In 1973, what a forty-hour week of work produced takes sixteen hours to produce today.
That’s according to Bureau of Economic Analysis (BEA) input output tables. An hour of work in American business now makes two and a half times what it made in 1973. For every forty hours that used to be needed, twenty-four have been freed.
You have probably not noticed them. The forty-hour week was written into American law by an act of 1938, and hasn’t changed. Counted for each head of population, the hours worked in American business in 2024 were 10 percent higher than in 1973. As a country, the US took the whole of fifty years’ progress as more output and none of it as time.
It was supposed to go differently. In 1930 the economist John Maynard Keynes wrote a short essay about the world his grandchildren would live in. He predicted that “the standard of life in progressive countries one hundred years hence will be between four and eight times as high as it is to-day.” And with so much plenty, he thought, we would hardly need to work at all: “Three-hour shifts or a fifteen-hour week” would be enough. He is usually said to have been right about the wealth and wrong about the leisure. Output per person in the United States in 2024 was 7.8 times its level in 1930. But look again at the sixteen hours. He was right about both. The fifteen-hour week arrived in terms of output, but not in the workweek.
I have spent a long time trying to find out where those twenty-four hours went, with a co-author, a small economy built on paper, and more national accounts than I’d like to admit. This essay is what I found. Eighteen of the hours never reached ordinary paychecks. The other six reached them in a form that couldn’t be taken as time. Both happened for the same reason, which has to do with what a wage is for. The same reason explains why people are frightened of machines that ought to be good news. And it points to a way out that is simpler, and nearer, than I expected when I started.
A note: The numbers are American, because the United States publishes the accounts this needs. I live in Sweden but it’s probably quite a similar story. The formal paper underpinning this work is here:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7226858
What is a wage for?
Three things:
A wage is a price. It tells a business what an hour of someone’s time costs, so that the hour gets used where it is worth most, and it tells people which work is wanted. A wage is good at this.
A wage is also a “ticket”. It is how nearly everyone gets a share of what is made. Bread reaches me because I have money, and I have money because I sold some hours. That isn’t a law of nature. It is an arrangement, and it has worked so well for so long that we’ve stopped seeing it as one.
And a wage is a place: a shape to the day, people to be among, a sign that somebody needs you.
One thing can do three jobs only while the jobs pull the same way. For two centuries they did, because of the kind of machine we had. The machines of the industrial age were strong and stupid. They could lift, spin, stamp and haul, and they couldn’t see, judge or decide. Nearly everything worth making took both kinds of ability, so every new machine needed people standing next to it, and the owners of machines had to bid for those people. The price of an hour rose with everything else, and a ticket made of that price was a good ticket.
The machines being built now can see, read, write and decide. I write software for a living and use them every day, and I find it harder each year to name a kind of work that is safely out of their reach.
My claim is that the price and the ticket have been pulling apart for fifty years, and that each is damaging the other, and that fundamentally this is driven by changes in technology.
How important is this?
Probably a lot. It surprises me just as much as you that some random 26 year old trans girl might work this out, but the maths has been checked, the paper is going through review, and I’ve done everything I can to try and break it. If you can see what’s really wrong with it: wilson@wilsoniumite.com
Follow a dollar you spend in a shop. Some of it pays the people who worked to make the thing. Some of it pays whoever owns what was used to make it: the oven, the truck, the building, the ground underneath. America’s national accounts are detailed enough to chase a dollar all the way back through every supplier, and when you do it for 2023, of what is paid to Americans for making the things households buy, 66 percent is pay for work and 34 percent goes to owners.
A third of the economy, in other words, already pays people for owning things, with no work asked of them. That would be a curiosity if we all held a bit of both. We don’t.
The poorest tenth of American households lives mostly on government benefits. From the second tenth to the ninth, about four-fifths of what a household earns in the market is pay for work. Then, in the richest tenth, it changes all at once, and pay and ownership are about half and half. That one tenth receives two-thirds of all the income from ownership.
Benefits look like a second source for everyone else. But 68 cents of every tax dollar in America is raised from pay for work, so a benefit is mostly a claim on somebody else’s wage. Follow the money far enough and nine households in ten are living on each other’s wages. We live off the earth and off the machines we’ve built on it. Our share of what they produce reaches us through one narrow channel, which is being paid for a job.
By the way at this point you may notice that “machines” sounds like capital, and yes, that would be the technical term. But to try and escape what we usually feel when we hear those terms, I will use work instead of labor, machines instead of capital, and land instead of non-produced inputs (the term used in our paper).
Where did the eighteen hours go?
Until 1973, what an hour of work produced and what an hour of ordinary pay could buy rose together. Between 1950 and 1973 the first rose by 81 percent and the second by 66 percent. (By ordinary pay I mean the hourly pay of everyone on a private payroll who isn’t a manager or a supervisor, measured against the prices of everything except housing.) Then they parted. Since 1973 the output of an hour has risen two and a half times, and an hour of ordinary pay buys 19 percent more. You might have heard of this as “the great stagnation” and yes, I claim to in large part be able to explain what happened, that technology is what changed. Consider that moors law is usually said to start around that period, and you begin to see what I mean.
So to buy what forty hours of ordinary pay bought in 1973 now takes thirty-four hours. Making it takes sixteen. The eighteen in between went somewhere else.
I would like to tell you they went to the owners of the machines. About a fifth of the gap did. About two-fifths of it is pay spreading out among workers, with the top pulling away from the ordinary worker. The rest I can’t separate cleanly. It is a tangle of prices, benefits and measures that don’t line up.
But look at what the largest piece is. Daron Acemoglu, who shared the Nobel prize in economics in 2024, and Pascual Restrepo estimate that between 50 and 70 percent of the change in America’s wage structure over the four decades to 2016 came from “relative wage declines of worker groups specialized in routine tasks in industries experiencing rapid automation.” Machines learned to do what ordinary workers did before they learned to do what the best-paid did.
So both pieces are the same thing happening. A wage is a price, and a price follows scarcity. It doesn’t follow effort, and it doesn’t follow need. The day a machine learns your job, you are working exactly as hard as you did the day before, and your hour is worth less. That is how a price ought to behave, and it is a fine way to decide who should do what. It is a strange way to decide who gets to live well.
That is the price shrinking the ticket. So far it has been slow.
And the other six hours?
Ordinary pay did rise, by a fifth. That could have been six hours off the week at the same standard of living. Instead the hours went up, and I don’t think it was because nobody wanted the time.
Think about what a price tells you. When I pay three dollars for a loaf of bread, the baker learns something true about me: the bread was worth at least three dollars to me, because I could have walked away. A wage is supposed to carry the same kind of information the other way. If I take a job at fifteen dollars an hour, the hour must have been worth less than that to me, or I would have kept it. A great deal of economics leans on this. It is why a job that somebody is willing to do is assumed to be a job worth doing.
It works only if I could have said no. If selling hours is the only way I eat, I will sell them for whatever they fetch, and the wage records how badly I needed the money. It says very little about what the hour was worth to me. It might have been an afternoon with my child, or a visit to my mother, or the Tuesday I would have spent running the club. Those lost a contest they were never allowed to enter. An auction in which one side can’t walk away tells you very little about what the thing was worth.
This is the ticket spoiling the price. Because the wage is the ticket, most people can’t refuse it, and a price that can’t be refused has stopped being a price. When we argue over “does money represent value”, this is one of the main reasons that isn’t true.
Two things then happen. The first is that some work gets done that isn’t worth doing. Take any job that produces a little more than the wage it pays and less than the hour is worth to the person doing it. The job covers its costs. The employer is doing nothing wrong. And every hour of it uses up something worth more than what it makes, with the whole of the loss falling on the worker. Nobody can see this, because all the prices look fine. Think uber or doordash. Would you ask your friend to pick you up because you are drunk and pay them a few dollars for it, knowing they’d rather be asleep? Maybe, just maybe, that job shouldn’t exist, at least not done by a person.
The second is that our economy has no way of saying “enough.” Other markets can. When there is more of something than people want, its price falls and less of it gets made. The market for work can’t, because for the person selling, a price that falls far enough means not eating. So we treat every job as a good in itself and every hour sold as proof that it was wanted. I once asked a friend why we assume that everyone ought to be working, and she said of course they should, she’d like to be able to buy more stuff. It’s a fair answer, and it contains an assumption, which is that every extra hour worked makes more of what we want. Labor is a resource, like water, and we don’t drain the oceans looking for a use for every drop.
I can’t put a number on what this costs, and the reason I can’t is part of the argument. Everything we measure about work is a record of what people did when they could not refuse. It is like trying to learn how much people like a food by watching what they buy when it is the only thing on the shelf. So let me ask you instead.
Do the people you know have more free time than their parents had at the same age? In a country that has made itself two and a half times as productive within one working lifetime, do they seem less pressed? When they do take time, for a new child or a dying parent, does it feel like something they are free to choose, or like something they will pay for later? And of the things that seem to have thinned out around you, the visits, the clubs, the neighbors who knew each other, how many were things that anyone was ever paid to do?
The harder evidence points the same way. Where ordinary people did gain free time over the last century, most of it came at the two ends of life, as longer schooling and longer retirement. Those are the two stretches that come with a ticket that isn’t a wage: a parent or a loan at one end, a pension at the other. And when researchers in the United States gave 1,000 low-income people $1,000 a month for three years, with no conditions, those people took roughly a sixth of the money as time. They received $11,400 a year more than a comparison group, and their other income ended up about $1,900 a year lower. The authors call that “a moderate labor supply effect.” People who could refuse a little did refuse a little.
What happens if machines get much better?
So far both failures have been slow. Nobody knows whether machines are about to get much better, or how fast. What I can do is show you what happens in an economy where they do.
An economic model is a small economy on paper, simple enough to solve and fitted to the real one wherever the two can be compared. In mine, businesses give each task to whoever does it more cheaply, a person or a machine. People are far better than machines at some tasks and only a little better at others. Machines have to be built, out of other machines, some human work and some land. Whoever pays for a machine wants a return of 10 percent a year. And there is only so much land. Fitted to the United States of today, this economy pays 60 percent of its income for work, 31 percent to the owners of machines and 9 percent to the owners of land.
Now turn one dial. Let machines get steadily better at everything, until they are a hundred times better relative to us than they are now, and let them learn to build one another.
For a long while very little happens. When machines are ten times better, work is still paid 52 percent of all income, because people move to what they are best at. Then, at about 35 times, the machines that could replace even the hardest hour of human work take less than an hour of human work to build, and from there no wage can stay above what the replacement costs. By the hundredfold point work is paid 17 percent of all income. At ten thousand times it is paid less than 1 percent.
Here is something fun though about this economic model. Technically, measured in “stuff”, nobody in this economy gets poorer. But, what stuff you can buy has changed:
At the hundredfold point total income is 32 times what it was, and an hour of work buys 14 times as many goods as it does today. What an hour stops buying is land: an hour’s pay rents about a twelfth of the land it does now. And the income of those who own machines and land is 66 times what it was.
Look at what has happened to the two jobs. As a price, the wage is working perfectly. It is telling everyone, accurately, that human hours are no longer what production is short of. As a ticket it has all but failed. It delivers a sixth of the income, in the middle of the largest boom there has ever been.
Who would that fall on? If ownership stayed as it is today and work were paid 17 percent of income, the richest tenth of American households would receive 62 percent of all market income, up from 46. The poorer half would go from 14 and a half percent to 10. People sometimes hope that an economy like that would seize up for want of customers. It wouldn’t. The owners are the customers, and the economy would turn to face them. Nothing in economics says an economy has to be for everybody.
Keep three cautions in mind. The dial measures technology and not time: a hundredfold improvement takes 117 years at 4 percent a year, the pace at which machinery improved for its price after the Second World War, and 25 years at 20 percent, and nobody knows the pace. If people turn out to have abilities that no machine can match at any price, the same dial stops with work at 48 percent. And it is a model, with one kind of worker and no politics.
So what would fix it?
There are two problems here with one cause. We ask the wage to be a price and a ticket. Being the ticket, it can’t be refused, so it has stopped pricing time. Being a price, it follows scarcity, so it is a ticket that shrinks whenever a machine learns something. The repair is to stop asking. Let the wage be a price, which it is good at, and give people a second ticket.
The question is what a second ticket can be made of.
It can’t be made of wages. A benefit paid out of today’s taxes is, for 68 cents in the dollar, wage earners paying each other. It moves the first ticket around.
It can’t be made by taxing what machines earn, which was my own first idea. In the model, a 20 percent tax on what machines earn their owners raises 6.3 percent of national income, and wages fall by 6.1 percent of national income. The reason is that investors build machines until the last one just pays the return they want. Tax the return and they build fewer, until the ones that remain earn the old return after tax. The owners keep their rate of return on a smaller pile. The people who work alongside the machines have fewer of them to work with, and are paid less. The tax goes straight through the owners and lands on the wage.
What works is owning the machines, and there is a way to do it that disturbs nothing. Let a company deduct the full cost of every investment from its taxable profit on the day it makes it. Then on every new machine the government is, in effect, putting up a fifth of the cost and collecting a fifth of what the machine earns. That is a partnership. The investor’s return on their own four-fifths is untouched, so just as many machines get built. The public gets its fifth, which is 6.2 percent of national income in the model, and wages don’t change. The honest catch is the machines that already exist, whose owners lose about a fifth, once.
Land is different, for one reason. Nobody makes it. Tax machines and there will be fewer machines. Tax land and there is as much land as before. Thomas Paine, the pamphleteer of the American Revolution, argued in 1797 that “Man did not make the earth,” and that every landowner therefore owed the public a ground-rent, to be paid out to everyone. “It is not charity but a right,” he wrote. I reached that idea on my own before I’d heard of him, which I’ve decided to find encouraging, and for a while I thought it was most of the answer. The numbers talked me out of it. The rent of the ground under America’s homes came in 2024 to between 4.3 and 7.1 percent of households’ disposable income, or between $2,776 and $4,627 a year for every person in the country. That is real money, and it isn’t what a country can live on. It grows, though. In the model, land’s share of income more than triples. And it can be collected by taxing only the growth from here on, which leaves today’s owners the rent they have now, though not the rise they were counting on.
So the second ticket has to be made of the other third of the dollar. It has to be ownership.
What would that take?
Less than you’d think. For the United States it comes to three things.
Hold a share in common, and pay everyone its income. A fifth of all the machines and land in the country would have paid $4,493 a person in 2023, or $374 a month, for every man, woman and child. Most of that fifth is within reach without buying anything: the tax America already levies on company profits, turned into the partnership I described, and the growth in land rent together come to about a seventh of all ownership income, and a public fund could buy the rest. It isn’t free money. The profits tax is already being spent, so paying it out means finding $608 billion a year somewhere else, and whatever a fund buys, somebody pays for.
Let people buy the rest out of wages. A common share of about a seventh would be enough to keep the poorer half of households whole, even if wages went to nothing. It can’t do that for the comfortable middle, who earn nearly half of all the pay in the country. They have to buy their ticket, and with the common fifth in place, saving 5 percent of pay and reinvesting what it earns at 5 percent a year would do it in 34 years, which is a working life.
And move taxes off the wage, toward spending, so that the safety net stops depending on the same thing as the incomes it is there to protect.
None of this is untried. Since 1982 the state of Alaska has paid every resident a yearly dividend out of a fund built from its oil revenues. Alaska’s dividend paid $1,312 a person in 2023 and $1,702 in 2024, to more than 600,000 people each year, and the fund behind it is worth $88 billion. Norway’s national fund owns, in its own words, “almost 1.5 percent of all shares in the world’s listed companies.” And Sweden puts 2.5 of the 18.5 percent of pay that goes toward the public pension into invested accounts that belong to the worker.
It starts small. A dividend of $374 a month is 29 percent of the poverty line for a person living alone. But it is made of the thing that is growing. In the model, the income of a common fifth is 13 percent of all wages today and 96 percent of all wages at the hundredfold point. And it answers the question I left hanging. A floor paid out of something everyone owns is mostly people paying themselves. As it grows, refusing becomes possible, a little and then more, and the wage goes back to being a price that means something.
There is a deadline. Today, everything that pays ownership income in the model economy, every machine and every acre, is worth about 6.7 years of the whole country’s wages. That is the price of a second ticket in the only currency most households have. At the hundredfold point it is 48 years. The saving that takes a household in the comfortable middle 34 years today takes 76 years there. The households that most need a second ticket are the ones who would have to buy it out of wages, and wages are the thing losing its value.
Is it fair?
The standard most often offered in defense of things as they are is that people are entitled to what they contribute, and that what you are paid measures what you add. I think that is a reasonable standard. My complaint is that its only reasonable if it’s wholly true.
Hold the kinds of income up to it. A wage is paid for work. The return on a machine is paid to someone who went without something in order to build it. The rent of land is paid for something that nobody made. Now look at what America taxes. It raises 68 percent of its revenue from pay for work, the most plainly earned income there is, and its tax on what machines earn lands, in the model, largely on wages again. By its own favorite standard, the country taxes most heavily what is most clearly earned. What I’ve described is that standard applied all the way through.
Now push it. Suppose machines can do nine-tenths of all work. The standard doesn’t break. It still says that those who work should keep their pay and those who saved should keep their return. It has less and less to say, because less and less of all income is anyone’s contribution. In the model with the program in place, pay for work goes from 60 percent of all income today to 17 at the hundredfold point and to less than 1 in the end. What is held in common, the public’s fifth of the machines plus the growth in land rent, goes from 6 percent of all income today to 40 at the hundredfold point and 48 in the end. That is about half, and nobody chose the number. It is where the contribution standard arrives by itself when work becomes scarce.
Is there any hope in this?
I think there is
The first reason is the one I started with. The hard problem, the one every earlier generation faced, was making enough. That problem is very nearly solved. We make in sixteen hours what took forty within living memory.
In 1955, paying every American a poverty-line income would have cost between 27 and 52 percent of everything the country produced. In 2024 it would cost between 9 and 18 percent. What is left is a problem of arrangement, of who holds which ticket. Arrangements are made by people, and they can be remade.
The second reason is that most of this can still be done as a trade, in which nobody is made worse off. Today, a household gets a second ticket by putting aside some of its pay and buying ownership from someone willing to sell it, and each side is paid in full for what it gives up. After a large shift, the same move is a transfer from those who gained to those who lost, which is a much harder thing to ask. We are still in the before. None of us knows for certain which side of the line our own work will fall on, and I can’t think of a better condition for agreeing on a rule.
The third is that every piece of it already exists somewhere, and works.
The fourth is what the machines themselves are offering. In the small economy on paper, the same hundredfold improvement produces two different worlds, and the only difference between them is who holds the tickets. In one, a tenth of the households receive three-fifths of the income, and everyone else lives on a wage that has stopped meaning much. In the other, people own a share of what makes everything, the wage is one of two incomes, and work is something you choose. That world is 32 times richer than ours, and your time in it is your own. It is the world Keynes was describing. He assumed that a society eight times as rich would simply work less, and it turns out that working less is something you can do only if you hold something besides your work. The wealth came to us as a country. The time has to come one household at a time, and most households have had no way to receive it.
I would be glad to be wrong about where the machines are heading. Very little of this depends on that. The twenty-four hours are already there. They were freed by people who are still working forty. What stands between those people and their own time is a second ticket, and it will not get cheaper by waiting.
I also made a video going over the maths, if you’d like to hear more:
For the careful: how sure am I?
Taken from public data, with every number and figure reproducible: that an hour of work makes two and a half times what it did in 1973, that an hour of ordinary pay buys a fifth more, and that Americans work more hours per head than they did then; that two-thirds of what Americans buy is paid for by work; that nine households in ten live mostly on pay and one in ten receives two-thirds of the ownership income; that 68 percent of taxes rest on pay; the size of land rent; and the falling cost of keeping everyone out of poverty.
From a model, and true to the extent that the model is correct: what happens when machines improve a hundredfold; that a tax on what machines earn lands on wages; that ownership costs more in years of wages the further the shift goes; and that the common part of income rises to about half. The model has one kind of worker, no risk and no politics. The result on taxing machines is a long-run result. In the short run the machines already exist, and their owners pay.
From argument, with modest evidence behind it: that a wage which cannot be refused says little about the worth of an hour, that some work is done which isn’t worth doing, and that our economy has no way of saying “enough.”
From nothing firmer than my own judgment: that machines will keep taking over tasks, and faster than they have; that a good deal of today’s work would not be done if the people doing it could afford to refuse; and that most of us would take more of our wealth as time if we could.
I have left things out. Every household is given the same average shopping basket, and renters are worse placed than that makes them look. The richest tenth is lumped together, though ownership is concentrated within it. Everything here is American.
How this was made. I wrote this with Claude Fable 5.2, Anthropic’s AI model. If you don’t trust AI, just know that these are the same models that are currently wiping the floor with every other field, and I really didn’t try and force fable to write something I already agreed with. Actually, this whole experience has been me realising just how wrong I am about some things. Anyways, Claude did the measurements with me, drew the figures and drafted most of the text from my earlier essays. The model economy is joint work with Johan Båge. Every number comes from code in a public repository, wilsoniumite/labor
Sources
The measurements are my own calculations from the national accounts of the U.S. Bureau of Economic Analysis: the input–output accounts, the distribution of personal income by tenth of households, and the accounts of government receipts. Output per hour, hours worked and the wage series are from the Bureau of Labor Statistics. The poverty line is that of Mollie Orshansky, “Children of the Poor,” *Social Security Bulletin* 26(7), 1963, updated for prices.
– John Maynard Keynes, “Economic Possibilities for our Grandchildren,” 1930, in *Essays in Persuasion*.
– Fair Labor Standards Act of 1938, section 7; 29 U.S.C. § 207.
– Daron Acemoglu and Pascual Restrepo, “Tasks, Automation, and the Rise in U.S. Wage Inequality,” *Econometrica* 90(5), 2022. Nobel prize citation: nobelprize.org, economic sciences, 2024.
– Eva Vivalt, Elizabeth Rhodes, Alexander Bartik, David Broockman, Patrick Krause and Sarah Miller, “The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States,” NBER Working Paper 32719, 2024; *The Quarterly Journal of Economics* 141(4), 2026.
– Jason Cummins and Giovanni Violante, “Investment-Specific Technical Change in the United States (1947–2000),” *Review of Economic Dynamics* 5(2), 2002, for the pace at which machinery improved.
– Thomas Paine, *Agrarian Justice*, 1797.
– Alaska Department of Revenue, Permanent Fund Dividend Division, “Summary of Dividend Applications and Payments,” pfd.alaska.gov. Alaska Permanent Fund Corporation, apfc.org, fund value as of September 30, 2026.
– Norges Bank Investment Management, “About the fund,” nbim.no.
– Swedish Pensions Agency, “National public pension,” pensionsmyndigheten.se.