The coordination of multiple AI agents that divide labor, pass results, and check each other's work to complete tasks too complex for a single agent.
Broad technical consensus on what this term means. Disagreements that invoke it are about the world, not the word.
Multi-agent systems allow subtasks to be parallelized or specialized, with one agent acting as a planner and others as executors or verifiers. This architecture is central to enterprise agentic deployments and directly affects reliability metrics cited in claims about agentic ROI. The scaffold has agent but not the coordination layer that makes agents economically meaningful at enterprise scale.