Chapter Nineteen: The Combine

When the mechanical combine harvester spread across the American farm in the nineteenth and early twentieth centuries, it must have looked to the farmer like deliverance. Here was a machine that could reap and thresh and winnow in a single pass, doing the work of dozens of men, turning weeks of brutal stoop labor into a few days riding high on a seat. Surely a tool this powerful would make the farmer who owned it rich. It multiplied his output many times over. How could the result be anything but prosperity?

Look at what actually happened. In 1870, roughly half of all American workers were farmers. Today the figure is around one or two percent. The combine, and the machines like it, did not make the average farmer rich. They made food astonishingly cheap — a gift to everyone who eats, which is to say everyone — and in doing so they made the farmer himself almost entirely unnecessary. The output stayed; in fact it exploded. The price of that output collapsed. And the people who used to produce it, the great mass of them, were simply no longer needed. The winners were the consumers, who got cheap food, and a small number of large landowners, who could now farm thousands of acres with a skeleton crew. The losers were the tens of millions of farmers who discovered that the wonderful machine had not enriched them but erased them. They did not ride the combine to wealth. They left the land.

I begin here because this is the single most important and most misunderstood pattern in the history of technology, and we are about to live through the largest instance of it ever — while almost everyone gets the direction of its effect exactly backwards. Artificial intelligence is the combine. And the knowledge workers cheering it on as the dawn of limitless prosperity are the farmers of 1870, admiring the gleaming machine that is about to cheapen their output to nothing and make most of them redundant.

The Combine for the Mind

Everything the combine did to the economics of harvesting grain, artificial intelligence is now doing to the economics of knowledge work.

Consider what AI is genuinely good at — and it is genuinely, startlingly good at these things: legal research, financial analysis, writing and editing, computer programming, graphic design, translation, the drafting of reports and memos and contracts, the whole vast category of sedentary, information-based work that the educated classes have spent the last half-century moving into. For each of these, AI does what the combine did to the wheat field: it takes a task that used to require many trained, expensive human beings and performs it at a tiny fraction of the cost, at enormous scale, in a fraction of the time. And the economic consequence is identical. The price of the output collapses. Just as the combine drove the price of harvested grain down toward the cost of the diesel it burned, AI is driving the price of a legal memo, a financial model, a block of working code, a competent essay, down toward the cost of the electricity and silicon required to generate it — which is to say, toward nearly nothing.

This is the thing to hold onto, because it cuts against everything you are being told. The defining economic property of artificial intelligence is not that it creates. It is that it cheapens. It collapses the price floor under everything it can produce.

Deflationary, Not Generative

The technology is being sold to the world under a particular word — generative — and the word is doing a great deal of quiet work. "Generative AI" suggests creation, abundance, the bringing-into-being of new value, the greatest wealth-creation event in human history. That is the marketing. The economics tell a different and opposite story, and the right word for the economics is deflationary. AI does not, in the main, create a new thing with a new price. It collapses the price of an existing thing. And the difference between those two descriptions is the difference between the future that is being promised and the future that is actually arriving.

Now, you might reasonably say: so what? Cheaper is good. Cheaper food made everyone richer; cheaper information should too. And in a certain light, and in a certain kind of economy, that is exactly right — falling prices are one of the great engines of rising human welfare. But there is a shadow side to "cheaper" that we are trained not to see, and it is the hinge of this entire chapter. Every price is someone's income. The price of a thing, from the other direction, is the wage of the person who makes it. When the price of legal research falls toward zero, that is a windfall for everyone who needs to buy legal research — and it is the evaporation of the livelihood of every person who used to earn a living producing it. AI, in this sense, does not so much create wealth as transfer it: from the producers of knowledge work, whose incomes vanish, to the consumers of it, whose costs fall. And — this is the part the cheerleading misses entirely — it shrinks the total pool of income that the knowledge economy used to support, because the salaries it eliminates were large and the compute that replaces them is cheap.

This is what people mean, or ought to mean, when they say the unfashionable thing that money is, at bottom, a measure of human labor. You do not have to accept it as an iron law to see the mechanism: an economy's income is, in the end, payment for human effort, and when you drive the price of a whole category of human effort toward the cost of a machine that never gets paid, the income that category supported does not transfer somewhere else. It substantially disappears. The combine did not pay the displaced farmers; it just stopped needing them. The model does not pay the displaced analyst; it just stops needing her. Wealth, in the sense of dollars flowing to human beings as payment for what they do, is precisely the thing this technology destroys.

The Hidden Condition

At this point every economist in the room is ready with the decisive objection, and I want to state it in its full strength, because it is correct as far as it goes and because answering it is the whole point.

We have heard this before, the objection runs, and we have always been wrong. The Luddites smashed the power looms in terror that the machines would destroy their livelihoods — and weaving employment, after a wrenching transition, grew. The scribes were ruined by the printing press — and the press created publishing, journalism, and a literate economy employing orders of magnitude more people than copying manuscripts ever had. The farmers were displaced by the combine — and they and their children flowed into the factories and offices of a booming industrial nation, into jobs that had not existed before, and the country grew vastly richer. This is the iron historical pattern: labor-saving technology destroys particular jobs and creates more jobs in total. The pessimists who cry that this time the machines really will take the work have been wrong every single time for two hundred years. Why should AI be different?

It is a serious argument and it deserves a serious answer, and the answer is this: notice the hidden condition that made the happy ending happen every single time — the condition so constant that it became invisible, like water to a fish. In every one of those transitions, the displaced workers were absorbed because the economy was growing. There was a rising population creating ever-new demand, opening ever-new industries, generating ever-new categories of work into which the redundant weavers and scribes and farmers could pour. The factories that absorbed the displaced farmers existed because the population was exploding and needed everything a factory could make. The absorption mechanism — the thing that turned every technological catastrophe into a technological blessing — was growth. It was, quite literally, the same population explosion we began with. The machines never created the new jobs by themselves. Growth created the new jobs, and the machines simply freed up the workers to fill them.

And growth, as the entire first half of this book has labored to establish, is over.

Structural, Not Transitional

This is why this time really is different — not because AI is a crueler or more powerful technology than the loom or the combine (the mechanism is exactly the same), but because the demographic condition that always rescued us has been removed. The optimists are right about the technology and wrong about the world it is arriving in.

Economists distinguish two kinds of unemployment, and the distinction is everything here. Transitional unemployment is what happens when a technology displaces workers but a growing economy generates new positions to absorb them — painful, disruptive, but temporary; the farmer's son becomes a factory hand. Structural unemployment is what happens when the displacement occurs and there is no absorption mechanism, no new industry being born downstream to catch the fall, because there are no new people to demand new things. The displaced worker is not between jobs. He is simply surplus. Every prior wave of automation produced transitional unemployment, because it landed in a growing population. AI is landing in a flat and aging one. Same machine; opposite world. The combine worked out for America because America was about to triple in size and pour its displaced farmers into a hundred new industries. AI is arriving in a society that has stopped growing — and a labor-displacing technology in a non-growing economy does not redistribute work. It destroys it.

There is a feature specific to AI that sharpens this further, and it is worth naming precisely, because it is why the coming displacement will show up not as falling salaries but as vanishing jobs. Call it the multi-threading effect. AI does not merely make a given knowledge worker cheaper; it lets a single worker do the work of several at once — running parallel streams of work simultaneously, drafting while researching while revising, occupying the cognitive bandwidth that used to require five separate people. So the effect on a workforce is not that everyone's wage drifts gently downward. It is that the number of positions collapses. The firm that needed four junior associates needs one; the content team of ten becomes two; the office that ran on a dozen analysts runs on three. The credential gate that the young generation went into such debt to pass through still stands, but it now opens onto a smaller, more crowded, lower-paid room than the one it promised, with the AI itself installed as the new gatekeeper. The slots are not getting cheaper. They are disappearing.

The Two Deflations Meet

And now watch the two great forces converge, because they converge on a single population with a precision that would be elegant if it were not so brutal.

A declining population deflates assets — fewer people every year to want the houses, the stocks, the stores of value, so their prices must eventually fall. Artificial intelligence deflates labor — driving the price of knowledge work toward the cost of compute. These are two separate deflations, arising from two separate causes, demography and technology. And they are arriving, at the same moment, on the same people: the aging professional and credential class, who hold the assets that the shrinking population is deflating and perform the knowledge work that the AI is deflating, and who are therefore losing the value of what they own and the value of what they do, simultaneously, in the same decade. Meanwhile — recall the displacement gradient, the inversion by which the physical and embodied work machines find hardest is automated last — the young tradesperson, the plumber and the electrician and the nurse, retains the value of her labor longest, precisely because hers is the work the combine for the mind cannot touch.

The technology that was sold to the developed world as its rescue from demographic decline turns out to be the second blade of the very scissors closing on it. AI does not save the shrinking, aging society. It accelerates the part of that society's reckoning that involves the credential class losing everything at once.

So if AI deflates wages and headcount and assets, where is all the wealth it is supposed to generate? Surely, you will say, even granting everything in this chapter, the enterprise as a whole still makes sense — the companies building these systems will be the most valuable in history, the productivity gains are real, the investment is sound. Trillions of dollars are being staked, right now, on exactly that proposition. And that investment thesis rests, every dollar of it, on a single buried assumption: that there will be a large and growing market to sell all this cheap output into. Capital only pays off if the future is bigger than the present. AI produces — but it does not consume. A machine that builds a house does not then need to buy one.

Why the largest investment boom in human history is built on a market that will not be there is the next question.