In this short announcement, Lydia, who works on Claude Code at Anthropic, introduces Claude Tag: the ability to bring Claude into a Slack conversation by tagging @Claude, so that Claude collaborates right alongside your team where the work already happens. Anthropic has been tagging Claude in for most of the year, and Claude Tag now opens 65% of Anthropic's product pull requests. The video walks through a realistic scenario, a team about a week from shipping a product, to show how Claude follows the conversation, takes real action across tools, and respects team and channel boundaries.
Claude Tag turns Claude into a genuine multiplayer teammate inside Slack. Instead of switching to a separate tool, anyone on the team can @Claude directly in a thread. Claude reads the surrounding conversation, keeps up with decisions made in real time, and then acts, opening pull requests, editing documents, and moving work forward without the usual cross-team handoff lag.
When tagged, Claude keeps up with the group thread, reacting to product decisions as they are made in real time. In the example, an engineer (Drew) learns from a sales rep that a feature called scheduled exports is blocking the biggest deals. He does not own that feature, so he asks the person who does, and Claude follows along. Because Claude already understood what the feature was and where in the codebase to go, it opened the PR and landed the change. Claude is scoped to each channel and the context each team has.
Claude does not piggyback on a human's session. It has its own account and its own permissions, and it makes edits itself. In the example, a teammate (Priya) did not have to open Drive, because Claude made the document edit directly. Claude has access to the systems it needs on a per-team, per-channel basis:
Claude builds memory as the work happens over time. In the scenario, Claude recognized that the change it just pushed affects launch marketing, and surfaced that connection, keeping the team moving without the typical cross-team communication lag. This persistence is what lets Claude carry context forward across a multi-day effort rather than starting from scratch each time it is tagged.
Memory follows the same access rules as tools. What Claude learned in a private channel or a DM stays there. It does not leak that knowledge into other channels. Combined with per-channel tool scoping, this means Claude only ever knows and acts on what a given channel is actually allowed to see.
Because Claude has its own account, every credential it uses is logged. Actions are attributable to Claude's own identity rather than blurred into a human teammate's account, which supports auditability and accountability.
Claude Tag is launching in beta on Team and Enterprise plans. Lydia's suggested way to start is low-commitment: add Claude to a single channel, tag it in, and see what happens. Anthropic invites feedback from early users.
@Claude, so collaboration happens in the tools teams already use.