Anthropic has unveiled Claude Tag, an always-on version of its AI assistant Claude that integrates directly into Slack channels. The tool is designed to function as a persistent, agentic teammate that can read conversations, join threads, remember context, and complete tasks with minimal human intervention. According to Anthropic, 65% of its own product team's code is now generated by an internal version of Claude Tag.

What is Claude Tag?

Claude Tag is an instance of Claude that resides within a specific Slack channel. It can be reactive, responding only when summoned via @Claude, or it can be set to ambient mode, where it proactively participates in discussions. In ambient mode, Claude observes all channel activity, learns from the ongoing dialogue, and can follow up on tasks, answer questions, or move work forward without explicit prompts. This makes it similar to a human team member who reads the room and contributes at appropriate moments.

Each channel gets its own isolated Claude identity. This means a Claude in the engineering channel cannot access information from the legal channel unless explicitly allowed. Administrators can tightly scope permissions, granting access to specific tools, data sources, or knowledge bases per channel. This isolation is critical for maintaining data security and compliance in organizations that handle sensitive information.

Key features of Claude Tag

• Always-on presence in Slack channels
• Reactive and proactive (ambient) modes
• Channel-specific identity and permissions
• Agentic task execution with multi-step reasoning
• Contextual memory that persists over time
• Asynchronous work: users can assign tasks and receive results later
• Token spend limits and admin oversight

Multiplayer and agentic capabilities

Anthropic describes Claude Tag as operating in a multiplayer mode. A single Claude instance interacts with multiple human participants in a channel, keeping track of who said what and to whom. This shared context allows Claude to provide coherent responses that reference previous interactions, much like a human colleague who remembers the history of a project.

The agentic nature of Claude Tag is particularly noteworthy. It can break down complex tasks into stages, use available tools (such as code repositories, databases, or APIs), and report back in a Slack thread. For example, a user could ask Claude to research a technical issue, find relevant documentation, and summarize the findings—all while the user focuses on other work. This asynchronous delegation is a step toward true AI coworking.

Claude Tag also learns over time. It builds a persistent memory of channel activity, which means users don't have to re-explain context repeatedly. If a thread goes quiet or a deadline is missed, Claude can proactively follow up, acting as a task master. While this may sound intrusive, it could significantly boost team productivity by reducing dropped balls and forgotten commitments.

Scoped permissions and administrative control

Anthropic has designed Claude Tag with enterprise governance in mind. Each Claude identity is isolated by channel, ensuring that sensitive data does not leak between teams. Administrators can set fine-grained permissions, specifying which tools and data sources each Claude can access. For instance, the legal team's Claude might have access to contract templates and case law databases, while the engineering team's Claude can only access code repositories and bug trackers.

Token spend limits are another key feature. Because ambient Claude consumes tokens continuously, organizations can set overall and per-channel caps to control costs. Administrators also have access to a full audit log of Claude's actions, including which user triggered each task. This transparency helps maintain accountability and allows teams to monitor how AI resources are being used.

Availability and pricing

Claude Tag is currently available in beta for Claude Enterprise and Claude Team customers. It replaces the existing Claude in Slack app, and administrators have 30 days to migrate. Anthropic has also announced an introductory launch credit for eligible organizations. The company plans to expand Claude Tag to other platforms beyond Slack in the future, though no timeline has been provided.

The pricing model is based on token consumption, which could be significant for always-on instances. However, the ability to set spend limits gives organizations predictability. Early adopters report that the productivity gains often outweigh the costs, especially for teams that rely heavily on Slack for asynchronous collaboration.

Implications for teamwork and privacy

The introduction of an always-on AI coworker raises important questions. On the positive side, Claude Tag can reduce context-switching, automate routine inquiries, and ensure that no task falls through the cracks. It can also serve as a knowledge base that grows with the team, storing and retrieving information that might otherwise be lost in crowded threads.

On the privacy front, Anthropic emphasizes that Claude does not report from private channels unless explicitly invited, and all data remains within the organization's Slack workspace. The channel-specific isolation further reduces cross-contamination. However, some users may feel uneasy about an AI that constantly watches and archives conversations. Organizations will need to establish clear policies about where and how Claude Tag is used, balancing productivity with employee comfort.

The broader trend is clear: AI is moving from a tool you call upon to a persistent presence that collaborates alongside humans. Claude Tag represents one of the most ambitious implementations of this vision within a widely used communication platform. As remote and hybrid work continue to grow, such agents could become essential for maintaining cohesion and momentum across distributed teams.

Anthropic's decision to launch Claude Tag for Teams and Enterprise tiers rather than general availability suggests a focus on organizations that can manage the governance and security requirements. The beta phase will likely refine the product based on real-world feedback, addressing concerns about noise, relevancy, and cost before a wider rollout.

Source: ZDNET News