Anthropic launches Claude Tag for Slack teams

Anthropic launched Claude Tag on June 23, 2026, giving Claude Enterprise and Team customers a shared AI agent inside Slack. People in an authorized channel can mention @Claude, assign work and follow the result in a public thread. The beta is powered by Opus 4.8 and replaces the existing Claude in Slack app; Anthropic says administrators can opt into the migration within 30 days.

Claude Tag differs from a one-to-one assistant because one channel shares the same Claude identity. Anthropic says the agent can retain relevant channel context, break requests into stages, run long or scheduled tasks and use connected tools, data or codebases that an administrator has approved. An optional ambient mode can surface relevant information or revisit unresolved work without a new request.

Reuters and TechCrunch independently confirmed the launch, availability and core shared-agent behavior. Anthropic also said an internal version produces 65% of its product team's code. That figure is a company-reported usage claim, not an independently measured productivity result.

Anthropic says administrators choose the channels, tools and information each Claude identity may access. Memories remain scoped to those configured channels, and admins can set organization- or channel-level token limits and review activity logs. Claude can also respond privately in direct messages using a user's personal connectors.

Those controls define the practical deployment boundary. Giving an agent access to Slack history and external tools can reduce repeated context-setting, but it also expands the data and action surface that teams must govern. The retrieved evidence describes access controls and logs; it does not show that every downstream connector or customer configuration is safe by default.

The immediate product shift is from occasional assistant use to a persistent, multiplayer agent embedded in a collaboration channel. Teams evaluating the beta should start with a narrow channel and limited tools, verify what appears in the audit log, set spending limits and test how retained context behaves before widening access.

Those steps are LDS interpretation based on the announced controls, not additional claims from Anthropic. The key question is whether shared context and asynchronous work save enough coordination time to justify the added permission, monitoring and cost-management work.