Published on August 12, 2026
Claude Tag allows Team and Enterprise users to bring Claude into Slack conversations, delegate multi-step work and receive completed results without moving the discussion into a separate AI workspace.
Anthropic introduced Claude Tag on June 23, 2026, as a new way for teams to work with Claude directly inside Slack. Instead of copying a conversation into another application and rebuilding its context, a user can mention @Claude in a channel or thread and assign it a task where the work is already being discussed.
Claude can examine the surrounding conversation, use tools and data sources approved for that channel, divide the request into stages and post the result back into the same Slack thread. The Claude Tag announcement presents this as a shift from using AI privately to delegating work to an agent that participates alongside an entire team.
The main value is continuity. The discussion, instructions, progress and final output remain visible in Slack, reducing the need to move repeatedly between workplace conversations and a separate AI assistant.
Claude Tag is designed for more than answering a question in a channel. Mentioning @Claude with a request begins a working session associated with that Slack thread.
For a simple question, Claude may respond directly. For a larger assignment, it can create a checklist, work through the required stages and update that checklist as it progresses. Depending on the task and connected systems, the final result could be a written summary, document, chart, hosted page or pull request.
A team discussing a delayed product launch could ask Claude to review open threads, compare them with a project plan and identify unresolved decisions. An engineering team could ask it to investigate a performance problem, examine monitoring data and prepare a proposed code change. Marketing or sales teams could use it to assemble campaign information, summarize account activity or convert a decision thread into a more structured document.
According to Anthropic’s documentation on how Claude Tag works, each Slack thread receives its own temporary cloud environment. Claude can read relevant conversation context, run code and use the connections assigned to that channel before returning the result to the discussion.
This makes the Slack message the beginning of a workflow rather than the final request in a short exchange.
Claude Tag differs from conventional workplace assistants because the work belongs to the channel rather than to one person’s private chat.
Anyone in the channel can see what Claude is doing, review its checklist and continue the interaction. A colleague can add missing information, redirect the approach or ask Claude to extend the result without requiring the person who started the task to remain involved.
Anthropic describes this as a “multiplayer” form of AI collaboration. Within a channel, the team works with the same Claude identity, the same approved connections and the same shared context. This creates a closer resemblance to delegating work to a visible colleague than privately consulting a chatbot.
The distinction could be useful for accountability. When AI work happens in a private assistant, colleagues may see only the final output and not the assumptions, instructions or corrections that shaped it. Claude Tag keeps more of that process inside the shared conversation.
It may also reduce duplicated effort. A person joining a project does not necessarily need to ask Claude the same questions from the beginning if the earlier work, results and follow-up discussion remain available in the channel.
Claude Tag can use more than the text contained in one message.
It reads the relevant Slack thread and can use channel history, pinned information and workspace search when permitted. Administrators can also connect it to repositories, ticketing systems, cloud documents, data warehouses, monitoring tools and custom services.
The access available to Claude depends on the channel, not on the individual who mentions it. A product channel might receive access to project documents and issue trackers, while an engineering channel could use code repositories and monitoring platforms. A sales-focused Claude identity could be configured with a different set of systems.
This channel-based model allows organizations to create separate working contexts. Information and tools intended for one department do not automatically become available to another.
Claude Tag also builds memory around team activity. Anthropic says information learned from public channels can contribute to workspace memory, while memories created in private channels remain scoped to those channels. Administrators and users therefore need to understand that Claude may preserve selected context beyond one individual task. The purpose is to reduce repeated explanations, but it also makes information governance important.
Direct messages work differently. In a DM, Claude uses the individual user’s own Claude account and personal connectors rather than the tools assigned to a shared channel.
Claude Tag does not require users to watch every stage of a task. A team can delegate an assignment, continue with other responsibilities and return when Claude posts its result.
The agent can also support longer-running and recurring work. Anthropic says Claude Tag can schedule tasks, react to repository events, monitor selected channels or run routines using the same access assigned to the channel.
A project team could use it to prepare a recurring status digest, follow up on unresolved approvals or watch for issues reported in a support channel. An engineering team might connect it to monitoring alerts or repository activity. These workflows move Claude from responding only when mentioned toward supporting work that develops over time.
Anthropic has also described an optional ambient mode in which Claude can proactively surface relevant information or follow up on discussions that appear unresolved. This kind of proactivity may reduce manual tracking, although organizations will need to decide carefully where automated intervention is useful and where it would create distraction.
Putting an AI agent into shared workplace conversations raises different security questions from using a personal chatbot.
Claude Tag begins without access to external company systems. An organization owner provisions its identity, chooses its service accounts and decides which connections apply to specific channels, workspaces or the wider organization.
The agent performs channel tasks inside isolated sandboxes hosted by Anthropic. According to the Claude Tag security documentation, the sandbox does not hold the organization’s credentials directly. Outbound requests pass through a controlled proxy that injects the necessary credentials and blocks access to unapproved destinations by default.
Administrators can also set usage limits and review records showing what Claude did and who requested the work. These controls are important because an agent connected to repositories, customer data or internal systems may be able to take more consequential actions than an assistant limited to generating text.
The system is still not risk-free. Claude may misunderstand a request, rely on incomplete channel context or produce an output that appears more reliable than it is. Connected documents and messages can also contain inaccurate or potentially manipulative instructions.
Teams should therefore review significant outputs and require human approval before Claude sends external communications, changes important records, merges code or completes other high-impact actions.
Anthropic now offers several agentic work environments, but they serve different collaboration patterns.
Claude Code is primarily intended for people working directly with code through terminals, development environments and repositories. Cowork is designed for individual assignments involving a user’s own files, tools and research. Claude Tag focuses on shared work that should remain visible and steerable inside team conversations.
This makes Claude Tag less suitable for private drafting or deeply personal work. Its main strength appears when the project already lives in Slack and several people need to see, contribute to or review the result.
The products may overlap in underlying capabilities, but the interface changes the working relationship. Cowork and Claude Code generally support an individual operator, while Claude Tag gives a channel a shared agent.
Claude Tag is currently available in public beta for organizations using Anthropic’s first-party Team and Enterprise plans. It is not included with individual Free, Pro or Max subscriptions and is not currently available through third-party Claude deployments.
Anthropic transitioned its previous Claude in Slack experience to Claude Tag on August 3, 2026. Organizations must pair their Slack workspace with their Claude organization and have an owner configure the agent’s access before channel members can use it.
Adding Claude Tag does not create a separate per-seat fee. Shared channel and thread activity is billed according to usage from an organization-funded balance, with administrators able to apply spending limits. Direct-message activity follows the individual sender’s normal Claude usage limits instead.
Because the product remains in beta, its availability, behavior and administrative controls may continue to change.
Claude Tag brings AI delegation into one of the places where team decisions and operational context already exist.
Its significance is not simply that Claude can respond inside Slack. The agent can use shared context, work through multi-step assignments, connect to approved business systems and return completed outputs to a discussion that the entire channel can review and continue.
This could reduce context switching and make AI-supported work more transparent. Instead of privately prompting an assistant and manually transferring the result back to colleagues, teams can delegate and supervise the work together.
The larger challenge will be governance. Shared memory, connected tools and autonomous actions can make Claude more useful, but they also increase the consequences of poor permissions, unclear instructions and insufficient review.
Claude Tag therefore points toward a model of workplace AI in which agents become visible participants in team workflows—not independent employees, but shared systems that can be assigned work, redirected and held within organizational boundaries.
Anthropic — Introducing Claude Tag — Published June 23, 2026.
Anthropic Claude Documentation — Work With Claude Tag — Updated August 2026; accessed August 5, 2026.
Anthropic Claude Documentation — How Claude Tag Works — Accessed August 5, 2026.
Anthropic Claude Documentation — Security and Data Handling — Accessed August 5, 2026.
Anthropic Help Center — Claude Release Notes — Claude Tag announced June 23, 2026; accessed August 5, 2026.
