Most AI tools wait for you to ask them something. Claude Tag watches your channels, learns your team's context, picks up where the last person left off, and works on tasks for hours without anyone in the room. Anthropic already uses it to write 65% of their product code.
Introduction
Every AI assistant launched in the last three years follows roughly the same model: you open an interface, type a request, read the response, and close the window. The interaction is one-on-one, synchronous, and scoped to whatever you asked in that single session. Tomorrow, you start over.
Claude Tag is built on a completely different premise. On June 23, 2026, Anthropic introduced Claude Tag — starting on Slack — where Claude joins a team channel as a permanent member, builds ongoing context from the conversations it follows, can be tagged by anyone on the team, works through tasks asynchronously while the team focuses on other things, and proactively flags information it thinks the team needs to know without being asked.
The simplest description: it behaves less like a tool and more like a colleague.
Anthropic's own teams have been using an internal version long enough to have production data on it. The number they chose to lead with is not a benchmark score or a capability claim — it is an operational reality: 65% of Anthropic's product team's code is now created by their internal version of Claude Tag.
Quick Summary
| Detail | Information |
|---|---|
| Product name | Claude Tag |
| Launched | June 23, 2026 |
| Platform | Slack (beta) |
| Available to | Claude Enterprise and Team customers |
| Underlying model | Claude Opus 4.8 |
| Replaces | Existing Claude in Slack app |
| Migration window | 30 days to opt in |
| Key stat | 65% of Anthropic product team's code created by Claude Tag |
| Goal | Expand beyond Slack to other team collaboration platforms |
What Claude Tag Actually Is
The clearest way to understand Claude Tag is to understand what it is not.
A standard AI chatbot lives in its own interface. Every conversation starts fresh. Only one person interacts with it at a time. It has no memory of what your team discussed last week. It cannot take a task and work on it while you attend a meeting. When you close the window, it stops.
Claude Tag inverts all of those assumptions.
Grant Claude access to a Slack channel and it becomes a persistent team member in that channel. It reads the conversations flowing through. It builds an understanding of the work, the context, the terminology, the ongoing projects, and the people involved. Anyone on the team can tag @Claude with a task at any time. When they do, Claude breaks the task into stages and works through them using whatever tools and data sources the administrator has connected — a codebase, a database, a ticketing system, whatever the team has made available. The result comes back as a response in the Slack thread.
The work does not stop when the person who requested it goes offline. Claude can schedule tasks for itself and pursue a project autonomously over hours or even days.
Anthropic describes Claude Tag as the beginning of an evolution of Claude Code — making the model more proactive and significantly better suited to a full team environment rather than a single developer working alone.
The Four Capabilities That Make It Different
Multiplayer by Design
Every other AI tool on the market is fundamentally single-player. One person, one session, one conversation. The output belongs to whoever asked.
Claude Tag is built for the opposite situation. Within a given Slack channel, there is one Claude that interacts with the entire team. Anyone can see what it is currently working on. Anyone can pick up the conversation from exactly where the last person left off. If a product manager tags Claude to analyze some data and then goes into a meeting, an engineer on the same channel can follow up on the same task thread as if they had been there from the start.
Anthropic describes this as "much more like interacting collaboratively with a teammate" than working within a single chat. That framing is deliberate — the product is designed to feel like adding a team member, not deploying a software tool.
Memory That Builds Over Time
One of the most persistent frustrations with AI assistants in workplace settings is the repetition tax. Every session requires explaining the same context: what the project is, what the team calls things, what decisions have already been made, what the current priorities are. Teams end up spending a significant portion of every AI interaction just re-establishing ground that was covered in the last session.
Claude Tag eliminates this by staying in the channel continuously. As conversations flow through the channels Claude has access to, it builds an ongoing model of the work — the terminology, the team structure, the recurring problems, the decisions made and the reasoning behind them. A team member tagging @Claude two months into using it gets responses informed by two months of accumulated context, not a blank slate.
If granted permission, Claude can also learn automatically from other Slack channels and additional data sources the administrator connects. Critically, it does not report from private channels — context from channels outside its permitted scope stays out of its memory and out of its responses.
This accumulated knowledge is what Anthropic refers to as "tacit knowledge" — the kind of organizational understanding that normally only long-tenured team members carry.
Ambient Initiative
This is the most distinctive feature and the one most unlike anything in the current AI assistant market.
When "ambient" behavior is enabled, Claude does not wait to be tagged. It monitors the channels it is in and proactively keeps the team updated about things it thinks they need to know. If a thread related to an ongoing project has gone quiet without being resolved, Claude follows up. If information from one channel is relevant to a conversation happening in another, Claude surfaces the connection. If a task assigned earlier is approaching a deadline or has encountered an obstacle, Claude flags it.
This shifts Claude Tag from a reactive tool — it answers when called — to a proactive team member that exercises judgment about what information is relevant and when to bring it forward.
The ambient mode is optional and can be disabled. Teams that prefer to control every Claude interaction explicitly can leave it off. But for teams that want Claude to function as a genuine active participant in the channel rather than a passive resource, ambient mode is what enables that.
Asynchronous Autonomous Work
The fourth capability is what makes the 65% code creation statistic possible.
Claude Tag can receive a task and work through it independently over an extended period — hours or days — without requiring the person who assigned it to remain available or check in. It can schedule tasks for itself: if told to generate a weekly report every Monday morning, it will. If given a multi-stage project, it will plan the stages and execute them in sequence.
Anthropic has found this particularly useful for running many instances of Claude in parallel. Rather than one person working with one Claude on one task, a team can delegate ten different tasks to ten different Claude instances simultaneously, each running asynchronously. The team's time shifts from doing the work to reviewing and directing it — a fundamentally different relationship with the work than any previous AI tool has enabled.
What Anthropic Uses It For Internally
The 65% code figure is the headline, but the internal use cases Anthropic describes go well beyond engineering.
Product metrics and data: Teams tag @Claude to pull together metrics, query data sources, and surface trends without needing an analyst to run the numbers manually every time.
Support tickets: Claude works through support ticket queues, categorizing, researching, and drafting responses, allowing support teams to focus on the escalations and edge cases that require human judgment.
Bug investigation: When a tricky bug appears, Claude can trace through logs, review recent code changes, check related issues, and narrow down the likely root causes before a developer even looks at it directly.
These use cases share a common pattern: tasks that require gathering and synthesizing information from multiple sources, that benefit from context the team has already established, and that do not need a human to be present every moment they are running.
How Access and Permissions Work
Anthropic has designed the access model for the enterprise environments where sensitive data and strict controls are non-negotiable.
System administrators define what Claude can access and where. They specify which tools and data sources are available in which channels. This creates what Anthropic describes as separate Claude identities for different purposes.
The isolation between identities is complete in both directions. A Claude configured for sales work will not carry memories or context into engineering channels. Engineers working in their channel will not be able to access sales data or tools through Claude. The boundaries are set by administrators and enforced at the channel level.
An organization running Claude Tag in both a customer success channel and an engineering channel has two functionally separate Claudes — each with deep context about its own domain and no visibility into the other.
Spending controls can be set at both the organization level and the individual channel level, giving administrators fine-grained control over how much compute different teams consume.
Audit logging captures everything Claude has done and who requested each task. For regulated industries or organizations with compliance requirements, this provides the audit trail necessary to demonstrate what the AI did and under whose direction.
Getting Started: Four Steps
For Claude Enterprise and Team customers, the setup is intentionally straightforward:
Starting with pairing Claude Tag with the Slack workspace, then giving Claude access to the tools and data sources the team needs it to work with, then setting a monthly spend limit for the organization, then testing Claude in a private channel before rolling out to the wider team.
For organizations already using the existing Claude in Slack app, migration to Claude Tag is available within a 30-day window. Anthropic is issuing introductory launch credits to eligible Enterprise and Team organizations so teams can try it across the whole company before committing to ongoing usage.
What Claude Tag Means for the Broader AI Assistant Market
Every major technology company has released some version of an AI assistant for workplace productivity. Most of them share the same fundamental architecture: user asks, AI responds, session ends.
Claude Tag represents a genuine architectural departure. The persistent channel presence, the accumulated memory, the ambient initiative, and the asynchronous autonomous execution are not incremental improvements on the existing model — they are a different model.
The closest analogy is not any existing AI tool. It is what you would get if you could hire a team member who reads every conversation in every channel, never forgets anything relevant, does not need to sleep, and can be handed ten tasks simultaneously that it works through in parallel.
The 65% code figure matters not as a benchmark but as evidence of what happens when a team actually works this way over time. Anthropic is not describing a feature they hope will be useful — they are describing the tool they have been working with long enough that two-thirds of their product code now flows through it.
What Is Not Yet Available
Claude Tag launches in beta on Slack only. Anthropic has stated a goal to expand to other platforms where teams work — which tools and on what timeline has not been disclosed. The current availability is limited to Claude Enterprise and Team customers, with no announcement about broader access for individual Pro or free users.
The underlying model is Claude Opus 4.8. Whether Claude Tag will gain access to Fable 5 or future models as they release has not been specified.
Final Takeaway
Claude Tag is the most significant shift in how Anthropic thinks about deploying Claude in workplace settings. Moving from a one-on-one chat interface to a persistent team channel member — one that learns, remembers, acts proactively, and works while the team is focused elsewhere — changes the nature of the relationship between a team and its AI.
The proof that this works is not a benchmark. It is the fact that Anthropic's own product team now creates 65% of their code through a system they built on this exact model. The product they are shipping to Enterprise and Team customers is the product that already runs their own engineering organization.
For teams that have been using AI tools in the standard mode — ask, receive, repeat — Claude Tag is worth understanding carefully. The jump from a tool you query to a colleague you delegate to is not a small one.
