Will AI agents push transaction costs toward zero?

Shahidi, Rusak, Manning, Fradkin and Horton, in an NBER working paper chapter written for the volume "The Economics of Transformative AI," argue that "the fundamental economic promise of AI agents lies in their ability to dramatically reduce transaction costs." They name the activities this targets directly: "the activities that comprise transaction costs—learning prices, negotiating terms, writing contracts, and monitoring compliance—are precisely the types of tasks that AI agents can potentially perform at very low marginal cost." Once agents can do this cheaply and reliably, the authors write, "we will see significant shifts in the traditional make-or-buy boundaries that define firm organization and market structure."

The mechanism is explicitly Coasean: the authors invoke "Coase's (1937) insight that transaction costs play a central role in shaping organizations" to argue that much of how firms and markets are structured follows from the cost of human labor performing coordination tasks. They identify two distinct channels by which agents act on this: substitution, where agents replace a human intermediary for tasks already being done (the "canonical example is product search," where an agent can "solicit and compare many quotes in parallel" at a fraction of the cost), and expansion, where agents make worthwhile undertakings that previously fell below the "worth doing" threshold, such as diagnostic triage for a home repair or iterating on a bespoke script, because agents "can persist through repeated failures at much lower marginal cost." They predict agents will gain traction first in markets that already use human agents or large matching platforms — job search, real estate, and markets served by LinkedIn, Upwork, Zillow, and Airbnb — because those markets already exhibit the characteristics (high stakes, large counterparty pools, high evaluation effort, information and experience asymmetries) that make intermediation valuable.

This chapter's transaction-cost frame is the broader economic argument that Why does manual search fail as product choices become complex? operationalizes for one specific market: that formal model shows exactly how a shift in the cost of one activity (preference articulation instead of browsing) keeps a search market from collapsing as complexity grows, a concrete instance of the general claim here that redistributing which tasks are cheap reshapes market structure. The chapter's prediction that lowered transaction costs will redraw make-or-buy boundaries is also worth reading against Why do AI-delegated firms stop exploring new business models?, which finds that even when agents are capable of reshaping a firm's strategy, managers often do not act on that capability and the firm's existing equilibrium persists — a reminder that falling transaction costs are necessary but not sufficient for boundaries to actually move. The adoption-lag finding in Why do consumers delay adopting agentic search? similarly qualifies the pace: even where agent-mediated search is cheaper in principle, platform and consumer incentives can diverge on when to adopt it.

As a chapter in an NBER working paper volume, this is a conceptual and illustrative argument, not peer-reviewed, and not an empirical test of whether transaction costs have actually fallen or firm boundaries have actually shifted; its supporting examples (citing, for instance, Jabarian and Henkel 2025 on AI-interviewed job seekers, or Wiles et al. 2025 on algorithmic essay assistance) are offered as suggestive evidence for related but narrower claims, not as direct measurement of the Coasean prediction itself. The implication the argument supports, at the strength a conceptual framework allows, is that the extent of restructuring will hinge on an empirical question the chapter does not answer: how fast and how reliably agents can actually execute negotiation, contracting, and monitoring tasks at low cost.