In the nineteen-fifties, a plant breeder at the University of California, Davis, named Jack Hanna set out to solve a problem that was not, strictly speaking, a plant problem. Down the hall, an engineer named Coby Lorenzen was building a machine to harvest processing tomatoes. The machine worked, in the sense that it could move through a field and separate fruit from vine. It also worked in the sense that a bulldozer works. The tomatoes came out the other end as sauce.
Hanna's answer was not a gentler machine. It was a tougher tomato. What he eventually released was oblong rather than round, thick-walled, and bred to ripen all at once so that a field could be taken in a single pass. It could be shaken off the vine and dropped into a bin without splitting. It was, in every meaningful sense, a fruit designed around the tolerances of a piece of steel.
The harvester arrived in California fields at almost exactly the moment the Bracero program ended, in 1964, and the hand-picking crews it had supplied went away. Within about a decade the number of tomato growers in the state had fallen by something like eighty percent. The small ones could not afford the machine. The large ones could not afford to be without it. The whole thing became a minor scandal: a land-grant university, funded by the public, had spent public money engineering the public's own workers out of a job.
I keep coming back to this story because it inverts the usual account of mechanization. We tend to say that machines replace hands. What actually happens is subtler. The world gets rearranged until machines can work in it. Nobody built a harvester delicate enough for a tomato. They built a tomato coarse enough for the harvester.
Hold that thought.
There is an argument circulating right now that I find appealing, and it goes like this. Mechanized agriculture freed up farm labor, which went into factories. Mass production freed up factory labor, which went into services. Services threw off enough surplus to fund an entire economy of people sitting at desks manipulating symbols. Now artificial intelligence is coming for symbol manipulation. So where does that labor go?
Back to the land, says the argument. Not to the industrial farm, which is already automated, but to the artisanal one. Small-scale, labor-intensive, hands in the dirt. The argument has a nice symmetry on both blades of the scissors. On the demand side, people with money increasingly want food that has a story attached: the heritage breed, the raw-milk cheese, the single-origin everything. On the supply side, a generation raised on screens is visibly hungry for physical work, weather, animals, fatigue that comes from the body rather than the eyes. Full circle.
I want this to be true. I think large parts of it are. But the circle is the wrong shape, and the part that's true is true for a reason more unsettling than the argument lets on.
Start with why the intuition is sound, because it is. In 1966 the economists William Baumol and William Bowen were trying to explain something that had puzzled arts administrators for a century. Why does live performance keep getting more expensive relative to everything else? Their answer was simple. It takes four musicians roughly forty minutes to play a Beethoven quartet. It took four musicians roughly forty minutes in 1826. There is no productivity growth available. Meanwhile the rest of the economy is getting relentlessly cheaper at making things, and those musicians have to be paid enough not to go do something else. So the cost of the quartet rises, not because anything about the quartet changed, but because everything around it did.
This is the real engine under the whole argument, and it generalizes hard. Whatever cannot be sped up becomes, over time, the expensive thing. If AI does to cognitive work what the assembly line did to manufacture, driving the marginal cost of a competent memo or a legal summary or a decent illustration toward zero, then the scarce, dear, high-status goods will be the ones that still require a specific human body in a specific place for a specific span of time. That is not a sentimental prediction. It falls straight out of the arithmetic.
So far so good. The trouble starts when you ask where all those bodies will go.
In the eighteen-fifties a Prussian statistician named Ernst Engel noticed that as a household's income rises, the share it spends on food falls. Not the amount, the share. Rich people eat better, but they don't eat proportionally more, and the fraction of their money that goes to calories shrinks steadily. This turns out to be one of the most reliable regularities in all of economics, and it is brutal news for the back-to-the-farm thesis.
The shape of it is roughly this. Americans spend something like eleven percent of disposable income on food, and about half of that is restaurants, which is to say service rather than agriculture. Of the money spent on food eaten at home, the farmer sees maybe fifteen cents on the dollar. The rest is processing, packaging, transport, refrigeration, retail margin. Farm employment sits under two percent of the American workforce.
Run the numbers with a generous hand. Suppose the artisanal premium goes wild and everyone triples what they pay at the farm gate for everything. You've moved a very small number to a somewhat less small number. The sector simply is not large enough to absorb a meaningful share of displaced knowledge workers. There is no version of this where the accountants and the copywriters and the junior analysts go to the fields, because the fields, financially speaking, are a rounding error.
Which means the thesis has the mechanism right and the address wrong.
The better frame is that the scarce good isn't farming. It's embodiment.
Think about what the Baumol logic actually selects for. Not agriculture specifically, but any work where the human body is the irreducible input: care, craft, trades, hospitality, repair, teaching, performance, the whole enormous category of things that have to happen at a particular address. Someone has to be in the room. An AI can design a beautiful staircase and cannot come to your house and cut the stringers.
Inside that category sits a smaller one that is automation-proof by definition rather than by difficulty. These are the goods where human provenance is the product. A robot-made artisanal cheese is not artisanal. That is not a claim about robot cheese being worse; it might well be better. It's a claim about what the word means. When you buy the cheese you are buying, in part, the fact that a person you could theoretically go and meet stood in a room and made it. Automate that and the thing you were paying for evaporates, however good the cheese is.
It's a strange and durable little economic fortress. It is also, historically, a trap, which is the part of the argument I'd want anyone making this case to sit with.
Because we have run this experiment before. Twice, at least.
The first time was the Arts and Crafts movement. William Morris looked at Victorian industry, saw joyless work producing ugly objects, and proposed to fix both at once with handwork. The wallpapers and textiles that came out of it are still, a century and a half later, beautiful. They were also expensive, because handwork is expensive, which meant they were bought almost exclusively by the wealthy. Morris, a committed socialist, was tormented by this. He had set out to give ordinary people beautiful things made by dignified labor and found himself, by the logic of his own costs, decorating the houses of rich men. He said as much, bitterly, more than once.
The second time was the nineteen-seventies. Somewhere north of a million Americans went back to the land: communes, homesteads, goats, a great many failed attempts at self-sufficiency. Nearly all of it dissolved within a few years. The standard explanation is that hippies discovered farming is hard, which is true but lazy. The interesting failure was structural. They tried to opt out of the market entirely and be self-sufficient in everything at once, in food, shelter, energy, childcare, medicine. That is the most labor-intensive possible way to live, and it is precisely the arrangement every peasantry in history escaped the moment it was permitted to. You cannot get good at anything if you have to do everything.
Now hold that up against craft beer, which is the same impulse and which worked. In 1980 the United States had fewer than a hundred breweries. Today it has something like nine thousand. That is a sustained migration of labor into a lower-productivity, higher-margin, human-provenance sector, and it stuck.
Why? Homebrewing was legalized federally in the late seventies, which mattered. But the deeper difference is that the brewers did the opposite of the communards. They did not exit the market, they entered it aggressively. They did not try to be self-sufficient. They specialized ruthlessly, picked one product, and bought everything else from the industrial economy they were supposedly rejecting. They kept the price mechanism and used it, charging four times macro-lager prices and getting it. And they were selling something with a high value-to-weight ratio that people consume socially, in public, as a statement about themselves. It is very hard to signal with a turnip.
The lesson generalizes uncomfortably well. The artisanal sector that survives is not the one that rejects the industrial economy. It's the one that parasitizes it, taking cheap industrial inputs, adding human labor at the last mile, and selling the labor as the product. Every successful small farm I know of does exactly this. The romance is in the output. The supply chain is Amazon.
So it's a spiral rather than a circle, and the distinction matters more than it sounds.
Consider what actually changed. Peasant agriculture was not a lifestyle. It was low-status, involuntary, and, for most of history, the thing everybody was trying to get their children out of. The great migrations of the last two centuries all ran one direction, from the field toward the town, and they ran that direction for reasons the people involved could articulate perfectly well.
The contemporary small farmer performs many of the same physical motions from an inverted social position. She is high-status. She is often, quietly, subsidized by an off-farm salary, a spouse's health insurance, inherited land, a partner in tech. And the reason her work reads as meaningful rather than degrading is that she chose it and could stop. The exit option is not incidental to the dignity. It is most of the dignity.
Which is why "full circle" undersells the phenomenon. Nobody is going back to peasant agriculture. What's on offer is peasant agriculture's choreography, performed by people with capital, insurance, and an escape hatch. That isn't a return. It's something new, and it deserves to be described as new rather than dressed in a costume borrowed from people who were, by and large, miserable.
Now the part that decides everything, and that the economics is silent on.
The artisanal premium has to be paid out of somebody's income, and in this scenario that income is generated by the AI-productive sectors. Every dollar spent on the good cheese is a dollar earned somewhere the machines are doing the work. So the whole question of whether this future is attractive collapses into a distributional one. How broadly are the gains from automation shared?
If they're broad, you get the good version. A large middle class can afford human-made things, the people making them earn a real living, and a substantial share of the population does physical, skilled, locally-rooted work and is well paid for it. That's an appealing society, and I don't think it's fantasy.
If the gains are concentrated, you get the other version, in which a small class of artisans and service workers ministers to a very wealthy few. Here it's worth being precise, because we know exactly what that looks like. It is the world of 1900. In Britain at the turn of the century, domestic service was the single largest category of female employment, well over a million people. In the United States the figure ran to a similar order. That was an economy in which an enormous fraction of the workforce did embodied, in-person, non-automatable labor. Cooking, cleaning, dressing, driving, tending. It was not liberation. Embodied labor has already been the destination once, and the thing that determined whether it was decent work was never the work. It was who had the money and what obligations came attached to it.
This is the fork, and nothing in the underlying economics picks a branch. Baumol's logic tells you that human labor becomes relatively expensive. It doesn't tell you whether the humans doing it are craftsmen or servants. Those are the same job description under different distributions of power.
Which brings me back to Hanna's tomato.
The thing people usually take from that story is a lament about mechanization, and that isn't quite it. The tomato harvester was a good machine. The scandal wasn't that a machine existed. It was that when machine and organism failed to fit, the question of which one should change got settled by whoever was holding the design drawings, and the people who lost their jobs were not in that room. Neither was the tomato.
The coming decade is going to pose that question about a very large number of people at once. It won't present itself as a question about tomatoes. It will present itself as productivity, as efficiency, as the natural course of things. And there is a real possibility, on the other side, of a world in which more of us work with our hands, outdoors, on things we can see the end of at the close of the day, which would be, by most measures of human happiness, an improvement on the open-plan office.
But whether that world is Morris's dream or Morris's disappointment does not depend on the technology at all. It depends on how the surplus gets divided, which is a political question wearing an economic costume.
The interesting question was never whether we go back to the fields. It's who arrives as Hanna, and who arrives as the tomato.