Multi-agent system
멀티 에이전트
A way for several AI agents to split roles and work together on one job.
Overview
If you give one agent a big job, it can lose its way or forget what came earlier. So one agent hands out the work, and several sub-agents sit under it, each taking one part such as research, writing or review.
Work can happen at the same time, so it gets faster, and agents can check each other's results to reduce mistakes. The cost is that every extra agent uses more tokens, so it costs more, and if the hand-offs go wrong, the result can be off target.
An everyday example
It is like moving house by dividing the work among a packer, a carrier and an organizer, with one person directing everyone, instead of doing it all alone.
Why it matters
It is a way to give AI jobs that are too heavy for one agent, such as long research or a big programming task. News that "several agents worked together" is about this approach.
Also calledmulti-agent