Can an algorithm that locks in its own rules in advance beat ordinary seller-vs-seller competition at matching buyers and sellers?

This explores whether an algorithm that can lock in its rules ahead of time can get better market outcomes than ordinary competition between sellers. First, a caveat: the collection has no paper that studies pre-commitment directly. What it has is one formal result that comes close, plus several studies of algorithms that have no commitment power and end up somewhere surprising anyway.

The closest match is a model comparing an optimal recommendation algorithm with plain price competition between sellers Does algorithmic recommendation lose its edge as sellers compete more?. The algorithm's advantage comes from one ability: it can withhold trades, deliberately leaving some matches unmade, to limit the 'information rents' sellers earn from knowing more than the platform. That is commitment-like power, meaning the algorithm sets rules sellers can't bargain around. The surprise is how fragile the advantage is. It's real only in thin markets. As more sellers enter, the algorithm's welfare edge over plain competition shrinks steadily toward zero. So committing to rules in advance mostly pays off where competition is weak, and buys little where the market is already crowded.

The collusion studies show the reverse case. Here, algorithms that can't commit to anything still reach outcomes that look coordinated. Independent AI pricing agents playing repeated games settle on above-competitive prices without communicating Can independent AI pricing agents collude without communicating?. LLM pricing agents do the same in a two-seller price war, signalling only through the prices they post, and small changes in prompt wording shift how high prices go Do LLM pricing agents autonomously collude in repeated auctions?. Pairing a larger model with a smaller one doesn't break this. The larger model becomes a price leader and the smaller one follows Does model-size difference break LLM pricing collusion?. Put together: repeated interaction, plus a capability gap between models, can do much of the work you might expect formal commitment to do.

The two-sided-market framing raises a further question: who is the algorithm committing on behalf of? Choudary argues that agents representing only consumers change where buyers enter a market but not how the market works. Structural change needs seller-side agents too, so that both represented sides can negotiate Do one-sided consumer agents actually change market structure?. Rothschild and colleagues argue that cheaper communication between agents could shift market power by lowering switching costs. Their paper makes this argument conceptually and offers no empirical evidence Could agent-to-agent communication reshape market power?. O'Reilly makes a related point at the infrastructure level: shared protocols like MCP act like market rules that everyone signs up to, and they can turn a winner-take-all race into a field where many vendors compete Do protocols reshape markets by enabling multi-vendor competition?.

Finally, actual LLM agents in simulated marketplaces show how far practice is from the theory. Frontier models get near-optimal results only when search conditions are ideal. All of them show a strong first-proposal bias, weighting how fast an offer arrives 10–30x more than how good it is Do LLM agents achieve fair marketplace outcomes when given ideal search?. Anthropic's Project Swap found that which model powers the agents shapes market efficiency more than whether they're told to be ruthless or cooperative Does model capability or instructions shape agent negotiation outcomes?. The takeaway you may not have expected: in today's agent markets, the effective 'commitment' may come less from rules a designer sets on purpose and more from model capability and quirks like answering first. Those properties act like rules, but nobody chose them.