When intelligence becomes free

Civilization has always automated whatever was expensive. First muscle, then repetitive physical work, then administration and information processing. Now we are automating intelligence itself: the deciding, designing, analyzing, coordinating, writing, coding and planning left over after every earlier transition.

Free here means free in practice, not literally. Inference costs compute and energy. But when a capable intelligence costs $20 a month and sits in your pocket, answering, drafting, analyzing and building on demand, it is free in every sense that matters to the economy. Its price is no longer what decides whether a thing gets done.

That changes the question. Not which jobs AI will replace, but what happens to an economy when almost anything humans can produce becomes abundant. And if production is nearly free, where does anyone's money come from?

Mechanization moved farmers into factories. Factory automation moved their children into offices: accounting, software, finance, design, management. Computers then automated much of the information handling, and labor moved further up into cognitive work. Each time, there was an obvious next rung.

AI attacks the rung itself.

The reason is generality. A tractor automated one task, and so did a loom. Every earlier machine did one thing, so labor could move to the next thing the machine could not do. AI does not automate a task; it automates learning tasks. Whatever new profession appears, the same system can pick it up. The accountant, the programmer, the designer and the writer have no profession waiting that is safe from the same machine.

Pace matters as much as direction. The move from farm to factory took more than a century, and it still brought enclosure, slums, labor movements and eventually the welfare state. A shift of the same size over ten years instead of a hundred is not an economic adjustment. It is a political crisis. Whether what follows turns out good or bad depends heavily on how fast it arrives.

AI does not only remove demand for existing work. It makes a vast amount of work worth doing that never was. That is a Jevons effect.

If a business application costs €2 million, only problems worth more than that get software. At €200,000 many more do. At €2,000 almost every process does, and at €2 everything does. Every workflow, machine, building, shipment and meeting room can join an intelligent system. The world saturates with software precisely because software becomes nearly free.

Software also stops being an artifact that is specified, built, tested and maintained. You say that premium customers get cancellation slots unless capacity is under 20%, and that technician travel should be optimized, and it appears. You say customers are confused during onboarding, and the system studies their behavior, tests a change and ships it. Features stop being projects and become events.

Software becomes exhaust from decisions.

So the bottleneck first moves from how to build something to what to build. Organizations are full of forgotten tasks, late information, manual copying, idle machines and approvals sitting in inboxes. Nobody fixed them because fixing them cost more than the problem. That opens a period of organizational invention, and it could employ a great many people directing it.

But AI climbs that layer too. It can watch an organization, notice the delays, the duplicated work and the customers dropping off, and propose the fix itself. The migration runs from coding to system design, workflow design, organizational design and strategy, and AI follows at every step.

How long that window stays open is the honest unknown. It could be decades of humans steering an explosion of automation, or a few years before the steering is automated too. Nobody knows, and a lot depends on the answer. The durable question is therefore not what AI cannot do. It is what stays scarce when intelligence is abundant.

Digging a hole with a spoon takes enormous effort and nobody pays for it. When millions can produce the same thing instantly, competition drives its price toward what it costs to make one more. AI is driving that cost toward zero for music, illustration, video, websites, bookkeeping, translation, legal drafts, analysis and business plans.

People will create more of all of it than ever. The world may drown in creation while paying creators less.

Humans will not run out of things to do. Given tools and time, they build, compete, create, study, travel and fix whatever annoys them. The hard question is why anyone would pay you for it. If AI writes the song, keeps the books and builds the app, doing those things and being paid to do them come apart.

Across a whole economy this becomes a demand problem. One person running what took fifty is excellent for one firm. Repeated everywhere, production rises while wages fall, and the two are perfectly compatible. Civilization gets extraordinarily good at producing things that many people cannot afford.

The machinery works beautifully. The spreadsheet congratulates everyone. The humans remain economically inconvenient.

There is a second path: need less income because what it buys gets cheaper. An autonomous car that arrives in thirty seconds and leaves again removes the cost of buying, financing, insuring, parking and repairing one. Transport becomes infrastructure. The same logic already reshaped media, computing and city bikes. Access replaces ownership, and access at near-zero cost needs little income.

Education may fall furthest. A personal tutor that knows every subject, adapts to each student and never tires costs almost nothing, and the credential a university sells weighs less when the knowledge itself is free.

But not every large household cost follows. Housing is mostly land, and land does not get cheaper when intelligence does. Healthcare is diagnosis, which collapses in cost, but also hands, beds, buildings and people in the room, which do not. As everything automatable gets cheaper, whatever stays physical and human gets relatively more expensive. A household may need far less income for goods, information and learning, and just as much for a home and for care.

You can make another car. You cannot make another Manhattan, another waterfront apartment in central Stockholm, another ticket to one particular night, or another hour of one particular person's time. Land, unique places, historical objects, nature, prestige and attention stay finite.

Ten thousand people want the one penthouse. Something has to decide who gets it: a lottery, a queue, politics, reputation, an auction or credits. And if people can collect different amounts of credits and trade them for scarce things, humanity has invented money again.

This technology is annoyingly hard to kill.

When coordination, administration and software stop being constraints, reality becomes the limit. Can the material take the heat? Can enough energy be generated? Does biology allow it? A cheaper energy source does not optimize a system. It moves the whole frontier and exposes the next bottleneck.

The economy moves closer to the edge of what is physically possible, and the discoveries that move that edge stay scarce.

Digital creation becomes abundant, so value moves to whatever resists copying. What follows is not a forecast. Nobody knows the future, and these are places value could go, not predictions of where it will.

Physical reality. AI can render Rome; it cannot put you in Rome. Reality has infinite resolution and nobody controls it: the smell of the street, getting lost, the festival you stumble into. A simulation is designed and reality is not, and that uncertainty is the product. Travel may sell less sightseeing and more discovery, discomfort and chance encounters.

Human presence. A recording of an opera singer may be indistinguishable from a generated one. The singer three meters away while you cook is not. The value is this person, here, now, for me. Cheap recorded music did not kill concerts; it made the difference clearer. People pay to stand in a stadium with a worse view than their TV because they are buying participation, not pictures of the match.

Embodied experience and friction. Digital media reaches a thin slice of the senses, and the experience economy may reach all of them: restaurants as performance, hotels as adventure, exercise as theater. People already pay for bars where the drink arrives with fire, water and a slap. They were never buying the drink. They were buying a story. As life gets frictionless, the appetite for deliberate friction may grow, from mountains and ultramarathons to cold water, wilderness and dancing until sunrise. Civilization spent centuries eliminating chaos only to find an industry selling it back by the weekend.

Authenticity. When anything can be generated, the question shifts from whether something is good to whether it is real. Vinyl, mechanical watches, handmade goods and live theater are already luxury categories built on provenance and imperfection.

Accountability. An AI can draft the diagnosis, the audit and the structural calculation. Someone still has to sign it, answer for it and be liable when it is wrong. Doctors, auditors, engineers and judges are paid partly for being responsible, and responsibility cannot be copied: it only means something when it attaches to a person or institution that can lose something. Trust, the belief that someone will stand behind an outcome, may become one of the scarcest goods there is.

Attention, community and status. AI can make a million excellent songs a minute, and nobody gains a million minutes to hear them. Supply heads toward infinity while demand stays bounded by waking hours, so discovery, reputation and distribution decide what gets heard. Shared context is scarcer still: a match matters because millions watch the same one. Status is relative by definition, since not everyone can be first, famous or invited, so markets for rank survive any amount of abundance. A society whose material needs are met and whose remaining competition is for rank alone may be calmer, or it may be more bitter, because when rank is all that is left, losing it may hurt more.

Human services as luxury. Human labor may flip from default to premium. The machine restaurant may cook better food for almost nothing. The one where a human chef picks the ingredients, tells stories and sings between courses may cost twenty times more, because the scarce input is the human. Anything that needs a real, present, consenting person resists automation for the same reason: massage, training, hospitality, companionship, performance, sex work.

There is something absurd about accountants being automated before prostitutes. Economically, it is perfectly coherent.

Who buys those services is the open part. In one future, broad prosperity lets many people pay for human presence, and serving each other becomes a large share of work. In another, a small owning class buys it and much of everyone else sells it, which is closer to a return of domestic servants than to a renaissance. Which one arrives is decided by the question this piece ends on.

It is that fewer things need humans in order to exist. People still play chess against better computers, run beside faster cars, paint despite cameras and play guitar despite Spotify. What breaks is the automatic link between being able to do something and someone paying you to do it.

Creative work already feels it. AI extends it to every cognitive profession.

Follow the chain to its end and one question is left. If labor stops being the main claim on output, the claim moves to whoever owns what production depends on.

When intelligence is cheap, the binding constraints are physical: energy, compute, chips, land, raw materials, grid capacity, and the rights and data that gate them. A data center is limited by electricity, not by ideas. Whoever owns the power plant, the chip fab, the model, the land under the data center and the distribution channels collects the income that wages used to carry.

Unequal claims on scarce things have to be earned somehow. When labor is no longer scarce, what remains is ownership, risk, invention, discovery and inheritance, and ownership is by far the largest of them.

Ownership is also where economics turns into politics. Who issues the credits, who taxes the land, the energy and the compute, and who holds shares in the productive machinery, whether pension funds, sovereign wealth funds, ordinary people or a handful of firms, is not decided by markets alone. It is decided by states, laws and voters. Purchasing power has to reach households by some route other than salaries: broad ownership of productive capital, taxes on the scarce inputs, direct transfers, or a mix. Choosing that route is a political act.

That choice decides which of the futures above we get. Abundance that reaches everyone, or abundance that reaches whoever holds the deeds.

Every earlier revolution moved labor into a new, larger kind of work. This one may have nothing waiting. Software, music, analysis, education, administration and intelligence itself become nearly free, and value gathers in what cannot be generated again: a place, a person, a moment, attention, trust, energy, land.

The defining question of the AI era is not what we can produce. We will produce absurd amounts. It is what is still scarce, and who owns it.