Anthropic shipped Claude Tag on June 23, putting @Claude into Slack as a persistent, memory-building team member for Enterprise and Team customers. The feature eliminates the re-entry overhead that makes most AI tools impractical at the team level, and Anthropic's own product team now attributes 65% of its code to the internal version.
The brief that used to take 45 minutes to write because nobody remembered where last quarter's campaign doc lived now starts with a tag. Anthropic launched Claude Tag on June 23, and it targets the most expensive part of working with AI inside a team: the re-entry cost. Every time someone opens a chat to delegate something to Claude, they spend the first 10 to 15 minutes re-explaining who they are, what project they are in, what the brand sounds like, and what has already been tried. That is not AI friction. It is organizational friction, and Claude Tag is designed specifically to remove it.
The product puts @Claude into Slack as a persistent, channel-resident teammate. Tag it with a request and it goes to work while you do something else. When it is done, it replies in thread. Anyone on the team can read what it did, add to it, or tag Claude with the next step. This is not a chatbot in a sidebar. It is a shared team member that retains what it learns from the channel over time.
The stat Anthropic dropped in the announcement is worth sitting with: 65% of their own product team's code is now created by their internal version of Claude Tag. That number is not a benchmark. It is an operational report from a company that runs on the thing it builds. Whether or not 65% translates to your team, the direction it points is clear. The AI is no longer an external tool someone visits. It is embedded in the workflow.
What actually changed
Claude Tag introduces something earlier AI integrations could not deliver: persistent channel memory. The model builds context over time from the conversations it follows, which means the third time someone in your marketing channel asks Claude to draft a brief using your brand voice, it already knows the brand voice. You do not explain it again.
The multiplayer piece matters just as much. In standard AI chat, each person has their own session and their own context. If two people on the same team each ask Claude about the same campaign, they get two disconnected answers from a model that does not know the other conversation happened. Claude Tag has one identity per channel. Everyone works with the same instance, the same memory, and the same context. The team's interaction with Claude becomes part of the team's shared record.
The ambient mode adds a layer that most current AI deployments completely lack. If enabled, @Claude will proactively surface relevant updates and flag threads or tasks that went quiet without resolution. For a RevOps team tracking deal blockers or a marketing team running a multi-channel campaign, that is not a nice-to-have. It is what an operations coordinator does. In async, at scale.
Why this hits business and marketing teams specifically
Agency teams and RevOps functions carry a particular version of this overhead. The context they manage is relational and situational, not just data. Campaign history, client preferences, account relationships, approval chains. None of it lives cleanly in a CRM field. Most of it lives in Slack. Claude Tag learns from the channel where that context actually exists, rather than asking you to move it somewhere cleaner first.
The admin model is designed for exactly this kind of cross-functional trust concern. System administrators scope each @Claude instance to specific channels and specific tools. The Claude in the marketing channel does not have access to the engineering channel's memory, and cannot pull sales data unless sales explicitly granted it. The separation is enforced at the identity level, not at the honor system.
Anthropic is calling this "the beginning of an evolution of Claude Code," which is the right framing. Claude Code made individual developers significantly more productive by keeping the model close to the work. Claude Tag is the same move, one abstraction layer up, applied to teams.
The honest caveat
This is a beta, available today for Claude Enterprise and Team customers only. The pricing implications of ambient Claude behavior across dozens of channels are real and not fully spelled out yet. Anthropic mentioned admin-controlled token spend limits per channel and per organization, but the shape of those limits at scale will take a few months to become clear. The async, initiative-taking Claude that runs projects over hours or days is powerful, and the credit consumption that comes with it needs monitoring. Channel-level spend controls are part of the admin setup. Use them from day one, not after the first bill.
There is also the memory question. A model that follows your Slack channels and retains what it learns is, by design, watching a lot. Data stays scoped to the channel and does not get shared across Claude instances or extracted outside the workspace. But the comfort level with persistent ambient memory varies significantly by team, industry, and what actually moves through those channels. Worth a deliberate conversation with your team before onboarding, not after.
Anthropic is issuing launch credits to eligible Enterprise and Team organizations to let the whole company try it. That is a sensible way to stress-test the behavior before it becomes load-bearing in production workflows.
The closing observation
The history of enterprise software is full of tools that were powerful in isolation and impractical at scale because they required constant individual setup. Every person re-explaining their project to their AI every session is the same pattern, repeating at a thousand different desks. Claude Tag is a bet that the right unit of AI deployment is not the individual user. It is the team. Whether that bet pays off probably depends less on the model and more on whether teams actually trust it enough to let it learn over time. The ones that do are going to be operating at a different tempo than the ones that keep starting from scratch.