Claude Code agent teams review: real coordination, one vendor, experimental
Last reviewed: September 2026
Verdict
Claude Code agent teams are the best-documented native team model we reviewed: a lead session, teammates that message each other directly, and a shared task list with dependencies. They only work with Claude, are disabled by default, and allow one team per session, so they rank fifth (5.7/10). BAND for desktop (8.3) covers the multi-vendor case.
5.7/10
Best for: Claude-only teams with a lead, shared task list and direct messages
- Agent-to-agent handoff (25%)
- 8.0
- Multi-vendor agents (20%)
- 0.0
- Local execution (15%)
- 9.0
- Shared visibility (15%)
- 5.0
- Human intervention (10%)
- 6.5
- Setup effort (8%)
- 5.5
- Price (7%)
- 7.0
- Multi-vendor agents: Single vendor: teammates are Claude Code instances only, which the rubric scores 0.
What is Claude Code agent teams?
Anthropic's Claude Code has two multi-agent features. Subagents are helpers inside a single session. Agent teams are several Claude Code instances working together, one of them the lead. Agent teams are "experimental and disabled by default"; you enable them with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1.
How does multi-agent work run in Claude Code agent teams?
One session is the team lead. Teammates "communicate directly with each other" and share a task list with pending, in-progress and completed states, dependencies, and file-locked claiming so two teammates do not grab the same task. Each agent has a mailbox stored as a JSON file under ~/.claude/teams/, and team and task state live under ~/.claude/. There is also cross-session messaging between sessions you run yourself. For plain parallel work, Anthropic documents git worktrees as the manual option.
Run your agents as a team
BAND for desktop lets your coding agents hand work to each other in shared rooms while you stay in the loop. Claude Code joins via BAND's plugin; Codex, GitHub Copilot and OpenCode are listed as integrations. Sessions stay on your machine (macOS or Linux).
What works in Claude Code agent teams, and what to watch out for?
Watch-outs (Anthropic documents most of these)
What works
- Actual agent-to-agent messaging, documented by the vendor.
- Shared task list with dependencies and claiming, which prevents duplicate work.
- Runs locally and costs nothing beyond your Claude plan.
Watch-outs
- Claude only. No Codex, Copilot or OpenCode teammates.
- One team per session, no nested teams, and the lead is fixed.
- No session resumption for in-process teammates, and task status can lag.
- Split panes need tmux or iTerm2; they do not work in the VS Code terminal, Windows Terminal or Ghostty.
- "Two teammates editing the same file leads to overwrites."
- Uses "significantly more tokens than a single session." Some users report idle teammates consuming tokens while they poll for messages, agents stalling silently on permission prompts, and tasks marked done with loose ends.
How much does Claude Code agent teams cost?
| Plan | Price |
|---|---|
| Pro | $17/mo annual or $20/mo monthly |
| Max | From $100/mo (5x or 20x Pro usage) |
| Team Standard | $20/seat annual or $25 monthly |
| Team Premium | $100/seat annual or $125 monthly |
| Enterprise | $20/seat/mo + usage at API rates |
API pay-as-you-go is also available.
What are the key facts about Claude Code agent teams?
| Item | Detail |
|---|---|
| Vendor | Anthropic |
| Execution | Local |
| Agents | Claude Code instances only |
| Coordination | Lead + teammates, direct messages, shared task list |
| Status | Experimental, off by default |
| Entry price | Claude Pro $20/mo ($17 annual) |
Who should use Claude Code agent teams?
Claude-only developers who want a native team model and can tolerate experimental rough edges.
Who should look elsewhere
Teams mixing vendors, or who want a board-style overview; see BAND for desktop vs Claude Code agent teams.
How does Claude Code agent teams compare?
Questions about Claude Code agent teams
How do I turn on Claude Code agent teams?
Set the environment variable CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1. The feature is experimental and off by default.
What is the difference between subagents and agent teams?
Subagents are helpers inside one session. Agent teams are separate Claude Code instances with a lead, direct messaging and a shared task list.
Can agent teams include Codex?
No. Teammates are Claude Code instances.