Anthropic launches Claude Tag, a Slack-based team agent for Enterprise and Team customers
Anthropic just launched Claude Tag, a beta product that puts Claude into Slack as a shared teammate rather than a private chat assistant. This is worth treating as breaking builder news because it changes where Claude agents run day to day: inside team channels, with scoped access to tools, data, and codebases.
What Anthropic announced
Claude Tag starts in Slack. Admins can add Claude to selected channels, connect approved tools and data sources, and let people tag @Claude into threads when there is work to hand off.
Anthropic says it is available today in beta for Claude Enterprise and Claude Team customers. The company describes it as an evolution of Claude Code and says its internal version is already creating 65% of the code for Anthropic's product team.
The key difference from a normal chatbot is that Claude Tag is designed to be shared, persistent, and asynchronous:
- Each channel can have its own Claude identity, scoped to that channel's permissions and memories.
- Claude can follow messy threads, pick up context from the channel, and respond in-thread with finished work.
- Teams can give it standing instructions, schedule future work, or let it run long tasks over hours or days.
- Optional ambient behavior lets Claude proactively surface relevant updates, flag unresolved threads, or tag people when it needs a decision.
- Admins can set token-spend limits and review logs of what Claude did and who asked for it.
Anthropic's product page brands the feature as @Claude and says it can be added to Slack for eligible Enterprise and Team workspaces.
Why builders should care
This is another step away from “open a chat tab and paste context” and toward agents living where teams already coordinate work. If it works well, the operating model is closer to a monitored teammate in Slack: it sees the thread, has controlled access to internal systems, and can be delegated work without forcing everyone into a separate tool.
For founders and engineering teams, the practical questions are now less about prompt tricks and more about agent operations:
- Which channels should an agent be allowed to read?
- Which tools should it be allowed to call from each channel?
- Who owns the cost limits and audit logs?
- What work is safe to let run asynchronously without a human watching every step?
- How do you prevent channel memory from leaking across sales, support, product, and engineering contexts?
That last point matters. Anthropic is explicitly pitching separate Claude identities with scoped memories and permissions, which is the right security direction for team-wide agents. It also means builders will need to design around identity, authorization, logging, and least-privilege access instead of treating an AI assistant as one global bot.
Caveats
This is a beta, not a broad free-for-all launch. Anthropic says it is available today for Claude Enterprise and Team customers in Slack, with wider availability planned later.
The announcement does not include public pricing specifics beyond references to admin token-spend limits. Teams should also expect the usual hard parts of agent deployment: noisy channels, stale context, permission mistakes, and the need to verify important outputs before they hit production or customers.
