Agent Teams & Multi-Agent
!Experimental Feature
Agent Teams is an experimental feature in Claude Code. It may fundamentally change or be removed at any time. Activation requires the feature flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1.
In the previous module, you learned about subagents: specialized assistants that handle a single task and return the result to the main agent. This works excellently as long as tasks run sequentially or in short, isolated bursts.
But what happens when you want to implement three features in parallel? Or when one agent writes code while another simultaneously creates the tests for it? Subagents aren't enough for that -- you need a real team.
Subagents vs. Agent Teams
The difference can be illustrated with a simple analogy:
- Subagent: You send an intern to research something. They come back with the result, and you continue working alone.
- Agent Team: You have three experienced developers working simultaneously on different parts of the same project -- each at their own workstation, but all on the same codebase.
| Subagents | Agent Teams | |
|---|---|---|
| Relationship | Parent-child (delegation) | Peers (collaboration) |
| Lifespan | Temporary, for one task | Long-lived, for the entire session |
| Communication | Result back to main agent | Direct messaging between agents |
| Parallelism | Limited, mostly sequential | True parallel execution |
| Coordination | None needed | Shared Task List, Messaging |
When Do You Need Agent Teams?
Agent Teams are worthwhile in scenarios where you have multiple independent but related tasks:
- Parallel feature development: Implementing three features simultaneously, each in its own Git worktree
- Concurrent testing: One agent implements while another writes the tests in parallel
- Code review + fix: One agent finds problems, another fixes them immediately
- Large refactorings: Restructuring different modules simultaneously
*Learning Objective
After this module, you'll understand how Agent Teams are coordinated, when they make sense, and how to make the right decision between subagents and Agent Teams.
For individual, isolated tasks, subagents remain the better choice. Agent Teams shine when true parallelism and communication between agents is needed -- and when you're willing to bear the corresponding costs.