FROM BUYING A SERVICE TO RUNNING IT: DELEGATED AI, STANDING PERMISSIONS, AND THE SCOPE OF THE FIRM
Sipeng Zeng
Abstract
Firms often buy a managed service even when its inputs trade separately, because recovering from a disruption requires timely discretion over resources that other firms own. We develop a transaction cost theory of how artificial intelligence (AI) changes this choice, starting from the comparative-adaptation logic of governance and tracing what happens when the provider, the buyer, or the suppliers deploy AI. The boundary effect turns on who holds executable authority over suppliers’ resources. AI used by the provider, or by the buyer only for advice, keeps the service purchased. AI that constructs responses to unanticipated disruptions and enacts them under standing permissions from independent suppliers lets the buyer orchestrate the service itself while it continues to buy the inputs. Because the commitments in a response are complements, the last missing permission often decides the outcome, and sourcing shifts are discrete. Delegation then moves the adaptation problem from cases to rules. As more transactions run under shared permissions, a stale permission produces correlated losses, and hierarchy, and sometimes ownership of a bottleneck asset, regains value where revising permissions requires control over a supplier’s assets.