Multi-agent workflows assign distinct roles to separate agents, such as one retrieving documents, another extracting facts, and a third verifying citations. Coordination may be sequential, parallel, or managed by a supervising agent.
Each agent operates with its own instructions and tool access.
Alternative Names:
Agent Orchestration, Multi-Agent System
Why it Matters?
Separation of roles allows a verification agent to check another's output against sources, which is a structural improvement over a single system checking itself. The corresponding risk is error propagation, since a mistake in an early step becomes an input the later steps treat as established. Workflows handling legal work should surface the intermediate steps rather than only the final output, so a reviewer can locate where a wrong conclusion originated.
Frequently Confused with
Related terms
Frequently asked questions
What is the advantage of separating agents?
What is the main risk?
