Is Intelligence a Commodity?

Most conversations about AI begin with supply. How much work can a model do? Which jobs can it replace? How cheaply can a company produce code, analysis, images, or advice?

Those are important questions. But they leave out the other side of the equation: What will people want to do when they can draw on far more intelligence than they could previously afford?

Venkatesh Rao raises that question in his essay “Cognitive Mercantilism”. His starting point is Joyce Appleby’s The Relentless Revolution: A History of Capitalism. Appleby describes capitalism as a historical invention, shaped by changes in farming, trade, production, culture, and people’s ideas about human behaviour. It was not the inevitable result of an age-old instinct to buy and sell. The world we now take for granted had to be imagined and built.

Rao suggests that we may be living through a comparable change in how we understand intelligence. I find the analogy useful because it challenges a quiet assumption beneath much of the AI debate: that the need for intelligence is more or less fixed, so producing more of it will mainly make existing workers redundant.

I can already see a version of this in my own practice, and I’ll come back to that.

When wealth looked like something to capture

In Rao’s account of the older mercantilist worldview, wealth appeared to be a limited stock. Land, precious metals, and the output of controlled labour were the prizes. A state could enrich itself by claiming territory, hoarding bullion, or taking a larger share of trade from its rivals.

Ordinary people figured chiefly as labourers. Their capacity to buy new things, pursue their own ambitions, or help generate new kinds of economic activity was easy for elites to overlook. If most people spent their lives farming, it could seem natural that farming was simply what most people were for.

Appleby’s larger point is that capitalism unsettled such assumptions. She emphasizes the role of culture and changing ideas about human nature alongside material changes. In her own discussion of the book, she argues that England’s transformation did not result from its people having superior intelligence. Technological curiosity became connected to economic opportunities and incentives to improve production.

The change was not just that more goods could be made. People encountered goods they had not previously expected to own, developed new tastes, and found reasons to earn more. Demand grew alongside productive capacity. New possibilities changed what people wanted.

That history has a destructive side as well as a creative one. Appleby is explicit about the communities disrupted by economic change, oppressive labour conditions, and the expansion of imperial power. Greater wealth did not distribute itself fairly. The fact that an old view of human potential was too narrow does not mean the system that replaced it was automatically just.

That caution matters when we turn to AI.

A mercantilist view of intelligence

Rao’s phrase cognitive mercantilism describes a way of treating intelligence as a scarce treasure to be accumulated and controlled. The valuable assets become elite talent, intellectual property, proprietary models, and privileged access to knowledge. Nations and companies compete to secure those assets much as mercantilist states once competed for territory and bullion.

This view also assigns people familiar roles. A small group produces and governs intelligence. Everyone else performs routine cognitive labour or consumes the finished products. If AI automates that routine work, the immediate question becomes: What use will all those people be?

That question may tell us more about our limited imagination than about the people in question.

We have organized education, professional services, research, software development, and much of business around the fact that skilled intellectual work is expensive. That has shaped not only who provides it, but who gets to ask for it. Many worthwhile questions never become projects because the first serious step costs too much.

A small organization may have ideas it cannot afford to investigate. An individual may want to build a tool, examine a policy, learn a technical subject, or test a business concept, but have no practical route from curiosity to action. We see the projects that get started. We rarely see the demand that never had a chance to appear.

What happens when the first step gets cheaper?

Rao’s most interesting move is to ask whether easier access to intelligence might increase the appetite for it. When the cost of pursuing an idea falls, people may discover ambitions they previously set aside. A useful answer leads to a better question. A rough prototype makes a more ambitious project imaginable. Success creates new needs for analysis, design, judgment, and collaboration.

This is a different way to think about AI’s economic impact. Instead of assuming a fixed volume of cognitive work to be divided between humans and machines, it allows for the possibility that the volume and variety of work will grow. People may spend more effort thinking precisely because their effort can now lead somewhere.

That does not mean every new project will succeed, or that everyone will enjoy equal access. It does mean we should be careful about treating today’s pattern of work and demand as a permanent measure of human potential.

When solo no longer means alone

I can see this change in my own business. I have spent more than three decades as a solo practitioner. Years ago, when I was providing outsourced IT services, I tried expanding through a group of seven subcontractors. I found it difficult to supervise all their work closely, and in some cases subcontractors developed direct relationships with clients who then stopped working through me. I decided the possible extra income did not justify the problems and returned to working solo.

AI gives me a different way to explore growth. I can delegate parts of a project to tools and agents while continuing to manage the client relationship and review the work myself. I can investigate areas and consider projects that would once have required a larger staff. I am still responsible for quality, accuracy, and the promises I make to clients. But “solo” no longer has to mean that every stage of every task depends on my working through it alone.

My example is small. The wider question is what happens when millions of people and small organizations gain a similar ability to pursue work that was previously beyond their reach.

The politics of access

If Rao is right, the central public question will extend beyond how to protect existing jobs. People will also want a fair chance to use intelligence: to learn, create, start enterprises, solve local problems, and participate in decisions that affect them.

Access means more than an affordable subscription. It includes the ability to check a system’s output, protect sensitive information, move work between tools, and challenge decisions made with AI. It includes education that helps people exercise judgment rather than merely accept a polished response. And it includes institutions willing to recognize work and ideas from people who lack traditional credentials.

There is a risk that abundant AI capability could coexist with concentrated power. A few firms might control the best systems, the channels through which people reach them, and the terms on which businesses build upon them. In that case, intelligence could become cheaper to produce while remaining expensive or restrictive to use. Rao’s analogy helps name that danger: a world full of productive capacity can still be organized around hoarding.

The history Appleby tells also suggests that new opportunities create new conflicts. People may work harder because they see possibilities worth pursuing, only to find that others capture most of the value. Access, bargaining power, and accountability will matter just as much as technical capability.

Who gets to have ambitions?

Rao pushes his argument one step further. He asks whether AI agents might eventually be more than tools that produce intelligence on command. Could sufficiently capable agents pursue projects, develop new needs for information, and participate in an economy as sources of demand themselves?

That is the most speculative part of his essay, and it deserves to be treated as a question rather than a prediction. Today’s useful agents do not establish that future systems will have independent interests, nor does the possibility of greater autonomy remove the need for oversight. But the question exposes the limits of a purely supply-focused view. If we imagine AI only as a workforce serving goals fixed in advance, we may miss how new forms of activity and demand could emerge.

For now, the human version of the question is already large enough: How many people have ambitions that our present institutions make too costly to pursue?

Appleby’s history reminds us that an economic order can seem natural long after its assumptions have become restrictive. Rao’s essay asks us to notice the assumptions we make about intelligence before we build the next order around them.

AI may eliminate some work and create serious problems that require deliberate responses. It may also make entirely new work worth doing. We cannot know the balance simply by counting the tasks a model can perform today. We also have to watch what people begin to want, attempt, and demand when they have more capacity to act.

That is the possibility I find most compelling. The story of AI may turn out to be about more than producing intelligence at scale. It may be about discovering how much intelligence people can put to use once they are free to imagine a larger life or a larger business.