Claude Tag, explained — what Anthropic's Slack agent does, who can use it, and the 65% claim worth reading twice

On June 23, 2026, Anthropic launched Claude Tag — Claude embedded in Slack as a shared teammate you summon by tagging @Claude in a channel. Hand it a task in plain English and it breaks the request into steps, runs them with your connected tools and data, and reports the result back in the thread. It’s not the old 1:1 Slack bot wearing a new name; it’s a different shape, and it shipped with a stat designed to make you stop scrolling.

This is the practical version: what Claude Tag actually is, who can turn it on, how to read the “65% of our code” claim, how it differs from the things you already have, and whether it’s worth it.

The 30-second version

1. What it is: a shared Claude teammate inside a Slack channel. Tag @Claude, describe a task, and it plans, executes with connected tools, and answers in-thread. One instance per channel, visible to everyone — not a private chat.
2. The catch: it’s a beta, Enterprise and Team only. Pro and Free users don’t have it. It replaces the older “Claude in Slack” app, with a 30-day migration window.
3. The headline stat: Anthropic says 65% of its product team’s code comes from an internal version. Real signal, but read it twice — it’s an internal build, self-reported, with no methodology.
4. The new trick: an “ambient mode” that works proactively and asynchronously — surfacing info, chasing stalled threads, tracking tasks over hours or days without being pinged each time.
5. Worth it when your team already coordinates in Slack and wants an agent that acts on tasks there. Less so if you’re solo, not on a team plan, or wary of an always-on bot reading your channels.

What Claude Tag actually is

A normal Slack bot answers when spoken to. Claude Tag is built to do the task. When you tag it, it decomposes the request, uses whatever tools and data it’s connected to in order to work through the steps, and delivers a finished result back into the Slack thread — closer to handing work to a teammate than querying an assistant.

Three properties make it more than “Claude in a chat window”:

• It’s shared, not 1:1. A single Claude instance serves the channel and is visible to every member. Anthropic’s framing is a single Claude “identity” across the organization, so teammates can collaborate with the same agent and even hand off half-finished tasks to one another.
• It accumulates context. Rather than starting cold each time, it builds understanding from the channel’s ongoing activity — the thread history is the memory.
• It can act on its own. With ambient mode enabled, it proactively jumps in: surfacing relevant information, following up on forgotten threads, and scheduling work for itself that runs asynchronously over hours or days.

Who can use it

At launch: a beta for Claude Enterprise and Team customers. That’s the line to internalize before you go looking for it — individual Pro and Free users don’t have Claude Tag yet, and Anthropic hasn’t published standalone pricing for it.

It also replaces the older “Claude in Slack” app, with a 30-day migration window for teams moving over. If your workspace already had the previous integration, this isn’t an addition so much as a swap to a more capable, more autonomous successor.

The 65% number, read twice

The launch’s attention-grabber is Anthropic’s claim that 65% of the code on its product team comes from an internal version of the tool. It’s a genuinely interesting data point, and it’s also exactly the kind of number bestagent readers should slow down on.

What it does not say, on inspection:

• It’s Anthropic’s own claim, not an independently verified figure.
• It refers to an internal version — the build Anthropic runs on itself, not necessarily the beta a customer gets.
• There’s no timeframe and no methodology: “65% of the code” by what measure, over what period, counting what as “code,” is unstated.

So don’t read it as “Claude Tag will write two-thirds of your code.” The real signal is different and arguably stronger: Anthropic dogfooded this in production internally, for months, before anyone else got access. That’s a credibility marker about the product’s maturity — not a forecast of your output. Read the percentage as a flex, and the dogfooding as the actual evidence.

How it’s different from what you already have

The names in this space collide, so here’s the map. Claude Tag is a collaboration surface; it’s not the old chatbot, and it’s not Claude Code.

Short version: if you want an agent that coordinates and acts on team tasks where your team already talks, that’s Claude Tag. If you want agents that write and refactor code across your repo, that’s Claude Code agent teams — a different surface for a different job.

The controls — and the risks coverage skipped

The autonomy is the selling point and the thing to scrutinize. An always-on agent that reads channel activity and acts with connected tools is a real expansion of what a bot can touch, so the admin controls matter:

• Per-channel tool and data access — you scope what each channel’s Claude can reach.
• Identity separation by channel — a sales-channel Claude doesn’t share memory with an engineering one.
• Token limits and action logs — you can cap spend and review who triggered what.
• Privacy boundaries — it won’t carry private-channel data across channels.

The launch coverage was thin on the flip side, so here it is plainly. An ambient agent reading your channels broadens the permission and data-exposure surface — the controls exist precisely because the default reach is wide. Consumption is metered, and an always-on agent that schedules its own work is a way to spend tokens you didn’t explicitly request, which puts this squarely in the enterprise AI-spend conversation. Ambient mode can become noise — proactive follow-ups are useful until they’re a bot interrupting every stalled thread. And a single shared “identity” raises a quiet accountability question: when the channel’s Claude does something wrong, the action log tells you who triggered it, but the agent is shared. None of these are dealbreakers; all of them are worth deciding on before you flip it on org-wide.

What would change this page

The obvious one: broader availability. Today’s Enterprise/Team gate is the headline limit; if Claude Tag reaches Pro or individual plans, the “who can use it” answer flips for most readers. Also worth watching: standalone pricing (unannounced), how ambient mode behaves at scale once teams have run it for a few weeks, and whether the 65% claim gets any independent corroboration. We’ll update when the beta moves.

Companion reading

• Claude Code Artifacts, explained — the other June Team/Enterprise launch, same gating pattern
• Claude Code agent teams: worth the tokens? — the code-side cousin of a shared agent
• The enterprise AI-coding bill came due — why an always-on, metered agent is a spend question
• Best MCP servers for Claude Code — the connected tools an agent like this acts through
• Running a fleet of long-lived agents — the asynchronous, long-running pattern ambient mode borrows

Sources

1. Anthropic releases Claude Tag, a virtual employee that works within Slack — Fortune
2. Anthropic’s Claude Tag is learning your company, one Slack message at a time — TechCrunch
3. Claude Tag embeds Anthropic’s AI in Slack, already writes 65 percent of internal code, company says — The Decoder

FAQ

What is Claude Tag? A shared Claude teammate inside Slack, launched June 23, 2026. Tag @Claude in a channel, describe a task, and it plans, executes with connected tools, and reports back in-thread. One instance serves the whole channel and builds context from its activity — not a 1:1 chatbot.

Who can use it and what does it cost? A beta for Claude Enterprise and Team customers only; Pro and Free users don’t have it. No standalone pricing announced. It replaces the old “Claude in Slack” app, with a 30-day migration window.

Does it really write 65% of Anthropic’s code? Anthropic says 65% of its product team’s code comes from an internal version — its own claim, an internal build, no timeframe or methodology given. The real signal is that Anthropic dogfooded it in production for months before release, not the percentage itself.

How is it different from the old Claude in Slack? The old app was a private 1:1 chatbot. Claude Tag is a shared, channel-scoped teammate that persists context, executes multi-step tasks, and can act proactively via ambient mode. It replaces the old app.

How is it different from Claude Code agent teams? Claude Code agent teams write and refactor code across your repo from the terminal. Claude Tag is a single shared agent in Slack for team tasks and coordination. Different surface, different job.