What Is Claude Tag? Anthropic's Slack Agent, Explained

Claude Tag is Anthropic's next-generation Slack integration. The pitch, in Anthropic's words, is that you can "tag @Claude into a conversation and it takes on real work, using your organization's tools." That last part is the difference between this and a typical Slack bot: Claude Tag isn't just answering questions in the thread — it can reach into the systems your team actually uses (repositories, connected apps, plugins) and do the task.

It's the evolution of the older "Claude in Slack" experience. Anthropic is transitioning Claude in Slack to Claude Tag on August 3, 2026, and it's rolling out in beta on Team and Enterprise plans. For now it lives in Slack.

The mental model: less "chatbot in a sidebar," more "an autonomous teammate with a name, a memory, and access to your tools, who happens to live in Slack."

Claude Tag shows up in Slack in three places:

• Channel tagging — mention @Claude in any channel to hand it a task in front of the whole team.
• Direct messages — DM @Claude for a private, one-on-one conversation.
• AI assistant panel — reach Claude from Slack's AI assistant header.

When you tag it in a channel, Claude's work stays visible to everyone in that channel. That's deliberate: teammates can watch what it's doing, correct course, and build on its output — collaborative steering rather than a black box. A quick first test after setup is /invite @Claude followed by @Claude summarize this channel.

Three properties separate Claude Tag from a command-response bot:

• Its own identity. Claude operates under an organizational identity using service-account credentials, not an employee's personal login. Its access is granted to it, deliberately, by an admin.
• Persistent memory. It builds and retains context across conversations and days, remembering relevant information instead of treating every message as a cold start.
• Agency. With organization-level access to tools and repositories, it can do multi-step work and proactively follow up — checking in on the progress of tasks it was given.

Put together, these make Claude Tag an example of what people are starting to call an "ambient agent": an AI that persists in your workspace and acts, rather than one you open, prompt, and close.

Setup is an admin task — only a Claude organization Owner (or Primary Owner) can complete it, and a Slack workspace admin is needed for the pairing step. It runs as a four-step flow at claude.ai/admin-settings/claude-tag:

1. Pair the workspace. Install "Claude for Slack" from the Slack Marketplace, then a workspace admin sends @Claude connect in any channel to generate a pairing code (valid 15 minutes). Paste it into the dialog and choose whole-workspace (recommended) or a specific channel.
2. Configure access. Build an Access Bundle — a named set of credentials, repository grants, plugins, and instructions Claude uses on behalf of channel members. Apps connect via service-account credentials (not personal logins); GitHub connects via the Claude GitHub App. You can defer connections to after setup.
3. Set spend limits. Cap channel usage at $100, $250, $500, $1,000 (default), Unlimited, or a custom amount up to $1,000,000. (DMs run on individual accounts and bypass this cap.)
4. Review and launch. Confirm and activate Claude Tag for the paired workspace.

Everything stays editable afterward, and a common rollout pattern is a pilot — one bundle on a single test channel — before expanding org-wide. Note that new connections only apply to fresh threads, not ones already in progress.

Claude Tag is consumption-based, and where the cost lands depends on how it's used:

• Channel work (when @Claude is tagged in a channel) draws from the organization's usage balance, governed by the admin-set spend limit.
• Direct messages to Claude bill to the individual user's Claude account and aren't covered by the org spend cap.

This split matters for budgeting: shared, team-facing work is centrally controlled and capped, while private DMs are on each person's own plan. Setting a sensible org spend limit (and starting below "Unlimited") is the obvious guardrail.

Claude Tag is a concrete example of a broader shift in enterprise AI: from assistants you summon to agents that persist and act. The earlier generation of workplace AI was a chat box — you opened it, asked, copied the answer out. The new generation lives where work already happens (Slack, your repos, your tools) and takes action with standing permissions.

The defining ingredients of a workplace agent like this are consistent across vendors:

• An identity the agent acts under, with explicitly granted permissions.
• Tool access — connections to the systems where work gets done.
• Memory so it can carry context across time and tasks.
• Autonomy to take multi-step actions and follow up, within guardrails.
• Governance — spend caps, role-based access, and auditability.

Understanding those five ingredients lets you evaluate any agent product — Claude Tag included — on the dimensions that actually matter. If you're newer to the concept, our explainer on what AI agents are covers the fundamentals.

Claude Tag is well-designed on governance — service-account credentials instead of personal logins, Owner-only setup, role-based access on Enterprise, spend caps, and in-channel visibility for auditability. But an agent with standing access to your tools and repositories, acting autonomously in shared channels, deserves a clear-eyed review:

• Scope of access. An Access Bundle's repo grants and credentials define a real blast radius. Scope bundles to the minimum needed, and prefer per-channel bundles over org-wide ones for sensitive work.
• Data exposure in channels. Anything Claude can read or pull becomes visible where it's tagged. Be deliberate about which channels it operates in.
• Autonomous follow-ups. Proactive check-ins are useful, but "an AI that acts on its own" needs monitoring, especially early.
• Consumption billing. Channel work draws on org balance; without a sensible cap, costs can surprise you. Avoid "Unlimited" until you understand usage.
• Beta maturity. It's in beta — pilot on a test channel, validate behavior, then expand.

None of these are reasons to avoid it — they're the standard diligence for adopting any agent with real-world permissions. The honest framing: Claude Tag is a capable, governable workplace agent, and the work is in scoping it well, not in trusting it blindly.

Claude Tag is a strong fit for teams that already run on Slack and want AI embedded in their actual workflows — summarizing channels, working across connected tools, referencing repositories, and handling recurring tasks without context-switching to another app. Engineering, ops, and support teams with well-defined tools and repos will get the most from it.

It's worth more caution where data sensitivity is high or where you can't yet commit to scoping and monitoring access. The recommended path is the same either way: start with a tightly scoped pilot on one channel, confirm it behaves as expected, then widen access. Since Anthropic is moving Claude in Slack to Claude Tag on August 3, 2026, current Claude-in-Slack users will want to review their setup ahead of that date.