Anthropic: Claude Tag brings a multiplayer AI agent directly into Slack teams

Anthropic launched Claude Tag in beta on June 23, 2026, for Enterprise and Team customers — a new model of team collaboration with an AI agent that, unlike classic chat assistants, does not live in a single user’s private session but is available to the entire team within a Slack channel.

What is a multiplayer AI agent?

Claude Tag introduces what Anthropic calls a multiplayer approach — a single Claude instance shared by an entire team (a multiplayer agent, as opposed to the traditional synchronous one-on-one model). By tagging @Claude in any Slack message, the team activates the same agent, which sees the full channel context, retains relevant information from previous conversations, and can proactively flag messages or data it considers important to the rest of the team.

The key difference from classic AI assistants such as standard Claude.ai chat is the time horizon for tasks: while synchronous assistants are optimized for responses within seconds to minutes, Claude Tag is designed for asynchronous work that can last hours or days — the agent keeps working while users attend to other matters.

Opus 4.8 as the underlying model

Claude Tag is powered by Opus 4.8, Anthropic’s currently most capable model, positioning it above most existing Slack integrations that typically use lighter models to control costs. This decision reflects Anthropic’s emphasis on reliability in long-running agentic tasks, where errors are more costly than in short synchronous interactions.

Internal data: 65% generated code

The most concrete figure Anthropic cites is not a benchmark but an internal data point: 65% of code written by Anthropic’s own product team in the recent period was produced with the help of an internal version of Claude Tag. This figure is comparable to what GitHub Copilot reports for individual developers, but unlike Copilot — which operates per developer — Claude Tag scales those gains at the team level, where context and knowledge are shared.

Administrative controls for Enterprise

Anthropic has introduced four key governance mechanisms for organizational administrators:

• Token consumption limits per channel — preventing uncontrolled costs in high-volume channels
• Activity logs — an audit trail of all Claude Tag interactions for compliance purposes
• Isolated access per channel — Claude in one channel cannot see content from other channels even within the same organization
• Proactive flagging — the agent can initiate communication on its own when it detects information worth the team’s attention

Async work as a paradigm shift

The prevailing approach to AI assistants in business has assumed synchronous interaction: the user sends a query, receives a response, and the session ends. Claude Tag proposes a different model — the agent remains active in the team’s context, builds understanding of the project over time, and can take on tasks that require waiting for external systems or longer research.

Conceptually, this is closer to a dedicated team member than a tool invoked on demand — raising new questions about governance, trust, and the boundaries of AI agent autonomy in organizations.

Claude Tag beta is available as of today for Enterprise and Team plans; general availability has not been announced.