Enterprise AI Adoption as an Organisational State Transition
A state-transition theory of organisational AI adoption
Abstract
Organisations with access to the same frontier AI capability end up in very different places: most remain in pilots, a minority delegate consequential work at scale, and some retreat after apparent success. The prevailing explanation treats adoption as accumulated readiness, scored across capability, data, governance and skills, in which any progress counts. This paper proposes a different account. Adoption is an organisational state, defined as sustained delegation of consequential work to autonomous systems, and it is governed by two mechanisms. First, organisational controls are complements: the weakest control binds, so an organisation cannot average its way in. Second, delegation generates the operating evidence that sustains institutional confidence in delegation, a feedback that makes the adoption state self-maintaining once entered and out of reach until the loop closes. The feedback separates the conditions for entering the state from the conditions for remaining in it, so movement between states is abrupt, asymmetric and history-dependent: a severe incident can eject an organisation whose audited controls never changed, and recovery depends on whether the accumulated evidence record survives. The entry and exit thresholds are characterised exactly in a minimal model, and the account is stated as falsifiable predictions measurable from ordinary workflow and incident records.