EN / RU
← Back to essays
· Essay · 1 min

Agent Teams in Claude Code - Swarms of Agents

Now, instead of launching a single agent, you can deploy an entire team. One lead agent coordinates the work, while the others operate in parallel.

🤖🤖🤖 Agent Teams have appeared in Claude Code - swarms of agents.

Now, instead of launching a single agent, you can deploy an entire team. One lead agent coordinates the work, while the others operate in parallel and communicate directly with each other.

When this is useful:
- Research: multiple agents explore different aspects of a problem simultaneously
- Debugging: each agent tests its hypothesis in parallel
- New features: each agent owns its module
- Code review: one looks at security, another at performance, and a third at tests

The difference from subagents: there, workers only return results to the main agent. In Teams, agents communicate with each other, argue, and challenge each other's conclusions.

This functionality was available before but hidden - enthusiasts found it. Now it's officially available for everyone.

Currently experimental and consumes a lot of tokens. It can be enabled through settings.json.

🔗 code.claude.com/docs/en/agent-teams