Claude Tag

Overview

Freshness note: AI products change rapidly. This profile is a point-in-time snapshot last verified on July 10, 2026.

Claude Tag is Anthropic’s shared team agent for Slack. Instead of opening a private assistant and pasting in a thread, a team can tag @Claude inside a channel, let it read the conversation, and delegate work through tools and repositories approved by an administrator. The result stays in the thread so everyone can see, steer, and continue the task.

The product launched in beta for Claude Team and Enterprise and currently works in Slack. Anthropic says it replaces the older Claude in Slack experience, with that integration scheduled to switch to Claude Tag on August 3, 2026.

Key Features

Claude Tag has three interaction surfaces: channel mentions, direct messages, and Slack’s AI assistant panel. Channel work is the genuinely new part. Everyone in the channel works with the same Claude, and the agent can use shared context, remember relevant information across threads and days, and post its result back where the task started.

It can also run asynchronously, schedule recurring tasks, and—when proactive behavior is enabled—follow up on stalled work or surface relevant changes without a fresh mention. Admins can scope tools, credentials, repositories, instructions, and memory at organization, workspace, and private-channel levels.

Agent identity makes the operational boundary more explicit. Claude acts through its own account rather than borrowing an employee’s login. Anthropic documents an audit view for scheduled and one-off tasks plus network calls, while actions in connected tools remain attributable to Claude and link back to the originating Slack thread.

Strengths

Claude Tag fits work that is already collaborative. A channel can hand off a messy bug report, metrics question, backlog, or recurring digest without one person becoming the hidden operator behind a private AI chat. Shared threads also make review and course correction more natural.

The admin model is stronger than a casual bot installation. Scoped credentials, per-channel memory, spend caps, usage alerts, and visible agent attribution give teams practical controls around a tool that can otherwise become surprisingly powerful.

Limitations

It is still a beta and currently Slack-only. It also introduces a new kind of shared context risk: workspace and channel memory can influence later work, connected sources can contain misleading instructions, and a broadly permissioned agent can turn a bad interpretation into a real system change.

Billing has two paths that teams need to explain clearly. Channel work is consumption-based and paid by the organization; direct messages use the member’s Claude account. Proactive tasks can also create recurring spend and notification noise if standing instructions are vague.

Practical Tips

Pilot Claude Tag in a private channel with read-only tools, a low spend cap, and one workflow whose output is easy to review. Write down what it may read, what it may draft, and which actions still require a human. Inspect its stored memory and audit history during the pilot, not only after something surprises you.

Treat write access as a separate rollout. Connected documents, tickets, repositories, and monitoring systems can carry indirect prompt-injection risk, so keep destructive or external-facing actions behind explicit approval and narrow agent credentials.

Verdict

Claude Tag is more interesting than a standard Slack chatbot because it behaves like a shared, persistent team agent with its own identity and tools. It is a strong fit for Slack-centered teams that can administer permissions, memory, spend, and review discipline; it is a poor fit for organizations looking for zero-governance automation.