Technological hold-up

Fisher Body and General Motors: a ten-year contract signed in 1919, stamping dies dedicated to each model, then GM's purchase of Fisher in 1926. Klein, Crawford and Alchian made it the textbook example of hold-up in 1978, and Coase disputed it in 2000. Technological hold-up applies to the digital sector the fundamental transformation theory that Oliver Williamson developed in his work on the economics of organizations. Klein, Crawford and Alchian (1978) formalized the mechanism. When a company invests in assets highly specific to one commercial relationship (here, ultra-calibrated prompts, complex workflows, fine-tuned models), those assets lose much of their value if the relationship ends. The balance of power tilts toward the supplier. A relationship that started out competitive turns into a bilateral monopoly: the supplier can extract economic rent, and the company cannot change partners without prohibitive cost. With agentic AI, this risk is structural for companies whose operational architecture rests on large language model APIs. The provider can unilaterally change the neural architecture and make months of prompt engineering obsolete. The abrupt deprecation of a model (a documented practice at OpenAI and Anthropic) forces recalibration of all workflows. Opaque internal policies can censor sensitive business queries and interrupt critical processes. The oligopolistic concentration of the leading LLM market aggravates the risk, because it limits substitutes.