a lot of people who can choose not to work still choose to do so anyway because they like to feel useful, they want a standard of living that is higher, or they simply need to occupy their days.
At this point, you might say it is implausible to imagine a society rich enough to pay elderly Latinas a million dollars a year to play basketball. But people apparently find it plausible that AI will automate all jobs. These are the same thing! A world where machines can do everything humans can cheaply will be rich beyond our wildest imaginations. You would be able to find pocket change in your couch and use it to take a luxury vacation to the Bahamas, since it will be nearly costless to make an airplane, pilot it, staff the hotels and restaurants at tourist destinations, supply the food there, etc. You can set up a Latina grandmothers’ basketball league if the market demands it, or do anything else that involves competition, status seeking, and the transfer of money.
That’s the implication of unprecedented increases in productivity to the point that human labor isn’t worth hiring. The discourse is not taking this point seriously enough. If a few people capture all the gains, they’ll occasionally lose their wallets, which they won’t bother to look for because it won’t be worth their time. Then someone will find them, and use the money to finance enough games and amusements to power the economy of an entire state. Maybe here you’re thinking they won’t carry actual cash, but hopefully you get the point.
People are imagining everyone going out of work, and then maybe growth increasing to 5% a year and the gains going to the top, instead of GDP doubling or tripling each year. When growth is high enough, there’s no more need to talk about “winners” and “losers,” since everyone wins. That’s why the poorest Americans today live better than the rich of two centuries ago.
I don’t think that a world of Latina grandmother basketball stars is plausible. But that’s because I don’t think AI will automate all human work any time soon, if ever. If you do worry about technology being able to costlessly replace all human labor, then you need to take the implications of your prediction seriously enough, which include humans living better than billionaires do today. To worry about “jobs” at that point is beyond silly, but if your real concern is that we continue transferring ownership over green pieces of paper between one another, then you can rest assured that we’ll always be able to do that.
Why Animals Don’t Have Jobs
So “everyone plays games” is the extreme end of what is possible, but the more realistic scenario is that humans will still be needed, they’ll have much higher standards of living, and the most unpleasant kinds of work will become less and less common. It’s unlikely people will ever want to spend all their time playing or watching basketball, but everyone has something that interests them, and the wealthier societies become, the easier it is to monetize games and hobbies. We probably won’t see a world where 100% of workers are basketball players, but with superintelligence we should expect an economy where there are a lot fewer garbage men, accountants and lawyers, and more professional athletes, chess players, podcasters, Substack writers, escorts, prostitutes, yoga instructors, streamers, party planners, artists, and others doing more enjoyable work where the human touch is valued. Categories of work will open up that we can’t even imagine existing right now.
Maybe somebody will be a “professional family Thanksgiving planner.” Even if AI could theoretically do that job better, all it takes for professional Thanksgiving planner to be a job is for someone to come up with the idea and convince others their services are necessary before OpenAI or Anthropic gets around to doing it, or before people can realize that they can get Claude to do the same thing. In a world with super high productivity, individuals will have a lot more disposable income and their time will be more valuable, so they wouldn’t have an incentive to seek out every single way in which an AI is cheaper than a human. A lot of work involves bringing to people’s attention things that they didn’t know they needed, which humans will be incentivized to do more than AI agents. Building professional family Thanksgiving planner AI agents will be priority number ten million for Sam Altman, while it would be the main focus of an entrepreneur who just lost his job.
People sometimes make the analogy to the gap between us and apes, but we’re not even better than ants, horses, or lizards at every conceivable task. In fact, I think every animal likely does something better than humans, even from the perspective of things that our species values. As Katja Grace points out, the reason that animals don’t have jobs is because they’re incapable of coming to agreements with us specifying the terms of their employment contracts. Ants, for example, are better than humans at cleaning hard-to-reach places. We could pay them to be janitors, but we have no way to explain to them that such a transaction would make them better off. If I could communicate with a gorilla and trust it to fulfill its contracts, it would be useful to have one around the house that I could pay with bananas to move furniture around and play with my kids, conditional on it using the toilet and not ripping off the faces of any humans.
Sometimes people bring up the decline in the number of horses, which were once used extensively in farm work and transportation. Their population plummeted with the invention of cars and other forms of mechanized labor-saving technology. But the reason behind the decline of horses wasn’t that they “lost their jobs.” It was because humans were in charge and didn’t need as many of them anymore. If we wanted to keep the same number of horses alive and give them a much higher standard of living, we’re certainly capable of it. So yes, AI doom is a risk if we lose control, but the horse analogy tells us nothing about what happens in a world where AI is aligned.
It’s ridiculous to me how far ahead of ourselves we’re getting when we still have low-skilled workers giving massages, delivering food, and providing haircuts all around us, not to mention masses of people living in crushing poverty across the developing world.
Some kinds of manufacturing have been automated, but people have found other things to do because, again, human desires are unlimited. If there’s even a single spot in your house that your vacuum cleaner can’t reach, or one kind of hand movement AI struggles with, there will be jobs. If people have a slight preference for humans over machines, even for reasons of virtue signalling alone, there will be jobs. And if AI can do 100% of what we can do for cheap, then we will have basically unlimited resources to engage in redistribution and indulge even a slight preference for human-made products and services, and we can all just call our hobbies jobs and pass green paper around, treating the accumulation of money as one status game among many.
It might be the case that more people don’t work because leisure is too enjoyable and they don’t want to bother making the minimal effort to try to get green pieces of paper from others. Here we must distinguish between the unemployment rate – the percentage of individuals who want a job but can’t find one – and the labor participation rate, which refers to what portion of the public has or wants a job. The latter excludes students, housewives, NEETs, retirees, etc. Someone being retired or being a housewife generally isn’t considered bad just because they’re not working. More unemployment, not a decreased labor force participation rate, is what people worry about when they talk about job loss.
I can construct scenarios where AI fizzles out, where it leads to a utopia, or where it kills or enslaves us all. But the one thing I will guarantee is that a future with aligned superintelligence but without jobs for the people who want them is impossible. I would be willing to bet that, assuming there is no doom scenario or other kind of civilizational breakdown, two things will be true for the rest of our lives.
We will not see sustained unemployment rates higher than historical norms for reasons having to do with automation; and
People working will be earning higher and higher incomes.
We may see unemployment significantly increase at some point, but if so it’ll be for normal reasons related to economic crisis and government mismanagement. My prediction will be falsified if unemployment rates stay elevated at higher rates than they have been in previous decades and most economists agree that it is due to AI. There is zero evidence in the current data to indicate such a scenario is coming, or even possible.
Why We Shouldn’t Pace the Rate of Change
Sometimes the issue is presented as being about the pace of change. Here, it’s at least theoretically possible that this could be a problem, unlike the idea that we can permanently lack jobs. If tomorrow Anthropic announces the creation of an aligned superintelligence that makes all white-collar workers unnecessary, this argument says that society might be fine in the long run, but we need to worry about the adjustment costs in the meantime.
I think this radically underestimates how quickly markets adjust to changing conditions. Every white-collar worker has probably at some point thought to himself, “I should quit my 9-5 and open a cat cafe/become a podcaster/run my own dog-walking business.” After superintelligence, society as a whole would become a lot more productive, and those who capture the gains will have a lot of money to spend. They will increasingly devote themselves to entertainments and bespoke services that other humans can provide. People are resourceful, and they would probably be capable of convincing other, more powerful humans that they’re still necessary even when AI can do a job better.
Note also that white-collar workers are heavily invested in the stock market and own property. Their loss of jobs would be cushioned by massive increases in the value of stocks and other assets. Predictions of a massive political backlash to the decline of traditional white-collar work need to take this into account; as upper-middle-class workers are losing their jobs, their home values and stock portfolios will be seeing massive gains. Again, political forecasting is hard, and we shouldn’t adopt policies that harm productivity based on inherently speculative predictions.
The fact that Luddites have always been wrong should give us pause here. Throughout the late 1970s, the unemployment rate hovered around 6-8%. After nearly half a century of innovation, today it is 4%. The fact that we constantly get more and more automation and unemployment, when policy makers are doing a good job, always remains in the 4-10% range should tell us something. So we have:
Zero evidence from human history that technological innovation can cause massive spikes in unemployment that well-functioning market societies can’t adjust to.
Constant and irrefutable evidence that fears of technology taking jobs are harming living standards.
I see arguments for pacing job loss as a Trojan horse for Luddism. Not as dumb of a theory as permanent 100% unemployment, but still bad and an argument we should dismiss. Thinking any good can come of worrying about the pace of job losses rests on a belief that government can rationally foresee the effects of technology on the labor market and adopt wise policies that don’t end up just harming economic growth. I don’t believe either of these things is true.
Technological progress always has some losers in the short run. So I’m not saying literally every single person will be better off at every single point in time. But one has to understand that Luddism and risk aversion represent the natural state of public opinion. Attempts to pace the rate of change will almost mechanistically lead to laws and regulations that freeze the status quo in place long after the justifications for the original policy have stopped making sense, if those policies are even wisely chosen in the first place. To create new excuses for government to protect jobs against technological innovation is like leaving a recovering alcoholic alone in a room with a bottle of whiskey.
Right now, our welfare state overwhelmingly functions to transfer money to the elderly, the group that needs it least. I have no reason to believe that policies designed to slow the pace of job loss will be any more rational.
The fact that humans will be materially fine in a world of aligned superintelligence does not necessarily mean that they will all be happy or fulfilled. We see famous actors and musicians die of drug overdoses and suicides at disturbingly high rates. This is a different question than the jobs worry. Getting the economic issues right is hard enough, and people should think carefully about social problems that are caused by affluence.
But the answer to diseases of affluence is never less affluence! And just as I don’t trust government to come up with rational policy solutions to pace the rate of change, I don’t trust it to intervene in the economy to construct meaning for people in order to deal with scenarios that haven’t even occurred yet. This isn’t necessarily because government is inefficient and often irrational; even the wisest among us cannot begin to foresee what a post-scarcity world will look like. I can only tell you a world with aligned superintelligence will make humans materially better off. So we should work on alignment and improving productivity, then deal with other issues as they arise.
I’ve become increasingly obsessed with the lump of labor fallacy as the key to understanding bad economic ideas. When it comes to human progress and economic growth, the worst position of the right is hostility to immigration, while the worst positions of the left are support for labor unions and opposition to technology more generally. All of these views rest on the belief that there is a fixed set of jobs available that need to be hoarded.
To say that believing in a limited amount of labor in the world is wrong wouldn’t be going far enough. It reflects a fundamental misunderstanding of what a job even is. And to think that our political economy will ever allow a world with little to no redistribution, even as productivity skyrockets, is to express contempt for everything we know about human nature.
We need to purge every variation of the jobspocalypse scenario from public policy discussions, and use its continuing existence as evidence that even smart analysts sometimes get wrong fundamental things about the way that the world works.
I doubt we’re getting GDP growth of 100% or more any time soon. There’s an absurd disconnect between a reality where nearly half of sub-Saharan Africa lacks reliable electricity and a discourse that worries about how we’ll handle a world where America can easily give each African their own private jet for a fraction of what we currently spend on foreign aid. That’s not the world people think they’re imagining when they talk about AI costlessly replacing all jobs. But that’s only because they are not understanding the real connections between technology, productivity, and wealth.
This post is sponsored by Mechanize. AI alignment is perhaps the most important issue of our time, and if you’re interested in being part of that effort, Mechanize is hiring software engineers at a base salary of $300K-$400K/year, plus substantial bonuses. They also have openings for other positions, including Recruiter ($200K). Here is their website, and you can apply here. If you apply, make sure to use the links provided here so I can receive credit, and name me as a reference in the process.


It's insane how much we're arguably already living in the world that AI pessimists fear, i.e. only a minority of people are involved in important productive work that we actually need, and an even smaller percentage are involved in important research and development that has a realistic shot at improving our society.
Many people already waste substantial percentages of their life on unproductive endeavours, and/or choose service sector jobs that are as much a service to themselves as to others. Streamers or game developers are of course the purest examples, but many coffee shop owners, political activists, or almost anything relating to pets, particularly expensive ones like horses, and of course sports, as you mentioned. Anyone who has interacted with it knows that the bureaucracy is partially a place to generate make-work for people who wouldn't cut it anywhere else, and another part is actively inhibiting productive work of others. And most people I know who work in large companies will have plenty of stories of dysfunction, bloat and inefficiency, even if the company might at least in principle do something worthwhile.
If the percentage of work humans perform which is objectively important goes from 10% to 1%, would the average person even notice? 1% to 0.1%? Does it matter to their day-to-day? I don't think so. And as you say, entertainment, make-work and status games can be indefinitely generated for the rest.
Almost everyone is simultaneously too cynical and not even nearly cynical enough.
My bigger worry is the possibility of disenfranchisement if the AI really is that much better than us, becomes fully agentic and holds the majority of the levers of power, but there are quite a few more steps from here to there.
Excellent post. This is interesting to think about:
"He was only wrong in assuming that the work week would be shortened, when in reality we simply extended the periods of life during which humans don’t engage in the labor force."
People often make fun of Keynes's prediction, but (as you suggest) much of our current work is actually a disguised form of play. If you subtract all the time office workers spend surfing the web and chatting with coworkers, I wonder if we don't already have the very short work week that Keynes predicted.