Claude Tag Puts an Always-On Claude in Your Slack, and It Already Writes 65% of Anthropic's Code
TL;DR
Claude Tag is Anthropic's attempt to make Claude a coworker instead of a chatbot. It lives in Slack, you summon it by tagging @Claude in a channel, and it breaks your request into stages and works through them with whatever tools it has been granted. It keeps a persistent memory of the channel, can read other channels for context when an admin allows it, and has an ambient mode that jumps into the conversation unprompted to flag things or chase forgotten threads. It runs on Opus 4.8 and is in beta for Claude Enterprise and Team customers as of June 23, 2026. The line buried in the announcement is the real headline: 65% of Anthropic's own product team code is now created by their internal version of this thing.
What actually changed
Anthropic already had a Slack app, plus direct messages and Claude Code in Slack. Claude Tag is the consolidation of those into something that behaves less like a command line and more like a colleague. The framing comes straight from the team that built it.
"Claude Code, Cowork, and chat are very single-player, whereas Claude Tag is built to be interactive and multiplayer. When Claude Tag works in a channel, everyone can see it, and everyone can jump in, engage, and steer it in the right direction." Cat Wu, Anthropic's head of product for Claude Code and Cowork.
That word, multiplayer, is the design thesis. One Claude identity sits in a channel, several people talk to it, and the work is visible to all of them rather than trapped in one person's private session. If you have ever watched two engineers paste the same prompt into two separate chats and get two divergent answers, you understand why this matters.
How a task runs
You tag @Claude and describe what you want in plain terms. It plans, executes stage by stage, and reports back in the thread. The interesting part is the asynchrony: it can schedule and grind work autonomously over hours or even days while you do other things, then surface the result when it is done. This is the ambient-agent pattern, not request-response.
Memory and the ambient problem
Claude Tag builds context as it follows a channel, so you stop re-explaining what your project is every morning. With permission it can pull facts from other channels too (private channels excluded), which is how it gets "up to speed" on an org without a human briefing it.
Ambient mode is the spicy bit. When enabled, Claude proactively posts on its own: status updates, cross-org flags, follow-ups on threads everyone forgot. That is genuinely useful and also exactly the behavior that makes people nervous. An agent that reads your channels and decides when to speak is a different trust contract than one that only answers when asked. Anthropic's answer is permission scoping.
The permission model is the whole ballgame
Administrators define which tools, data, and memories Claude can touch per channel, and you get separate Claude identities so the memories and access of, say, a sales Claude never bleed into an engineering Claude. This is the right call. A single omniscient bot reading every channel is a data-exfiltration incident waiting to happen; siloed identities at least bound the blast radius. If you run this in production, the audit you care about is which identity can read what, not the demo.
Why a builder should care
If you build agents, Claude Tag is a reference design for the ambient, multiplayer, memory-backed pattern that everyone has been gesturing at. The 65%-of-our-own-code stat is marketing, but it is also a signal about where Anthropic thinks the value is: not in one-shot chat, but in a persistent agent embedded where the work already happens. Note the caveats before you quote that number at your standup: it is Anthropic measuring Anthropic, on a codebase and workflow tuned for exactly this, and "created by" is doing a lot of work (humans still review and merge). It is in beta, Slack-only for now, gated to Enterprise and Team plans, and ships with introductory launch credits. Expansion to other platforms is promised but not dated.
Key Takeaways
- What it is: Claude Tag is a persistent
@Claudeteammate in Slack that plans, executes, and reports in-thread, running on Opus 4.8. - The headline number: Anthropic says 65% of its product team's code is created by its internal version. Impressive, self-reported, and not your codebase.
- Multiplayer by design: one Claude per channel, visible to the whole team, steerable by anyone, replacing single-player chat sessions.
- Ambient mode: it can speak up unprompted to flag issues and chase forgotten threads, which is powerful and the reason permissions matter.
- Security boundary: per-channel tool/data scoping plus separate Claude identities keep sales and engineering memories isolated.
- Availability: beta for Claude Enterprise and Team as of June 23, 2026, Slack-only, with launch credits and undated plans to expand.
Sources: Anthropic, "Introducing Claude Tag", TechCrunch, Fortune, VentureBeat