Anthropic 发布了 Claude Tag，允许用户在 Slack 频道中 @Claude 并分配任务，Claude 将异步执行并回复结果。Andrej Karpathy 称此为 LLM 交互的第三次重大设计，但引发争议。关键在于云端 AI 接入公司系统后的开箱即用体验，而非 Slack 本身。
SynthePulse Insight · AI deep reading
Version 1 · 2 sources
When @Claude becomes a permanent member of a Slack channel, AI upgrades from personal assistant to team collaborator, but out-of-the-box integration, enterprise governance, and compliance issues will determine whether this paradigm can take hold.
In June 2026, Anthropic launched Claude Tag in beta for Claude Team and Enterprise users. The feature allows admins to configure the channels, tools, data sources, and code repositories Claude can access, after which any channel member can @Claude to assign tasks just like @colleague. Claude breaks down tasks in the background, executes asynchronously, and returns results in threads. It uses the Opus 4.8 model and replaces the old Claude Slack app; admins have a 30-day migration window, and accounts will receive startup credits.
On the day of release, Andrej Karpathy called this the third major redesign of LLM interaction: the first generation was websites (ChatGPT web), the second was desktop apps (Codex App, Claude desktop, Cursor), and the third is the persistent, asynchronous entity with organizational tools and context represented by Claude Tag. He believes that once the underlying integrations (tools, compute, permissions, memory) are in place, Claude becomes like a seamless team member.
Karpathy's post drew polarized reactions. Some see it as mere overhyping of a Slack bot, while others believe he captured the paradigm shift. After speaking with Anthropic insiders, Gergely Orosz pointed out that the focus is not Slack, but the out-of-the-box access of cloud AI to internal company systems—Slack is just the entry point, behind which is a combination of cloud execution environment, persistent memory, tool integration, and organizational permission controls.
Orosz cited an example of a startup that spent months building a similar system: by @mention in Slack, they could launch a cloud development environment and automatically connect internal tools. The company called it an 'absolute game changer.' The real beneficiaries are new hires, non-engineers, and developers who need to modify unfamiliar codebases, as they no longer need to configure local environments. However, integration remains the core challenge, and more vendors will follow with the combination of 'cloud development environment + agent + integration + Slack entry point.'
Claude Tag comes with enterprise-grade controls: admins can set channel-level tool and data source access, token consumption limits (at enterprise and channel levels), and view request logs. Different teams can have independent Claude identities, context is not shared across channels, and Claude does not access private channels. These designs partially address data isolation issues but do not eliminate the risk of sensitive data exposure in shared channels.
Australia ranks fourth in Anthropic's global AI adoption index (3.27 times expected) and accounts for 5.4% of global Slack enterprise deployments (fourth place), making it a key market for this product. However, Australia is undergoing privacy law reform, and regulated industries (finance, healthcare, critical infrastructure) need to assess whether Claude Tag creates new exposure points in shared channels. In particular, 'ambient mode' allows Claude to proactively flag information and follow up on unresolved issues, which changes the compliance risk from AI that only responds to requests.
Claude Tag is not without rivals. GitHub Copilot already supports @GitHub in Slack to trigger coding agents, OpenAI Codex is building cloud-based asynchronous execution capabilities, and Salesforce naturally holds the entry point as the owner of Slack. Claude Tag's differentiation lies in the combination of channel-level shared identity, persistent memory, and asynchronous execution, but there is still a gap between the promise of 'integration' and 'just works.' Whether Anthropic can make it out-of-the-box for enterprise customers, as that startup achieved after months of effort, is key to fulfilling Karpathy's vision.
Cat Wu, Anthropic's product lead, told Reuters that the ability to treat @Claude like a colleague significantly changes interaction dynamics. Slack GM Rob Seaman described the shift as 'making AI a multiplayer game.' In the future, AI may transform from a personal productivity tool into a shared work resource, and the economic model may shift from per-user pricing to per-team shared agent pricing. However, the impact on employment, job design, and team structure remains uncertain, and enterprises need to establish repeatable governance frameworks to handle this standard feature.
This article synthesizes information from Anthropic's official release, X posts by Karpathy and Orosz (both citing insiders), TechRepublic reporting, and third-party data (Visual Capitalist, TechnologyChecker). Product features, admin controls, Australia rankings, etc., are confirmed facts; views from Karpathy and Orosz are cited as source claims; competitive landscape and governance risks are analytical inferences based on available information.
Claude Tag represents a paradigm shift of AI from personal tool to team collaborator, but its success depends on whether enterprises can balance out-of-the-box integration, data governance, and compliance frameworks. Early adopters need to be wary of the 'integration trap' and establish reusable AI governance strategies.
Primary report
Primary source
Also covered by 1 sources