Agents & tools
Multi-Agent System
An AI architecture where multiple agents — often with different roles, models, or tools — collaborate on a task one agent could not handle alone.
01 ——
In plain English
A multi-agent system (MAS) is the broader engineering pattern of breaking a task across several specialised agents that coordinate. It's the umbrella that includes agent swarms, role-based teams, and assembly-line workflows.
Common patterns:
- Pipeline — agents run in sequence (researcher → writer → editor → publisher)
- Manager-worker — a coordinator agent delegates to workers and stitches results
- Debate / vote — multiple agents propose answers; a judge picks or merges
- Specialist team — domain-specific agents (legal, finance, design) collaborate on cross-functional work
Why split a task across agents:
- Smaller, sharper prompts — each agent gets a focused role and toolset
- Parallel execution — independent steps run at once
- Modularity — swap one agent without rewriting the system
- Quality — a critic agent catches what a single producer misses
Frameworks: LangGraph, CrewAI, AutoGen, OpenAI Swarm, Microsoft's Magentic, and Anthropic's Claude Agent SDK all support multi-agent patterns. Production teams usually start with one agent and add more only when complexity demands it.
02 ——
Related terms
Agent Swarm
A group of AI agents that work together — often with different roles — to solve a problem one agent could not handle alone.
Agent Orchestration
The coordination layer that decides which agent or tool runs next, manages state across steps, and handles failures in multi-step AI workflows.
AI Agent
Software that uses an LLM to plan and act — picking tools, taking actions, and adapting based on results to complete a user’s goal.
Agentic AI
AI systems designed to act, not just respond — they plan, use tools, and make decisions across multiple steps to complete a goal.
Autonomous Agent
An AI agent that operates with minimal human oversight — making and executing decisions independently across long time horizons.
Agent Harness
The scaffolding around an LLM — tools, memory, loops, and orchestration — that turns a model into an agent.
Back to glossaryLast reviewed May 2026