A-TIER technique
Claude Code agent teams — parallel sessions that message each other
100k ft — What is this?
Claude Code agent teams — parallel sessions that message each other
50k ft — Why care?
Claude Code's agent teams let separate Claude sessions talk laterally through a shared mailbox, unlike subagents which only report back a single result at the end. It ships disabled; you enable it with an experimental environment variable set to 1 in .claude/settings.json (paid plans only).
10k ft — Key takeaways
• Agent teams allow agent-to-agent messaging mid-task; subagents cannot coordinate
• Enable via the experimental agent-teams env var in settings.json
• Use for parallel work where agents must share findings, not just fan-out/fan-in
5k ft — Tools & techniques
Claude, Claude Code
AI agents, agent teams, multi-agent orchestration, subagents
1k ft — Deep-dive / try it
that new 1M context window too 😩🤌🏽
#claude #ai
Transcript excerpt
Claude just shipped this today. This is an agent team. No, these are not sub agents. It's a similar concept since these are also individual Claude sessions, but in an agent team, each agent is allowed to communicate with each other. Any agent is able to just message another agent through a shared mailbox, which allows lateral communication while you're doing parallel work. This is a game changer. Parallel sub agents do not work like this. When Claude starts many sub agents, they're not able to communicate with each other. You get one cumulative analysis at the end, but there's no communication happening at all between sub agents. Agent teams completely fix this, but it's disabled by default. So you have to set this environment variable to be equal to one in the settings.json file of your dot Claude folder. Unfortunately, this is also only available for paid users, but I have one more guest pass. Most like comment if you want it, you got it.