GUIDES
Claude Just Joined The Team 📚
Anthropic turned Claude from a personal assistant into an actual teammate that lives in your Slack. Here is what it does, why it acts as itself instead of as.
Source: Anthropic's @Claude in Slack page and the agent identity and access model write-up →
The Short Version
Up to now, AI at work has been a private thing. You opened a chat, you typed, it answered, and nobody else on your team ever saw it. Helpful, but lonely. It only ever knew your corner of the work.
This update flips that. With @Claude in Slack (the feature people are calling Claude Tag), Claude stops being your private chat and becomes something the whole team can talk to in the same place they already work. You mention it in a channel, it reads the conversation everyone already shares, and it jumps in to help right there.
That sounds small. It is not. Moving AI out of a private window and into the shared channel changes what it can actually do for a team, and it forced Anthropic to rethink something most people never think about: who exactly the AI is allowed to be. Let's break the whole thing down.
What It Actually Is
@Claude in Slack lets anyone on your team tag Claude directly inside a thread or channel, the same way you'd tag a coworker. The moment you mention it, it does three things:
- It reads the full thread. It picks up the context everyone already shares, so you don't have to re-explain the decision, the bug, or the back-and-forth that led here.
- It actually does the task. Run the calculation, pull the numbers and post the chart, turn a messy thread into a clean list of action items, even draft a pull request from a bug report, without anyone leaving Slack.
- It can stay on watch. Give it standing instructions and it works without being mentioned again. Ask it to watch a channel, run a weekly digest, flag anything urgent, or surface a thread that went quiet.
A few of the things teams are tagging it to do on day one:
Catch Up
Pull the decisions and open questions out of a 90-message thread you don't have time to read.
Pull Numbers
Query the data and post the metric or chart straight into the channel.
Draft a PR
Turn a bug report in the thread into a pull request without opening your editor.
Monitor
Run an automated digest and flag the urgent stuff so nothing slips.
Why It Acts As Itself
This is the part that's actually clever, and it's the part nobody's talking about. Stay with me, because it explains why this is more than a chatbot in a sidebar.
Normally, when AI does something for you, it acts as you. It uses your logins, your permissions, your access. That works fine in a private chat, because there's only one person: you.
But the second you drop AI into a channel with five people, that model breaks. If Claude has to borrow someone's identity to act, whose does it use? There's no single person who'd be the right choice every time. The intern shouldn't get the CFO's access just because they tagged Claude in the same channel.
So in practice, Claude shows up as itself everywhere it works:
- It posts in Slack as the Claude app, not as you
- It opens pull requests as the Claude GitHub App
- It queries your data warehouse under a service account an admin provisioned
An admin decides what that identity can reach: which repos it can read or write, which tools and connectors it can use, and at what level. The clever bit is that access can be scoped per channel. Claude can have read-only warehouse access in a general channel, and write access in the data team's private one. A private channel gets its own distinct Claude identity; public channels share one workspace-level identity.
Why you should care even if you're not the admin: it means a teammate who doesn't personally have database access can still ask Claude for a number in a channel that's been cleared for it, and every action Claude takes is logged under its own accounts. You get help without handing out keys, and a clean trail of what was done.
3 Ways Teams Win
Your team lands on a decision in the thread, and Claude just goes and does it while everyone keeps talking. "Let's ship the fix and post in #launches when it's live." Claude drafts the PR, and posts when the deploy is done. The busywork between the decision and the done handles itself, instead of becoming three follow-up tasks nobody wants.
Before this, even on a Team plan, everyone still had their own private Claude that only knew their slice of the work. Now the whole channel shares one Claude that has the full context and remembers what's happened there. Ask it something and it answers from the team's shared history, not from one person's private chat. That's the difference between five separate assistants and one actual teammate.
Because Claude acts on its own identity, it doesn't need you online to keep going. Give it a standing instruction, like watch this channel and run the weekly digest, and the work keeps moving even when the whole team is offline. Reliable AI tasks are getting longer every few months, so an agent that runs while you sleep is starting to matter a lot more than it used to.
How To Start
Right now @Claude in Slack is in beta for Claude Enterprise and Team plans. An admin adds the integration to your Slack workspace, sets what Claude is allowed to reach, and then anyone in an enabled channel can tag it.
Once it's in, here are real prompts you can paste into a Slack thread to feel it out. Notice the pattern: you're talking to it like a coworker, not writing a command.
Tag it on a noisy thread
Make it do the task, not just summarize
Give it a standing job (set and forget)
This isn't a better assistant. It's the first time your whole team has the same teammate, one that reads the room, does the work, and keeps going after everyone clocks out. Start small, point it at one noisy channel, and let it earn its spot.
The real unlock of AI agents is not full autonomy, but human-in-the-loop systems design. Structure your agent inputs with explicit constraints, negative prompts, and automated test checkpoints. When building tools, keep token consumption lean by caching system prompts and isolating tasks into specialized sub-agents.