Orchestrate teams of Claude Code sessions
code.claude.com
February 5, 2026
15 min read
🔥🔥🔥🔥🔥
65/100
Summary
Agent teams in Claude Code enable coordination among multiple instances, with one session acting as the team lead to assign tasks and synthesize results. To enable this feature, add CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS to your settings.json or environment, noting that there are limitations regarding session resumption, task coordination, and shutdown behavior.
Key Takeaways
- Agent teams in Claude Code allow multiple instances to work together, with one session acting as the team lead for coordination and task assignment.
- Agent teams are most effective for tasks requiring parallel exploration, such as research, debugging, and cross-layer coordination.
- Enabling agent teams requires setting the CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS environment variable to 1 in settings.json or the shell environment.
- Agent teams use significantly more tokens than single sessions and are best for complex tasks that require discussion and collaboration among teammates.
Community Sentiment
Positives
- The concept of orchestrating teams of agents like Claude Code could revolutionize AI workflows, akin to 'Kubernetes for agents', enhancing collaboration and efficiency in complex tasks.
- There is a growing interest in integrating multiple LLMs for orchestration, which could lead to more versatile and effective AI applications in software development.
Concerns
- There are significant trust issues with Claude Code handling large tasks independently, as maintaining code quality requires more human oversight and careful design.
- The cost of running these AI agents continuously is a major concern, suggesting that only large companies might afford such advanced tools, limiting accessibility for smaller teams.
