The Proportionality Principle: Governance Architecture as a Function of Autonomous Capacity

The governance of artificial intelligence systems is currently structured around a single categorical variable: whether a system uses AI. This paper argues that this classification is the root cause of two persistent and symmetric failure modes — Governance Theater, where structurally simple systems are subjected to unnecessary constitutional overhead, and Governance Insufficiency, where operationally autonomous systems are governed by reactive mechanisms that cannot enforce pre-commit constraints. Neither failure mode results from insufficient regulatory effort. Both are structural consequences of measuring the wrong variable. The correct variable is autonomous capacity: the effective ability of a system to alter state without human authorization at the commit boundary. This paper introduces the Proportionality Principle — governance architecture must scale with autonomous capacity, not with AI presence — and formalizes it through the Autonomous Governance Requirement (AGR), a calibration framework that quantifies minimum residual governance architecture as a function of seven structural dimensions. The framework is anchored through four empirical archetypes spanning the full autonomy gradient, from assisted human review to persistent multi-agent runtime. A system performing document extraction with human validation and a system executing autonomous operational decisions both use AI. They do not belong to the same constitutional class.