Multi-Agent System

A system of multiple independent AI agents that interact and coordinate to solve tasks or make decisions within a shared environment.

Also known as: MAS, agent-based system

In plain English

A multi-agent system is like a team of specialised helpers (think apps or workers) that each do part of a job and pass results between them. By sharing information and dividing work, the team can handle more complex tasks than a single helper could on its own.

In practice

You encounter multi-agent systems when products chain AI components (retrieval, summarisation, verification) or run specialised bots together. They can boost capability and reliability but increase integration, monitoring, and security overhead when adopted at work.

Under the hood

A multi-agent system consists of multiple autonomous agents, each with its own state, policy and action space, interacting in a shared environment through messaging or shared resources. Coordination approaches include centralised orchestration, peer-to-peer negotiation, market mechanisms and multi-agent reinforcement learning; key engineering issues are non-stationarity, credit assignment, partial observability and ensuring robust, explainable coordination.

Example

"Can we prototype a multi-agent system where one agent retrieves documents, another summarises them, and a third verifies factual accuracy?"

Related terms

Agentic AI · AI Agent · Agent Orchestration · Reinforcement Learning · Human-in-the-loop

•