Transaction costs are the expenses of using markets: the resources spent learning prices, finding counterparties, negotiating terms, writing contracts, and monitoring compliance. Ronald Coase argued in 1937 ("The Nature of the Firm") that these costs explain why firms exist and where their boundaries fall, with activities organized inside a firm when doing so is cheaper than transacting in the market. Shahidi, Rusak, Manning, Fradkin, and Horton (2025) use this frame to organize the agentic transition: because AI agents can perform many transaction-cost activities at very low marginal cost, the make-or-buy boundary shifts, the firm's optimal scope changes, and previously infeasible market designs become viable.
Background
Coase's 1937 account treats the boundary between firm and market as the outcome of a comparison: an activity is brought inside a firm when internal coordination is cheaper than contracting for it on the market, and left to the market otherwise. The expenses that tip this comparison are transaction costs. Shahidi, Rusak, Manning, Fradkin, and Horton (2025) adopt this as the central economic frame for AI agents, on the argument that the activities making up transaction costs are precisely those agents can carry out cheaply and at scale.
Coasean categories
Transaction costs decompose into three activities:
- Information costs — learning prices, finding counterparties, evaluating quality.
- Bargaining costs — negotiating terms, structuring contracts.
- Enforcement costs — monitoring compliance, dispute resolution.
As the marginal cost of these activities collapses, the frame predicts three consequences: markets that were previously too costly to organize become viable; firm scope contracts, with more activities done in markets and fewer inside firms; and new market designs become feasible, including matching algorithms that require full preference rankings and deferred acceptance.
Where the frame applies
Shahidi et al. (2025) identify several settings where falling transaction costs reshape existing arrangements.
In substitution for human intermediation, realtors, headhunters, brokers, and financial advisors charge percentage-of-transaction fees that approximate the human transaction-cost markup. Agents able to intermediate at near-zero marginal cost would collapse those margins.
In enabling previously infeasible markets, some markets do not exist because the transaction cost exceeds the gain from trade, as with diagnostic triage for small home repairs, custom one-off scripts, and micro-negotiations on small purchases. Agents push that threshold down.
In shifting make-or-buy, firms that vertically integrated to economize on transaction costs, for example by bringing customer service in-house, may find market provision more attractive once agents make outsourcing cheap and reliable.
In bargaining transformation, classic bargaining models such as Rubinstein's assume impatience because human time is scarce. For agents the binding constraint is compute rather than time, so agents can negotiate earlier, hold many concurrent negotiations, and persist longer, changing the equilibrium of bargaining games.
Limits and countervailing frictions
Shahidi et al. (2025) present the frame as descriptive of forces rather than deterministic, and identify several frictions that can blunt or reverse the predicted dynamics. Externalities across agents include congestion, such as job applications flooding employers (per Wiles & Horton 2025), as well as Sybil attacks and race-to-the-bottom dynamics in price obfuscation. Imperfect alignment, captured by Specification Hazard, means an agent's actions do not reliably reflect the principal's preferences. Persistent dispersion under product differentiation is documented by Ellison & Ellison (2018) and Kaye (2024), who show that lower search costs can increase price dispersion when products are differentiated. Market power, in the form of concentration among foundation-model and agent providers, can capture the gains rather than passing them to consumers. Information asymmetry arises because agents reveal preferences to firms, enabling personalized pricing that may benefit consumers through better matching or firms through better surplus extraction.
See also
- Agent Supply Archetypes (BYO/Bowling-Shoe × Horizontal/Vertical) — the make-or-buy frame applied to agents themselves (BYO vs. Bowling-Shoe as market vs. integrated supply).
- Pay-per-Crawl (Pigouvian Pricing of Agent Traffic) — Pigouvian pricing internalizing the externality that pure transaction-cost reductions do not capture.
- Principal-Agent Problem Applied to AI — what determines whether agents actually carry out the principal's intent, conditional on transaction costs being low.
- Algorithmic Pricing and Antitrust — agent-driven pricing collusion (Calvano et al. 2020) as one downstream consequence.
- AI Economic Primitives — the empirical measurement framework for how agents reshape labor and tasks at the activity level.
Relationships
- depends-on: Agentic AI — without capable agents, the cost reductions do not materialize
- supports: Agent Supply Archetypes (BYO/Bowling-Shoe × Horizontal/Vertical)
- supports: Pay-per-Crawl (Pigouvian Pricing of Agent Traffic)
- related: Principal-Agent Problem Applied to AI
- related: Specification Hazard
- related: AI Economic Primitives
- supports: The Coasean Singularity? Demand, Supply, and Market Design with AI Agents — Shahidi, Rusak, Manning, Fradkin, Horton (2025) — primary expository source for the Coasean frame