Claude Tag: Anthropic's Always-On AI Agent Living Inside Slack
In recent hours, Anthropic has introduced Claude Tag, an always-on agent that lives inside Slack as a permanent team member. The tool is available in beta for Claude Enterprise and Claude Team customers and runs on the Opus 4.8 model. The promise is to shift the bar beyond the interrogative chatbot to a collaborator that remains in the channel, accumulates context, and acts on its own initiative.
The novelty compared to previous connections between Claude and Slack lies in two elements: persistent memory and continuous presence. Until now, the options were the direct message to @Claude, the tag on request, and Claude Code within Slack for programming sessions. Claude Tag inherits that setup and adds the ability to remember and stay attentive over time. We already discussed this when Claude Code arrived on the messaging platform.
A distinctive feature is the “multiplayer” setting: within a Slack channel, there exists a single Claude identity, shared among all members. Anyone can see what the agent is working on and pick up a conversation from where a colleague left it off, without starting from scratch.
Growing Memory and Unsolicited Interventions
The more Claude Tag follows its channel, the more it learns about the work being done there. If permitted by the administrator, it can also gather information from other channels and the organization’s data sources, building a picture of the corporate context over time.
Additionally, there is an ambient functioning mode, where the agent intervenes proactively without needing to be explicitly tagged: it can signal relevant updates coming from other parts of the company and pick up threads or activities left unanswered. On the operational front, Claude Tag handles asynchronous tasks: it is assigned a task while you attend to other matters; it can also plan future activities on its own, progressing a project independently for hours or days.
Supporting the maturity of the tool, Anthropic states that today 65% of the code from its product team is produced by the internal version of Claude Tag, used by the Claude Code team all year round.
Controls for Administrators and Competitive Landscape
On the control side, system administrators decide granularly which tools, data, and channels each instance of Claude Tag can access. Each identity remains confined to defined channels: the Claude of the legal team, for example, cannot transfer its memories to the engineering channel. Spending limits in tokens can also be set at both the organizational level and for individual channels, and a complete log of everything @Claude has done is available, indicating who requested each task.
Claude Tag replaces the previous Claude app in Slack: administrators have thirty days to migrate, and Anthropic provides an introductory launch credit to eligible Enterprise and Team organizations so that the entire company can try it. Expansion to other platforms beyond Slack is expected in the coming weeks.
This move fits into a landscape where organizational context has become the true competitive front of AI for businesses. Microsoft is building Work IQ through Copilot and Microsoft Graph; Glean has surpassed $300 million in annual recurring revenue at a $7 billion valuation, with a knowledge graph respecting permissions between the model and business data; and startups like Viktor are raising tens of millions to bring virtual colleagues into Slack and Teams. Against this backdrop, Anthropic's bet is that the agent that knows the company best, and not just the most capable abstractly, is the one the team ends up using every day.