AI Won't End Scarcity. It Will Change What We Compete For.

Start with the thing I can actually see from my own desk. Over the last few years the models made writing software faster, and I know almost nobody working fewer hours. What moved was the expectation. The ticket that used to take a week is quoted at two days. The prototype is wanted before the meeting rather than after it. The finish that used to count as done now counts as a first pass.

I should be careful here, because my evidence is a feeling and feelings are unreliable in exactly this way. In 2025 METR ran a randomised trial with sixteen experienced open-source developers across 246 tasks on repositories they knew well. With AI tools allowed, they were about 19% slower. Afterwards, those same developers estimated the AI had made them roughly 20% faster. METR now marks the result as historical and no longer representative of current tools, and I would expect a rerun to look different. But it should permanently change how much weight anyone puts on "it feels faster", mine included.

So hold the modest version, which is all I need: whatever the real gain was, it did not turn into free afternoons. It got absorbed. That is the whole argument of this essay, arriving in the first hundred words, and the rest of it is about where absorbed capacity goes.

The promise that keeps arriving

Every technological revolution turns up carrying the same suitcase, and the same promise is inside it: this time the struggle is over. Steam was going to end exhausting labour. Electricity was going to hand us leisure. The assembly line was going to make necessities so cheap that want would simply stop. The computer was going to produce the paperless office — a phrase that now reads like a joke told by someone who has never watched a printer jam at a quarter to five on a Friday.

Every one of those was half true, which is the most dangerous kind of true. Each did make something genuinely abundant, and each quietly turned something next to it into the thing everybody competed for instead: not grain but land near a railway, not steel but the refinery, not airwaves but the spectrum licence, not software but the app store standing between a maker and his market.

So the claim I want to defend is narrow, and it is not a prediction about capability. Technology does not eliminate scarcity; it relocates it. AI is now driving the cost of producing information — code, analysis, drafts, images, plausible-sounding arguments — toward zero. What that makes scarce is everything the production of information used to smuggle in for free: judgment, accountability, and the physical plant that intelligence has to run on. Scarcity does not leave the economy. It changes address.

The prediction

In an interview with The Economist, Elon Musk predicted that within about ten years people will no longer need money: by roughly 2036, robotics and AI will produce such a volume of the goods we currently buy that the dependence simply ends. The editor interviewing him, Zanny Minton Beddoes, objected that handing out income that way would only cause inflation, and noted that she was an economist.

Miguel Morgado, writing in Observador, went after the objection rather than the prediction. Inflation belongs to a world of scarcity, and economics is the study of human affairs under scarcity; answering a claim about abundance by invoking it misses the distinction that precedes every other one in the discipline. He calls the ten-year deadline impossible, calls the very idea of a clean historical threshold problematic, then sets that aside and leaves standing the question nobody can engineer: what becomes of a person when the struggle against scarcity ends?

I want to treat that as a forecasting problem rather than an alarm. And we are not guessing from nothing, because parts of it have already happened.

What the last great transition actually did

The usual story is agriculture. In 1790 about ninety percent of the American workforce farmed; today it is under two percent. It is tempting to say machines took eighty-eight percentage points of humanity's oldest occupation, and I believed that until I looked properly.

The number is real and the reading is wrong. Absolute farm employment kept rising until around 1910. The share fell because everything else grew faster, and most of the fall had already happened by 1900, largely before the tractor. What did it was railways and refrigeration opening distant markets, fertiliser and better seed, the fact that people do not eat proportionally more as they get richer, wages pulling workers toward towns, and later agricultural policy and migration. Machines mattered, but they arrived to finish a job that other things had started.

And the two percent is an accounting artifact as much as a fact. The work did not evaporate; a great deal of it moved into making tractors and fertiliser, into processing, packing, refrigeration and freight. We stopped calling those people farmers. The category emptied faster than the labour did.

Which is a better lesson than the one I was going to draw, and it is the one this essay is built on. Work is reorganised and relabelled far more often than it is abolished. Keep that in mind every time someone counts the jobs a robot will take.

The boats

My grandfather fished for a living, his father before him almost certainly too, and some of my relatives still do. It is the case I can check by asking.

The boats that go out now carry sonar, GPS, weather models that arrive before the weather does, and motors that hold a position by themselves while the tide moves underneath. Nobody was replaced by a machine. What the instruments removed was a particular set of uncertainties, and less of it than you would think.

The sonar does not even report what is under the boat. Fish holding tight to the seabed sit in what acousticians call the dead zone, where their echo cannot be separated from the far stronger return of the bottom, and it is worst over rough, rocky ground, which is precisely where fish like to be. What you get is a partial, ambiguous picture that still has to be read by someone who knows the ground. It does not tell you which of your marks to run to at four in the morning, out of the several that have produced before and the several that have not, with fuel burning either way and a day that happens only once.

And a mark that another boat is already sitting on is worth less, sometimes nothing. Its value depends on who else knew and who got there first, which makes it scarce in a way no equipment can fix. Nor did the instruments end the experimenting: a cousin of mine is permanently running trials, a different method one week, a different bait, a different lure, then a comparison against whatever the last attempt produced. The sonar forms no hypothesis about why the fish took one thing on Tuesday and refused it on Thursday.

I would not bet on a robot taking that boat out alone any time soon. A kitchen is a controlled room and a warehouse floor is a solved one; a working boat is weather that changes while you are in it, a deck that will not hold still, gear that fouls, animals that do not cooperate, and a cost to being wrong measured in lives.

But the fleet did shrink, and I should not pretend otherwise or blame the wrong thing. Aquaculture is usually accused, and the accusation is too simple: farmed production only overtook wild capture in total tonnage around 2022, though it passed it for human consumption earlier, since much of the wild catch becomes fishmeal. Meanwhile wild landings have looked flat for decades at roughly ninety million tonnes, but flat is not healthy. The share of stocks fished within biological limits has fallen from about 90% in the mid-1970s to around 62%, and that steady tonnage has been held up by fishing further out, deeper, and further down the food web. Add quota, decommissioning and consolidation, and you get the same weight of fish landed by far fewer people. The causes tangle well beyond what one essay can unpick.

What is clear is that wild fish did not become worthless. Plenty of people prefer it and pay more for it, so the old way stopped being the default and became the premium. That is worth naming as its own pattern: sometimes you are not replaced and not restructured, you simply keep going at the same scale while something else takes all the growth, and you look up to find you are a niche with better margins and fewer colleagues.

The house

The case closest to what Optimus actually promises is housework, and it is the one that complicates everything above.

Cooking left the home first, and not into robot kitchens. It went into industrial food processing above all, and into restaurants, delivery and the freezer aisle. Cleaning divided along a line you could have predicted, though it has travelled less of it than the enthusiasm suggests: a machine has been able to vacuum an open floor since 2002, but most people still push the vacuum themselves, and where the robot exists it leaves the stairs, the edges, the corners and everything under the low furniture. More than two decades on, the easiest domestic robotics task there is remains partial in what it does and partial in who has it.

Laundry is the sharpest case. The washing machine removed one of the heaviest physical burdens in domestic life, and for the women who carried most of it that was liberation of a real kind. But Ruth Schwartz Cowan's More Work for Mother (1983) documented what happened next: standards rose to absorb the saving, clothes were washed after one wearing rather than several, and the machines took over work that had previously been shared with hired help, children and men, so what remained concentrated on one person. Her constancy is period-specific and I should say so, because time-use research since finds women's housework falling substantially from the 1960s onward. The hours did eventually come down. They just did not come down when the machine arrived, and they did not come down for whoever was left holding the residue.

There is also a whole category the machines have not touched at all, and I nearly wrote this section without noticing it. Nobody in the story above remembers anything. Somebody has to know the washing-up liquid is running out, that the shoes no longer fit, that the dentist needs booking and the school needs answering and the prescription needs collecting. Somebody holds the running model of a household in their head, and that load does not sleep. If you want the clearest existing example of the thing I claim survives automation — judgment, knowing where the boat should be — it is not a boat at all. It is whoever is carrying that.

Machines reduce effort. They do not absorb responsibility, and responsibility turns out to be the part that was scarce all along.

Which also means the automation order everyone predicts is ranked by the wrong criterion. Folding and ironing are usually named as the frontier because cloth is a hard robotics problem, and it is. But folding is the pleasant end of the day, done sitting down. What actually grinds people down is the transport and return: unloading the dishwasher, gathering laundry from around the house, putting clean clothes back in the right drawers, taking the bins out. Rank by difficulty and you get one list; rank by what costs a person their evening and you get another, and nobody is building for the second.

One more thing worth noticing. Your house is already full of robots. The washing machine, the dishwasher and the vacuum are robots; we stopped calling them that when they became furniture. Every one of them won by narrowing the problem rather than solving it in general. Nobody built a machine that washes clothes the way a person does. They built a drum that spins, and then redefined laundry around the drum. The humanoid is a bet against that entire history.

What the pattern says

Put those together and the shape is consistent. When the doing gets cheap, money does not evaporate. It migrates to whatever is still scarce. In farming it went to land, machinery, seed and freight. In my own trade, answers became free and the spending moved toward compute, subscriptions and the electricity behind them — though I should not overstate it: companies still spend far more on the people writing software than on the machines helping them, so what moved is a slice at the edge rather than the whole budget changing hands.

Two mechanisms make this more than an anecdote. Baumol's cost disease, from the study William Baumol and William Bowen published in 1966: a string quartet needs the same four players and the same forty minutes it needed in 1800, so as productivity rises everywhere else, the quartet gets relatively dearer. Relatively, not absolutely — Baumol's own conclusion was that we can still afford quartets precisely because we are richer. And positional goods, Fred Hirsch's term from Social Limits to Growth (1976): things valuable because not everyone can have them, whose scarcity is relational and therefore untouched by production. The fishing mark that is worth nothing once another boat is on it is exactly that. No fleet, however well equipped, can make everyone first.

Which is why I do not think there is a threshold. Musk and his interviewer were arguing about a line we cross into abundance or fail to reach. Abundance does not arrive everywhere at once; it arrives one domain at a time, and each arrival hands the same old economic questions to whatever has become the binding constraint. There is no Tuesday on which scarcity ends. There is a bottleneck that keeps changing address.

Beneath all of it sit energy and materials, and I would be wrong to promise those a smooth decline. Energy is priced by politics and conflict as much as by technology, and a war can undo a decade of progress in a season, as anyone who has opened a gas bill lately can confirm. That is not a caveat to the argument. If the abundance runs on energy, energy is exactly where the scarcity concentrates.

So what gets expensive

On the physical side, the bottleneck moves down the stack: energy, materials, land, compute. There is a version of the abundance argument that treats intelligence as though it were weightless, and it is not. It runs on chips a handful of firms can design and fewer still can manufacture, in buildings that need land, cooling and water, and — the constraint fewer people mention — a grid connection and a transformer, both of which now carry multi-year queues across much of the developed world. When one link in a chain gets cheap, the value does not evaporate; it migrates to whichever link is still constrained. I have argued before that the models themselves are nobody's moat, and I still think that — everyone orders from the same short menu. But the layer underneath the model is not a commodity, and that is where the ownership question lands.

On the human side, the survivors are the things this series keeps circling: judgment, testimony, and the willingness to be responsible for an outcome. I want to state it more precisely than "trust", though, because trust is a soft word for something commercially specific. What buyers actually purchase when they cannot verify is recourse — references, guarantees, warranties, insurance, an audit trail, somebody who carries the cost of being wrong. Trust does not scale, because it is per-relationship and non-transferable. It accrues slowly and collapses all at once, usually not through counterfeit sincerity but through one broken promise about delivery.

Which is the honest version of a story I like. A gardener films himself rescuing strangers' overgrown gardens and gives that work away; the videos win him contracts. It is tempting to say the money moved from the labour to the trust. Really it moved one step downstream: the free job is customer acquisition, it qualifies the lead and demonstrates competence, and the margin still sits in the paid work afterwards. The demonstration is not the product. It is the reason anyone believes in the product.

And one prediction I would stand behind: as verification gets harder, spending shifts toward whatever makes a claim checkable. The machines that made the goods cheap are extremely good at producing the appearance of sincerity, so the premium goes to the things sincerity cannot fake — a name on a contract, a track record someone can call, a person who loses something if it goes wrong.

Who owns the next scarcity

Here is the corollary that should concentrate the mind: whoever owns the next scarcity owns the abundance. Cheap goods produced by machinery held by a dozen firms is not a world without money. It is a world with a very short list of landlords.

That is a question of justice rather than physics, and it has a serious body of thought behind it that most people building this technology have never opened. Leo XIV's first encyclical, Magnifica Humanitas, signed in May 2026 on the 135th anniversary of Rerum Novarum, extends the principle of the universal destination of goods — the old claim that created things are meant for everyone — to "patents, algorithms, digital platforms, technological infrastructure and data", warning that when these stay "concentrated in the hands of a few, without adequate forms of sharing and access, a new imbalance is created". John Paul II had already pointed that way in 1991; what is new is naming the assets.

Two things are worth taking from it even if you share none of its premises. First, it locates the danger correctly: not in the machines, but in the fact that control of platforms, infrastructure, data and computing power "does not rest with States, but with major economic and technological actors" who set the conditions of access. Second, it refuses to let efficiency be the only measure, insisting that automation be judged against "the dignity of the worker, the right to sufficient remuneration and the genuine possibility of participating in society". That third clause is the one with teeth. Keeping displaced people alive is something a transfer payment can do. Letting them participate is not.

I would add a warning of my own, narrower than the ones usually offered. The losses and the gains in a transition like this do not arrive on the same schedule, and they do not land on the same people. The losses come first and are concentrated: this trade, this town, this person who was forty-eight and expert at the thing that stopped being needed. The gains come later and are spread thin — slightly cheaper, slightly easier, for millions who never notice, and for some who are not yet born. Both halves are true at once, which is why the verdict on a transition can be positive in total while the people who fought it were telling the truth about their own lives. Being right about the total is not the same as being right about anyone inside it.

What to do on Monday

None of this is a reason to stop, and I want to be exact about why.

If the pattern holds, the safest thing you can do is what you already suspect you should be doing: get genuinely good at something, work where the reasoning is visible, and put your name on outcomes. Not because craft is romantic, but because the forecast says the scarce goods are judgment, framing the problem, and being answerable for the result. My cousin choosing a mark is doing tomorrow's expensive work. So is anyone building the picture that lets them tell right output from wrong.

And decide in advance what a freed hour is for, because something will claim it otherwise. That is the practical lesson of the laundry: the saving is real and it is quietly reassigned unless somebody chooses. Nobody announces the new standard. It simply becomes what normal means.

The last part is the one I hold as a Catholic, offered rather than argued. Laborem Exercens draws a distinction the AI conversation keeps missing. Work has an objective dimension — what gets produced, the output, the entry in the ledger — and a subjective one: what happens to the person while they do it. John Paul II's claim is that the second matters more, not because output is unimportant but because the person doing the work is worth more than the work being done. Man is the subject of work, never merely its instrument.

That cuts straight through the abundance argument. Automation is extraordinarily good at the objective dimension; that is the entire point of it. It is structurally incapable of touching the subjective one. A machine can take over producing a thing. It cannot take over what producing the thing was doing to the producer — the discipline, the responsibility, the ordinary experience of being needed by somebody else. When we say AI will do the work, we nearly always mean the objective half, and then act surprised that delivering more goods to more people does not leave them better off in the way we assumed.

In Genesis, work is given before the Fall; what the Fall adds is toil. It is tempting to conclude that removing scarcity removes toil and leaves work intact, but Laborem Exercens will not let me have it that cleanly: John Paul II says work is a good for man in some sense because it is hard, and treats endurance of it as a share in the Cross. So the honest version is smaller. Ease is not the same as meaning, and a life arranged to minimise effort is not obviously a life arranged well.

The old name for what goes wrong in comfort is acedia, which Aquinas defines not as laziness but as sorrow at the divine good — a listlessness toward the very thing that should delight you. It is a spiritual condition rather than an economic one, which is why no machine addresses it, and why Morgado is right that we cannot look to Musk for the answer. Magnifica Humanitas opens by quoting Augustine on the restless heart. A world that has automated every necessity will have removed the last excuse for that restlessness and will find it still there.

Which is why the public conversation about this often feels like it is missing a leg. We spend an enormous amount of energy asking what machines will be able to do for us, and almost none asking what will still be worth doing ourselves.

So: money will still be here in 2036, moved to whatever we cannot yet make cheaply, and my guess is that it lands on energy, on land, and on being answerable. Whether it lands on a dozen balance sheets or on something wider is being decided now. And the right response is not to wait for a threshold that will never arrive on a particular Tuesday, but to get good at the things that keep their value, argue about the ownership while it is still open, and know what you would do with a free afternoon before somebody hands you one.