Anthropic introduces Claude Tag: an AI teammate that lives in Slack

On June 23, Anthropic announced Claude Tag: a persistent AI agent that lives inside Slack. The old "Claude for Slack" integration was a question-and-answer box; this one is positioned as a standing team member that sits in a channel, accumulates context, and works through tasks on its own.

What was announced

You summon Claude Tag in a Slack channel with `@Claude`. It runs as a shared identity that belongs to the whole channel rather than a single user — everyone sees what it does, and anyone can pick up where it left off. It breaks a task into stages, executes step by step with the available tools, and delivers the result back to the channel.

How it differs

Three things stand out. First, persistent memory: it learns continuously from the channel's conversations and, with permission, can look across other channels. Second, ambient mode: without being asked, it can proactively flag organizational updates and stuck points. Third, scoped access: admins control which tools, information, and channels each Claude instance can reach — an attempt to prevent cross-team data leakage.

Who can use it

For now it's a research preview / beta open only to Claude Enterprise and Claude Team customers. Anthropic says it aims to bring the capability to other platforms in the coming weeks. It sits in the same "agentic" family as Claude Code and Cowork.

What it means for a solo builder

I work alone, so the "a channel full of people plus a shared Claude" scenario isn't my daily flow directly. But the direction is clear: Anthropic is moving the agent from a single assistant into an actor that lives inside a team and works asynchronously. This is a continuation of the same line as the dynamic workflows in Claude Code: the agent is no longer something that answers one command but something that owns a process. The same "agent reliability" conversation was on the table around GPT-5.6 — the whole industry is shifting from single answers toward agents that take steps and follow through.

Singrey's Note

I haven't gotten my hands on it yet — the beta is Team/Enterprise only and I'm on the solo side. So I'm writing this as a "first impression," not a hands-on test. The part that makes me think most is "ambient mode": an agent that acts without being asked and watches a channel is a powerful idea, but it makes taking access control seriously a requirement. The takeaway for a solo builder: even without a team tool, I can mimic the approach of "placing the agent permanently inside a conversation flow" in my own tooling. I'll keep watching and try it when it opens up to me.