The following article covers Claude Code Agent Teams and how you can set up this new feature from Anthropic in only ten minutes.
How Claude Code Agent Teams Actually Work
Most people who use Claude Code are still running single agents or subagents. Subagents run in parallel but they don’t talk to each other. Agent teams are different. Each agent has a role, they share information mid-task, and the output reflects that coordination.
That distinction matters more than it sounds. When agents can’t communicate, you get duplicate angles, wasted tokens, and outputs that don’t build on each other. The team model fixes that.
According to Anthropic’s official Claude Code documentation, agent teams use a shared .claude/agents folder. Each agent is defined by a Markdown file with its role, tools, and instructions. Claude Code reads those files and coordinates from there.
Setting Up Agent Teams (Step Zero)
If agent teams aren’t configured yet, that’s the first thing to fix.
Open Claude Code (VS Code extension or terminal both work) and ask: “Do you know how to spin up agent teams and do I have them set up correctly?” Claude will check for a .claude/agents folder. If it doesn’t exist, it’ll tell you.
From there, go to the official agent teams setup page, copy the full contents, paste it into your Claude Code chat, and tell it to set everything up. It takes about 60 seconds.
I tested this on a fresh VS Code instance. No folder existed. Claude read the docs, created the folder structure, and confirmed setup was complete. Done.
A Real-World Agent Team Use Case
Once the setup is done, here’s where it gets interesting.
I built a content repurposing team with three agents: one for LinkedIn, one for X (Twitter), and one for email newsletters. Each agent had a specific job tied to a corresponding skill file I had already written. The LinkedIn agent used my LinkedIn post writer skill. The X agent used my social media skill. The email agent used my newsletter skill.
My Claude Code Skills Stack has 20+ skills I use daily to automate marketing and content creation with a one time fee and lifetime updates.
Here’s the key instruction that made it work as a team rather than parallel subagents:
“Before writing, each agent should read the full transcript and identify the three most compelling insights. Have them share chosen insights with each other so no two platforms lead with the same angle.”
That one instruction is what separates agent teams from subagents. The agents actually negotiated angles before writing. The result was three posts with distinct hooks pulling from the same source material.
Credit to Mark Kashef here. He’s one of the best at explaining Claude Code mechanics in practice. I pulled from one of his prompt structures and adapted it for this workflow.
What the Output Looked Like
The LinkedIn post opened with: “GPTs and projects are already outdated. Skills are what replace them.” The X post went a completely different direction: “Most people using AI are making the same mistake every single day.” Neither hook overlapped.
That’s the point. Three agents, one source, three unique angles. No duplication.
I also added a synthesis step: after all agents finished, Claude summarized what each platform covered and flagged inconsistencies. That saved a full manual review pass.
The Cost and Speed Reality
Here’s what I won’t sugarcoat.
This took longer than I expected. Running three coordinated agents on a full transcript is not a fast process. Plan for it. If you’re expecting a 30-second turnaround, you’ll be disappointed.
Token usage was also higher than anticipated. I’m on the Claude Max plan at $100 per month. For that use case, it was fine. If you’re running agent teams on API keys, costs add up fast. Run a test with a short transcript before you commit to high-volume workflows.
You can also check in mid-task. Type “How much longer?” and Claude messages the active agents, then reports back. During my test it returned: “All three agents are actively working. None have saved their files yet.” That real-time feedback tells you whether the process stalled or is still running.
Taking the Output Further with Blotato
After the agents finished, I connected Blotato to post directly from Claude Code.
Blotato is the social media tool I use to schedule and automate posts. The owner recently released an MCP setup for Claude Code. That means agent-generated content can go live without leaving your workspace.
I told Claude to post the X output using the Blotato MCP. It found my account, confirmed the post, and returned a live link. About 10 seconds total. No copy-paste, no tab-switching.
Ryan’s Final Thoughts
Claude Code agent teams are genuinely useful when you want platform-specific output from a single source. The coordination between agents is what makes them worth using over standard subagents.
Set realistic expectations on time and token cost before scaling. Test on short transcripts first. Once output quality checks out, build it into a repeatable workflow. If you want to keep that token spend under control, these Claude Code hacks from a top-tier user cover the planning and model-switching habits that make heavier runs affordable.
Claude Code Agent Teams FAQs
What is the difference between Claude Code agent teams and subagents?
Subagents run in parallel and do not communicate during execution. Agent teams are coordinated agents that share information mid-task. In a content workflow, that means agents can compare angles before writing, preventing duplication. Subagents are faster for independent tasks. Coordinated teams pay off when output quality depends on what other agents are doing.
How do you set up agent teams in Claude Code?
Open Claude Code, paste the Anthropic agent teams setup page into your chat, and tell Claude to configure it. Claude creates the .claude/agents folder and confirms setup. Under two minutes on a fresh install.
Are Claude Code agent teams expensive to run?
More than a single agent, yes. Each agent uses tokens independently and inter-agent communication adds overhead. On Claude Max, a three-agent workflow is manageable. On API keys, costs add up fast. Test with short inputs first.
Can I check on agents while they are running?
Yes. Type a status check mid-task and Claude messages the active agents for an update. The response shows which agents are still working and whether files have been saved. Useful for longer tasks where you want confirmation nothing stalled.
What is the best use case for Claude Code agent teams?
Content repurposing is one of the clearest fits. Feed a transcript to platform-specific agents, have them coordinate on angles, and get publish-ready output for LinkedIn, X, and email in one pass. The coordination layer is what makes the added token cost worth it.
