What are multi-agent systems?
Several agents can cooperate, compete or handle different parts of a task.
What is it?
In a multi-agent system, several AI programs cooperate, compete or split up a task: one gathers information, another writes, a third checks. The division of labour can increase capability, but it also complicates oversight and can multiply errors.
An everyday example
For a report, one agent collects sources, another writes the draft and a third checks the figures before it reaches a person.
Why does it matter?
Multi-agent systems are increasingly common in programming and research. When several agents work together, it is harder to tell who did what and why.
A common misconception
That more agents automatically mean a better result. Agents can confirm each other’s mistakes.
In more detail
Why use several agents?
To split up work, to check each other or to coordinate several tasks.
Do they correct each other?
Sometimes. But they can share the same mistakes or reinforce each other in a wrong conclusion.
Is that the same as “collective intelligence”?
It is a description of some systems, not a guarantee of a better result.
What should be measured?
Task success, costs, safety, traceability and the benefit compared with a simpler system.
Remember: More agents are not automatically better than one.
Reliable sources
Last reviewed 1 November 2026. An explainer, not professional, legal or medical advice. For rules and figures that change, check the official version.