It’s 9:40 on a Tuesday. Someone on your team drops a line in the #growth channel — @Claude can you pull last month's signup numbers, compare them to May, and draft the summary for the leadership update? — and then walks into a meeting. No new tab, no separate app, no hunting for the right prompt. An hour later they come out, and the summary is sitting in the thread: numbers pulled, the month-on-month change called out, ready to edit.
That small moment is the whole idea behind Claude Tag, the team product Anthropic announced on 23 June 2026. It’s the most approachable front door Claude has had yet — and the first one where you don’t open anything. Claude comes to where your team already works.
What Claude Tag actually is
Claude Tag puts Claude inside the place your team already coordinates work: your Slack channels. Instead of opening a chat window and talking to an assistant that forgets you the moment you close it, you tag it — you @Claude it, exactly the way you’d @mention a colleague — and hand it a task.
The word “tag” is doing real work there. You’re not starting a private session you have to manage. You’re flagging a teammate who’s already in the room. That’s the mental shift: Claude Tag isn’t a chatbot you sit and babysit, it’s a persistent teammate you delegate to and then get on with your day. It stays in the channel, it sticks around between conversations, and it’s there the next time anyone needs it.
If you’ve read our explainer on what an AI agent is, this is that idea made concrete and given a desk in your Slack: something you give a goal, not a click-by-click script.
How it works: tag it, then walk away
When you tag @Claude with a request, it doesn’t just blurt back the first thing that comes to mind. It breaks the job into stages and works through them in order — using whatever tools and data it’s been connected to, like your metrics, your documents, or your codebase — then replies in a Slack thread with what it produced and how it got there.
The key word is asynchronously, which is a fancy way of saying: you don’t wait around. You tag it and leave. It works in the background while you’re in a meeting or heads-down on something else, and the result lands in the thread when it’s ready. For anything beyond a quick answer, it can even schedule work for itself and pursue a project over hours or days — checking things off, coming back, picking up the next stage — rather than needing you to nudge it through each step.
That’s the difference between an assistant and a teammate. You don’t manage the steps. You hand over the outcome you want.
One Claude, the whole channel
Here’s the part that makes it feel less like a tool and more like a colleague: within a given channel, there’s one Claude that everyone shares.
It’s multiplayer. When you tag it, everyone in the channel can see what it’s working on. If you start something and step away, a teammate can read the thread and pick up exactly where you left off — ask a follow-up, redirect it, approve the next stage — without you in the loop. Compare that to everyone having their own private AI chat that nobody else can see: the work, the context, and the back-and-forth all live in the open, in the channel, like any other piece of teamwork.
And it gets better at your work the longer it’s around. As Claude follows along in a channel, it builds up context — your team’s shorthand, your conventions, what “the weekly report” actually means to you — so you stop re-explaining yourself from scratch every time. With permission, it can also draw on other channels and data sources you’ve connected, picking up the unwritten, tacit knowledge that makes the difference between a generic answer and one that fits how you actually operate.
Crucially, that memory is scoped. What Claude learns in one channel stays in that channel unless an admin says otherwise — which is what makes the next part, the safety story, hold together.
What you’d actually hand it
Forget code for a second — this is built for the everyday work that clogs up a team’s week. Real examples of what people tag it for:
- Chase down a number. “Pull our active-user count for last week and compare it to the week before.” It finds the metric, does the comparison, and posts the answer instead of waiting on whoever owns the dashboard.
- Work through a support queue. Hand it a batch of tickets and let it draft replies or triage them, so your people spend their time on the hard ones, not the repetitive ones.
- Write the recurring thing. The weekly update, the campaign recap, the “what changed this sprint” note — the report you write every single week becomes a tag-and-edit job.
- Dig into a problem. Even on the technical side, teams use it to trace a tricky bug back to its root cause — the kind of investigation that used to eat an afternoon.
How far can that go? At Anthropic, 65% of their product team’s code is now created by their internal version of Claude Tag. That number is striking, but read it the right way: it’s not “the AI replaced the engineers.” It’s that a capable teammate handles a large share of the mechanical work so the humans spend their time deciding what to build and checking that it’s right. The same shape applies whether your team writes code or campaign briefs.
Ambient mode: when it speaks first
Everything above is Claude responding when you tag it. There’s also an optional ambient mode, where it’s allowed to speak first.
Switch it on and Claude will proactively keep you posted — flagging something across your channels it thinks you’ll want to know, or following up on a thread that went quiet without a resolution. It’s the difference between a teammate who only answers when asked and one who taps you on the shoulder with “hey, this looks like it needs you.” Useful, but it’s opt-in by design — you decide whether you want a colleague who volunteers, or one who waits to be asked.
The guardrails that make it safe to invite in
Inviting an AI into the channels where your real work happens is a reasonable thing to feel cautious about. Claude Tag is built so the cautious answer is the easy one, and the controls sit with an admin, not with every individual:
- An admin decides where it lives and what it can touch. They connect Claude to the Slack workspace, choose which tools and data each channel can reach, and — the recommended first move — test it in a private channel before letting it loose. Access is something you grant deliberately, channel by channel.
- Separate identities, separate memories. You can run different Claudes for different purposes, and their knowledge stays walled off. A Claude set up for sales won’t pass its memories to one set up for engineering, and it won’t hand engineers access to sales data or tools. The compartments are real.
- A spending cap you set. Admins set a monthly limit on how much the organisation spends, so there’s no runaway bill — costs stay inside a number you chose.
- A log of everything. Admins can review what Claude did and who asked for each task. Nothing happens off the books.
And then the house rule that applies to every way you use Claude, not just this one: Claude drafts, a person verifies. It’s brilliant at the first 90% — the pulling, the drafting, the first pass. Anything involving money, personal data, or a message going outside the company should be drafted by Claude and checked by a human before it ships. That last 10% is your job, and it’s the most important 10%. Our team rollout playbook goes deeper on setting those expectations across roles.
How it’s different from chatting with Claude on Desktop
If your team already uses Claude Code on Desktop, you might be wondering where Claude Tag fits. The clean way to think about it: same engine, different room.
We’ve written before about how starting a team on Desktop removes the biggest barrier to AI adoption — the terminal you’d otherwise ask people to learn first. Claude Tag removes the next barrier: now there’s no separate app to open at all. Desktop is where one person sits down to do focused work with Claude — open a folder, work through a problem together, accept changes with a visual diff. Claude Tag is where the whole team delegates in the place they already talk all day.
Most teams will want both, and they don’t compete. One is your workshop; the other is the colleague in the group chat.
Is it for your team yet?
Today, Claude Tag is in beta for Claude Enterprise and Team customers, and it runs on Opus 4.8, Anthropic’s most capable model. For those plans it replaces the older Claude in Slack app, and eligible organisations get introductory credits to try it without committing budget up front.
If that’s you, the sensible first step is the small one: have an admin add it to a single low-stakes channel, connect one or two tools, and tag it for the recurring report nobody enjoys writing. That’s the whole pilot. Watch what it gets right and where it needs a human, then widen it — the same way you’d onboard any new teammate, which is exactly what our rollout playbook is built around.
Step back and the direction is hard to miss. First the AI lived behind a terminal. Then it moved into a normal app anyone could open. Now it’s moving into the channel where your team already spends its day — no app, no install, just an @mention away. The tool isn’t asking you to come to it anymore. It’s showing up where the work already happens.