Back to Search View Original Cite This Article

Abstract

<p>Transaction cost economics has long treated information technology as a force that lowers the cost of transacting and therefore shrinks firm boundaries. Generative and agentic systems break the assumption on which that prediction rests. They collapse the cost of producing an artifact without lowering, and in several documented cases while raising, the cost of establishing that the artifact is fit for use. This paper develops the consequences of that wedge. Three results follow. First, as unit production cost approaches zero and per-unit verification cost remains bounded away from zero, verification's share of total activity cost converges to one, so the make-or-buy question ceases to be about production and becomes a question about verification alone. Second, verification cannot be recursively delegated to automated verifiers whose errors are correlated with those of the generator, and residual liability cannot be assigned to agents that hold no assets and no legal personality; recent decisions in Mata v. Avianca and Moffatt v. Air Canada, together with the allocation of duties under Regulation (EU) 2024/1689, place that residual on the deploying firm. Third, the human judgments produced during verification are the scarce, distribution-specific labels required to improve a firm's own models, which makes verification simultaneously a cost and an appropriable asset. Together these mechanisms predict selective reinternalization of high-consequence knowledge work rather than the continued unbundling that the electronic markets hypothesis anticipated. The paper states six falsifiable predictions with proposed operationalizations, and identifies the conditions, principally the availability of cheap machine-checkable success criteria, under which the opposite prediction should hold.</p>

Show More

Keywords

cost verification which firm prediction

Related Articles

PORE

About

Connect