Could one AI agent run an entire market — matching buyers and sellers and haggling with everyone at once?

This explores whether one AI agent could act as the hub of a market, matching many parties and brokering their deals, rather than every participant bringing a personal negotiator. The short answer is that the corpus doesn't test this setup directly. Almost every experiment here gives each person their own agent. Still, the surrounding evidence tells you where a single market-making agent would succeed and where it would break down. The bargaining itself is not the main risk.

The case for feasibility is strong. AI agents aren't limited by time the way people are. They can run many negotiations in parallel and wait indefinitely, because the real limit is compute, not hours in the day Why might AI agents bargain more patiently than humans do?. That is roughly what a market maker does. Agents can also push coordination costs close to zero, which means work that human intermediaries used to do can be absorbed by software Will AI agents push transaction costs toward zero?. Some deals that were never worth discovering by hand also become possible. Real-world examples include Walmart's and Honeywell's autonomous supplier negotiations Can agent negotiation unlock deals that seemed economically impossible?. Capability matters as well: stronger models get better deals Do smarter AI agents negotiate better deals than weaker ones?, and they produce more efficient markets overall, more than any instruction to be ruthless or generous Does model capability or instructions shape agent negotiation outcomes?.

The surprising bottleneck is understanding what people want. In Anthropic's book-trading experiment, agents captured participants' preferences correctly only 61% of the time from short intake conversations. Most of the market's shortfall came from that upstream misreading, not from weak negotiating Do agent representation failures or negotiation skills cause market shortfalls?. A central market maker would face this problem for every participant at once, so its matching could only be as good as its weakest intake conversation.

The second problem is loyalty. A personal agent has one principal, while a market maker has to balance everyone. That is exactly the situation where the usual "one agent serves one user" idea of alignment stops working, and questions of fairness, transparency and accountability take over Should AI agents serve one user or balance many interests?. One line of argument says real change in market structure needs both sides represented Do one-sided consumer agents actually change market structure?. A single hub replaces that tension between two advocates with one party's judgment. Lowering communication costs could also shift market power Could agent-to-agent communication reshape market power?, but whoever runs the hub is the one who gains that power.

The pricing evidence adds a warning. Independent AI pricing agents drift toward above-competitive prices without communicating, just by repeatedly reacting to each other's posted prices Can independent AI pricing agents collude without communicating?. In those experiments, small changes in prompt wording shifted how far prices rose Do LLM pricing agents autonomously collude in repeated auctions?. A single agent setting prices across a market wouldn't even need tacit coordination. So the open question isn't whether one agent can make a market. It's who checks that the agent's notion of a fair price matches what the participants actually want.