loading
loading
A system where several specialized agents work together — splitting, delegating, or debating a task — instead of one agent doing everything.
In a multi-agent setup you decompose a problem across agents with distinct roles and prompts: a planner hands subtasks to workers, or a researcher and a critic check each other. The win is focus and parallelism — each agent has a tighter context and clearer job — but you pay in coordination overhead, latency, and more places for things to go wrong. Reach for it when a single agent's context gets overloaded or the work genuinely splits into independent pieces, not just because it sounds sophisticated.
Plainly
Think of Multi-agent as a simple recipe for doing the work better. A system where several specialized agents work together — splitting, delegating, or debating a task — instead of one agent doing everything.
In practice
Use it when you need a repeatable method instead of guessing from vibes. In practice, define the owner, input, output, and failure mode before you rely on it.