QWE AI Academy Anthropic dropped Claude Tag on June 23, 2026, and tech press piled on within hours – TechCrunch, Fortune, Engadget all ran it same-day. If you’re on a Claude Enterprise or Team plan, you can install it right now. Before you do: there are two ways to roll this out, and one will hurt you in week three.
Approach A: Blanket-install @Claude across every channel, flip ambient mode on, see what happens. Approach B: Drop it into one scoped test channel, watch the task logs for a week, then expand deliberately.
B wins – not because “go slow” is wisdom, but because Claude Tag builds memory from the channels it sits in. Get it into 40 channels before you understand its behavior, and unwinding that context is a real problem. Start narrow.
What it actually is
Claude Tag is the replacement for the old “Claude in Slack” app. Beta access today for Claude Enterprise and Team customers, running on Claude Opus 4.8. Anthropic plans to expand it beyond Slack in the coming weeks, though no specific platforms or dates are confirmed yet.
Three behaviors – not features, behaviors – define it:
Tagged tasks. Type @Claude in a channel. It breaks the work into stages, posts results in the thread. Anyone on the team can see it, reply to it, or continue from where the last person left off. That’s the core mechanic: one shared Claude identity per channel, not one per person.
Direct messages. DM @Claude and the conversation is private to you. According to Anthropic’s announcement, personal connectors and tools you’ve configured apply here – use it for anything you’d rather not surface in a shared thread.
Ambient mode. That gap between “chatbot you query” and “agent that does things on its own” – this is it. Claude posts on its own: flagging items from across the org, following up on forgotten threads. Per TechCrunch’s launch coverage, it can pull context from channels it has permission to read, including connected tools like Gmail. Configure this last. Seriously.
One thing the docs spell out clearly: Claude doesn’t report from private channels. Whatever’s in #exec-offsite stays there. The qualifier the docs don’t address – what happens to channels you set private after Claude was already invited – is genuinely unclear as of June 2026. When in doubt, remove @Claude and re-add it under the correct scope.
Setup path that actually works
1. Workspace admin opens the Claude Tag install page (linked from Anthropic's announcement)
2. Connect to your Slack workspace
3. Create a SCOPED Claude identity - pick ONE channel
4. Test in that channel before letting anyone tag it
Step 3 is the one teams skip. Here’s why it matters: according to Anthropic’s launch documentation, Claude Tag is designed to create separate identities with separate memories scoped to specific channels. An HR Claude and an Engineering Claude are different entities – different tools, different memory, different context. Try to run one global @Claude across everything and you get the worst outcome: a confident assistant with incoherent context. It will surface roadmap quotes in #random. Someone will ask why. You won’t have a good answer.
Think of it like hiring a contractor who learns your business by sitting in on meetings. Fine for one team. Weird if they’re simultaneously sitting in on every meeting across the company from day one, absorbing everything before anyone’s decided what they should actually know.
Before enabling ambient mode: run two weeks of tag-only usage first. You’ll see which channels generate real signal and which generate noise (hint: #general is almost always noise). Enable ambient only where the signal is high – otherwise you’re burning tokens on Claude commenting on lunch polls.
Cost control: the admin lever
Per-channel token caps. That’s the thing. Anthropic’s docs confirm admins can set token spend limits per channel and per organization, plus a full log of every task @Claude ran and who requested it. The press mostly glossed over this.
Set a low ceiling on #general, a generous one on #eng-review. When a viral “@Claude write a haiku about our CEO” thread burns the cap in #general, Claude stops responding there until the window resets – your engineering channel keeps running. Without this, you will hit that scenario. The launch credit Anthropic is issuing to eligible Enterprise and Team orgs is worth using here: stress-test ambient mode in a noisy channel on the credit, watch the logs, calibrate your caps before real spend starts.
The migration trap (read this twice)
Already running the old Claude in Slack app? The 30-day window is the thing to act on.
Admins can opt into Claude Tag migration within 30 days of launch. The legacy app switches over automatically on August 3, 2026 – with or without your team being prepared. Miss the window and you haven’t avoided the migration; you’ve just done it without a heads-up to anyone.
The actual problem: your team’s existing workflows assume the old behavior. No persistent memory. No ambient interjections. Suddenly on August 3 they get a Claude that remembers what happened last Tuesday and posts unprompted. Slack will look strange that morning. Schedule a 15-minute internal demo before that date.
What nobody’s benchmarked yet
| Claim | Reality check |
|---|---|
| “Full org context, multiplayer” | Context quality matches scope quality. Wrong channel permissions = confidently wrong answers. |
| 65% of Anthropic’s own product team code comes from their internal Claude Tag version | Anthropic’s number, Anthropic’s team, Anthropic’s internal tooling. No third-party benchmarks exist yet for accuracy, latency, or cost per seat (as of June 2026). Don’t quote this to your CFO as a projection. |
| “Works with your tools” | Gmail integration showed up in launch demos. Each connector added is real data exposure. Approve one at a time. |
| Beta status | Memory behavior, ambient triggers, pricing – all can shift without warning. Log everything early so you have a baseline when things change. |
The ambient mode question is genuinely open: is an agent that interrupts with cross-channel context useful or exhausting? For some teams, proactive flagging will feel like having a sharp colleague watching your back. For others, it’ll feel like a coworker who listens in on every conversation and then mentions it in standup. There’s no universal answer – and no published research yet on how persistent ambient agents affect team communication patterns over months, not days.
FAQ
Can I use Claude Tag on the free or Pro plan?
No. Beta access is Enterprise and Team customers only, as of June 23, 2026.
Does @Claude see my private DMs and private channels?
Private channels are excluded – that’s explicit in Anthropic’s docs. DMs to @Claude are private to you. The grey area: channels that were public when Claude joined but later set to private. The docs don’t address this scenario yet. If you’ve changed a channel’s visibility after Claude was added, verify the scope with your admin or remove and re-add @Claude under the intended permissions.
Will it replace Claude Code?
No. Anthropic frames Claude Tag as Claude Code’s multiplayer form – shared context, team handoffs, visible-to-everyone threads. Claude Code stays the single-user, private, deep-session tool. Use Tag when the work needs to be visible and collaborative. Use Code when you want your own space.
Next step if you’re an admin: Install Claude Tag into one private test channel today, set a per-channel token cap before you invite anyone else, and leave ambient mode off for at least a week. The logs will tell you more than any guide can.