TL;DR
Anthropic introduced Claude Tag, an always-on Slack AI that monitors conversations, learns their context, and proactively joins updates and tasks.
Anthropic applies the Claude Tag in a research review, a “always in ClaudeIt lives inside Slack and acts as a persistent AI teammate. This feature allows users to tag @Claude in conversations to get ideas and assign tasks. Available to Claude Enterprise and Claude Team customers starting today.
Claude Tag is an evolution Anthropic’s existing Slack integrationsit now allows users to message Clo directly or tag him in channels for on-demand help. Claude Code on Slack can also direct coding tasks from channel notes to full coding sessions on the web, posting updates back into the thread. But Claude Tag adds a persistent layer of context and memory that previous tools couldn’t maintain.
“As Claude watches his channel, he learns more about the business,” Anthropic said in a statement. Claude can also automatically gather facts from elsewhere in the organization if given permission to read other channels. The result is an AI that accumulates institutional knowledge over time, rather than starting from scratch with each interaction.
💜 of EU technology
The latest rumblings from the EU tech scene, a story from our wise founder Boris and some questionable AI art. Free in your inbox every week. Register now!
Everyone in a given Slack channel can log into a single Claude identity, meaning everyone can see what Claude is working on and pick up where the last person left off. System administrators define which tools, data, and channels each Claude identity can access, and each identity remains in scope for the channels defined by the admins. A Claude set up for legal work, for example, cannot plant memories in an engineering channel.
When given a task, Claude Tag breaks it down into phases and uses which tools to work on them, responding in a Slack thread with what he’s created. But a more notable feature is the environment mode, which proactively joins conversations to update teams, record relevant information within the organization, and follow up on forgotten topics or tasks.
Ambient mode is the feature that separates Claude Tag from a regular chatbot. Instead of waiting to be asked, Claude monitors his assigned channels and intervenes when he decides a team would benefit from a reminder, summary or context from another part of the company. Anthropic says that this “working with a real colleague who can create work in public view with greater context and understanding than before“
The organizational context is increasingly the battleground for enterprise AI. Microsoft is building Work IQ, a layer of intelligence expressed through Copilot that uses Microsoft Graph to understand roles, collaboration patterns, and organizational structure. Startups like Victor have raised tens of millions to put AI workers directly into Slack and Teams..
Glean, which recently surpassed $300 million in annual recurring revenue at a valuation of $7 billion, builds a permission-aware knowledge graph that sits between model and enterprise data.
Claude Tag is Anthropic’s answer to the same problem, but its approach is narrower and arguably more disciplined. Instead of creating a horizontal layer of intelligence in each enterprise application, it embeds AI inside a surface where multiple knowledges work: the team conversation. The bet is that a persistent presence on Slack, combined with cross-channel storage and admin-driven coverage, is enough to gather the institutional context that makes an AI agent useful.
The privacy implications are significant. Always-on AI that monitors workplace conversations and autonomously decides when to intervene will face scrutiny from both employees and compliance teams. Anthropic’s admin scope controls are a structural answer to this concern, but the real test will come when enterprise customers deploy it at scale and discover how employees respond to always-listening AI.
Anthropic said it is working to bring Claude Tag to other platforms in the coming weeks. For now, Slack is the only surface that limits the feature’s availability, but also limits its complexity, a purposeful trade-off for research previews.
