A colleague arranges several Claude Code sessions as an agent team: one lead, a few teammates, and a task list shared between them. What does that arrangement add, compared with a set of workers that each hand their findings back to the session that called them?

The principle — a team is not a bigger subagent; what changes is who talks to whom.

Why the key is correct — three documented sentences fit together. Agent teams let you coordinate multiple Claude Code instances working together. One session acts as the team lead, coordinating work, assigning tasks, and synthesizing results. And teammates work independently, each in its own context window, and communicate directly with each other. The glossary names the two channels: a shared task list and peer-to-peer messaging.

Why the others are wrong — a relabelling would leave the topology alone, and that is what moves. A lead that relays every finding is not what is described: it assigns and synthesises while teammates talk between themselves. Capability is not what separates the surfaces.

Remember this — a team buys peers with a shared list, not better workers.

Sources — Anthropic's agent teams documentation and the glossary.