Anthropic launched Claude Tag today, available now to Claude Enterprise and Team customers through Slack. The basic idea is simple: instead of asking an AI a question and waiting for an answer, you tag @Claude in a channel the same way you would tag a colleague.
The product has three things that set it apart from what existed before.
First, it handles tasks end-to-end. When you assign Claude Tag a job, it breaks the work into steps, carries them out using whatever tools the administrator has connected, and posts the result back in the thread. You do not need to manage the steps yourself.
Second, it builds memory. Previous AI tools in Slack forgot everything between conversations. Claude Tag follows along with its channel over time and accumulates understanding of what the team is working on. Administrators can also grant it permission to read other channels, so a single Claude identity for, say, the procurement team can draw on context from across the organization.
Third, it works without being asked. An ambient mode lets Claude Tag jump into conversations on its own to flag relevant updates, surface forgotten tasks, or keep people informed. This is a meaningful shift: the tool is no longer passive.
Anthropics's own product team says Claude Tag is already handling 65 percent of code changes submitted internally, which is a striking claim from a company about to file for a public listing. Anthropic's run-rate revenue has climbed to around $47 billion as of May 2026, with around 80 percent of that coming from business customers. It recently overtook OpenAI in enterprise market share for the first time.
The broader race here is not really about features. It is about which AI company earns the right to sit at the center of how your organization thinks and operates. Microsoft is building the same kind of persistent context layer, called Work IQ, which reads emails, meetings, calendar entries, and chats to give Copilot a standing understanding of how your organization actually works. Snowflake, Databricks, and a startup called Glean are all competing to be the layer that connects AI to company knowledge.
The reason this matters for non-technical managers and operators is straightforward. An AI that remembers everything your team has discussed, knows who owns which project, and can act without being prompted is a fundamentally different kind of tool than a chatbot. The switching costs rise sharply once a system like this is embedded and has built up six months of organizational memory.
The privacy and governance question is real. When an AI agent reads every channel it is given access to, stores context over time, and acts on its own, sensitive information can travel further than intended. Anthropic says administrators control which channels Claude can access and what tools it can use. That is the right answer, but it places a genuine responsibility on whoever sets those permissions. Poorly scoped access in a legal team's channels leaking into an engineering channel, for example, is the kind of incident that is difficult to untangle after the fact.
The practical question for any business operator today is not whether this category of tool is coming. It is which version you want embedded in your organization, and who controls its boundaries when it arrives.