On 23 June 2026, Anthropic launched Claude Tag, an AI agent that works inside Slack and that you call on the same way you would tag a coworker. It is a small change in form that turns out to be a big change in feel: instead of switching to a separate chat window, you bring Claude into the conversation you are already having, and it gets to work. This explainer walks through what Claude Tag is, how it works, what makes it different from the old Claude in Slack app, how it is governed and priced, and how to start using it.
What is Claude Tag?
Claude Tag is a collaborative AI agent that lives in your Slack workspace. When someone types @Claude in a channel or thread, the agent reads the surrounding conversation, breaks the request into stages, carries them out using whatever tools it has been granted, and posts its work back into the thread for everyone to see. It runs on Anthropic's Claude Opus 4.8 model, and it replaces the earlier "Claude in Slack" app with something far more capable.
The clip below, from Anthropic's announcement, shows the idea in under a minute: tag Claude right where the work is already happening.
Why it matters
Claude Tag is the latest move in a race to put AI directly inside the tools teams already use, where the enterprise market has become a key battleground for AI companies. By dropping the agent into Slack, Anthropic collapses the distance between talking about work and doing it: the bug report, the context and the fix can now live in the same thread. As Cat Wu, Anthropic's head of product for Claude Code, put it, "a lot of the capabilities did exist, but actually the form factor of being able to tag it the same way that you would a coworker is really powerful."
23 Jun 2026
the day Claude Tag launched in research preview for Claude Enterprise and Team customers.
Opus 4.8
the Claude model Claude Tag runs on, Anthropic's most capable at launch.
65%
of Anthropic's own product team code is written by its internal version of the agent.
How Claude Tag works
Under the surface, each tag kicks off a short, agentic loop. Here is the path from mention to result.
-
1
You tag @Claude
Someone mentions @Claude in a channel or thread with a request, exactly as they would loop in a colleague.
-
2
It reads the context
Claude reads the surrounding conversation, and draws on the channel history and approved data sources it has built up over time, so you rarely need to re-explain.
-
3
It plans the work
The request is broken into stages. For longer jobs it can work asynchronously and even schedule tasks for itself, pursuing a project while you focus elsewhere.
-
4
It executes with granted tools
Using the tools, data and codebases an administrator has allowed for that channel, it carries the task out, from running an analysis to opening a pull request.
-
5
It posts the work back
The result lands in the thread where everyone can see it, so the discussion and the output stay in one place and teammates can pick it up.
What makes it different
Three things set Claude Tag apart from a normal chatbot. First, it is multiplayer: rather than a private bot for each person, a single Claude lives in each channel and works with the whole team, so context and progress are shared. Second, it keeps persistent context, building an understanding of the work from the channel history and the data sources it is allowed to see, which means less repetition for you. Third, it can be proactive: with ambient mode switched on, it can flag relevant updates, follow up on unresolved threads, and surface things it thinks you should see, all without being tagged. Ambient mode is off by default, so the agent stays quiet until you decide otherwise.
Claude in Slack vs Claude Tag
Claude Tag replaces the older Claude in Slack app, which retires on 3 August 2026. The table below shows how the new agent differs from what it replaces.
| Factor | Claude in Slack (legacy) | Claude Tag |
|---|---|---|
| Identity | A personal assistant tied to each user | One shared agent per channel that everyone collaborates with |
| Context | Mostly limited to the current conversation | Persistent: builds context over time from approved channels and data |
| Initiative | Responds when prompted | Works asynchronously, schedules its own tasks, and (ambient on) flags updates proactively |
| Tools and data | Limited integrations | Admin-granted tools, data sources and codebases, scoped per channel |
| Governance | Basic controls | Per-channel spend caps, full audit logs, and a scoped agent identity |
| Pricing | Tied to seats | Consumption-based, with organisation and per-channel budgets |
| Status | Retiring on 3 August 2026 | Research preview for Claude Enterprise and Team customers |
Governance, security and pricing
For an agent that can touch real tools and data, control matters as much as capability, and Anthropic has made governance a central part of Claude Tag. Each channel runs the agent under its own scoped identity, and administrators decide which tools, data sources and codebases that identity can reach, channel by channel, so configuration does not bleed between departments. There are token spend caps at both the organisation and channel level, and full audit logging of the tasks it carries out, attributed to whoever requested them.
On cost, Claude Tag uses consumption-based pricing: you pay for usage rather than per seat, with organisation-wide budget visibility and per-channel limits. It is available through Claude Enterprise on AWS Marketplace, where setup involves an administrator provisioning the agent identity in the Claude admin console, which Anthropic says takes about an hour. Eligible organisations also receive an introductory launch credit.
What teams are using it for
Because it can act, not just answer, the early use cases are practical ones. Anthropic and early write-ups point to Claude Tag opening pull requests, running data analysis, and helping with incident response, all from inside the Slack thread where the conversation is already happening. The strongest signal of how far this can go is internal: Anthropic says its own version of the agent now writes roughly 65% of its product team's code. The open question, as some analysts note, is how useful proactive ambient mode proves in practice, since poorly tuned notifications could become noise rather than help, something each team will need to tune for itself.
How to get started
At launch, Claude Tag is in research preview for Claude Enterprise and Team customers. Getting going follows four steps: pair Claude with your Slack workspace, grant it access to the tools and data it should use, set spending limits at the organisation and channel level, and test it in a private channel before rolling it out more widely. Starting small in a low-risk channel, with ambient mode off, is a sensible way to learn how it behaves before you give it more reach. And while it is Slack-only today, Anthropic has said it plans to bring the same experience to other platforms in the coming weeks, so it is worth watching even if Slack is not your main hub.
Key takeaways
- Claude Tag is an AI agent from Anthropic that lives inside Slack. You summon it by typing @Claude in a channel, and it reads the thread, breaks the request into steps, does the work with the tools it has been granted, and posts the result back. It launched in research preview on 23 June 2026.
- It is "multiplayer": one shared Claude lives in each channel rather than a private bot per person, so it builds context over time and teammates can pick up where each other left off. It runs on Claude Opus 4.8.
- It can work asynchronously, schedule tasks for itself, and, with ambient mode switched on, proactively flag updates across connected channels. Ambient mode is off by default.
- Governance is built in: each channel agent has its own scoped identity, admin-controlled tools and data, organisation and per-channel token spend caps, and full audit logging. Pricing is consumption-based rather than per seat, and it is available via Claude Enterprise on AWS Marketplace.
- Claude Tag replaces the older "Claude in Slack" app, which retires on 3 August 2026. Anthropic says it plans to bring the same tag-a-coworker experience to other platforms in the coming weeks.