Anthropic Ships Claude Tag - Slack Gets a Teammate

The Claude in Slack app most enterprise teams have been using since 2024 is being retired. Its replacement, which Anthropic launched on June 23, isn't an incremental update - it's a different model of what AI inside a workplace should look like.

Claude Tag gives every Slack channel a shared @Claude identity that carries context across days, works through tasks while you move on to something else, and keeps all its work visible to the whole team. It ships on Claude Opus 4.8, replacing a product that ran on whatever model version happened to be current.

TL;DR

Beta available now for Claude Enterprise and Team customers; old Claude in Slack app retires August 3, 2026
One @Claude per channel, visible to everyone, context persists across conversations and team members
Runs on Claude Opus 4.8 with admin-controlled tool access, per-channel token spend limits, and audit logs
Anthropic's own product team creates 65% of its code via their internal version of Claude Tag

Old App vs Claude Tag

The table below captures what actually changed between the two products:

Dimension	Claude in Slack (old)	Claude Tag
Identity model	Per-user, per-session	Single shared org identity per channel
Memory	No persistent context	Builds over time from channel history
Collaboration	Private per-user	Multiplayer - whole team sees the work
Task execution	Synchronous, wait for response	Async - delegate and return later
Billing	Against individual users	Billed to the org centrally
Ambient behavior	Not available	Optional proactive monitoring
Admin controls	Minimal	Channel-scoped tools, token limits, audit logs
Underlying model	Various Claude versions	Claude Opus 4.8
Status	Retiring August 3, 2026	Beta for Enterprise and Team

That retirement date is the operational detail that matters most for current users. Administrators have a 30-day window from launch to opt in, and Anthropic is offering introductory launch credits to eligible Enterprise and Team organizations.

Claude Tag overview - Slack-native, Team and Enterprise beta, Opus 4.8 Claude Tag is Slack-native, runs on Opus 4.8, and is currently in beta for Team and Enterprise plans. Source: aiandapi.com

What Claude Tag Actually Does

The core mechanic is straightforward: type @Claude in a channel with a description of what you need, and Claude breaks the task into stages, works through them using whatever tools the administrator has connected, then posts results in a thread where everyone can read, follow up, or redirect.

Multiplayer by Design

Cat Wu, Anthropic's head of product for Claude Code and Cowork, drew the distinction clearly: "Claude Code, Cowork, and chat are very single-player, whereas Claude Tag is built to be interactive and multiplayer."

That framing separates Claude Tag from the AI chatbot assistant category. When someone in an engineering channel asks @Claude to draft a migration plan, the whole team sees the staged output, can pick up the thread, or steer Claude toward a different codebase. No context is buried in someone's private session.

Context That Persists

Claude Tag builds working memory from a channel's history and, with administrator permission, from connected data sources across the organization. It learns what the team does, what tools it uses, and what the recurring questions are. Users stop re-explaining context from scratch on every request.

Memory is scoped by channel and by administrative permission. A Claude instance configured for engineering can't read from a legal channel, and vice versa. Administrators define which tools, codebases, and data sources each instance can reach - separately per department if needed.

Four operating modes: channel mention, ambient, async tasks, direct message Claude Tag operates in four modes - direct channel mention, optional ambient monitoring, async long-running tasks, and private direct message. Source: aiandapi.com

Ambient Mode and Async Tasks

Ambient mode - One of two opt-in behaviors that extend Claude Tag beyond direct @mentions: when enabled, Claude monitors the channel and posts when something needs attention, whether that's a thread that went quiet, a finished deploy, or a decision waiting on a call. No tag required.

Async tasks - For complex multi-step work, Claude schedules its own follow-ups and works over hours or days, returning with results rather than requiring the user to wait in real time. This is close to true task delegation, and it runs under the same admin-controlled permission scope with per-channel spend limits capping how much any instance can consume.

The Numbers Behind the Pitch

Anthropic didn't ship Claude Tag based on theory. Its own product team has been running an internal version, and the number they chose to publicize is striking: 65% of the code changes on Anthropic's product team are now produced by that internal deployment. Usage has spread beyond engineering into product metrics tracking, support ticket triage, and bug diagnosis.

65% of Anthropic's product team code is now created by their internal Claude Tag deployment - including, they say, most of the code that built Claude Tag itself.

The broader market context explains why Anthropic is betting on an enterprise-first product right now. According to Ramp's May 2026 AI Index, Anthropic surpassed OpenAI in business adoption for the first time, reaching 34.4% of firms against OpenAI's 32.3%. The engine behind that gain was mainly Claude Code - the agentic coding product that has become the fastest-growing tool in Anthropic's history. Claude Tag is the natural extension of that momentum from individual developer use into team-level coordination.

Rob Seaman, EVP of Slack, positioned Salesforce's stake plainly: "Slack is the only layer in the AI stack where teams work together."

Anthropic filed confidentially for an IPO in June and raised its Series H at a $965 billion post-money valuation. Claude Tag's ability to create measurable, org-wide productivity data is exactly the kind of enterprise adoption story that supports that valuation in public market terms.

What It Doesn't Tell You

Claude Tag's launch materials are polished, but several details are missing.

No Pricing Beyond Credits

Anthropic mentions introductory launch credits for eligible Enterprise and Team organizations but gives no per-token, per-seat, or per-channel pricing for Claude Tag beyond standard Claude Enterprise terms. For budget planning, that's a gap - especially given that ambient mode and async tasks produce token spend without anyone explicitly requesting work.

No Independent Performance Data

The 65% code-generation figure at Anthropic is internally measured and self-reported. There are no independent benchmarks, no latency numbers for channel-level tasks, and no comparison to what teams using GitHub Copilot Enterprise or Microsoft Copilot for Teams are seeing on equivalent workloads.

No Clarity on Context Limits

How much channel history does Claude Tag actually index? What happens when a channel has three years of messages? The announcement says Claude "builds context from channel history" but doesn't define what that context window looks like at organizational scale.

Comparison: Claude in Slack (old) vs Claude Tag (new) across identity, memory, collaboration and more Feature-by-feature comparison showing the shift from per-user session model to shared org identity with persistent memory. Source: buildfastwithai.com

The Real Competition Is Larger Than Slack Bots

Anthropic positions Claude Tag against its own previous Slack app. The actual competition is: Microsoft Copilot for Teams and Copilot Studio, Salesforce Agentforce inside Slack, and enterprise knowledge layers from Glean, Databricks, and Snowflake - all of which also claim to learn company context over time. Claude Tag's strength is the underlying model capability; its weakness is that it's still in beta without customer-verifiable performance data.

For teams on Claude Enterprise or Team plans, the migration decision isn't optional for long - the old app retires August 3 regardless. The question is whether ambient mode, async task scheduling, and cross-channel memory hold up at scale after the launch credits run out. Anthropic's own product team is the only reference customer right now.

Sources: