Anthropic’s product team does not use Claude as a chatbot. They use it to review and approve production code, and it handles 65% of submitted changes autonomously. That is the signal engineering teams should pay attention to when evaluating Claude Tag, the new Slack integration Anthropic launched on June 23, 2026. This is not a Q&A tool dressed up with a new interface. It is a persistent autonomous agent with its own identity, ambient monitoring, and the ability to execute multi-step tasks across your workspace.

What Claude Tag Actually Is

The previous Claude-in-Slack integration was stateless. You asked a question, you got a reply, and the thread ended. Claude Tag breaks from that model entirely. It has a persistent identity inside your Slack workspace, meaning it remembers context across an entire conversation thread, not just the most recent message. It does not wait to be asked. In ambient mode, it monitors stalled threads and proactively surfaces relevant information or follows up on work that has gone quiet.

The distinction matters. A stateless Q&A tool is a lookup service. Claude Tag is closer to a team member with a defined role. Anthropic built it in partnership with Salesforce, and it is available to Claude Enterprise and Team plan customers. It does not run on free or Pro plans. This positioning signals Anthropic’s intent: Claude Tag is aimed at organizations with real workflows, not individual productivity use cases.

How It Works in Practice

The interaction model is straightforward, but what happens underneath is not. Here is the full loop from trigger to delivery:

• You @-mention Claude Tag in any Slack channel and assign it a task, for example: “Claude, review this PR and flag any edge cases in the auth flow.”
• Claude Tag breaks the task into sequential sub-steps and executes them independently, without further prompting from you.
• It maintains full context across the conversation thread, so it understands the discussion history when it begins working.
• When it finishes, it delivers structured results back to the channel, not a generic summary but an actionable output tied to the original request.
• In ambient mode, it continues monitoring the thread after delivery. If the conversation stalls or relevant context emerges elsewhere, it resurfaces that information without being asked again.
• You can assign Claude Tag ongoing responsibilities, not just one-off prompts. It holds that role across sessions.

This architecture puts Claude Tag in a different category than any previous Slack AI integration. It is not answering questions. It is executing tasks and managing follow-through.

The Internal Proof: Anthropic’s Own 65%

The most credible data point in the Claude Tag launch is not a benchmark or a demo. Anthropic runs Claude Tag in production on its own codebase. The Anthropic product team submits code changes, and Claude Tag reviews them. It approves and incorporates 65% of those changes autonomously. That figure is not aspirational. It describes current operations at a company that builds and trains frontier AI models.

Engineering teams evaluating Claude Tag should treat this as a reference deployment, not a marketing claim. Anthropic is using this tool on the code that ships Claude itself. That context gives the 65% figure its weight. It also sets a realistic performance baseline for teams considering similar use cases, such as PR review, edge case analysis, or documentation generation.

Anthropic describes Claude Tag as “a tool that works like a virtual employee within Slack.”

Fortune

That framing is accurate. A tool that autonomously approves production code changes at a 65% rate is not functioning as a reference assistant. It is functioning as a participant in the engineering process.

The August 3 Migration Deadline

Anthropic is retiring its previous Claude-in-Slack integration on August 3, 2026. Admins have 30 days from the June 23 launch to complete the migration. Here is what that involves:

• The old integration stops working on August 3. Teams still using it after that date lose Claude access in Slack until they migrate.
• Workspace admins enable Claude Tag through Slack workspace settings. This is an admin-level action, not a user-level toggle.
• Claude Tag is available now in public beta for Enterprise and Team plan customers. Admins can enable it immediately without waiting for the deadline.
• The old integration offered only stateless @-mention replies. It does not carry forward any context, task history, or ambient capabilities into the new system. There is no data migration, because the old integration had no persistent state to migrate.
• Teams on free or Pro plans are not eligible for Claude Tag. They lost access to Claude in Slack on August 3 without a plan for an upgrade path to Claude Tag.

August 3 is a hard cutoff. Admins who enable Claude Tag before then avoid any disruption to their teams.

How Claude Tag Compares to Microsoft Copilot in Teams

Microsoft Copilot in Teams is the closest competitor in terms of native messaging platform integration. The comparison is worth making precisely because the products differ in scope.

• Platform lock-in: Copilot in Teams requires a Microsoft 365 subscription and is tightly coupled to the Office ecosystem. Claude Tag is Slack-native and does not require additional platform subscriptions beyond the existing Slack plan.
• Task scope: GitHub Copilot focuses on code-specific PR review and is not designed for general task execution across engineering and non-engineering channels. Claude Tag operates across both contexts inside the same workspace.
• Ambient behavior: Copilot in Teams does not have a documented equivalent to Claude Tag’s ambient mode. Copilot responds to explicit prompts. Claude Tag monitors threads and follows up independently.
• Persistent identity: Claude Tag holds context across a full thread. Copilot interactions are generally scoped to individual queries within a meeting or chat window.
• Production evidence: Anthropic has published its own internal usage data. Microsoft has not released equivalent autonomous task completion rates for Copilot in production engineering environments.

Neither tool is strictly better for every team. But Claude Tag’s ambient mode and multi-step execution model represent a meaningfully different design philosophy than Copilot’s prompt-response loop.

The Bottom Line

Claude Tag is the first time Anthropic has put a genuinely autonomous agent directly inside Slack with a persistent identity, ambient monitoring, and real task execution. The 65% internal code approval rate is not a feature claim. It is a usage report from a team that builds the model itself. Engineering teams on Enterprise or Team plans should enable Claude Tag before August 3, not because the deadline forces it, but because the old integration is functionally obsolete next to what replaces it. Ambient mode alone changes how on-call and code review workflows operate. Teams that adopt Claude Tag early will calibrate its behavior during the beta window, when iteration costs are low, and tolerance for rough edges is high.