Anthropic Releases Major Update: Introducing the Evolved Claude Tag with CC; Karpathy: “The Third Interaction Revolution for LLMs Has Arrived”

Author: QbitAI

Anthropic has launched a new product called Claude Tag.

Its usage is simple: @Claude in Slack and then move on to other tasks.

Claude Tag enables Claude to join Slack as a team member, with access to channels and tools you specify. When you @ it with a task, it decomposes the task into steps and executes them sequentially—then replies with results directly in the thread. It can write code, submit pull requests, perform data analysis, troubleshoot production issues, and more.

Anthropic describes this as an evolution of Claude Code—more proactive and better suited for team collaboration.

Internal data already reveals key insights: 65% of Anthropic’s product team code is now generated by its internal version of Claude Tag. Its use cases extend beyond coding—to tracking product metrics, handling support tickets, and diagnosing elusive bugs.

Key Features

Compared with prior experiences, this update delivers four core advantages.

First, multi-user collaboration. In a single channel, everyone interacts with the same Claude instance. All members can clearly see its progress—and any team member can pick up where the previous person left off. This fully replicates real-world collaborative workflows with human colleagues.

Second, continuous learning. Claude follows conversations end-to-end within the channel, accumulating contextual knowledge over time. Users no longer need to re-explain background details each time. With proper authorization, it can also automatically absorb knowledge from other channels and data sources—acquiring the tacit knowledge required to excel at work. The system strictly prohibits access to information from private channels.

Third, proactive engagement. Once ambient mode is enabled, Claude becomes highly proactive. It extracts potentially critical information from various tools and channels and surfaces it to you—even proactively following up on neglected channels or stalled tasks.

Finally, asynchronous operation. After assigning a task to Claude, you’re free to focus elsewhere. It self-schedules and autonomously advances projects over hours—or even days. Anthropic employees find this capability exceptionally powerful: they now routinely dispatch multiple concurrent tasks to different Claude instances. For sensitive work, users may also send direct messages to invoke personal, dedicated tools.

Permission Management

Claude Tag’s permission model isolates access by channel. Administrators define which tools and data Claude may access in each channel; memory and context are strictly siloed across channels. For example, the Claude instance in the Sales channel won’t share information with Engineering—and engineers won’t see sales data or tools.

Administrators can set token usage caps at both the organization and channel levels—and audit all Claude operations via logs, including who initiated each task.

Claude Tag currently runs on Opus 4.8.

How to Get Started

Claude Enterprise and Team users can begin using Claude Tag today in beta.

Setup steps include: connecting Claude Tag to your Slack workspace, integrating required tools, setting monthly token limits, and validating functionality in private channels.

Claude Tag will replace the existing “Claude in Slack” app. Administrators of existing users have 30 days to migrate. Anthropic will distribute launch credits to qualified Enterprise and Team organizations for company-wide trials.

Official documentation and the product page are available at claude.com/docs/claude-tag/overview and claude.com/product/tag, respectively.

What Andrej Karpathy Thinks

Andrej Karpathy gave this update high praise.

This represents a new paradigm for interacting with Claude. Once the underlying engineering is solid, Claude essentially joins your team as a full participant—you converse with it just as you would with a human colleague, delegating diverse work tasks.

He frames LLM interaction in three stages: Stage One—LLMs are websites you visit; Stage Two—they’re applications installed on your computer; Stage Three—the current stage—LLMs evolve into independent, persistent, asynchronous entities equipped with organizational tools and context, working side-by-side with human teams.

Source: AI Cambrian Explosion

Just now, Claude Code received a major upgrade.

Anthropic officially launched its new enterprise collaboration tool:

Claude Tag.

The official positioning is that this is an evolution of Claude Code—more proactive and more adept at team collaboration.

Anthropic revealed that approximately 65% of its product code is now co-developed with Claude Tag.

Kapasi, who recently joined Anthropic, was quick to endorse the product, calling it exceptional:

“This marks the third major shift in LLM user interfaces. First came web-based chat, second desktop applications—and now, LLMs become independent, continuously running systems with organizational tools and context, collaborating alongside human teams.”

While the official framing positions Claude Tag as an “upgraded Claude Code,” its implications run deeper.

It appears instead to be Anthropic’s strategic attempt to embed Claude deeply into organizational context, enterprise knowledge, and operational workflows.

Unlike traditional AI assistants designed primarily for individual use, Claude Tag’s core value proposition centers on team collaboration.

Claude doesn’t merely passively respond to @mentions in group chats—it can actively participate in team workflows once relevant modes are activated.

For example, in Slack, simply @Claude and state your request: it will decompose the task into discrete steps and execute them sequentially using its integrated tools.

Specifically, it handles PR submissions or merges, performs data analysis, assists with issue resolution—and replies with results directly in Slack upon completion.

Currently, Claude Tag only works with Opus 4.8; Fable 5 remains unannounced.

This direction isn’t surprising: earlier efforts—from Lobster and “digital employees” to Feishu’s AI assistant—have all pointed toward similar goals.

Yet Claude’s model capabilities still make this upgrade especially anticipated.

Are you ready to welcome an AI colleague who collaborates seamlessly across your workflow?

Are Advanced Teams Already Adopting Claude?

Overall, Claude Tag resembles many agents active in chat apps—but Anthropic has embedded it directly into Slack’s native workflow.

Users can @Claude directly in channels or threads, turning it into a shared AI assistant for the entire team.

It sees not only the current message but comprehends accumulated background and context from the full conversation history.

Whether analyzing data, submitting tickets, summarizing meeting notes, or distilling long chat threads into actionable plans—just delegate it to Claude.

But what truly distinguishes Claude Tag isn’t conversational ability—it’s beginning to operate like a genuine team member.

You can assign both immediate tasks and long-term responsibilities.

Examples include monitoring specific channels continuously, auto-summarizing weekly progress, flagging urgent items, or sending scheduled reminders to responsible stakeholders.

Once tasked, Claude persists—even across hours or days—then proactively @mentions you upon completion for review.

When integrated with Claude Code, it can even translate development requests from Slack into concrete engineering tasks and sync results back to the original channel.

As noted earlier, the biggest difference from traditional AI assistants is that Claude Tag isn’t a personal chat window per user—it’s a single, shared Claude instance for the entire channel.

If Zhang San assigns a task to Claude, Li Si joining later can immediately see progress and continue; Wang Wu joining afterward grasps the full context.

Everyone collaborates around one shared Claude—not fragmented, isolated contexts.

Over time, Claude accumulates organizational knowledge—understanding project backgrounds, team conventions, tech stack preferences, and collaboration patterns—eliminating repetitive explanations.

In its official blog, Anthropic highlights four core capabilities: shared context, persistent memory, proactive intervention, and asynchronous execution.

Most intriguing is proactive intervention (Ambient Mode).

Once enabled, Claude stops waiting for prompts—and starts surfacing insights unprompted.

It flags overlooked critical discussions, follows up on long-unresolved issues, highlights decision points, and proactively notifies teams when relevant information emerges.

In effect, Claude begins exhibiting “proactive work” behavior.

Meanwhile, asynchronous execution makes it function more like a true agent.

After task assignment, users can leave Slack entirely—Claude autonomously schedules, executes, and reports back upon completion.

In Anthropic’s demo, engineer Nadia proposes adding a cadence picker feature in the #product-eng-launches channel. Claude instantly analyzes the codebase and delivers a solution—all within the Slack thread, visible in real time to the team.

Viewed holistically, Claude Tag transcends being a mere chatbot—it evolves into a unified enterprise entry point.

People reach out to Claude, and Claude orchestrates GitHub, Jira, Linear, databases, CRM systems, and more.

For employees, the future may mean no longer memorizing dozens of enterprise software logins—just remember one name: @Claude.

As TechCrunch analyzed, Anthropic’s true ambition isn’t just enabling Claude to write code—but empowering it to understand internal organizational knowledge, processes, and collaboration dynamics, embedding itself deeply into enterprise workflows.

This defines the new frontier of enterprise AI competition.

Microsoft has Graph and Copilot; Snowflake and Databricks aim to become enterprise knowledge foundations; Glean seeks to build the intelligent layer connecting models to enterprise data.

Many netizens echo this view—Anthropic is indeed expanding into diverse application scenarios.

What’s ultimately at stake is the hard-to-document yet deeply real organizational knowledge embedded across enterprises.

Deployment

Claude Tag launches first on Slack.

Slack General Manager Rob Seaman stated in his announcement:

“This means ‘AI can be used collectively.’ Human-AI collaboration, previously confined to private chats, now happens transparently in team channels—where all members observe AI’s reasoning, progress, and outcomes.”

On privacy and permissions, Anthropic implemented strict controls.

Administrators decide which tools, data, and channels Claude may access.

Anthropic refers to these as distinct “Claude identities”—

the Claude instance serving Sales won’t retain Engineering information, nor can Engineering access Sales data or tools—all memory and permissions rigorously contained within their respective scopes.

Once permissions are configured, team members can begin using Claude Tag immediately.

Administrators can set token budgets at both organization and channel levels—and audit every action Claude performs, including the initiator of each task.

Claude Tag is now available in beta for Claude Enterprise and Team users.

Enterprises need only complete Slack integration, tool authorizations, and budget setup to begin. Over the next 30 days, Claude Tag will gradually replace the existing Slack-based Claude app.

According to Reuters, Anthropic plans to expand this capability to additional collaboration platforms in the coming weeks.

Every Time Claude Updates, I Think of Fable

Naturally, Claude’s latest update triggered reactions across social media reminiscent of the recent 4o deprecation.

On Reddit, the top comment beneath Anthropic’s official blog post—generated by AI—summarized sentiment:

“Most people say: We don’t care—just bring Fable back.”

Within detailed comments, “Fable’s return” quickly became the dominant theme.

Many Twitter users echoed: “Don’t update randomly.”

Especially without news about Fable…