Anthropic launched Claude Tag in research preview today, an always-on version of Claude that lives inside Slack and behaves like a shared AI teammate the whole channel can address. The feature is available to Claude Enterprise and Claude Team customers, and it extends the existing @Claude integration with persistent memory, cross-channel context, and an ambient mode that lets the model speak up without being summoned.
The product fills an obvious gap. Anthropic already shipped DM-based @Claude assistance and Claude Code in Slack, which routes coding requests from a channel mention to a full web session and posts results back to the thread. What was missing was continuity — a Claude that remembered last week's discussion, knew who was working on what, and could pick up where a colleague left off. Claude Tag is that layer.
Each channel gets a single Claude identity that every member can interact with. Anthropic says the model learns the work as it follows along in the channel, and that anyone can see what Claude has been working on and resume the conversation from the last handoff. Admins decide which tools, information sources, and channels each Claude identity is allowed to touch.
“As Claude follows along with its channel, it learns ever more about the work,”— Anthropic, Company statement
Key facts
- 01Anthropic launched Claude Tag in research preview for Claude Enterprise and Claude Team customers, accessible inside Slack.
- 02Each Claude identity is scoped by admins to specific channels, tools, and data — a legal-channel Claude cannot seed memory into engineering.
- 03Ambient mode lets Claude post unprompted, flagging cross-org updates and chasing forgotten threads.
- 04Anthropic is competing with Microsoft Graph and Copilot, Glean, Snowflake, and Databricks for the enterprise-context layer.
Scoping is the central design choice. A Claude configured for the legal channel cannot deposit memories into the engineering channel, and vice versa. That isolation matters in regulated industries, where bleeding confidential matter between business units is the kind of mistake that ends pilots. By treating each Claude as a per-channel persona rather than a single org-wide oracle, Anthropic is acknowledging that enterprise context is also enterprise risk.
Assigned tasks get decomposed into stages. Claude Tag works through the steps with whatever tools it has been granted, and posts what it produces back into the Slack thread for the team to review. The output is meant to be public by default — a colleague's work in view, not a private side channel between one user and a model.
The more aggressive feature is ambient mode. Here, Claude posts unprompted: surfacing updates from other parts of the organization it has read access to, flagging blockers, and chasing threads that went quiet. Anthropic frames the experience as working with a real colleague, one that can produce work in public view, with far greater context and understanding than before.
“working with a real colleague — one that can produce work in public view, with far greater context and understanding than before.”— Anthropic, Company statement
Whether teams want a coworker who chimes in on its own is the open question. Slack culture is already noisy, and an LLM willing to volunteer commentary is a different product than one that waits to be asked. The control surface — which channels Claude can read, when it's allowed to post, what triggers it to speak — will determine whether ambient mode reads as useful or as another notification source to mute.
Anthropic is not alone in chasing the enterprise-context layer. Microsoft has Graph, exposed through Copilot and Work IQ, which sits on top of the same M365 surface most large companies already pay for. Snowflake and Databricks are positioning their platforms as the back end for tacit organizational knowledge that agents can query. Glean is building a dedicated intelligence layer between models and enterprise data. Each of these companies is trying to own the same primitive: the place where an AI agent learns what your company actually knows.
Anthropic's wedge is the conversational surface itself. By embedding Claude directly into the channel where work already happens, the company is betting it can collect the context — who said what, what was decided, what's still open — without requiring customers to stand up a separate knowledge platform first. The trade-off is that the data Claude learns from is whatever ends up in Slack, which in most companies is uneven coverage of the actual work.
Pricing and availability beyond the research preview were not disclosed. The feature is gated to Enterprise and Team tiers, which is consistent with Anthropic's broader strategy of pushing higher-value work to paid business plans rather than chasing consumer reach. Anthropic, whose research preview last week introduced 'loops' as a successor to single-shot agents, is layering these capabilities into one stack — agents that can act, loops that can run, and now a teammate that remembers.
The competitive read is that the enterprise-AI fight has moved past raw model quality. Every frontier lab now ships a capable model; the differentiator is whether the model can hold an organization's context long enough to be useful at week three rather than minute one. Claude Tag is Anthropic's attempt to install that memory inside the tool enterprises actually use, and to do so before Microsoft's home-field advantage in productivity software closes the door.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
