Anthropic has introduced Claude Tag, an AI agent that operates as a full-fledged team member in Slack. The tool works for entire organizations at once, builds persistent memory per channel, and performs tasks independently. Claude Tag is now available in beta for Enterprise and Team subscribers.
Whereas the existing Claude integration responded to direct questions from individual users in Slack, Claude Tag functions as a shared team member active across an entire channel. Everyone in that channel can use the agent. It’s also possible to check on progress and pick up a task that a colleague left halfway through.
This is a major difference from other AI chatbots. Claude Tag is assigned a single shared identity per channel, allowing it to build context around the projects taking place there. If Claude Tag is granted access to multiple Slack channels, it will also scan those channels to better understand a task—provided the administrator has given permission to do so.
Ambient Mode and asynchronous work
Specifically, Claude Tag works as follows: an employee tags @Claude with an assignment, after which the agent automatically breaks it down into steps and carries it out independently. The result appears in a Slack thread. Tasks can also take hours or days to complete, allowing employees to work on something else in the meantime.
There is also a so-called ambient mode. In this mode, Claude Tag is proactively activated in conversations to provide updates, flag something important, or send reminders for forgotten tasks. Anthropic describes this as working with “a real colleague who can present work publicly, with much more context and understanding than before.”
Anthropic is not alone in the race to develop AI assistants for the workplace. In early April, Salesforce announced thirty new AI features for Slack and Slackbot, including reusable AI skills and integration with Agentforce. Slackbot also functions as an MCP client and works with external services. Claude Tag, however, focuses specifically on persistent context and asynchronous collaboration within teams.
Why Claude Tag is not a chatbot
With Claude Tag, the line between a communication tool and active IT infrastructure disappears. Although the software runs in Slack, the term “chatbot” is technically outdated. This marks the shift from generative AI to agentic AI. Traditional chatbots are reactive: they wait for a user to type something, respond, and forget the context as soon as the window closes. AI agents like Claude Tag, on the other hand, work proactively, asynchronously, and with context awareness.
For IT architectures, this fundamentally changes how data is handled. Instead of isolated chat sessions, the agent builds a collective memory for each channel. The AI understands a project’s history and ongoing discussions. As a user, you therefore don’t have to keep providing the context over and over again. The interaction also changes. It’s no longer about immediate question-and-answer exchanges, but about delegating tasks. The agent breaks down a task on its own and carries it out autonomously over a period of hours or days. AI thus becomes an invisible force in the background.
Technically, this autonomous operation increasingly relies on open standards such as the Model Context Protocol. As a result, Slack transforms into a central layer through which AI agents directly control systems like GitHub, databases, or CRM packages. At the same time, this setup resolves a major governance issue: the proliferation of individual AI accounts often leads to “Shadow AI” and data risks.
The battle for the enterprise market
The launch is part of Anthropic’s broader strategy to gain ground with large enterprises. The tool is already in use internally within the company itself. Currently, 65 percent of the product team’s code comes from the internal version of Claude Tag. Meanwhile, competitors such as Microsoft, with Copilot and Work IQ, and Glean Technologies are developing their own enterprise AI layers.
Claude Tag is now available in beta for Claude Enterprise and Team subscribers. Administrators determine which channels, tools, and data the agent can access. Each Claude Tag identity remains strictly limited to its assigned channels. For example, an agent for the legal department cannot access data from the technical teams. Anthropic plans to expand availability to other platforms in the future, but details have not yet been announced.