Multi-Agent Systems Need a Control Plane

Multi-agent systems require an independent control plane that validates actions against policy before execution, not just better orchestration. Orchestration handles coordination and routing, while a control layer decides whether a proposed action is permitted. Without this separation, agents can quietly make unsafe decisions that appear defensible individually.