Multi-agent system: A multi-agent system is an AI setup in which several agents, each with its own role, instructions, and tools, work together on a task, usually with one agent coordinating the others. It splits complex jobs into pieces the way a team would.
A single AI agent is a model in a loop: it plans, calls tools, checks results, and repeats until the job is done. A multi-agent system runs several of those loops and lets them hand work to each other. A common shape is an orchestrator that breaks a request into subtasks and assigns them to specialists: one searches, one drafts, one checks the draft against the sources, one formats the result.
There are practical reasons to split the work. Each agent gets focused instructions and only the tools it needs, which improves reliability and limits the damage if one goes wrong. Subtasks can run in parallel. Each agent keeps its own context, so a long research job does not clog a single window with every page it read. Frameworks such as LangGraph, CrewAI, Microsoft's AutoGen, and the OpenAI Agents SDK exist to wire these systems together, and several coding assistants now spawn subagents on their own.
The costs are real too. Every extra agent multiplies token spend and latency. Errors compound: a researcher agent that misreads a source feeds a confident wrong fact to the writer. Handoffs are where things get lost, and debugging a conversation among five agents is harder than debugging one. Many products marketed as multi-agent are one model with several prompts, which is fine, but it is not magic. Start with a single agent, split only for a clear reason, log everything, and keep a human approval step before anything irreversible.
Example at work
A consulting analyst sets up a due-diligence workflow: one agent pulls a target company's filings and news, a second extracts financial figures into a table, a third writes a one-page summary, and a fourth compares every number in the summary to the table and flags mismatches. She reviews the flags before the memo goes to the partner.
Why it matters
Multi-agent systems are how AI moves from answering questions to completing multi-step work. Knowing how they are built helps you judge vendor claims and design workflows where the checks are as strong as the automation.
Related terms
- AI agentAn AI agent is a system that uses an AI model to pursue a goal on its own by planning steps, using tools such as web browsers, files, and software APIs, and adjusting based on results, rather than answering a single prompt.
- Agentic AIAgentic AI describes AI systems designed to act with autonomy: they take a goal, break it into steps, use tools, and carry the work through with limited human direction. It is the broader approach; an AI agent is a specific system built that way.
- Tool useTool use (also called function calling) is the ability of an AI model to request that an outside tool be run, such as a web search, a calculator, a database query, or a calendar lookup, and then use the result in its answer.
- Human-in-the-loopHuman-in-the-loop is a design approach in which a person reviews, approves, or corrects an AI system's output at defined points before it takes effect, combining the speed of automation with human judgment and accountability.
- Workflow automationWorkflow automation connects apps and steps so that a trigger, such as a new form response or an incoming email, automatically runs a sequence of actions, increasingly with an AI step that reads, decides, or writes along the way.