The Benjamin Button Problem: Ashby's Constraint and the Agentic Enterprise

Abstract

The industry tried to be born old—deploying agents before earning the governance to sustain them. For eighteen months, the narrative held that LLMs reason like humans, therefore agents everywhere. That story skipped the hard parts: governance, credibility, institutional risk, coordination cost. Enterprises didn't reject agents. They priced them correctly. This paper argues for aging backwards: beginning with GenAI-augmented automation, graduating to agentic patterns, and instantiating agents only as a reward for earned governance. The theoretical foundation combines Ashby's Law of Requisite Variety with Blackman's insight on stability: you cannot govern a high-variety system with a low-variety regulator, and low-frequency governance cannot stably control high-frequency agent behavior. Together, these constraints define a regime boundary—below which deterministic governance suffices, above which agentic governance becomes mandatory. Crucially, GenAI itself enables the crossing: by augmenting governance at computational frequency, it solves the frequentist problem that makes agentic governance possible.