Claude Tag explained: @Claude as a team member in Slack

TL;DR: Claude Tag is a new Slack app from Anthropic that lets you tag @Claude as a shared team member. One Claude per channel, shared context, asynchronous work, and its own agent identity separate from individual user permissions, under the control of your own admins.

Reality check: your team already works with AI, but probably still in separate chat windows. Everyone has their own Claude conversation, nobody knows what others have already asked, and every task starts again with providing context from scratch. Anthropic has launched Claude Tag, and that's precisely the pattern it breaks.

With Claude Tag, @Claude becomes a team member in Slack: anyone in a channel can tag it, it builds its own context over time, and it continues to work while your colleagues move on to their tasks.

Direct answer: Claude Tag is a new Slack app that lets teams tag @Claude as a shared team member. It builds context across channels, works asynchronously, and has its own agent identity separate from individual user permissions.

What exactly is Claude Tag?

Claude Tag is a new way to work with Claude, launched on Slack. You give Claude access to selected channels and link it to the tools, data, and codebases you choose. After that, anyone in the channel can tag @Claude and delegate tasks while they continue working on something else.

The difference from the existing Claude in Slack app is fundamental. Claude Tag completely replaces that old app and adds three things that elevate it from 'loose tool' to 'team member': shared context, asynchronous work, and an independent identity separate from the people in the channel.

Quick Take: Claude Tag is multiplayer. There is one Claude per channel that interacts with everyone. Colleagues can take over each other's work with Claude without starting from scratch.

In my experience, this is precisely the problem many SMB teams now face with AI adoption: everyone experiments in isolation, no one builds on what someone else has already figured out. Claude Tag structurally solves this by placing collaboration within the existing Slack channel, rather than in a private chat window.

How does @Claude work in practice?

Multiplayer: everyone sees what Claude does

Within a channel, there is one Claude that works with everyone. Colleagues can see what Claude is working on and pick up the conversation where the previous person left off. This feels more like collaborating with a teammate than using a standalone tool.

Learning over time

As Claude follows the channel, it builds more context about the work. You don't have to explain things repeatedly. Claude can even automatically learn from other Slack channels and data sources if access has been granted. It does not report from private channels.

Taking initiative

If 'ambient' behavior is enabled, Claude proactively keeps you informed. It highlights relevant information from the channels and tools it's connected to, and follows up on threads or tasks that have been left unfinished.

Asynchronous work

Give Claude a task and turn your attention to something else. It can even schedule tasks for itself and work autonomously on a project over hours or days. At Anthropic, this has been a direct change: teams now spend much more time delegating tasks to multiple Claudes in parallel.

Quick Take: You can also send direct messages to Claude. In a DM, it works privately, with your own tools and integrations, separate from the shared channel behavior.

See Claude Tag in action

Past vs. Present: from separate chats to a team member

Before: everyone has their own Claude conversation. Context needs to be explained repeatedly. Work happens synchronously: you wait for the answer before moving on. Nobody knows what a colleague has already figured out.

Now: one Claude per channel that builds context over time. Tasks are delegated and handled asynchronously. Colleagues see each other's work with Claude and build upon it. Claude itself follows up on threads that have been left unfinished.

Agent identity: how Anthropic manages permissions

This is the part where most IT managers are rightly cautious: who has access to what if Claude works with multiple people simultaneously? Anthropic has built a separate model for this, called agent identity.

Why 'acting as the user' doesn't work

With a personal AI assistant, you link your own accounts, and the model operates with your permissions. That no longer works as soon as an agent autonomously takes on longer tasks, and as soon as multiple people provide direction in a shared channel. Whose permissions apply then? There is no single correct choice that always works.

Claude acts as itself

In a channel where Claude Tag is active, Claude does not act on behalf of one specific user. It has its own account in every system it touches: it posts messages in Slack as the Claude app, opens pull requests as the Claude GitHub App, and queries your data warehouse under a service account set up by an administrator.

Permissions per channel, not per user

Administrators define a baseline identity at the workspace level, which each channel inherits by default. You can then override this per channel. For example, by giving the engineering channel access to GitHub and the data warehouse, while a CRM integration remains limited to one private channel.

Quick Take: Each private channel gets its own Claude identity. What Claude learns in a private channel never appears in the broader workspace. Public channels share a single, workspace-level identity.

Audit and control

Administrators can set limits for token usage, organization-wide or per channel, and view a log of everything @Claude has done, including who requested each task. Every routine, every memory write, and every network call with agent credentials is logged.

Concrete use cases for SMBs and scale-ups

Investigating bugs and support tickets

Tag @Claude in your engineering channel with a tricky bug, and it will break the problem down into steps, search logs and codebase, and return with a possible root cause. At Anthropic itself, Claude is already tagged to identify the root cause of complex bugs.

Retrieving product metrics and data

Instead of sifting through dashboards yourself, tag @Claude with the question, and it will retrieve the relevant data from the tools it has access to. I use this myself for clients who need quick insights without calling in a data analyst.

Async development for smaller teams

For SMB teams without large engineering capacity, this is a direct way to get more done: give Claude a task in the channel, let it work while your team is on something else, and come back to a completed result in the thread.

Onboarding new team members

Because Claude builds context across channels, a new team member in an existing channel can immediately benefit from what Claude has already learned, without anyone having to explain everything again.

Want your dev team to work agentically with Claude Code and Claude Tag? Discuss what fits your stack and your codebase. → Schedule a call

What Claude Tag DOES NOT do? (transparent section)

Let's be honest: Claude Tag is not a plug-and-play solution that should be given access everywhere without careful consideration.

Claude Tag is only available in beta for Claude Enterprise and Team customers. Smaller teams on Pro or individual subscriptions do not yet have access to this. Anthropic indicates a desire for wider rollout, but a concrete date is not yet available.

Setting up agent identity requires conscious choices from an administrator. Granting overly broad access is a risk. Anthropic themselves advise starting with a basic profile in a few channels, reviewing the audit log, and then expanding purposefully.

Just-in-time credential grants, where a user approves a single sensitive action without permanently expanding the agent's permissions, have been announced but are not yet available. For organizations with more complex authorization structures, this is currently a promise, not a feature.

Pro tip: start small. Give Claude Tag access to one well-defined channel with a limited set of tools, and build from there based on what you observe in the audit log. Don't open everything at once.

My take: this is the next step after Claude Code

Bottom line: Claude Tag is not just another new Slack integration. Anthropic calls it the beginning of an evolution of Claude Code.

What this means for SMB teams: AI collaboration is shifting from 'I use a tool' to 'we work together with a team member who thinks along, remembers, and keeps working while we're away.' That's a different way of organizing work, not just a new button.

No bullshit, just results. Want to know if your team is ready for this way of working with Claude? Read how a company brain gives your team the knowledge of your company, so Claude works with the right context.

Prefer to discuss a safe rollout of Claude Tag in your organization first? Schedule a non-binding consultation.

And do you want your dev team to truly work agentically with Claude Code and Claude Tag combined? Schedule a call about your stack, your codebase and your team.