How to Set Up Claude Tag for Your Team (and Keep It on a Leash)

What is Claude Tag, and who can actually get it today?

Claude Tag, launched today, June 23, 2026, is a beta way for Claude Enterprise and Team customers to work with Claude inside Slack. You tag @Claude into a channel, hand it a task, and walk away while it works autonomously on Claude Opus 4.8. One shared Claude per channel serves the whole team.

The one-line version is the one Anthropic uses: it’s “a new way for teams to work with Claude.” Practically, Claude joins your Slack as a team member. Under the hood it breaks your request into stages and executes them in order using whatever tools it has access to.

Here’s who can actually get it: it’s in beta, for Claude Enterprise and Claude Team customers, and it starts on Slack. So if you’re on a free or Pro plan, this isn’t for you yet. If you’re an Enterprise or Team admin, it’s available now.

The part I find most interesting is the multiplayer model. Within each Slack channel, “there’s one Claude that interacts with everyone.” It’s not a private chat tab per person — it’s a shared teammate. Anyone in the channel can see what it’s working on and pick up the conversation from where the last person left off. That’s a genuinely different mental model from the assistant-in-a-sidebar pattern most of us are used to.

The primary source is Anthropic’s announcement, and I’d read it before you touch the admin panel.

My opinion, flagged as opinion: scope @Claude like a new hire, not a chat window. You wouldn’t give a new hire root access to every system on day one. The rest of this post is the practical version of that instinct.

How do you add @Claude to a channel and delegate your first task?

The mechanics are simple: add @Claude to one specific channel, give it a clearly staged task, and walk away. It’ll break the work into steps and execute them with the tools you’ve granted.

Two capabilities make this more than a fancy chatbot. First, persistent learning: @Claude builds context about the work as it follows the channel, so you don’t re-explain from scratch every time. Anthropic says it “can automatically learn from other Slack channels and data sources, IF it’s granted permission” — note the IF, because cross-channel learning requires explicit permission. Second, async autonomy: you can set it a task and leave. Anthropic says it can “schedule tasks for itself, pursuing a project autonomously over hours or days,” and plan tasks for the future.

The multiplayer payoff lands here too. Because there’s one Claude per channel, a teammate can pick up the same in-progress thread mid-task — no handoff document, no re-briefing.

My opinion: pilot exactly one well-scoped task in one channel before you roll this out widely. You learn more from watching a single autonomous run end-to-end than from reading any docs — including mine.

How do you configure the admin access controls (the leash)?

This is the section that matters most, so I’ll be direct. Anthropic gives administrators control over which tools, data, and channels @Claude can access — per channel. That’s the lever. Use it.

Apply least privilege. A channel where @Claude drafts release notes doesn’t need access to your finance data. Admins can also create separate Claude identities for different uses — a tightly-scoped identity for sensitive channels, a broader one for general work. Do that. Don’t run one all-powerful @Claude across the org.

Then set token spend limits, which apply both org-wide and per-channel. This is your brake on autonomous runs that go longer than you expected. If you want to understand what an Opus 4.8 run actually costs before you set those numbers, I broke down the math in how much it costs to add AI to your app. And there’s full activity logging — review what it actually touched.

My opinion, and it’s the one I most want you to take away: the per-channel access control is the only governance lever Anthropic highlights. The questions a security team would ask next — what does it retain long-term, how is it stored, what are the deletion policies — aren’t answered in the announcement. Until they are, your access scoping isn’t just a convenience setting. It’s your entire data-governance posture. Scope tight.

Should you turn on ambient (proactive) mode?

Claude Tag has an optional ambient mode. When enabled, Claude proactively keeps you updated, flags relevant information across the channels and tools it’s connected to, and follows up on threads or tasks that have gone quiet or unresolved. It stops being something you tag and becomes something that’s always watching.

When does that help? Active project channels where a nudge on a stale task is genuinely wanted — a standup channel, a launch war room. There, proactive follow-up is a feature.

The risk is the flip side of the same coin. An always-on agent that reads across channels and speaks up on its own introduces the possibility of misinterpretation, false confidence, and — the one that should worry you most — unintended cross-channel disclosures. An agent that “helpfully” surfaces something from a private channel into a public one is a real failure mode, not a hypothetical.

My opinion: keep ambient mode off by default. Enable it per channel, only where it’s wanted and only where the data is non-sensitive. Proactivity is the highest-leverage and highest-risk setting in the whole product. Earn the trust before you turn it on.

How do you migrate from “Claude in Slack” before it sunsets?

If you already use the existing Claude in Slack app, this isn’t an add — it’s a replacement. Claude Tag replaces it.

The primary fact is the 30-day migration window Anthropic gives admins. Reporting says the old app goes away on or around August 3, 2026 — treat the date as reported and the 30-day window as the firm thing to plan against.

The sequence I’d run: audit how you currently use Claude in Slack, stand up a scoped Claude Tag in parallel, move your workflows over, then retire the old app. Don’t wait for the window to nearly close before you start scoping — scoping is the slow part, not the install.

One bit of forward context: Anthropic plans to expand beyond Slack to “many other places teams work.” So whatever scoping discipline you build now is an investment, not a one-off.

Should you use Claude Tag or build your own Slack agent?

This is the question I get from every engineering team, so here’s my honest framing. The axis is control vs speed.

Claude Tag is the managed path: fast, supported, admin-scoped, running on Opus 4.8. You get a capable Slack teammate today with near-zero engineering lift. What you give up is visibility into and control over retention, storage, and behavior — exactly the governance questions the announcement leaves open.

Building your own with tool/function calling plus MCP is the other path. It’s more work, but you own everything: which tools it calls, what it retains, where data is stored, how it behaves. If you’ve never connected an agent to your own systems, I walk through it in connect an AI agent to your data with MCP, and the managed-vs-roll-your-own tradeoff in AI agents vs automation tools.

For context, this is bigger than Anthropic. Claude Tag is part of an industry-wide push toward “organizational context” — competitors in this space include Microsoft Graph and Copilot, Snowflake, Databricks, and Glean. Everyone is racing to make AI that knows your company.

My opinion: start managed to learn the workflow. Tag @Claude into one channel, watch how your team actually uses an autonomous Slack teammate, and let that teach you what you need. Then build custom when — and only when — you need governance guarantees the managed product doesn’t give you. If you want the deeper engineering playbooks, my AI engineering guides collect them in one place.

FAQ: Claude Tag setup, access, and migration

Who can get it, and is it free? It’s in beta for Claude Enterprise and Team customers, starting on Slack. Not on free or Pro plans.

What model does it run on? Claude Opus 4.8.

Does it replace Claude in Slack? Yes. There’s a 30-day migration window; reporting puts the old app’s sunset on or around August 3, 2026.

Can it see my other channels? Cross-channel and external data-source learning requires granted permission — Anthropic says it learns from other channels and data sources only if permitted.

Can admins cap the cost? Yes — token spend limits apply org-wide and per channel, alongside full activity logging.

Is it Slack only? It starts on Slack. Anthropic plans to expand to many other places teams work.

The bottom line

Claude Tag is genuinely useful: a shared, persistent, autonomous Slack teammate on Opus 4.8, where one Claude serves a whole channel and anyone can pick up where the last person stopped. As a builder who ships agents in production, I don’t dismiss that — it’s a real step.

But there are real open questions, and I won’t paper over them. The governance story — what it retains, how it’s stored, how you delete it — isn’t answered yet, and ambient mode introduces failure modes worth taking seriously. The dogfooding number Anthropic cites is striking — they say 65% of their product team’s code is created by their internal version of Claude Tag — but their internal trust and their controls aren’t yours by default.

So the practical move is the same one I’d make on any agent I put into production: one channel, tight scope, capped spend, ambient mode off until it’s earned trust, and migrate off the old app within the 30-day window. Onboard it like an employee, not a toy. Do that, and Claude Tag is a sharp tool. Skip it, and you’ve handed an autonomous, always-learning agent the keys to your company’s Slack on launch day. Your call — but make it deliberately.