Claude Tag: Anthropic puts a persistent AI agent inside Slack

On June 23, 2026, Anthropic made Claude Tag available in research preview — a persistent AI agent that lives inside Slack as a full channel member for Claude Enterprise and Claude Team customers. Tag the @Claude handle, and the agent takes on a task or answers a question with full awareness of the channel's conversation history. This isn't another on-demand chatbot integration. It's a bid to embed organizational memory directly inside the communication tool where millions of employees already spend their working hours.

Key takeaways

Claude Tag is in research preview, available only to Claude Enterprise and Team customers via Slack
The agent runs continuously — monitoring channels, retaining context, and proactively joining conversations without being explicitly tagged
Each Claude identity is channel-scoped — admins define which channels and tools the agent can access
Ambient mode lets the agent independently surface updates, flag stale threads, and pull context from across the organization
All agent activity is visible to the entire team in public Slack threads — no private sessions

How Claude Tag works

Claude Tag builds on earlier Anthropic integrations with Slack, including the ability to DM @Claude directly and route coding tasks through Claude Code. The core new capability is persistent context. The agent joins channels as a regular member, reads conversation history, and — with admin permission — can pull data from other channels in the organization to complete tasks. When an admin creates a Claude identity for a specific channel, that instance is strictly scoped: context cannot bleed into other channels.

Task assignment works as expected. A user tags @Claude, the agent breaks the task into stages, executes it using available tools, and reports results in the Slack thread. Every team member in the channel can see the full activity log — there are no private sessions, and context accumulates collaboratively.

Ambient mode is the more ambitious feature. Without being explicitly tagged, the agent can interject to surface forgotten tasks, flag incomplete threads, or share relevant information gathered from elsewhere in the organization. Anthropic frames this as working with "a real colleague." That's a high bar — and exactly what many enterprises deploying AI are trying to achieve.

The strategic layer: organizational context as competitive advantage

Claude Tag enters a crowded space. Microsoft is expanding Microsoft Graph as the context foundation for Copilot and Work IQ, giving its model awareness of relationships between documents, emails, and employees. Glean has been building an intelligence layer positioned between AI models and enterprise data for months. Data platforms like Snowflake and Databricks are positioning themselves as tacit knowledge repositories for agents.

Anthropic's approach is different. Rather than building a separate context layer, it enters Slack — where employees already work — and lets the agent learn the organization as a byproduct of daily communication. The adoption barrier drops significantly: companies don't need to deploy new software or migrate data. Claude becomes part of the existing communication workflow.

Technically, Claude Tag draws on the existing Claude Enterprise infrastructure, which supports in-session memory and tool access via API. The key shift is that the agent isn't invoked on demand — it's always present, observing and building an organizational knowledge model.

Risks: privacy, scope, and the limits question

Continuous monitoring of Slack channels by an external AI model raises legitimate questions about privacy and data security. Anthropic addresses this through admin-controlled permissions: each Claude instance is limited to the channels and tools the admin defines. Even so, companies operating in regulated industries — finance, healthcare, law — will need to assess what data reaches Slack in the first place and what retention policies apply.

A second risk is structural lock-in. If organizations build dependence on Claude Tag as their primary record of institutional knowledge, the dependency becomes architectural, not just contractual.

Anthropic is launching Claude Tag as a research preview, which means agent behavior, usage limits, and data policies may change before general availability.

Why this matters

Claude Tag is a strategic signal, not just a product feature. Anthropic is moving directly into the space Microsoft is trying to dominate through Graph and Copilot — persistent organizational context as the foundation for enterprise AI agents. Whoever builds a durable representation of corporate knowledge secures enterprise loyalty for years, not months.

Previous AI integrations with Slack operated primarily on-demand: the agent answered when asked. Claude Tag proposes a fundamentally different model — the agent is always present, actively building contextual understanding, and acting proactively. If this proves reliable in practice, it represents a genuine shift in how AI operates in the workplace.

For the market, it's also a test of whether employees and organizations will accept a model in which an AI continuously processes internal communications. The result of that test will matter more than the product itself.

What's next

Claude Tag research preview is Slack-only — Anthropic has not announced integrations with Microsoft Teams, Google Chat, or other platforms, nor a general availability date
Anthropic stated it will collect feedback during the preview phase to refine agent behavior before full Enterprise rollout
Regulatory questions around external AI models processing internal corporate communications remain unresolved, particularly for EU companies operating under GDPR