Anthropic is putting Claude in the Slack channel - you tag it, it builds up context, takes on tasks and works asynchronously (not in real time). It's closer to a colleague than to a chat.
- Anthropic unveiled Claude Tag (June 23, beta for Enterprise/Team): you tag Claude in a Slack channel, it takes on tasks.
- It builds up shared context across the team and works asynchronously; there's also a proactive ambient mode.
- Runs on Opus 4.8 with per-channel control, monthly limits and logs; replaces Claude in Slack from August 3.
You tag Claude in the channel. It builds up the team's context, takes on a task and carries on by itself. This is Claude Tag - in beta for Enterprise and Team customers, launched on Slack, with a proactive ambient mode that reminds you about forgotten tasks and flags problems.
This isn't another integration. It's an attempt to put AI where the work already happens - in the team's channel, with memory in that context, with logs (records of every action) and admin control.
Let's not sugarcoat it: AI no longer stands aside waiting for you to ask. It enters the channel, has limited access, remembers the context, and can take on a task without you standing over it. That's the behavior of a colleague who drives on their own, as long as you give them the frame.
And that's exactly why the controls around it aren't decoration. Give an agent access to the team channel, and the logs, limits and access levels carry exactly as much weight as the capability itself.