Anthropic’s Claude Tag Slack integration, launched on Monday in research preview, does something its predecessors did not: it stays in the room. Rather than responding to individual queries and forgetting them, Claude Tag builds persistent memory across a channel, learns the team’s work over time, and can act of its own accord without waiting to be asked.
The Claude Tag Slack Integration and What Sets It Apart
Users on Claude Enterprise and Claude Team plans could already direct-message @Claude within Slack or tag it in a channel for on-demand help. Claude Code in Slack handled the narrower job of routing coding tasks from channel mentions to full sessions on the web, posting results back into the thread.
Claude Tag absorbs both of those capabilities and adds a layer of shared, persistent context. ‘As Claude follows along with its channel, it learns ever more about the work,’ Anthropic said in a launch statement. ‘Claude can also automatically gather facts from elsewhere in the organisation, if it’s granted permission to read other channels.’
The result is a single Claude identity that every member of a channel can see and interact with. Anyone can check what Claude has been working on and pick up a conversation from where a colleague left off.
The product also introduces what Anthropic calls an ambient mode: Claude monitors channels without being tagged and proactively flags items, follows up on forgotten threads, and surfaces updates from across the organisation. That is a meaningful step beyond the on-demand model most enterprise AI tools still operate under.
Inside the ‘Multiplayer’ Design
Fortune spoke with Cat Wu, Anthropic’s head of product for Claude Code and Cowork, who framed Claude Tag as an evolution of Claude Code, describing it as ‘multiplayer’ in contrast to Claude Code, CoWork, and chat, which she characterised as ‘very single player.’ The distinction matters for enterprise buyers who have grown frustrated watching individual employees extract value from AI while team-level coordination remains manual.
Wu also disclosed that within Anthropic itself, Claude Tag is already approving and incorporating 65% of the code changes the product team submits. That figure suggests a degree of autonomous workflow integration that goes well beyond a chatbot assist.
When Claude Tag receives a discrete task, it breaks the work into stages, processes each using whatever tools the administrator has permitted, and posts its output into the Slack thread. The administrator defines those boundaries per channel, so a Claude instance configured for legal work cannot write memories into the engineering channel.
Engadget notes that Anthropic says access to sensitive data and task-specific tools ‘can be very tightly controlled,’ with a systems administrator required to specify permissions for each channel. The Anthropic Claude and Slack page confirms that Claude searches only the public and private Slack channels, direct messages, and shared files that the user has already been granted access to.
The Broader Context Race
Persistent organisational context is the feature every enterprise AI vendor is now competing to own. Microsoft expresses it through Copilot and Work IQ via its Graph layer. Snowflake and Databricks are positioning their platforms as the back-end repositories of tacit company knowledge that agents can draw from. Glean is building an intelligence layer that sits between the model and enterprise data.
Anthropic is pursuing the same territory through integrations rather than infrastructure. Beyond the Slack launch, the company has an expanded partnership with Salesforce that makes Claude a preferred model for the Agentforce platform across regulated industries including financial services, healthcare, cybersecurity, and life sciences. Salesforce is also deploying Claude Code across its global engineering organisation.
Slack is only the start of where Claude Tag will appear. VentureBeat reports that Anthropic plans to extend the tagging model to Microsoft Teams, email, and project management tools, widening the surface area of that persistent context well beyond any single platform.
How quickly teams actually hand Claude Tag the kind of access that makes persistent memory useful, rather than a liability, is the question administrators will be weighing. The permissions architecture exists; the willingness to use it fully is the variable Anthropic cannot control.
