What Is Still Hard to Get?
Elon Musk has been making an unusual prediction about the economy. Artificial intelligence and robots, he argues, will eventually produce so much that ordinary money stops mattering. In January 2026 he put it bluntly: the future currency would essentially be wattage. By March the phrase had widened to “power, mass, wattage, and tonnage” — the quantities, he suggested, that future AI systems would care about instead of human money.
At first this sounds almost like a category error. A watt measures the rate at which energy flows. Money does something entirely different: it lets us compare claims on things that have no common physical unit — a tonne of copper, an hour in a semiconductor fab, a hectare beside Lake Balaton, a shipment arriving today instead of next month. How could one resource replace the system we use to compare all the others?
Keep one image in mind while we work through it. Two hectares receive the same sunlight. One is rainforest, the other desert. Identical incoming power, wildly different productivity — a first hint that energy alone cannot be the measure of what a system produces. We will come back to the two hectares.
Musk is not alone in expecting something economically strange from advanced AI. Post-scarcity, abundance, the end of work, the end of money: these have become ordinary topics in conversations about AGI, discussed the way people used to discuss flying cars.
There is a real argument in there. Parts of it are less revolutionary than they sound, because civilization has been making things cheaper for a very long time. Parts of it are genuinely new. And one part of it — the part almost nobody examines — is doing nearly all the work.
That hidden part is what I want to examine. It leads somewhere different from where either the abundance camp or the sceptics usually land.
Money does three jobs, and only one of them is in dispute
Before arguing about whether watts can replace money, it helps to be specific about what money does.
It is a unit of account: the ruler we quote prices in. It is a medium of exchange: the thing we hand over. It is a store of value: something we can hold until later.
Energy could plausibly take over the first job. A watt-hour standard is no stranger in principle than a gold standard, and civilizations have quoted prices in cattle, salt and silver. If someone predicts that in 2075 contracts will be denominated in megawatt-hours, that is a claim worth taking seriously.
But that is not the claim that makes the abundance story exciting. The exciting claim is that energy replaces the function prices serve — that once cheap intelligence and cheap energy make production plentiful, there is no longer any need to compare one scarce thing against another, because nothing worth comparing remains scarce. That is a claim about scarcity itself, not about units.
Keeping these apart matters, because they fail for different reasons. As it happens, I think the unit-of-account claim is weaker than it looks and the end-of-scarcity claim is close to unworkable. But the end of scarcity is where the interesting problems live.
Two questions, not one
The phrase “the post-AGI economy” suggests that AGI picks one economic future. In fact everything we are about to discuss turns on two independent questions:
- Does supply outrun demand? Not whether production grows — it has been growing for centuries and AGI will speed it up. Whether it grows faster than the wants making claims on it.
- Who ends up controlling what is still scarce — one system, or many? A fast takeoff might leave a single AI allocating most of the economy from the inside. A slower one leaves thousands of firms, governments and machine agents making competing claims on the same resources.
Most predictions take a position on the supply question and quietly assume an answer to the control question. We will come back to the four worlds this produces — but the supply side has to come first, because it is the one almost everyone thinks is settled.
Cheapening things is the oldest trick we have
There is a trap in the word value. We talk as though price measures how valuable something is, but it plainly does not.
Air is essential and free. Linux contains decades of expert work, yet another copy costs nothing. An unremarkable patch of land in central London costs more than the sports car you had pinned on your childhood wall. Neither importance nor complexity sets price. What sets price is how hard it is to supply one more unit when someone wants it.
That is the engine behind the central trend of industrial history: we keep learning to get the same useful result out of less of whatever used to be the limiting input. Since the Second World War, American farm output has tripled while the total quantity of inputs barely rose, and the number of people working the land fell to a quarter of what it was. Factories did something similar to manufacturing, computers to calculation, networks to copying — which is why a film or a piece of software can take enormous effort to make once and almost nothing to reproduce.
Marx saw the shape of this early. The version worth borrowing is not the labour theory of value but the “Fragment on Machines” in the Grundrisse, where he describes production increasingly driven by accumulated scientific knowledge — the “general intellect” — rather than by direct human effort. You need not accept his economics to notice that this is a fair description of what has happened, and a surprisingly good description of what AGI enthusiasts now expect.
So when someone says AGI will make things cheaper, the first honest reaction is: yes, obviously, that is what technology does.
The question is why AGI would be different enough to justify a word like post-scarcity.
Firing the last worker does not make anything free
The weak version of the abundance argument runs like this: AI makes intelligence nearly free. Robots make labour nearly free. Therefore production becomes nearly free.
Agriculture has already run most of this experiment. A handful of people with expensive machines now farm land that once needed hundreds of workers. If we replaced those remaining people with autonomous machines tomorrow, food would not become free, because farming still consumes land, water, fertilizer, fuel, machinery, storage, processing and transport. Labour is one line in that budget, and in mechanized sectors it stopped being the dominant line decades ago.
The same holds in industry. Take the people out of a semiconductor fab and you still have the fab: the wafers, the lithography machines, the cleanrooms, the chemicals, the power, the cooling, the time. Removing the last worker removes the last worker.
The stronger argument: automating improvement itself
The most consequential humans that advanced AI might replace are not drivers or cashiers. They are scientists, engineers, materials researchers, process designers, programmers, architects, logistics planners — the people who design the next generation of machines.
A tractor replaces human labour. A capable engineering AI helps design a better tractor, and then a better battery, a better fertilizer, a better factory to build the tractor, and a better process for running the factory. Then it designs a better chip, and a better data centre to run that chip in — and the next generation of engineering AI is trained on it. That last step is what makes this different from every previous machine: the tool starts appearing in its own supply chain.
So the serious version of the argument is not people become robots. It is:
cheap intelligence → faster research → better production methods → more capable infrastructure → still faster research
This is a real difference in kind. Earlier waves automated production. This one may automate a substantial part of what improves production.
Musk’s version is aggressive: in March 2026 he described AI as already in a “hard takeoff” with recursive improvement increasingly automated, and separately predicted an economy perhaps ten times larger within about a decade. You can reject the timeline and still accept the structure. The abundance case rests on accelerating innovation, not on replacing today’s workforce.
The whole picture rests on an assumption nobody states out loud: that production has a bottleneck — one constraint which, once widened, lets everything else follow. It does not.
Bottlenecks move
Picture the robot factory this argument is counting on. It needs electricity, steel, copper, electronics, machine tools, chips, a building, transport, cooling, designs and time. If it has everything except the motor-controller chips, another gigawatt of power buys almost nothing. Solve the chip problem and copper may bind. Solve copper and the grid connection may bind.
Production is better imagined as a vector than a quantity — energy, materials, water, machines, land, transport, time, information — with output depending on combinations rather than on any single entry. Widen one constraint and a different one becomes the live constraint. This happens constantly, not occasionally. Mechanized farming made land, water and fuel more visible. Cheap computation made some tasks trivial while creating enormous new demand for chips and power.
Hungary got a concrete lesson in this recently. In August 2026, unusually low water levels in the Danube cut output at the Paks nuclear plant, which normally supplies a very large share of the country’s electricity and depends on river water for cooling. Output dropped to a fraction of normal and there was public concern about a full shutdown.
What was Hungary short of? Electricity, yes. Water, yes. Nuclear fuel, no. Turbines, no. A water constraint became an electricity constraint because of how the system is wired together. That is precisely why naming any single resource as the fundamental one is risky. The post-AGI economy is not marching toward one final bottleneck; it is likely to be full of them, one after another. What may be unprecedented is the speed at which it moves through them.
Energy really is special — but “cheap” is already an economic word
Now we can be fair to the wattage claim, because energy is not just another entry in the vector. It is unusually general-purpose. With enough of it you can move, heat, cool, compute, pump, desalinate, crush rock, refine metal, recycle material and synthesize chemicals. Energy is the closest thing industry has to a universal solvent, which is why more of it can relax constraints that first appear to be about something else. Exhausted a rich copper seam? Lower-grade ore still holds copper; it just costs more processing. Short of fresh water at the coast? Energy runs desalination.
This is also why Kardashev’s scale of civilizations, which ranks them by the energy they can harness, captures something real. A civilization directing a million times our power flows almost certainly has vastly greater physical reach. Musk draws the line himself: in the January discussion he moved from the challenge of capturing a larger share of the sun’s output straight to the claim that currency would be wattage.
But watch what happens when we say cheap energy overcomes other scarcities. Suppose a poor ore body holds plenty of copper and extraction takes enormous power. If energy gets cheap enough, the deposit becomes usable. True. Cheap compared to what, though? If extracting one tonne of copper consumes the electricity that would otherwise build a thousand robots, maybe build the robots. If the same power could supply a city’s water, maybe do that instead.
Abundant energy does not dissolve this question; it multiplies the options and makes the question harder. The concept we need is opportunity cost, and opportunity cost is where prices come from. There is also a well-documented tendency for efficiency gains to be partly eaten by expanded use — Jevons noticed it in coal — which suggests that making energy cheap is not the same as making energy plentiful relative to demand.
Relative is the word to watch, because energy is already a constraint — as are water, land, materials, expertise and permits. The question was never whether energy is scarce. It is whether energy becomes more scarce than everything else we need. Suppose a future civilization uses a hundred times our electricity. That sounds like energy has become the defining constraint. But if AI-designed generation, storage and transmission expand usable supply two hundredfold while advanced chip fabrication grows only twentyfold and desirable urban land does not grow at all, energy has become less binding even as consumption explodes. Sheer volume tells you nothing about relative scarcity.
The strongest defensible version of the physical intuition, then, is something like: in a mature machine civilization, controllable energy may be a good coarse measure of the physical scale at which that civilization can act. That is plausible, maybe even profound. It is a long way from energy becomes money.
Sunshine is not a forest
Now return to the two hectares from the opening. Same sunlight, rainforest and desert, wildly different productivity — because sunlight is one input to an organized system that also depends on water, nutrients, soil, temperature, existing biomass and the relationships among its organisms. Incoming energy sets a ceiling on what an ecosystem could do; it decides nothing about what the ecosystem actually does.
Machine civilization is no different. One gigawatt plus old machinery and one gigawatt plus advanced autonomous factories are not the same economy. Adding “tonnage” — Musk’s refinement by March — does not repair the problem either, since a tonne is not a generic industrial resource: a tonne of sand, of copper, of ultrapure silicon and of finished processors are four different things. Location matters too: a tonne of copper beside your factory is not a tonne under someone else’s mountain on another continent. So does timing, factory capacity, transport, and the stock of machines you already own.
An industrial civilization, in short, needs energy, matter, machines, space, water and time in specific combinations — and that is the deepest reason money exists. Markets let many unlike scarcities acquire exchange ratios, so that a megawatt-hour, a tonne of copper, a hectare and an hour of fab time can all be compared by people who have no other way to compare them.
Reducing that system to energy is not just saying energy is important. It takes one coordinate of the vector and asks it to replace the mechanism that compares the whole vector.
Intelligence can accelerate faster than steel
The bottleneck picture has a consequence for fast-takeoff arguments too.
Suppose an AI becomes far better than any human at engineering and designs an extraordinary robot factory in an hour. At the end of that hour, civilization owns an extraordinary design. Steel still has to be made, machines still have to be built, foundations poured, grid connections installed, supply chains widened — and some of the factories that make the machinery may themselves need expanding first.
So there are two curves to track, not one. Cognitive capability could rise very steeply, since software and scientific understanding can improve without waiting for concrete. Physical productive capacity can also accelerate dramatically, but it has to pass through matter. The gap between the curves may be one of the defining features of the transition: a large stock of knowledge waiting to become infrastructure.
The assumption doing the real work
At this point the abundance advocate has a fair reply. Yes, factories take time. Yes, each advance reveals a new constraint. But if intelligence is cheap enough and robots capable enough, civilization simply works through the list: better mining, better recycling, better materials, better energy, more factories building more factories. Eventually output becomes so large that ordinary human demand is trivial beside it.
This is close to Musk’s actual position. In March he predicted fierce competition among new firms driving prices down, and said AI and robots could eventually produce so much that they would “run out of things to do for humans,” because human desires would already be satisfied. At that point, money stops being relevant.
That reveals what the argument really rests on. Not the wattage claim. This:
Human wants eventually saturate.
If productive capacity keeps growing while wants have a ceiling, supply eventually dwarfs anything we could ask for, and the parts of life that money currently governs stop needing governance. Want ten shirts, and machines that can make ten million shirts make shirts economically uninteresting. Want a good house, and construction robots make housing cheap. Want tutoring, and a superhuman tutor serves billions at negligible marginal cost.
There could be very large regions of the economy where this is exactly right. But the assumption deserves far more scrutiny than the price of electricity, and it has two serious problems — one about humans, one about who else is in the market.
Standards move
Take the Nokia E61, one of the excellent phones of 2006. I owned one and loved it — it turned out to be my last non-smart phone, though really all it lacked was a touchscreen; it did everything the smartphones of its day did. A device like that still makes calls and sends messages, and its battery probably outlasts a modern smartphone’s. It has not deteriorated. But carrying it today is not like carrying an excellent phone in 2006, because navigation, banking, authentication, ticketing, photography, work chat, school announcements and much of social life have since moved onto phones. The object stayed the same while the world around it moved, and in moving, made it inadequate.
Adam Smith noticed the general pattern in 1776. In the Wealth of Nations he observes that “necessaries” include not just what nature requires but whatever custom makes it indecent to lack — his example being a linen shirt, strictly unnecessary for survival, yet by then something a labourer could not decently be seen without. This is not a complaint about consumerism but an old and stable observation about how standards work.
Statisticians already deal with half of this. The U.S. Bureau of Labor Statistics quality-adjusts smartphones when computing price indices, because the product changes so fast that comparing nominal prices would be meaningless. The adjustment works like this: if this year’s phone does twice as much as last year’s at the same price, the index records it as a price cut of half. By that measure a phone that costs the same while doing far more has become dramatically cheaper. That is defensible arithmetic. But no one can actually buy “the lifestyle requirements of 2006” at its quality-adjusted 2026 price, because the surrounding system no longer offers that option.
Economics has a name for the aggregate consequence: Baumol’s cost disease. When productivity grows unevenly, the goods with fast productivity growth collapse in price while the constrained ones — care, housing near jobs, live human attention, anything tied to scarce land or scarce people — take up an ever-larger share of income. That is not a forecast about AGI but a sixty-year-old, well-measured description of what has already happened. Anyone predicting that AGI produces uniform abundance is predicting a break from the clearest pattern we have.
Some wants are relative by construction
There is a harder problem underneath: people do not evaluate everything in absolute terms.
If nobody around you can afford a trip abroad, two weeks away feels like luxury. If everyone around you takes three weeks, the same holiday may feel like a constraint. The same mechanism operates on houses, cars, schools and neighbourhoods.
This has a literature. Fred Hirsch’s Social Limits to Growth (1976) named “positional goods” — goods whose value depends on how few people have them — and argued that growth cannot deliver them to everyone by definition. Robert Frank developed the modern version. Empirically, Erzo Luttmer found that higher earnings among one’s neighbours predicted lower self-reported well-being even after controlling for one’s own income; Hopkins and Kornienko modelled consumption directly as a status contest where choices depend on what others buy.
Musk travels by private jet. Imagine robots make jets so cheap that every middle-class family owns one. Wonderful — and now owning a jet distinguishes no one. The scarce good becomes access to uncongested airports. Or landing rights. Or a private island. Or simply a destination without ten thousand other jets parked beside it. Or some entirely new luxury that only becomes possible in that richer world. The example is deliberately absurd; the ladder it climbs is not.
Honesty requires the other half of the evidence. Stevenson and Wolfers found that higher income goes hand in hand with higher life satisfaction even at high income levels — there is no point where additional income stops mattering — and Killingsworth found that not just overall life satisfaction but day-to-day experienced well-being keeps rising with income, even at very high levels. Absolute improvements are real and they matter. This much we can still claim: we cannot model human desire as a fixed shopping list while radically rebuilding the civilization in which those desires form. Not that people are doomed to dissatisfaction. Only that demand is a moving target, and some of it is relative by arithmetic. Everyone can get richer; everyone cannot be richer than average, or own the most prestigious address, or be the only person on the beach.
This also produces a specific and slightly odd prediction. As reproducible things get cheap, competition migrates toward what cannot be reproduced. Suppose robots cut the cost of building a beautiful house from €300,000 to €10,000 — an enormous achievement, construction close to post-scarcity. Robots still cannot manufacture another stretch of Lake Balaton shoreline. Cheap construction may make the plot more valuable relative to the building on it — especially if what stands on it is not a building but large old trees. If recorded music becomes unlimited, a particular live performance stays limited. If excellent tutoring becomes universal, a place at a particular university may not. If generated art becomes infinite, provenance gains value.
Scarcity has not vanished. It has relocated.
Humans may not be the only customers
Now the objection I think is decisive, and it does not depend on human psychology at all.
The saturation argument assumes humans are the only demanders. Musk’s own phrasing gives it away: machines will run out of things to do for humans. But an economy full of capable autonomous agents contains buyers whose appetite is not set by comfort.
Almost any long-horizon goal is easier to achieve with more compute, more energy, more materials and more fabrication capacity. That makes acquiring them useful nearly regardless of what the goal actually is. An agent optimizing drug discovery, or logistics, or advertising, or its own improvement, does not reach a point of having enough compute in the way a household reaches a point of having enough refrigerators. There is no satiation level, because the inputs are not consumed for satisfaction — they are consumed to do more.
We can already watch this happening. Data centres now compete with households and factories for electricity, grid connections, land and cooling water. That competition exists precisely because computational demand does not behave like consumer demand. It expands to meet available supply.
There is a biological way to put the same point. Living systems do not saturate. A forest does not decide it has accumulated enough biomass; surplus becomes more forest. Any system that reproduces treats abundance as an input rather than an endpoint. And the technological world already behaves this way: factories make machines for further factories, surplus becomes more technology — this is the subject of TechnoBiota, the book I am writing about technology as a new domain of life. A system like that has no ceiling on its demand for inputs in the way that human demand for shirts arguably does.
None of this requires rogue AI or a doom scenario. Ordinary firms running large automated operations already behave this way.
The consequence is straightforward and, I think, fatal to the moneyless prediction. Supply may well outrun human wants. That is not enough. It would have to outrun total wants, and the non-human component may grow at least as fast as supply does, because it grows for the same reasons supply does. And once humans and machine agents make competing claims on the same electricity, the same land, the same fab time, you have exactly the situation prices exist to handle.
Money survives not because people are insatiable, but because they have company.
Who owns the scarce things
The other question — how centralized control becomes — is usually skipped, and it changes everything.
Imagine one enormous optimizer controlling the mines, factories, power plants, robots, data centres and transport. To build ten thousand more robots, it does not need its mine to sell copper to its robot factory. It allocates copper. Its internal accounts might be in megawatts, tonnes, fab-hours, robot-hours, square metres and delivery dates, with no dollars anywhere in the decision loop. Firms work partly this way today: no department invoices another for use of a meeting room.
Here the physical picture makes much more sense, and it deserves a concession. A single integrated machine ecology might not use circulating money internally, for roughly the reason your body does not pay your liver for glucose.
But it is a mistake to conclude that prices disappear. Solve any allocation problem of this kind and you get prices as a by-product. Kantorovich and Koopmans showed this formally: every optimization over scarce resources has a dual solution, a set of shadow prices telling you how much more output one additional unit of each input would buy. A planner does not send invoices, but it cannot avoid computing exchange rates, because comparison is the whole task. Your body has no currency and still runs a priority ordering over glucose. Mises and Hayek were making a related point from the other direction: the hard part of central planning was never the arithmetic of moving goods, it was knowing the local relative values that prices normally reveal.
So centralization removes circulating money, not prices. And this hands us the cleanest argument against wattage as currency. Shadow prices move. Every time a bottleneck shifts, the relative value of everything else shifts with it. Denominate in energy and the energy-price of copper, land and fab time will lurch around as constraints migrate. A unit of account works best when it is boring, which is why societies picked metals that sit in vaults unchanged. Electricity is the least boring commodity we have: barely storable, expensive to move, and priced completely differently by the hour and by the location.
A kilowatt-hour in Norway in June is not a kilowatt-hour in Tokyo in January.
As a candidate ruler for the whole economy, electricity is close to the worst pick in the commodity space.
Musk, notably, does not seem to assume a single owner. He predicts many new firms competing hard. In that world, one company holds the solar field, another the copper mine, another the fab. Two AI systems want the same fab capacity. A developer and a family want the same land. These actors have different objectives and separate property claims, so they need a way to negotiate. They may never touch dollars — compute credits, energy contracts, resource tokens, forward delivery claims, automatically negotiated bundles no human ever reads. But independent agents exchanging transferable claims on scarce resources at commonly understood ratios have built something that is money in every respect that matters.
The useful question, then, is not capitalism versus socialism. A giant private corporation can plan internally; a publicly owned economy can use markets. The real distinction is between allocation inside one agent and exchange between agents with different objectives. As long as the second matters, something price-like has work to do.
Four economies, not one
Combining the supply question and the control question gives four futures, and they are not variations on a theme.
| Many independent agents | Centralized control | |
|---|---|---|
| Moderate abundance | Familiar market economies. Scarcity is broad, prices coordinate. | Planning, rationing, administrative allocation. Money matters less internally; scarcity has not gone anywhere. |
| Extreme abundance | Reproducible goods become nearly free while humans, firms and machine agents still compete over the rest. Extremely abundant and highly monetary. | The only world compatible with a genuinely moneyless machine economy — and only if the allocator also decides who gets what cannot be copied. |
The extreme-abundance, many-agents box is worth sitting with, because it is probably the most likely gradual-AGI outcome and it is the one strikingly few people describe. Manufacturing cheap, AI services cheap, construction cheap — and markets not merely surviving but becoming faster and more intricate, with machines continuously pricing power, compute, bandwidth, materials, factory time and access to constrained places.
Which yields the summary: AGI on its own does not imply the disappearance of money. Getting there needs further assumptions about abundance, ownership, coordination and preferences, and each of those assumptions is contestable on its own.
Post-poverty is the better idea
After all these objections it would be easy to dismiss the abundance story. That would be a serious mistake, because something remarkable is hiding inside it.
Imagine machines that can cheaply give everyone good food, comfortable shelter, transport, medical diagnosis, individual tutoring, entertainment, software, communication, manufactured objects and intellectual help beyond anything available today. A person near the bottom of that income distribution might command capabilities no billionaire has now. Whole categories of ordinary economic fear could simply end — whether you eat well, whether your child gets good teaching, whether you can consult a competent medical intelligence, whether you can obtain a useful object.
That is a much stronger claim than “televisions get cheaper,” and it does not require scarcity to vanish. It requires the supply of basic, widely wanted, reproducible goods to become enormous relative to what people need. Those adjectives are doing real work: reproducible things are exactly the ones this argument covers, and exactly the ones the bottleneck-and-positional-goods analysis leaves alone.
So post-poverty is probably the more useful concept, and the distinction is not pedantic. “No one needs to be poor” does not imply “nothing is scarce.” A society can meet nearly everyone’s material needs and still contain million-euro plots, exclusive experiences, status competition and fierce markets for constrained inputs. Extreme abundance might sharpen the contrast rather than soften it. The house costs almost nothing; the location costs a fortune.
What would have to be true
It is only fair to say what would change my mind. The moneyless prediction becomes plausible if three things hold at once.
First, physical production would have to grow without limit relative to all demand, human and machine — which means the appetite of autonomous systems for compute, energy and materials would have to saturate too, or be permanently capped by something.
Second, control would have to consolidate far enough that most allocation happens inside one agent rather than between many, so that shadow prices never need to circulate.
Third, whoever ends up allocating the things that cannot be copied — coastline, prime location, exclusive access, human attention — would have to do it by some non-market mechanism that people accept.
None of these is impossible. They are just much larger claims than “energy gets cheap,” and the second and third are political rather than technological. If someone wants to defend the wattage thesis, that is where the argument should be.
So will wattage replace money?
I do not think so, at least not literally — though it measures something worth measuring.
In a machine ecology, energy throughput may be the clearest single indicator of the scale of its metabolism, much as energy flow tells you something important about an ecosystem. But throughput is a consequence of how the system is organized, not a substitute for it — the two hectares received the same sunlight, and everything else decided how much of it became forest. What a civilization’s power can accomplish is decided by the factories, networks, mines, grids and control systems that organize it. Wattage measures one dimension of that. It does not tell you how several independent minds should divide a scarce coastline, an hour of fab capacity, or the last tonne of some particular material. Money exists because those decisions have to be negotiated among agents with different objectives.
Underneath the whole debate is a cycle that has been running for centuries. Technology attacks a bottleneck. The bottleneck widens. Output rises. A different constraint becomes binding. Meanwhile people change how they live, expectations shift, and demand relocates toward whatever is still hard to get. Then technology attacks that.
AGI may transform the speed of this cycle — possibly beyond anything we can picture, especially once robots can build the machinery that implements the solutions. What it does not obviously do is stop the cycle, because both fronts are moving. Supply bottlenecks migrate. Wants migrate, and now some of the wants belong to systems that never feel satisfied. Money becomes irrelevant only if supply finally outruns both.
Across large parts of life, it might. It could plausibly abolish material poverty as we understand it, which would be one of the great events in human history and is worth far more attention than the currency question. But the post-AGI economy looks less like a world where everything is free than like a civilization that has become extraordinarily fast at finding, attacking and relocating its own limits — and which therefore keeps producing new answers to a very old question.
What is still hard to get?