For years the promise of AI at work meant one person, one chat window, one conversation at a time. You got answers, then switched back to Slack to share them with your team. The AI and the team existed in separate worlds.
Claude Tag collapses that gap.
Launched by Anthropic on June 23, 2026, Claude Tag puts Claude directly inside Slack as a persistent team member — one that teammates tag by name, share context with, and delegate to while they focus on other work. It is not a bot that answers questions in a sidebar. It is a Claude instance that follows your channels, remembers what it learns, plans multi-day tasks, and ships output directly into your threads.
The internal numbers are striking: Anthropic's own product team now gets 65% of their code from their internal version of Claude Tag. That figure has spread well beyond engineering — the same tagging pattern now handles product metrics, support tickets, and incident investigation at Anthropic.
What Claude Tag Actually Is
Claude Tag is Anthropic's next evolution of Claude Code. Where Claude Code is a single-user CLI, Claude Tag is designed from the ground up for a full team working together in Slack.
The core mechanic is simple: an admin connects Claude to selected Slack channels and tools (including codebases, data sources, APIs, and external services). Anyone in those channels can then type @Claude followed by a task in plain language. Claude breaks the task into stages, works through them using the tools it has access to, and posts the results back in a Slack thread.
What makes this different from other Slack bots is the architecture underneath:
One Claude per channel, shared by the whole team. There is not a separate Claude instance for each person. There is one Claude in a channel that all teammates interact with. A colleague can pick up a thread you started with Claude and continue it without re-explaining context. Claude remembers what it already knows about the project.
Memory that builds over time. As Claude follows a channel, it accumulates context about the work — the codebase conventions, the ongoing initiatives, the team's preferences. You do not need to explain things from scratch every time you tag it. It can also be granted permission to learn from other Slack channels and external data sources (excluding private channels it has not been given access to).
Proactive ambient mode. When "ambient" behavior is enabled, Claude does not wait to be tagged. It monitors the channels it is in and proactively surfaces information it thinks the team needs — flagging relevant updates, following up on threads that have gone quiet, connecting dots across tools and channels.
Async and long-horizon tasks. Set Claude a task and walk away. It works while you do other things, and can schedule follow-on work for itself over hours or days. Anthropic describes using it to run many Claudes on parallel tasks simultaneously — a fundamentally different way of working than one sequential conversation.
The Four Core Behaviors
1. Multiplayer
In a traditional AI chat, your colleague cannot see what you asked or what Claude answered. Every person starts from zero. Claude Tag makes the AI interaction a team artifact.
Within a Slack channel, there is one Claude that everyone addresses. This means:
- A product manager can tag Claude to draft a spec, then an engineer picks up the same thread to ask implementation questions without re-loading context
- Standup updates Claude gave yesterday are still in the thread — everyone saw them
- When someone goes on vacation, their delegated Claude tasks keep running
This is the "multiplayer" framing Anthropic uses: Claude Tag is designed for collaboration, not solo use.
2. Learns Over Time
Claude Tag's memory is channel-scoped and persistent. As it follows along with a channel's conversations, it learns:
- What the team is working on and why
- The conventions used in the codebase or workflow
- Who owns which areas
- What has already been tried and rejected
This tacit knowledge is what makes Claude useful for more than simple one-off tasks. The longer Claude Tag has been in a channel, the less onboarding each new task requires.
Administrators can also grant Claude permission to pull in context from other channels, documents, or integrated data sources — expanding its working knowledge without manual briefing.
3. Takes Initiative (Ambient Mode)
When ambient behavior is enabled, Claude Tag shifts from reactive to proactive. It monitors what is happening across its connected channels and surfaces things the team should know:
- Relevant information from other connected channels
- Threads that have been unresolved for too long
- Updates from integrated tools (deploys, metrics, alerts)
- Connections between things different teammates have been working on separately
This is closest to how a sharp teammate operates — not waiting to be asked, but flagging things before they become problems.
4. Works Asynchronously
Perhaps the most practically valuable behavior: Claude Tag keeps working after you stop watching it.
Tag Claude with a research task, a codebase investigation, or a multi-step data pull — then go to a meeting. When you come back, the results are in the thread. For tasks that span hours or days, Claude can schedule follow-on work for itself: run the analysis at midnight when compute is cheap, check the deploy logs tomorrow morning, follow up on a ticket if it has not been resolved by Thursday.
Anthropic explicitly mentions using this to run many Claude instances in parallel, which unlocks a qualitatively different way to use AI at work.
How Anthropic's Own Team Uses Claude Tag
The @ClaudeDevs account published the specifics on June 23, 2026: the Claude Code team has been using Claude Tag internally all year, and it now writes 65% of their product team's code — including most of what built Claude Tag itself.
Here are the five workflows they shared directly:
1. Incident Response
When a page lands, tag Claude in the incident Slack thread. It pulls the relevant graphs, diffs the deploy, and comes back with a root cause analysis and the author already tagged. The team approves in-thread. Claude opens the fix, lands it, watches the metric recover, and resolves the page.
The entire loop — detection to resolution — happens inside the Slack thread, with humans approving at the decision points and Claude doing the investigation and the fix.
2. Bug Triage
Let Claude sit in your feedback or bug-report channel and automatically pick up reports as they arrive. It finds the relevant code path, reproduces the issue, runs git blame, writes a fix, and tags the code owner. What is left for the team: code review before Claude merges the PR.
The key word is automatically. Claude is not waiting to be tagged — it is monitoring the channel and acting on signals it decides are actionable.
3. Dependent Work
Hand Claude work that is blocked on something else completing first. The example from @ClaudeDevs: wiring up the frontend once the backend ships to production. Claude waits. It watches the backend deploy. Days later — whenever it actually ships — Claude shows up in the thread with the PR, already adjusted for whatever changed during the backend review.
This is the async scheduling behavior that Anthropic describes as one of Claude Tag's core differentiators. Claude holds context across multiple days and acts when the blocking condition clears.
4. Thread to Postmortem
When an incident wraps, tag Claude to write the postmortem. It reads the incident thread back, rebuilds the timeline from the messages and tool outputs, drops the postmortem into your docs system, and files the action items as issues.
What used to be a 30-90 minute task for an on-call engineer — reconstructing the timeline while still tired from the incident — becomes a tag.
5. Background Watchers
Give Claude a threshold to watch instead of a dashboard to monitor. Example: ping the team when CI stays red for longer than a defined window. Claude stays quiet until the threshold is crossed, then posts in the channel with the failing test and the culprit commit already identified. The team can put up the fix from the same thread without switching contexts.
This is ambient mode made concrete: Claude is not responding to questions, it is watching a condition and notifying when action is needed.
Key Use Cases Beyond Engineering
Claude Tag's strongest workflows are in engineering — that is where Anthropic's own 65% figure comes from. But the same patterns extend across functions:
Data Analysis and Metrics
Connect Claude to your data warehouse or analytics tools and it can pull and interpret product metrics on request, track down anomalies in dashboards, and surface trends without requiring a dedicated analyst for every ad-hoc question. Anthropic mentions "chasing down product metrics and data" as one of the patterns that has spread beyond their engineering team.
Customer Support and Tickets
Claude can sit in a support channel and triage incoming tickets — finding the relevant code path for bug reports, drafting response templates for human review, identifying recurring issues across ticket volume, and following up on tickets that have gone quiet.
Research and Documentation
Tag Claude to summarize a long thread for a teammate who missed context, draft a spec or proposal given a brief and relevant docs, or maintain a living summary of a project's status updated from channel activity.
How to Set It Up
Claude Tag is available today in beta for Claude Enterprise and Team customers. Setup is four steps:
Step 1: Pair Claude Tag with Your Slack Workspace
In the Claude admin console, navigate to the Claude Tag section and connect your Slack workspace. This is an OAuth flow — standard Slack app installation.
Step 2: Give Claude Access to Your Tools
This is where you define what Claude can do. Options include:
- Code repositories (GitHub, GitLab) — for PR creation, code review, codebase exploration
- Data sources (warehouses, analytics platforms) — for metrics and analysis
- External APIs and services — for any custom integrations your team uses
- Slack channels — specify exactly which channels Claude can join and learn from
Each Claude instance is scoped. A Claude set up for sales has no access to engineering tools or data, and vice versa. This channel-level isolation is by design.
Step 3: Set Spend Limits
Configure your organization's monthly token spend limit and, optionally, per-channel limits. This gives you cost control before users start experimenting heavily.
Step 4: Test in a Private Channel
Before rolling out to the full team, add Claude to a private test channel, give it a representative task, and verify it has the right permissions and produces useful output. Then open access to the broader team.
Migration from Claude in Slack
Claude Tag replaces the existing Claude in Slack app. If your workspace already has the previous integration installed, administrators have a 30-day opt-in window to migrate.
Anthropic is issuing an introductory launch credit to eligible Enterprise and Team organizations, so teams can trial Claude Tag without immediately hitting spend limits.
Claude Tag runs on Opus 4.8 — Anthropic's most capable model — compared to whatever the previous Slack integration used.
What This Means for How Teams Work
Claude Tag represents a shift in how AI productivity tools are positioned. Most AI tools today are personal — you use them in your own workflow and share outputs manually with the team. Claude Tag makes the AI interaction itself a shared, team-visible artifact.
This has practical implications:
- Institutional memory — Claude's channel memory means the team's AI interactions accumulate context over time, rather than resetting each session
- Parallel delegation — instead of one person using Claude sequentially, multiple teammates can have Claude working on different tasks simultaneously in the same channel
- Accountability — the admin audit log means every Claude action is traceable to a requester, which matters for compliance and for understanding what the AI has actually done
- Gradual adoption — because Claude lives in Slack where the team already works, adoption friction is lower than standalone tools that require context-switching
The 65% code statistic from Anthropic's own team is a credible signal, not a marketing claim — it is a company publicly committing to a usage pattern they are building on internally.
The Third Paradigm: Karpathy's Take
Andrej Karpathy — AI researcher, former OpenAI and Tesla AI director — posted his reaction to Claude Tag, and it is worth reading carefully:
"This is a new paradigm for interacting with Claude that is significantly more 'inline' with all the other human activity org-wide. Once you do all of the under the hood engineering work to make this 'just work' (e.g. across tools, integrations, compute environments, memory, security, etc.), Claude basically joins the team in a seamless way — you can talk to it as you would talk to a person and it can help with a very large variety of workloads."
He then framed it as the third major redesign of LLM UI/UX:
- First paradigm — The LLM is a website you go to (ChatGPT, Claude.ai — you visit them)
- Second paradigm — The LLM is an app you download to your computer (Claude Code, Cursor, Copilot — local installs)
- Third paradigm — The LLM is a self-contained, persistent, asynchronous entity with org-wide tools and context, working alongside teams of humans
Karpathy's framing is clarifying because it explains what is architecturally new about Claude Tag beyond the Slack surface. The prior two paradigms both treated the AI as something you go to. You context-switch to the chat window, to the IDE extension, to the terminal. The AI waits for you.
The third paradigm inverts this. The AI is already in your environment. It has access to your tools, your channels, your codebase. When you need it, you tag it — the same gesture you use to involve any teammate. When you do not need it, it is still there, watching, learning, potentially flagging things you should know.
Karpathy noted that the underlying engineering to make this "just work" — tools, integrations, compute environments, memory, security — is substantial. The basic idea is easy; a v0 is a hackathon project. A production-grade, enterprise-secure version of it is a different problem. That is what Claude Tag represents.
His bottom line: "it really takes a while to wrap your head around it, but it works and it is awesome."
What to Watch
Claude Tag is launching on Slack specifically because that is where collaborative work already happens and where Anthropic's internal rollout succeeded. But the blog post is explicit that Anthropic plans to expand beyond Slack.
The multiplayer, channel-scoped, tool-connected architecture is the pattern — Slack is the first surface it ships on.
For teams evaluating it now: the strongest signal to watch is how quickly the "learn from channel context" capability matures. Persistent, accurate tacit knowledge about a team's work is what would differentiate Claude Tag from a smarter Slackbot. Early access gives you the chance to calibrate how well it actually accumulates that context in your specific channels.
Getting Started
If your organization is on Claude Enterprise or Team, Claude Tag is available now. Visit anthropic.com to access the setup flow from the Claude admin console.
For organizations not yet on Enterprise or Team plans, Anthropic has not announced a timeline for broader access. Given that Claude Tag is framed as a foundational shift rather than a feature addition, wider availability seems likely to follow relatively quickly — but the beta starts with paying enterprise customers.
The full documentation for Claude Tag, including tool connector setup and admin controls, is available alongside the launch.
Frequently Asked Questions
What is Claude Tag? Claude Tag is Anthropic's new Slack integration that adds Claude as a persistent team member in your workspace. Admins connect it to selected channels and tools, then anyone in those channels can tag @Claude to delegate tasks. Unlike a one-on-one chat, there is one Claude per channel that all teammates share — it builds context from channel history and picks up conversations where the last person left off.
How is Claude Tag different from Claude Code? Claude Code is a single-user CLI tool. Claude Tag is its multiplayer, team-native evolution. It runs in Slack where your whole team already works, is shared across teammates in a channel, builds persistent memory from channel history, can take initiative proactively in ambient mode, and can schedule tasks for itself over hours or days. Anthropic describes it as Claude Code "made more proactive and built to work with a full team."
Who can use Claude Tag today? Claude Tag launched in beta on June 23, 2026, for Claude Enterprise and Team plan customers on Slack. Anthropic plans to expand availability to more plans and platforms in the future. Eligible organizations receive an introductory launch credit to trial it across the company.
Does Claude Tag replace the existing Claude in Slack app? Yes. Claude Tag replaces the previous Claude in Slack app. Existing administrators have a 30-day opt-in window to migrate. The new app runs on Opus 4.8 and brings the full Claude Tag feature set including memory, ambient mode, and async task scheduling.
How does Claude Tag handle privacy and data security? Administrators define exactly which channels and tools each Claude instance can access. Memory and context stay scoped to the channels an admin defines — a Claude set up for sales work cannot see engineering channels. Claude does not report from private channels. Admins can view a full log of every action Claude has taken and set monthly token spend limits per organization and per channel.
Can Claude Tag write and merge pull requests? Yes, if you connect Claude Tag to your code repository and give it the appropriate permissions. Anthropic uses it internally to write and merge pull requests, run data analysis, work through support tickets, and investigate bug root causes. At Anthropic, 65% of their product team's code is now created by their internal version of Claude Tag.