Claude Tag puts an asynchronous Claude agent inside Slack channels

Anthropic released Claude Tag, which lets a team mention @Claude in a Slack channel and hand off work that the agent then does on its own. The framing is collaborative rather than one-to-one: each channel has a single Claude that everyone can see, it remembers the channel's history, and it can reach connected data sources and codebases. It runs asynchronously, so a request can kick off a longer task and Claude can schedule its own follow-ups. There is also an optional ambient mode where it watches a channel and surfaces relevant information without being asked.

The control story is the interesting engineering choice. Administrators decide per channel which tools and data Claude can touch, which creates separate Claude identities for different uses, with token spending limits set at both the organization and channel level. Anthropic backs the pitch with an internal number: it says 65 percent of its product team's code is now written by its internal version of this tool, with the same pattern spreading to metrics analysis, support tickets, and debugging. Claude Tag runs on Opus 4.8.

Why it matters

If your team already lives in Slack, this lowers the friction of using an agent to almost zero, but it also puts an agent with real data and tool access into a shared channel. The per-channel permission and spend controls are the part to configure carefully before you turn it loose, because the same setup that makes it convenient is what decides how much damage a bad request can do.