Anthropic on Tuesday launched Claude Tag, a product that gives enterprise Slack teams a persistent, always-on AI agent that accumulates institutional knowledge over time, works asynchronously on multi-step tasks, and can act without waiting to be asked. Available today in beta for Claude Enterprise and Team subscribers, Claude Tag replaces Anthropic's existing Claude in Slack application and raises the stakes considerably: this is not a chatbot bolted onto a messaging sidebar, but an AI that holds its own service accounts, tracks what a team has decided, and keeps working long after everyone has logged off.

The most striking evidence of its maturity is internal. Anthropic says its own product teams now route approximately 65% of all code changes through an internal version of Claude Tag, drawing on it not only for software development but for tracking product metrics, triaging support tickets, and investigating bugs. That figure comes from Cat Wu, Anthropic's head of product for Claude Code and Cowork, and has been confirmed by multiple independent outlets covering today's announcement.

"Claude Tag is built to be interactive and multiplayer," Wu told Fortune. "When Claude Tag works in a channel, everyone can see it, and everyone can jump in, engage, and steer it in the right direction."

What Agent Identity Actually Means

The part of Claude Tag that deserves the most technical attention is not the feature list — it is the access architecture that makes the feature list possible.

Every prior AI Slack integration, including Anthropic's own previous Claude in Slack app, ran under individual user credentials. When you tagged Claude, it acted under your personal permissions, used your connected accounts, and billed usage to you. That model breaks down immediately in a multiplayer environment: if three engineers and a product manager are all steering the same agent, whose permissions apply? There is no single answer that works correctly across all situations.

Anthropic's answer is what it calls agent identity. In a Claude Tag channel, Claude does not act as any particular user. It has its own service accounts in every connected system: it posts in Slack as the Claude app, opens pull requests as the Claude GitHub App, and queries connected data warehouses under a service account that an administrator provisioned. The result is a clean audit trail — every action Claude takes is attributable to a known service account rather than to whichever employee happened to type the last message. A shared channel can no longer serve as a backdoor into someone's private documents, because no personal user credentials are ever in play.

Credential storage is handled by a component called Agent Proxy. When an administrator connects a tool — GitHub, a data warehouse, a CRM — the credential is stored independently and mapped to that channel's identity. At the moment an outbound request is made, Agent Proxy injects the credential at the network boundary and checks the destination against administrator-configured allow rules. Outbound traffic to any host not explicitly permitted is blocked outright. The entire exchange happens at the network edge; neither the model itself nor the task sandbox ever holds the raw credential.

Four Things That Make Claude Tag Different From Prior Slack Bots

Multiplayer by design: every member of a Slack channel interacts with the same Claude identity. Anyone can see what Claude has been working on, pick up a task someone else started, and continue it without re-briefing the model. The context is not personal — it belongs to the channel.

Accumulating memory: Claude follows conversations as they unfold and builds a continuously deepening understanding of the team's work, vocabulary, and decisions. With administrator permission, it can draw on context from other channels across the organization to fill in facts it needs for the task at hand. Private channels remain strictly isolated and are never visible to other channel instances.

Ambient initiative: when ambient mode is enabled, Claude does not wait to be tagged. It monitors the channels it has been assigned to and intervenes when it judges that a reminder, summary, or piece of cross-organizational context would benefit the team. It also follows up on threads or tasks that have gone quiet without resolution — what The Next Web described as the feature that "separates Claude Tag from a conventional chatbot."

Asynchronous execution: users can assign a task and walk away. Claude works through multi-step requests in stages, using connected tools to execute each one, and posts results back to the Slack thread when complete. Anthropic says its teams spend much of their time dispatching tasks to multiple Claude instances running in parallel. Tasks can be scheduled to run over hours or days, with Claude pursuing projects autonomously over extended periods.

How the Architecture Keeps Data Separate

The question enterprise security teams will ask first is not what Claude Tag can do — it is what it can reach, and whether those boundaries hold.

Channel-scoped identities are the structural answer. An administrator sets up a separate Claude identity for each channel or use case, each with its own tool connections, its own memory, and its own token-spend limit. A Claude configured for sales work has no access to engineering data and cannot seed memories from one channel into another. Administrators can view, edit, and delete channel memory from a centralized audit console, and every scheduled or one-time task Claude performs across the organization appears in a log that identifies which user initiated the request.

Anthropic describes the separation as moving the permission question from "what can this user do?" to "what can this agent do in this compartment?" That reframing is not just semantic. It allows an organization to deploy Claude across departments with fundamentally different data sensitivity levels — legal, engineering, HR — without any risk that a query in one channel exposes information from another.

Rob Seaman, Slack's general manager, described the strategic logic at the launch: making AI "multiplayer" means moving it from a private back-and-forth between one person and one assistant into the open channel where a whole team can see what the AI has been working on and collectively redirect it.

A Crowded Race for Organizational Context

Claude Tag enters a market where multiple well-funded competitors are chasing the same prize. Salesforce, which owns Slack, announced more than 30 new capabilities for its own Slackbot in March — a sweeping overhaul that also runs on Anthropic's Claude. OpenAI introduced Workspace Agents in April, allowing enterprise subscribers to delegate tasks across third-party apps. Viktor, a Slack-native AI coworker startup, raised $75 million in a Series A led by Accel in May. Microsoft has woven GitHub Copilot into Teams.

The logic behind this convergence is straightforward: the average enterprise now uses more than a thousand applications, and employees lose significant productive time to context-switching between them. Every major AI lab has concluded that the right place to intercept that problem is not a standalone tool but the team-chat surface where work is already being coordinated. Claude Tag's bet, as TechCrunch framed it, is Slack-native depth over a horizontal intelligence layer — planting the AI inside the one surface where most knowledge work already happens.

Enterprise IT teams weighing the decision should also consider a second-order implication: as Claude accumulates months of channel context and institutional memory, the cost of replacing it grows substantially. A Claude that has tracked every architectural decision made in the engineering channel over the past six months is considerably harder to swap out than a chatbot that starts from scratch every session. That stickiness is a feature for users who trust the tool and a risk calculus for procurement teams who value flexibility.

What Enterprise Admins Should Know Before Deploying

The initial setup requires an administrator to connect Claude Tag to a Slack workspace, link it to the tools and data sources appropriate for each channel, set a monthly token-spend limit at the organization and channel level, and test in a private channel before rolling out more broadly. Anthropic recommends keeping ambient mode off until the team understands the failure modes — a sensible precaution given that ambient mode gives Claude authority to intervene in conversations without being explicitly invited.

Claude Tag runs on Claude Opus 4.8, Anthropic's most capable publicly available model, which launched in late May with improved agentic coding scores and a higher benchmark on knowledge work tasks. Enterprise customers whose compliance or regulatory requirements are governed by the EU's AI Act should note that the Act's high-risk system enforcement provisions take effect on August 2, 2026 — 40 days from today — though Anthropic signed the EU General-Purpose AI Code of Practice in July 2025 as part of its compliance framework.

Existing Claude in Slack users are not immediately affected by today's launch. Administrators have until August 3, 2026, when Anthropic will automatically migrate the workspace to Claude Tag, to opt in and configure the new system on their own timeline. Anthropic says it is issuing launch credits to eligible Enterprise and Team organizations to allow company-wide trials.

Is an AI That Never Forgets What Your Team Decided What Enterprise Software Was Always Building Toward?

The ambient, always-on AI teammate that Claude Tag describes is not a new idea — it is the fulfillment of a vision that ambient computing researchers and enterprise software architects have been describing for nearly three decades. Mark Weiser's 1988 conception of ubiquitous computing imagined technology that would weave itself into the fabric of everyday life, responding to context rather than waiting for commands. What has changed is that large language models provide the reasoning layer those earlier frameworks lacked: an agent that can follow an engineering thread for a week, understand what was decided and why, and surface the relevant context when a new team member asks a question that has already been answered in a different channel.

What Anthropic is claiming with Claude Tag is that LLMs have now crossed the coherence threshold required to make ambient enterprise AI useful rather than intrusive. Whether that claim holds at the scale of organizations with thousands of employees, sensitive data across dozens of channels, and existing security and compliance obligations is the question enterprise buyers will spend the next several months answering. Anthropic has built the governance architecture — channel-scoped identities, Agent Proxy credential injection, full audit logs — to give enterprise security teams a structural answer to those questions. Whether that architecture performs as documented under production load is the next test.

Frequently Asked Questions

What is Claude Tag and how is it different from the old Claude in Slack?

Claude Tag is an always-on AI agent that lives inside Slack as a shared team member with its own service accounts and persistent channel memory. The previous Claude in Slack integration operated under each individual user's personal permissions — when you tagged Claude, it acted as you. Claude Tag introduces agent identity: Claude acts under an organization-level account provisioned by an administrator, not any single user. That shift makes it possible to have a shared AI teammate with consistent access, full audit logging, and memory that accumulates across the team rather than resetting with every conversation.

How does Claude Tag handle data privacy when it monitors Slack channels?

Each Claude Tag instance is scoped to the specific channels an administrator assigns it. What Claude learns in one channel — including memories and context — never appears in any other channel instance. Private channels are excluded from any cross-channel context gathering. Every action Claude takes appears in a centralized audit log that identifies which user initiated each request. Admins can view, edit, and delete channel memory at any time from the organization settings console.

What happens to the existing Claude in Slack app?

Claude Tag replaces the existing Claude in Slack application. Administrators who want to migrate immediately can do so by opting in through their workspace settings. For organizations that take no action, Anthropic will automatically complete the migration on August 3, 2026. The new identity and billing model moves usage from individual user accounts to organizational billing, meaning each connected tool requires an administrator-provisioned service account rather than individual user credentials.

Does Claude Tag work anywhere besides Slack?

Claude Tag launched in beta exclusively inside Slack on June 23, 2026. Anthropic has said it plans to expand the capability to other platforms but has not disclosed a specific timeline or a list of target platforms as of the launch date.