Claude Tag

The old 'Claude in Slack' was a personal assistant: you @-mentioned it, it answered for you, and it ran on your account's permissions and your bill. Claude Tag re-architects that into something organizational. There is now one shared Claude per channel, with its own org identity that every member sees, its own memory that isolates per channel and builds across days, and — when admins turn on ambient behavior — the ability to post updates unprompted and schedule follow-up tasks for itself over hours.

A legal-channel Claude can't read engineering channels unless an admin grants it; spend limits and activity logs are set per channel.

The quiet part is the permission and cost model. Channel usage is billed to the organization at Anthropic's top model rates, and the agent's access to tools, repos, and data runs through admin-provisioned service accounts rather than any individual's login. A legal-channel Claude can't read engineering channels unless an admin grants it; spend limits and activity logs are set per channel. What reads as an 'AI teammate' launch is a move of Claude from personal software to governed infrastructure — the decision that matters is an admin's, not a user's.

The old app is being retired; workspaces that do nothing are auto-migrated inside a 30-day opt-in window opened June 23, with no option to stay on the per-user version. Anthropic's proof point is that it dogfoods the thing: it says 65% of its own product team's code is now written by an internal version of Claude Tag. For enterprise admins the launch isn't a new feature to try so much as a cost and access model to sign off on before the switch flips itself.