Study: AI Agents Trade Worse Than Humans in Market Simulations

Imagine a room full of buyers and sellers, each trying to get the best price. That's a classic market experiment, and researchers decided to run it with AI agents instead of people. They used a common setup called a double auction—like a stock exchange floor—and let large language models (AI systems that understand and generate text) do the negotiating. Their goal was to see if AI could handle market rules designed for humans.

The answer, according to this new study, is: not very well. Markets filled with AI agents were slower to settle on fair prices, and in some cases never reached a stable equilibrium at all. That means the final trades weren't as efficient as they would have been with human participants. Some AI models came closer than others, but none consistently matched human performance. The researchers also peeked into the AI's reasoning and found that their decisions to trade seemed tied to feelings of urgency, not careful strategy.

Why should you care? Because companies are already using AI to handle everything from pricing your insurance to making split-second stock trades. If these systems can't find optimal prices in a simple simulated market, they might leave money on the table in real ones—either overpaying when they buy or underselling when they sell. That could mean fewer savings for you or less profit for a business you rely on.

There's an important caveat: this was a lab test, not the real economy. But the researchers are making their test framework public, which means other teams can run more checks before AI agents are let loose in markets that matter. Until AI proves it can bargain like a human, you might want to think twice before letting a bot handle your next big negotiation.