How to put Claude
inside your team.
Anthropic just turned Claude from a private chatbot into a shared teammate that lives in Slack. Here is exactly how it works, how to set it up, and how to make your team actually use it.
Until now, using AI at work meant opening a tab, giving it a task, and closing it. Every person had their own private Claude that only knew their corner of the work.
Claude Tag flips that. Instead of a private session per person, one Claude joins a Slack channel as a shared team member — with its own identity, its own memory, and access to the tools and data your admin allows. Anyone in the channel types Claude, hands off a task, and Claude works through it in stages, posting updates in the thread while everyone keeps moving.
It runs on Claude Opus 4.8 and launched on June 23, 2026, in beta for Claude Enterprise and Team plans. Slack is the first home for it, with other platforms planned next.
The old Claude-in-Slack app acted under your personal login and billed to your account, starting cold every time. Claude Tag introduces an agent identity: Claude acts under the organisation's identity within a channel, with access defined by an admin before it ever starts working.
That one change is what makes everything below possible — shared memory, shared context, and clean separation between teams, all controlled at the channel level.
It's multiplayer
One Claude per channel that everyone shares. A teammate can pick up exactly where the last person left off — no re-briefing, no duplicate work.
It learns over time
Claude remembers relevant context from the channels it's in, so you stop re-explaining the same background every week. With permission it can pull from other channels and data sources too — but never from private channels.
It can take initiative
Turn on “ambient” mode and Claude flags things you might need to know, surfaces relevant updates, and follows up on threads that have gone quiet — without being tagged.
It works asynchronously
Set a task and walk away. Claude can schedule work for itself and run a project over hours or days. You spend your time delegating instead of doing.
Setup is an admin job, and it's deliberately short. From your Claude admin settings, an administrator does this once for the workspace:
Pair Claude Tag with your Slack workspace
Connect Claude to Slack from the admin settings. This is the handshake that lets Claude appear as a member your team can tag.
AdminGive Claude access to your tools
Decide which tools, data sources, and codebases Claude can reach — and in which channels. This is where you scope its power. Grant the minimum it needs to be useful, nothing more.
AdminSet a monthly spend limit
Cap token spend for the organisation, and set per-channel limits where it matters. This keeps an autonomous agent from quietly running up cost.
AdminTest it in a private channel
Before turning it loose on busy channels, confirm it works in a quiet one. Check that permissions, data access, and output all behave the way you expect.
AdminOnce permissions are set, everyone can start tagging right away. No per-person setup needed.
The feature is only as good as the habits around it. The teams getting real value treat Claude like a colleague they delegate to — clear ask, right channel, let it run. Here's where it earns its keep:
“@Claude, open a PR that adds rate-limiting to the auth endpoint.” It writes code, raises and reviews pull requests, and helps trace the root cause of tricky bugs. Inside Anthropic, 65% of the product team's code is now created by their internal version of this.
“@Claude, pull last week's signups by channel and flag anything unusual.” Use it to chase down metrics, run analysis, and turn a thread decision into a finished output while the conversation keeps going.
“@Claude, draft a reply to this ticket using our refund policy.” It works through support tickets with the context of the channel, so answers stay consistent with how your team actually responds.
“@Claude, watch this launch channel and summarise blockers each morning.” With ambient mode on, it becomes a teammate that never logs off — surfacing what stalled and what needs a decision before you even ask.
One rule that makes it click: give Claude tasks the way you'd brief a new hire — say the outcome you want in plain terms, point it at the right channel, and let it break the work into stages itself. You can also DM Claude directly for a private answer using your own connected tools.
Claude Tag means giving an AI agent access to your team's data and tools. That's the whole point, and it's also the thing to be deliberate about. Anthropic built the controls; using them well is on you.
Your governance checklist
- Scope channel by channel. Grant access per channel, not blanket. A Claude set up for sales keeps its memory and tools separate from one set up for engineering — they don't share data.
- Start in a private channel. Test permissions, data exposure, and cost in a quiet channel before rolling out widely.
- Set spend limits. Cap tokens at the organisation and channel level so autonomous work can't surprise you on cost.
- Use the audit log. Admins can see everything Claude did and who requested each task. Review it, especially early on.
- Mind private channels. Claude won't report from private channels — but be intentional about which channels and systems it can connect to.