A multi-agent system consists of several specialised AI agents, each handling one part of a process and exchanging results with one another, rather than a single agent handling the entire flow.
Distinction from a single agent. A single AI agent works through a task chain itself. A multi-agent system splits the task across several agents with different specialisations.
Typical setup in sourcing. An orchestrating agent distributes sub-tasks, specialist agents handle them, a human reviews the combined result before anything goes out.
Advantage. Specialised agents produce better results per sub-task than a generalist agent.
Risk. As agent count grows, so does coordination complexity — a clean handoff protocol is a prerequisite for traceability.