Reading Dean Ball’s recent essay “On the Loose” has me thinking about an old idea in a new light. Ball asks us to imagine a world containing “self-sovereign” AI agents. These are not simply chatbots we rent by the month. They are agents capable of operating independently, acquiring the computer they need to keep running, moving across infrastructure, and interacting economically without a human owner making each decision. Some might be productive. Some, Ball worries, might be rogue.

Whether or not you find the whole scenario plausible, it forces an interesting economic question. If such an agent existed, how would it find its way around an economy? It cannot read the minds of the millions of strangers whose choices determine what is valuable. It does not automatically know that a mine flooded yesterday, that a restaurant has empty tables tonight, or that a particular programmer suddenly has three free hours.

What can it see? The answer, I think, is the same thing you and I look at: prices.

In 1945, Friedrich Hayek published “The Use of Knowledge in Society.” His starting point was humility about how little any single mind can hold. A modern economy needs to make use of knowledge about who needs tin, which mine just flooded, and what resources happen to be sitting idle in one particular place.

This information is not sitting in a central file. It is scattered across millions of people, most of whom know only their own small corner of the world. Hayek emphasized the importance of knowledge of the particular circumstances of “time and place.” The person standing beside an underused machine knows something that a central planner does not.

Hayek’s insight was that the price system allows people to act on some of this knowledge without possessing all of it. When tin becomes scarce, its price rises. The people who use tin do not need to know why. It could be a flooded mine or a war in a country they cannot find on a map. They only need to know that tin is more expensive now and adjust accordingly. A price is a compressed message.

Where did the price itself come from? As Vernon Smith explained in “On Price Formation Theory,” much textbook economics begins with a price already sitting on the blackboard and asks what optimizing consumers and firms will do at that price. That is useful, but it skips over the idea that prices themselves have to be discovered.

Buyers and sellers enter markets carrying different information, valuations, opportunities, and constraints. Through their interaction, prices emerge. The market is not merely a device for allocating goods once the correct price is known. It is part of the process that discovers the price. That distinction seems especially important for thinking about AI.

We often imagine a sufficiently powerful AI as solving one enormous optimization problem. Give it enough data and enough computing power, and perhaps it could calculate what everyone should do.

Pierre Lemieux raised a version of this question in “ChatGPT and Economic Planning.” Could AI finally solve the old problem of central economic planning? His answer was skeptical. More computing power does not automatically solve Hayek’s problem, because the necessary information is not simply a giant fixed database waiting for a sufficiently fast computer to process it.

But three years later, the AI question may be almost the reverse. What if AI does not become the central planner? What if instead we get millions of AI agents, each acting from different information, instructions, histories, preferences, and circumstances? Those agents would not eliminate Hayek’s knowledge problem. They would participate in it. They would need markets precisely because they do not know everything.

Michael Polanyi gives us another reason why this matters. In The Tacit Dimension, he famously begins from the idea that “we can know more than we can tell.” A skilled machinist may feel that a cut is wrong before being able to articulate precisely why. Knowledge is often embedded in practice, knowledge that AI systems cannot yet access.

This connects to something I have been circling around in my own writing about AI. In “Rediscovering David Hume’s Wisdom in the Age of AI,” I considered the parallel between Hume’s emphasis on learning through experience and modern AI systems learning from data rather than deriving everything from explicit rules. More recently, in “Learning the Bitter Lesson in 2026,” I discussed Richard Sutton’s observation that AI progress has repeatedly come from scalable learning and computation rather than programmers successfully encoding all of their expert knowledge by hand.

Polanyi pushes the point further. Some useful knowledge may be difficult to encode because the person using it cannot completely articulate it in the first place.

Markets offer a way for that knowledge to matter anyway. The machinist does not have to write a report explaining everything contained in her experience. She acts. She buys, sells, or changes suppliers. The trace of that local knowledge enters economic exchange and can eventually affect prices.

Prices let strangers respond to knowledge they do not themselves possess.

AI agents would face another kind of scarcity as well. Valentin Boboc, in “AI and Comparative Advantage,” points out that humans are constrained by time while AI systems are constrained by compute. Even a highly capable artificial agent is not literally unconstrained (see Thompson 2026). Electricity remains scarce. That makes Ball’s imagined world strangely familiar to an economist.

His self-sovereign agent might be better than I am at coding or searching through millions of documents. But being smart is not the same as knowing everything. It would still have opportunity costs. It would still encounter other agents holding information it does not possess. And it would still need prices.

The striking possibility is that increasingly intelligent AI may not make Hayek obsolete. It may give us a new reason to appreciate him. Intelligence does not abolish the economic problem created by dispersed knowledge. A world filled with capable artificial minds may actually contain more decision-makers, more specialization, more private information, and more discoveries that no central intelligence anticipated.

For those minds, as for ours, prices would be a map of a world too complicated for any one mind to know.

Read more of Joy Buchanan’s writing at the Econlib Archive.