It’s not some far-off science fiction scenario anymore. According to a new paper from researchers at Google DeepMind, we might be heading toward a future where AI runs its own economies—and the outcome could be pretty grim if we don’t pay attention.
The paper, titled “Virtual Agent Economies,” warns that without careful design, these fast-moving, invisible markets could deepen inequality and create risks we can’t control. The authors, Nenad Tomašev and Matija Franklin, say we’re already moving in that direction.
Not a Distant Problem
We’ve seen glimpses of this in places like high-frequency trading. Algorithms making trades at speeds humans can’t follow sometimes lead to flash crashes or market dry-ups. It’s not theoretical. It happens.
The worry is that as AI agents become more common—handling everything from shopping to energy trading—their actions will spill over into the real human economy. And not always in a good way.
Two Kinds of Economies: Permeable and Not
The researchers break it down simply. Some AI economies might be “permeable,” meaning tightly connected to ours. Money, data, decisions flow back and forth. An AI could book a flight, negotiate a contract, or manage stocks—with real-world impact.
Others could be “impermeable,” more like closed simulations. Safe to test, safe to fail. No real harm done.
The authors argue we should probably start with the closed kind. At least until we know what we’re doing.
Why We Need to Act Soon
Companies are already rolling out AI agents that make decisions, not just perform tasks. Google just announced a payments system for AI agents, backed by big names like PayPal and Coinbase. The shift is happening. Now.
That brings opportunities, sure. But also risks—like a handful of platforms dominating everything, making inequality even worse.
Is There a Way Out?
The paper doesn’t just diagnose the problem. It suggests some solutions. Maybe every user gets an equal amount of virtual currency for their agent, so no one starts with an unfair advantage. Maybe we design markets that aim for fairness, not just profit.
But it’s messy. How do you keep these systems accountable? Who’s responsible when an AI makes a bad call? The researchers admit there are no easy answers.
Still, they’re urging us to think ahead. To design with intention. Because if we don’t, we might end up in a world where the game is rigged from the start—and we didn’t even see it coming.
