The transaction cost structure of pre-AI software development was elaborate. A person with an idea needed to translate intuitive knowledge into technical specification — a transaction. The specification was interpreted by a programmer — another transaction, involving the cost of communication between two people with different cognitive frameworks. Each stage was a handoff with costs: scheduling, context-switching, information loss in transmission. Large organizations developed elaborate institutional structures — agile methodologies, code review processes, integration testing — to manage these costs. A significant fraction of any software project's cost was coordination, not code.
The language interface collapsed these costs. A person could describe an idea in natural language and receive a working implementation. The translation transaction was eliminated. The coordination transactions were compressed. What remained was a conversation between a single human and a machine, producing functional output in real time. The Trivandrum training account in You On AI is a concrete measurement of this collapse — twenty engineers each achieving the leverage of a full team.
But North's framework warns that transaction costs shift rather than disappear. Three new categories deserve identification. First, quality evaluation cost: when the machine writes the code, the human's relationship to the output changes fundamentally. The senior engineer described by Segal making architectural decisions with diminishing confidence because she had lost the incidental learning of manual implementation illustrates this new cost.
Second, human capital maintenance cost. The language interface reduced the cost of producing software but not the cost of producing the judgment required to direct the machine effectively. The traditional pathway for developing this judgment — years of hands-on implementation that deposited understanding through struggle — has been disrupted. Third, social adjustment cost: if twenty engineers can do the work of four hundred, the institutional question is what happens to the three hundred and eighty. Labor market adjustment, retraining, and social safety nets for the displaced are transaction costs of the AI transition that do not appear on any company's balance sheet.
Ronald Coase introduced the concept in his 1937 paper 'The Nature of the Firm' to explain the existence of hierarchical organizations within a theoretically efficient market economy. The paper sat largely unused for decades until North and others recognized that the same logic explained not only the theory of the firm but the theory of the economy itself. Coase was awarded the Nobel Prize in 1991, two years before North received the same honor for extending the framework.
The 1960 publication of Coase's 'The Problem of Social Cost' made explicit the connection between transaction costs and property rights. North's contribution was to demonstrate that institutional quality determined the magnitude of transaction costs across entire economies, and that institutional differences — rather than technology or resource endowments — explained the divergent development paths of nations.
Transaction costs shift, never disappear. The reduction of one category creates or reveals another. Net economic welfare depends on the ratio of new costs to old.
Institutional frameworks determine magnitudes. Good institutions reduce transaction costs; bad institutions raise them to prohibitive levels that inhibit exchange.
The AI collapse is uneven. Implementation costs fell dramatically; quality evaluation, human capital maintenance, and social adjustment costs rose in ways that balance sheets do not capture.
Invisible costs still bite. The social adjustment cost of AI-driven displacement is borne by real people even when no corporate accounting system records it.
Institutional capacity is the variable. A society with strong institutions for managing quality evaluation, human capital, and social adjustment absorbs the new costs productively. A society without them accumulates costs as strain.