Multi-Agent System

Definition

A Multi-Agent System is an architecture where multiple specialized AI agents collaborate on a shared task, each handling a subset of the work according to its specialization or the type of information it has access to. Rather than one generalist agent trying to do everything, a multi-agent system decomposes complex tasks across a network of focused agents—a research agent, a writing agent, a validation agent—coordinated by an orchestrating layer. This mirrors how human teams organize around specializations to tackle complex problems.

Engineering Context

Multi-agent systems enable parallelization and specialization but introduce coordination complexity. Key design decisions include topology (orchestrator-worker vs. peer-to-peer), shared state management (how agents share information), failure isolation between agents (one agent's failure should not cascade), and cost control (LLM costs multiply with each additional agent). Use multi-agent architecture only when a single agent genuinely cannot handle the task complexity—premature multi-agent decomposition adds overhead without benefit. In LangGraph, multi-agent systems are implemented as a parent graph that invokes subgraphs, with each subgraph representing an agent with its own state and tools.

Related Terms