Anthropic launched Claude Tag on June 23, putting a shared version of Claude inside Slack channels for Claude Enterprise and Team customers in beta. The product lets employees mention @Claude in a channel, assign work, and watch the agent respond in the thread after using the tools, data, and codebases an administrator has approved.
The launch is a notable shift in enterprise AI because Claude Tag is not just another private chatbot window. It is designed to sit inside a team’s normal Slack conversations, build context from the channels where it is allowed to operate, and carry work forward asynchronously. Anthropic frames it as an evolution of Claude Code, but the initial use cases go beyond engineering into product metrics, support tickets, incident investigation, and other team workflows.
In its launch post, Anthropic said Claude Tag starts in Slack because Slack is already where much of the company’s day-to-day work happens. The company also made a striking internal claim: 65% of its product team’s code is created by an internal version of Claude Tag. That number should be read as a company-provided benchmark rather than an independent productivity measure, but it shows the kind of usage Anthropic is trying to package for customers.
How Claude Tag Works
The basic workflow is familiar to anyone who has used a bot in Slack. A user tags Claude with a request, and Claude breaks the task into stages before working through it with the tools it can access. When it finishes, it posts the result back into the Slack thread. Users can also direct-message Claude for private responses tied to their own connected tools.
The difference is persistence. Anthropic says each channel can have a single Claude identity that everyone in that channel can interact with. That makes the AI’s work visible to the group, lets another employee pick up a thread without restarting the conversation, and reduces the repeated context-setting that happens when every user has a separate chatbot session.
Claude Tag can also learn from other Slack channels and data sources if administrators grant permission. Anthropic says those memories stay scoped to the channels defined by administrators, so a Claude configured for sales work should not carry memories into an engineering workspace or expose sales data to engineers. The system can also run with “ambient” behavior enabled, allowing Claude to flag relevant information, follow up on quiet threads, or surface updates from connected tools without waiting for a direct prompt.
Why Slack Is Becoming an Agent Surface
Slack is becoming an obvious place for AI agents because it already contains a running record of decisions, handoffs, unresolved questions, project context, and links to external systems. That makes it more useful than a blank chat box, but also more sensitive. A Slack agent that can see the right history, fetch the right document, open the right ticket, and report back in public can remove real coordination friction. The same agent with overly broad access can quietly become a data exposure risk.
That tension is already visible in Slack’s own AI strategy. In a February post on rebuilding Slackbot, Slack described its bot as a personal AI agent powered by Claude and connected to Slack’s search and workplace context. Slack said more than 42,000 Salesforce employees were already using the rebuilt Slackbot internally, saving a combined 138,000 hours per week, according to the company’s calculation. Claude Tag adds a related but different model: not just a personal assistant inside Slack, but a shared channel agent that works where the team can see it.
TechCrunch also noted that Claude Tag extends earlier Claude integrations. Users could already message Claude in Slack or use Claude Code from Slack threads, but Claude Tag adds persistent context, shared channel memory, and more proactive behavior. That puts Anthropic in the same enterprise race as Microsoft Copilot, Google Workspace AI, Slackbot, Glean, Databricks, Snowflake, and other systems trying to turn company knowledge into agent-ready context.
The Admin Controls Matter More Than the Demo
The most important part of Claude Tag may be its administrative model. Anthropic says administrators decide which tools and information Claude can access in each channel, set monthly spend limits for the organization and individual channels, and view logs of what Claude has done and who requested each task. Enterprise customers also have a 30-day opt-in window to migrate from the existing Claude in Slack app, and eligible Enterprise and Team organizations are receiving launch credits.
Those controls are not cosmetic. Shared agents change the risk profile of workplace AI. A private chatbot leak is one kind of problem; a channel-level agent with tool access, memory, and initiative is another. Teams will need clear boundaries around which channels get Claude, which repositories or business systems it can touch, which tasks require human approval, how logs are reviewed, and when ambient behavior should be disabled.
There is also a social question. If Claude can keep track of quiet threads, chase unresolved work, and flag context from other systems, it may become useful quickly. But a proactive AI in the middle of workplace chat can also create notification fatigue, unclear ownership, or misplaced confidence if employees treat its summaries as authoritative without checking the underlying work.
What To Watch Next
Claude Tag is available now in beta for Claude Enterprise and Team customers, works with Claude Opus 4.8, and starts in Slack. Anthropic says it wants to expand the product to other places where teams work, which is where this launch could become more consequential. A shared AI teammate in Slack is useful; the same pattern across project management tools, code hosts, ticketing systems, CRM records, analytics dashboards, and document stores would make Claude a more persistent operating layer for work.
The near-term test is not whether teams can tag Claude and get a polished answer. It is whether companies can safely give a channel-level agent enough context and tool access to do meaningful work without blurring permissions, leaking sensitive information, or creating a trail of AI-made decisions no one owns. Claude Tag is one of the clearest signs yet that workplace AI is moving from private assistants toward visible, shared agents embedded in the messy places where work already happens.
