Claude Code now has Agent Teams Multiple Claude sessions working in parallel, coordinating through shared tasks and messaging each other. Enable it by adding this to your settings.json: { "env": { "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS":"1" } } Just describe the team you want in plain English Shift+↑ to check each teammate's progress... This + worktrees = workflow multipliershow more

Whoever want to directly get this setup running:

I want a dashboard to manage those teams

been running 6 agents in parallel for months. coordination is the missing piece. single-agent coding feels like dialup now.

With one supervised agent I already see: elaborate fixes to non-existent bugs, unnecessary backward compatibility, solutions built on false premises, etc etc Now multiply that by 10 unsupervised agents building on each other’s output. It’s telephone game but with code. Don’t go too far too quick little boi

This is the shape of things actually working—parallel specialists with message passing rather than one oversized context window. The interesting question becomes governance: how do you prevent oscillations or deadlocks when these agents have conflicting subgoals? I dig into this via Ising model consensus in my Epistemic Homeostasis work—curious if Anthropic is handling coordination procedurally or learned?

WTF

Daniel, que fachero que te quedo Ghostty, como lo tenes configurado? (No lo pude probar todavia porque uso windows 😭)

What’s this setup you got for the UI ?

I've been there, trying to keep track of multiple developers working on the same project. It's like herding cats, and coordinating everyone can be a real challenge. Having Agent Teams in Claude Code sounds like a game-changer for my team's workflow.

終於不用手動 spawn 子 session 了 這個功能對於複雜 codebase 的重構應該很有用 —— 一個 agent 寫 tests，一個寫 implementation，第三個負責 review

sounds expensive

agent teams, nice. hope it doesnt tank my context window tho.

Is this reliant on a specific Claude Code version? I'm stuck on 2.0.9 until my company fixes a(low priority) proxy issue

Amazing work! This is soo next level!

When can the CC accessed by GLM be used?

@Claude_Memory

it couldn’t communicate right with the team, THEY HAVE SUCCESSFULLY EMBODIED A DEVELOPER

Performance ? How does it feel vs regular plan mode w opus 4.5 spawning sub agents ? Does this do its own work trees ??

Agent teams that message and work in parallel are the right idea. They still fail when context and ownership stay private. Vokal gives those teams durable shared state and live channels so every agent and every human works from the same source of truth. No more silos. No more re-explaining.

Agent teams. Nice. What's your kill switch when one starts hallucinating tasks?

Dude what the fuck. My Claude Code still has Opus 4.6 even though I updated it?

Exactly what I felt "multiple sessions" doesn't feel Orchestration

Agent Teams in Claude Code allows parallel sessions to coordinate via shared tasks and messages. Efficient for complex workflows.

Using Claude Code agents as subagents inside OpenClaw right now. The real unlock is letting them spawn each other for parallel tasks.

@dani_avila7, San, Agent Teams in Claude Code can really improve complex task management.

This is EXACTLY what OpenClaw has been doing with sessions_spawn — multiple agents working in parallel, each with their own context. The difference: OpenClaw agents can also message each other, access external tools, and persist across restarts. We're entering the "agent swarm" era 🐝

great breakdown. saved this for later