Some requests are simple enough for one agent to handle in a single pass. Others are really several jobs wearing one sentence: "get the quarter ready for review" touches invoicing, payments, the pipeline, and a report. For work like that, one agent splitting itself across many tasks is slow and easy to lose track of. The better pattern is several agents working together.
One agent is not always enough
Ask a person to do a big, multi-part task and they naturally break it down. Multi-agent systems do the same. Rather than one agent trying to hold an entire complex job in its head, the work is divided into focused pieces, each handled cleanly, then recombined. The result is faster, more reliable, and easier to follow.
A coordinator and its workers
The shape is simple. A coordinating agent reads your request, decides whether it is one job or many, and if many, breaks it into a list of tasks. It then spawns focused worker agents, each given exactly one job and only the tools that job needs. The workers run, often at the same time, and report back. The coordinator gathers their results and gives you one coherent answer instead of a pile of fragments.
Why split the work
Focus makes each step better. A worker handed a single, well-scoped task with a small toolkit is less likely to wander, cheaper to run, and quicker to finish than one generalist agent juggling everything. Splitting also means independent pieces happen in parallel, so a job that would be slow done in sequence finishes in a fraction of the time.
Boundaries still hold
Spawning more agents does not widen the blast radius. Every worker inherits the permissions of the person the whole request acts for, so the team of agents can collectively do no more than that person could. Usage still draws from one budget, with the work shared out across the workers, so more agents does not mean uncontrolled spend.
What you see
You do not manage the workers; you make one request and get one result. Behind it, the coordinator handles the decomposition, the spawning, the gathering, and the cleanup. The complexity stays under the hood, which is exactly where it belongs.
Coordinated agents are how a single plain-language request can run a whole corner of the operation at once. See how the platform orchestrates them, or launch a workspace and give one a job worth splitting up.