Agents
Multi-agent system
An AI architecture in which several specialised agents collaborate, delegate, or compete to solve a task more reliably than a single monolithic model could.
A multi-agent system (MAS) is an architecture in which two or more autonomous AI agents interact to solve problems that are too large, too varied, or too risky for a single agent. The term predates modern AI — it has roots in distributed-systems and game-theory research from the 1980s — but it has been revived by the rise of large language models capable of acting as general-purpose reasoners.
In a typical LLM-based MAS, a coordinator agent decomposes the task and assigns subtasks to specialised workers — for example a planner, a researcher, a coder, a tester, and a critic. The agents communicate through structured messages, share intermediate state, and may use external tools through protocols like MCP. Frameworks such as AutoGen, CrewAI, LangGraph, and the Anthropic Agent SDK make this pattern reusable.
Multi-agent designs are popular for code generation, research and writing pipelines, data analysis, and customer-support automation. They can outperform a single-agent baseline by enforcing role separation, enabling self-critique, and parallelising independent steps — but they also introduce new failure modes: cascading errors, runaway loops, and coordination overhead.