Do AI agents actually reduce friction in electronic markets?

Martin Bichler, writing in Electronic Markets (Springer, open access), models "agentic markets" — electronic markets where autonomous AI agents search, evaluate, negotiate and transact on behalf of buyers and/or sellers — and argues they "do not eliminate market frictions; they transform them." Even though AI agents can "drastically reduce technical search costs," delegating search and negotiation to "probabilistic, content-generating systems introduces new behavioral and informational frictions." Bichler states this directly challenges "the common presumption that AI-mediated markets necessarily approximate Bertrand-style competition as search costs approach zero" — the textbook expectation that cheaper search forces competing sellers down to cost-based pricing.

The argument rests on five capabilities Bichler attributes to GenAI-based agents, each of which modifies core primitives of standard microeconomic models. Agents performing "autonomous delegated search" infer latent preferences from loosely specified natural-language goals rather than acting on fully specified utility functions, replacing "explicit preference revelation with probabilistic inference" and introducing noise into match evaluation. Agents can inspect far more alternatives, shifting the "locus of search from human cognition to algorithmic mediation," and can summarize and compare offerings, which reduces information overload but "may also compress or distort match-value signals." Automated negotiation and execution let agents "trigger renegotiations" and adapt to prices continuously, and low-cost content and offer generation lowers entry barriers, increasing variety but potentially inducing "excessive entry and low-value proliferation." In the frictionless limit, Bichler's model still predicts an efficient market with intensified competition and reduced markups, but he treats that limit as the case these five capabilities push real markets away from.

This is a market-level argument underneath agent-level findings already in the library: Do shopping agents respond more strongly to price and rating cues? and Do AI shopping agents show persistent position bias in text interfaces? document exactly the kind of "behavioral and informational friction" Bichler's framework predicts should survive falling search costs — manipulation susceptibility and position bias are the micro-level mechanism his five capabilities describe in the abstract. It also sits alongside Why does manual search fail as product choices become complex?, which Bichler's "autonomous delegated search" capability echoes: both treat preference specification, not search cost, as the new scarce resource.

The excerpt is a theoretical model paper, not an empirical study of real agentic marketplaces, and the "five capabilities" form an analytical taxonomy rather than measured behaviors; the claims about dispersion, inefficient entry, and matching distortions describe predicted equilibrium tendencies, not observed market outcomes. The implication, at the strength a model paper supports, is that falling technical search costs should not be read as a proxy for falling total friction in agentic markets — the composition and reliability of the agents doing the searching matters as much as the cost of search itself.