Anthropic launches Claude Tag, a permanent agent that lives inside Slack and learns how your company actually works. Tag @Claude in any thread, and the agent gathers, remembers and executes. The target is clear: capture the organizational memory before Microsoft does.
Key Takeaways
- Claude Tag is an always-on agent inside Slack with persistent memory scoped per channel
- The product attacks Microsoft Copilot, Glean, Snowflake and Databricks head-on on the enterprise context layer
- It ships in beta, restricted to Claude Enterprise and Claude Team customers
How Claude Tag actually works
Claude Tag is described as an always-on Claude that lives in Slack at all times. A team member writes in a channel, mentions @Claude, and the agent replies in the same thread with analysis or a deliverable. The conversation stays shared, every team member sees Claude’s output, and anyone can pick up where the last person left off.
The break with a basic bot is memory. Claude Tag learns by following the channel it sits in. The more the team writes, the more the agent understands projects, people and past decisions. With admin permission, it can also pull facts from elsewhere in the organization to complete its reply.
Each channel runs a single Claude identity, available to everyone inside that channel. Administrators define which tools, information sources and channels each Claude can reach. A Claude attached to a legal channel cannot seed memories into an engineering channel, and the reverse holds too. The default isolation is strict.
The agent also handles assigned tasks. When a user hands Claude a job, the agent breaks it down into stages, uses whichever tools are available, and posts the work back into the Slack thread. The format is built so that an entire team can follow progress without leaving the chat for another tool.
The most advanced layer is ambient mode. Claude jumps into the chat on its own to keep the team updated, flag information coming from elsewhere in the organization, or follow up on tasks left dormant. The product moves from passive responder to autonomous operator running quietly in the background.
A head-on fight with Microsoft, Glean and the data clouds
The terrain Anthropic targets is not the chatbot space. It is the enterprise context layer, the infrastructure that learns what an organization does and serves it back on demand. After locking down Korea with NAVER, Samsung and LG, Anthropic opens a second front inside internal workflows.
Microsoft has held this territory through Microsoft Graph, Copilot and Work IQ, where corporate memory is fed by the Office suite and Teams. Glean has built a transverse intelligence layer that reads enterprise context across vendors. Snowflake and Databricks defend the storage and reasoning side of business data.
Anthropic does not attack through storage or office tooling. It attacks through the conversation. Slack remains where day-to-day decisions get made in tens of thousands of companies. Settling inside Slack means capturing context as it forms, before it ever lands in a wiki, a ticket or a shared drive. The full picture is available in Anthropic’s announcement introducing Claude Tag for Slack.
The approach extends the product strategy already in place around Claude. Claude Code Artifacts opened the door to live shareable pages generated by the agent, and Claude Tag pushes the same idea into the chat layer. The agent’s memory grows while the company works, with no one having to feed it manually.
The rollout stays controlled. Claude Tag is in beta, available only to Claude Enterprise and Claude Team customers. The goal is to test the product on mature organizations already deep in the Anthropic stack, and to validate the per-channel scoping format before a wider release. For context, see our earlier piece on Horizon: Anthropic Dreaming: Claude Agents Can Now ‘Dream’.
Also on Horizon:
- OpenAI Starts Patching Open Source Bugs
- Sam Altman Says Researchers Held the Whole Field Back
- Samsung Korea Goes All-In on ChatGPT and Codex
What changes short term and medium term
Short term, the impact concentrates on Claude Enterprise and Claude Team customers, especially those whose workflows already run mostly through Slack. Teams that turn on ambient mode will see an agent monitor their channels, revive dormant tasks and surface information they had forgotten. The time saved will show within the first few weeks.
The first friction will come from governance. Per-channel scoping protects against vertical leaks, but the question of cross-channel permissions will surface fast. An administrator who opens access too widely hands the agent a view that may exceed what most human employees can see.
Medium term, the enterprise context fight will sharpen. Microsoft owns the installed position, Glean owns the transverse layer, Anthropic builds the conversational layer. If Claude Tag delivers on persistent memory, Microsoft 365 customers will start comparing the quality of replies between Copilot and Claude on the same Slack perimeter.
Snowflake and Databricks stay in ambush on structured data. Their read is that the useful corporate memory will end up anchored in the data cloud, not in the chat. Anthropic bets the opposite, that the chat captures the decision layer before the data layer catches up.
The real test will play out in team behavior. If ambient mode holds on busy channels without saturating the conversation, Claude Tag becomes a standard. If the agent piles up off-topic interventions, the beta exit will be pushed back, and competitors will get the time they need to answer.
Follow the story on Horizon.
