Transaction Costs

In the AI Story

Carl Dahlman consolidated transaction costs into three canonical categories in 1979. Search costs are the costs of finding the right person or resource — discovering who can do what you need, at what price, with what quality. Bargaining costs are the costs of negotiating terms — agreeing on specifications, ownership, timelines, and the hundred details a contract must settle. Enforcement costs are the costs of ensuring compliance — monitoring whether work meets standards, resolving disputes, pursuing remedies when parties fail to perform. The distinction between transaction and production costs matters: when economists claim markets are efficient, they implicitly assume exchange is costless. In reality, exchange consumes substantial resources.

John Wallis and Douglass North estimated in 1986 that the transaction sector accounted for over forty-five percent of U.S. national income. Nearly half the economy was devoted not to making things but to facilitating agreements about making things. This estimate included legal services, financial intermediation, real estate brokerage, advertising, management consulting — activities whose primary purpose is coordinating exchange rather than producing goods. The order of magnitude is not disputed: a very large share of what knowledge workers do every day exists not because it produces value directly but because it facilitates production by others.

AI's effect on transaction costs is dramatically uneven across categories. Search costs approach zero in many domains: the engineer who once spent hours reading documentation can now ask an AI and receive synthesized answers in minutes. Bargaining costs collapse when the person who conceives a product can also produce it — the specification meeting that consumed an afternoon is replaced by direct conversation with an AI system. Translation costs — the expense of converting intention into a form someone else can execute — are eliminated when natural language becomes the programming interface. Enforcement costs present a more complex picture: monitoring whether implementation matches specification disappears when one person does both, but new verification costs emerge in evaluating whether AI output meets standards the user may not fully understand.

Origin

The concept emerged from Coase's 1931 traveling scholarship to study American industry. Twenty-one years old and armed with a question his London School of Economics professors could not answer, Coase visited factories and asked businessmen why they organized activities within firms rather than contracting for everything on the market. The businessmen found the question peculiar — of course firms existed. But economics had built an elaborate structure assuming all coordination happened through price mechanisms, without noticing that vast activity occurred inside organizations operating by different principles. Coase's 1937 paper "The Nature of the Firm" provided the answer: firms exist because using the price mechanism is costly. When transaction costs are high enough, internal coordination becomes cheaper than market exchange.

Key Ideas

The forty-five percent economy. Wallis and North's estimate that nearly half of U.S. GDP in 1986 was devoted to transaction rather than production activities — a finding that reframes what knowledge workers actually do.

AI as transaction-cost collapse. The natural language interface eliminates the most expensive transaction cost in knowledge work: translating intention from the person who knows what should exist into the language of the person who knows how to build it.

Mutating, not vanishing costs. Dave Friedman's observation that AI introduces new costs — compute friction, verification burdens, plausible-but-flawed output evaluation — meaning transaction costs transform rather than disappear.

The boundary follows the costs. Coase's foundational insight that the firm's boundary sits where marginal coordination costs equal marginal transaction costs — and shifts whenever that equilibrium moves.