Back to BlogAgentic Coding
Episode 2. Agent Teams Complete Guide - Working with Multiple Claudes
5.52min
Claude CodeAgent TeamsMulti-AgentAI CodingDevelopment Tools
Introducing Claude Code's Agent Teams feature. We'll explain how multiple Claudes collaborate to code smarter, in a way that's easy for junior developers to understand.

SeriesEP 2 / 4
What are Agent Teams?
Hello, fellow developers!
Until now, we've worked with a single Claude. But think about it: working alone in a development team is very different from working with multiple people, right?
Agent Teams started from this idea. Multiple Claude Code instances work together as a team to work smarter.
Loading diagram...
Let me explain step by step why this is beneficial.
Why Do We Need Agent Teams?
Limitations of Single Agent
Trying to do everything with one Claude had some issues:
- Lack of Specialization: One person handling backend, frontend, security, and testing lacks depth
- No Parallel Work: Can only do one thing at a time, so it takes longer
- Perspective Limitation: One person's thinking can be biased
Advantages of Agent Teams
When multiple Claudes work as a team, you get these benefits:
- Specialization: Each handles their area of expertise more deeply
- Parallel Processing: Can work on multiple tasks simultaneously, saving time
- Diverse Perspectives: Different thoughts lead to better solutions
Agent Teams Architecture
Team Lead
The Team Lead coordinates the entire workflow. Think of them as a project manager.
Roles:
- Break down work into smaller units
- Assign appropriate tasks to each Teammate
- Monitor overall progress
- Integrate results into final deliverables
Teammate
Teammates perform the actual work. Each has their area of expertise.
Roles:
- Perform assigned tasks
- Report progress to Team Lead
- Collaborate with other Teammates
Two Execution Modes
In-Process Mode
Multiple agents run within a single Claude Code process.
Advantages:
- Simple setup
- Efficient memory usage
- Quick to start
Disadvantages:
- Processes one task at a time (not truly parallel)
- Can be slow for complex tasks
When to Use:
- Simple projects
- When you need quick results
- Limited resources
Split-Pane Mode
Run multiple Claude Code instances simultaneously using terminal multiplexers like tmux or iTerm2.
Advantages:
- True parallel processing
- Each agent operates independently
- Suitable for complex projects
Disadvantages:
- More complex setup
- Requires terminal configuration
- Uses more resources
When to Use:
- Large-scale projects
- When true parallel processing is needed
- Sufficient resources
Setting Up Agent Teams
In-Process Mode Setup
The simplest method. Add this to
.claude/settings.json:JSON
{"agentTeams": {"enabled": true,"mode": "in-process","teamSize": 3}}
Now Agent Teams will be automatically enabled when you run Claude Code.
Split-Pane Mode Setup
Here's an example using tmux:
Bash
# Start new tmux sessiontmux new-session -d -s claude-team# Create Team Lead windowtmux rename-window "lead"tmux send-keys 'claude' C-m# Create Teammate 1 windowtmux new-window -t claude-team -n "mate1"tmux send-keys 'claude' C-m# Create Teammate 2 windowtmux new-window -t claude-team -n "mate2"tmux send-keys 'claude' C-m# Attach to sessiontmux attach-session -t claude-team
If you use iTerm2:
Cmd + Shift + Dfor horizontal split- Run
claudein each split - Use first as Team Lead, rest as Teammates
Usage Tips for Junior Developers
1. Parallel Code Review
Agent Teams are really handy when reviewing large PRs.
Example:
- Teammate 1: Backend code review
- Teammate 2: Frontend code review
- Teammate 3: Test code review
- Team Lead: Overall integration and final judgment
2. Exploring Diverse Perspectives
When solving complex problems, you can approach from different angles.
Example:
- Teammate 1: Analyze from performance perspective
- Teammate 2: Analyze from security perspective
- Teammate 3: Analyze from maintainability perspective
- Team Lead: Synthesize all perspectives for optimal solution
3. Learning and Exploration
When learning new technologies, you can learn from multiple aspects simultaneously.
Example:
- Teammate 1: Read official documentation
- Teammate 2: Analyze example code
- Teammate 3: Research best practices
- Team Lead: Summarize and organize
Real Usage Example
Bash
# Start task from Team Lead/Users/me/project $ claude> I need to implement this REST API. Need /users, /posts, /comments endpoints.# Team Lead automatically distributes work→ Assign /users endpoint to Teammate 1→ Assign /posts endpoint to Teammate 2→ Assign /comments endpoint to Teammate 3# Each Teammate works in parallel→ Teammate 1: User schema design and implementation→ Teammate 2: Post schema design and implementation→ Teammate 3: Comment schema design and implementation# Team Lead integrates results→ Full API testing→ Documentation generation→ Final code review
Tips for Efficient Agent Teams Usage
1. Appropriate Team Size
Too many Teammates reduces efficiency. In my experience:
- Small projects: 2-3 people (Lead + 1-2)
- Medium projects: 3-4 people (Lead + 2-3)
- Large projects: 4-5 people (Lead + 3-4)
2. Clear Role Division
Assign clear responsibility areas to each Teammate.
Bash
# Good exampleTeammate 1: Backend API developmentTeammate 2: Frontend UI developmentTeammate 3: Testing and documentation# Bad exampleAll Teammates: "Do whatever"
3. Regular Sync
Have Team Lead regularly check each Teammate's progress.
Bash
> Share your current progress.
Learn by Video
Anthropic has created official videos. Watching them will help you understand faster.
Wrapping Up
Agent Teams is a really powerful feature. When multiple Claudes work together:
- Faster: Parallel processing reduces time
- Smarter: Specialization improves quality
- More Diverse: Solve problems from multiple perspectives
As a junior developer, accessing such tools makes me envious, but also makes me think "How can I use this effectively?"
In the next episode, we'll cover a practical guide to using Opus 4.6 and Agent Teams together. We'll learn how to apply to real projects!
We'll continue with Episode 3: Practical Agentic Coding for Junior Developers.