AI agent governance moves to the data layer

The article argues that as AI agents gain autonomy, governance must shift from abstract policies to the data layer, where the context of each action is known. The car door example illustrates that rules without context can be counterproductive, and agents lack the judgment to override them. This implies that effective governance requires real-time, context-aware enforcement, likely embedded in the infrastructure where data and actions occur. The stakes are high because enterprises bear the responsibility for agent actions, and this cannot be managed in hindsight. The shift to data-layer governance suggests a need for technical solutions that can evaluate actions against dynamic rules, rather than static policies.