Claude Code Agent Teams: Teammates That Message

Two of the most talked-about launches in agentic AI this month share the exact same name: Agent Teams. MiniMax shipped a desktop version where multiple agents collaborate in parallel, and Claude Code added an experimental Agent Teams mode of its own. The reason everyone is suddenly excited is simple: a single agent, no matter how strong the model, hits a wall on work that branches in several directions at once. A code review, a multi-layer feature, a bug with three competing explanations — these don't get faster by thinking harder in one context window. They get faster when independent workers explore at the same time and then compare notes.

Claude Code's Agent Teams is the first time the official CLI gives you that out of the box. One session becomes the lead. It spawns teammates, each a full Claude Code instance with its own context window. They share a task list, claim work off it on their own, and — the part that makes this different from subagents — they message each other directly instead of funneling everything back through one orchestrator.