CHAPTERS
Launch announcement: Claude Tag brings Claude into team workflows
Lidia from Anthropic introduces Claude Tag, a new way to collaborate with Claude directly inside team communication channels. She frames it as a workflow Anthropic has already been using internally for most of the year.
Internal adoption and impact: Claude opens most product PRs
The video highlights Claude Tag’s real-world usage at Anthropic, emphasizing how embedded it is in day-to-day engineering. A key metric is shared to show scale and maturity of use.
Shipping pressure example: sales feedback drives a last-minute feature push
A scenario is set up where the team is about a week from shipping, and a sales rep surfaces that scheduled exports are blocking major deals. The request lands with someone who doesn’t own the feature, prompting coordination across the team.
Multiplayer collaboration: tagging Claude into the group thread
The workflow is shown as a shared, real-time thread where multiple teammates coordinate. Nadia tags Claude, and Claude stays in the loop as decisions happen, participating alongside the group.
From request to code change: Claude opens a PR and lands the fix
Claude doesn’t just advise—it executes by making a change in the repository, opening a pull request, and getting it merged. The emphasis is on speed and end-to-end follow-through.
Codebase understanding and context scoping per channel
Claude is described as knowing what the feature is and where to work in the codebase. It also scopes context by channel/team, and builds memory over time as work occurs.
Keeping cross-team work moving: anticipating downstream dependencies
Claude connects engineering changes to downstream stakeholders, like launch marketing, and reduces typical cross-team coordination delays. The promise is smoother execution across functions.
Document work without leaving the thread: Claude edits directly
The video points out that teammates don’t need to open separate tools like Drive to make updates. Claude can perform edits itself, streamlining documentation and coordination work.
Permissions model: Claude has its own accounts and logged credentials
Claude operates with its own accounts and permissions, with access granted per team and per channel. Every credential used is logged, emphasizing auditability and controlled access.
Hard boundaries by channel: legal vs engineering access separation
A concrete example explains that Claude in a legal channel can see contracts, while Claude in engineering can edit code. If asked to cross boundaries (e.g., legal channel requesting code edits), Claude can’t access what it’s not permitted to see.
Memory boundaries: private context stays private
The memory system follows the same access rules: what Claude learns in a private channel or DM remains contained there. This reinforces compartmentalization of sensitive information.
Call to action: add Claude to a channel and start tagging it in
Lidia closes by describing how Claude Tag has changed how she and teams at Anthropic work, and invites viewers to try it. The simplest onboarding step is adding Claude to a channel and tagging it to begin collaborating.
