When Intelligence Comes Off an Assembly Line: Eight Predictions

Once something is manufactured, it gets a unit price. Muscle got horsepower and the kilowatt-hour. Intelligence already has the token, and the price of a token trends toward the cost of the electricity that made it, plus a margin. That is what a commodity is.

The billable hour was an honest way to price scarce thinking: you paid for the time of a rare mind. When the mind is not rare, the hour becomes absurd. Professional work splits into two products that used to be sold as one. The work, which is manufactured and priced like a utility, and the signature, which is a human or firm taking accountability for it. The work goes toward free. The signature holds its price and may rise.

Already true

Firms that deliver finished AI work are pricing by outcome, not by hour, and clients are asking why a first draft still costs what it did in 2022.

What has to happen

A large firm in law, accounting or consulting publicly drops the billable hour as the default in a major practice. I expect that before the end of 2029.

Where I am probably wrong

Regulated professions have enormous inertia, and the hour survives in part because it protects margins. If insurers and bar associations keep tying liability to hours, the split could stay hidden inside the old invoice for a decade.

The first work to be manufactured is not the senior partner’s. It is the junior associate’s: the research memo, the first draft, the spreadsheet cleanup, the summary of the deposition. That work was never valuable for its output. It was valuable because doing it is how a twenty-four-year-old becomes a forty-year-old who has judgment.

So the disruption shows up first as a missing generation. Companies will find that manufactured intelligence can do the junior work cheaper than any junior, stop hiring juniors, and then discover around 2032 that they have no one ready to be senior. The professions have been running a hidden apprenticeship system financed by billable grunt work. Take away the grunt work and the apprenticeship stops paying for itself.

My prediction is that apprenticeship becomes a deliberate, budgeted expense, the way corporate R&D did, and the companies that build it on purpose own the next generation of talent. For young people, this is the moment for the strategy I give every student I work with: let your first employer be your first customer. Don’t wait to be trained on junior work that is no longer being assigned. Treat whoever pays you first as a customer buying a finished outcome, and learn judgment by owning results early.

Already true

Entry-level hiring in white-collar fields is where the pressure is showing first, and new graduates feel it before anyone else does.

What has to happen

By 2028, major employers begin describing junior hiring as a training investment rather than a production cost, with formal programs to match.

Where I am probably wrong

New entry-level roles could appear faster than the old ones vanish (people who operate, check and direct AI systems), and they may teach judgment just as well. If that happens, the rung is replaced, not removed.

For a century, knowledge work clustered where the knowledgeable people lived: New York, San Francisco, London, Boston. Talent was the scarce input, so industry went to talent.

Manufactured intelligence has a different scarce input: electricity, plus the land, water and transmission to use it. So the production of intelligence moves to where power is: Abilene, Texas. Richland Parish, Louisiana. Memphis. Rural counties with a substation and a permissive planning board are the new industrial heartland, the way coal country and river towns were once.

That makes the least glamorous officials in America suddenly consequential. State utility commissions, which decide who pays for new generation and transmission, become the most important economic regulators most people have never heard of. And the politics of AI arrive not through fear of robots but through the household electric bill. When a family’s bill rises and the reason is a campus down the highway, the argument about who benefits from AI stops being abstract.

Already true

Gigawatt-scale campuses are being sited by power availability, and electricity prices have already become a live issue in state politics.

What has to happen

By the 2028 presidential cycle, who pays for data-centre power is a national campaign issue, not a regional one.

Where I am probably wrong

Efficiency could outrun demand. The human brain runs on about 20 watts, a fifty-millionth of a gigawatt, which is proof of how far the engineering can still go. If chips close even part of that gap quickly, the power problem could shrink faster than the politics can form around it.

When one input becomes abundant, everything made mostly of that input gets cheap, and everything made of something else gets relatively expensive. That is the whole economic story in one sentence.

This quietly reverses a forty-year status order. The economy has told a generation that the safe path was the desk and the risky path was the trade. Manufactured intelligence says the opposite. The hibachi chef flipping a shrimp in front of a birthday table in Charleston is doing something no data centre will ever do, because the value is the performance in the room. The receptionist answering the phone is not, which is why that job is already being done by software in restaurants I work with.

Hospitality, care, skilled trades and anything experienced in person will gain relative value and dignity. The places that win are the ones that use manufactured intelligence for everything behind the counter and spend the savings on the people in front of it.

Already true

AI phone agents, drafting tools and automated bookkeeping are in use at ordinary small businesses, not just in tech companies.

What has to happen

Wages in hands-on, in-person work rise faster than wages in routine desk work across a sustained period in the late 2020s.

Where I am probably wrong

Robotics. If manufactured intelligence gets a cheap, capable body sooner than I expect, the “hands” half of this goes the way of the “mind” half, and the only durable premium is presence and trust.

Here is the part almost nobody is ready to say out loud. Governments are funded by taxing work. Income tax, payroll tax and the programs built on them assume that most economic value passes through a human paycheck on the way to being spent.

Manufactured intelligence is capital. It is owned the way a factory is owned. When cognitive work moves from people to machines, the returns move from wages to whoever owns the compute, the power and the models. That is not a moral claim. It is what happened to the returns on muscle when factories arrived, and it is why the twentieth century had to invent the income tax, labour law and the welfare state.

I expect the fight of the 2030s to be over how to tax manufactured thinking. Expect proposals to tax inference, compute or the energy that feeds it, and expect them to be ugly, contested and eventually unavoidable. A major economy will pilot some form of compute or AI-output levy by 2032.

Already true

The companies building intelligence factories are among the most valuable and capital-intensive on earth, and their spending is measured in the hundreds of billions.

What has to happen

The labour share of income falls measurably in the economies adopting AI fastest, forcing the tax question onto the table.

Where I am probably wrong

Abundance may create so many new, human-centred jobs that wages keep up and the tax base simply moves along with them, as it eventually did after electrification. History says that is possible. It also says it took a long time and a lot of pain.

Schick’s own thesis carries a condition most people skip. She describes intelligence as manufactured from energy and compute, and dependent on the public’s trust. She has also argued that Americans have judged AI before most of them have adopted it. I think that last clause is the most important sentence in her whole framework.

When the supply of intelligence is effectively unlimited, the scarce thing is knowing which intelligence to believe. Output is no longer the product. Provenance is. Who made this, who checked it, who is liable if it is wrong. The winning brands of the next decade will not be the smartest. They will be the most accountable.

I expect a visible “human-accountable” standard to emerge in professional services, the way “organic” did in food: a label that says a named person stands behind this. Not because humans made every word, but because a human answers for it.

Already true

Public opinion of AI is running ahead of personal experience with it, and provenance tools for synthetic content already exist and are funded.

What has to happen

Clients begin demanding, and paying for, a disclosed chain of accountability on AI-assisted professional work.

Where I am probably wrong

People may simply stop caring. If manufactured work is reliably good, trust could become ambient, the way nobody asks who assembled their phone, and accountability labels would fade into fine print.

For fifty years, the educated class watched manufacturing towns lose their factories and told them to retrain. Now the substitution is coming for the educated class, and the shock will not only be financial. It will be about identity.

A great many people built their sense of worth on being the smart one: the one who knows the code, the case law, the tax rule, the answer. When that becomes something you can rent by the million tokens, the loss is not just income. It is status and meaning. Displaced factory towns turned that loss into a political force that remade both parties. The credentialed will do the same, and they are far better at organising, writing and arguing.

I expect a new populism of the displaced professional: articulate, angry at the owners of compute, and demanding protection for licensed work. It will look strange because it will come from people who used to be the winners.

Already true

Anxiety about AI is concentrated among educated knowledge workers, the group that historically felt safest from automation.

What has to happen

Licensing bodies and professional associations begin lobbying openly to restrict AI in their fields, framed as consumer protection.

Where I am probably wrong

Professionals may adapt faster than factory workers could, precisely because they can use the new tool themselves. If most of them become the signature from prediction one rather than the displaced, the backlash stays small.

If intelligence is manufactured from quartz, chips, power and models, then it has a supply chain, and anything with a supply chain can be controlled, embargoed and fought over. Nations will start measuring intelligence self-sufficiency the way they once measured energy independence.

We have already had a preview. In June 2026, US export controls briefly forced one frontier lab to suspend access to its newest models, until the controls were lifted at the end of the month. Whatever you think of that decision, it established the category: a model can be a controlled good. From here, expect sovereign compute programmes, national model champions, and alliances built around who can make intelligence and who has to import it.

Already true

Chips are export-controlled, fabrication is treated as strategic infrastructure, and at least one frontier model has been subject to export restrictions.

What has to happen

Governments begin treating domestic compute capacity as a national security metric, reported and targeted like energy reserves.

Where I am probably wrong

Open models could make intelligence too diffuse to control, the way the internet made information hard to embargo. If a capable model runs on a laptop anywhere, control shifts back to chips and power alone.