Anthropic brings Claude Tag AI into Slack channels

Anthropic launched a beta of Claude Tag for Enterprise and Team tiers that embeds its Claude AI into shared Slack channels. Team members can type @Claude to assign tasks, review outputs and resume work from earlier threads while execution steps remain visible to the channel.

The integration runs on Anthropic’s Opus 4.8 engine. The agent divides requests into sequential execution phases and can access linked corporate databases, internal tools and code repositories to carry out tasks without constant human prompting. Any channel member with access can interact with the agent, review its responses and continue a paused conversation.

When configured in an ambient mode, Claude Tag monitors inactive threads, flags priority items from connected extensions and tracks unresolved assignments across multiple days. Cat Wu, head of product for Claude Code, described the format as: “The form factor of being able to tag it the same way that you would a coworker is really powerful.” Wu said she connected a personal Claude Tag agent to her email archive to prioritise incoming messages and post alerts into Slack.

Rob Seaman, general manager of Slack, described the operational shift as: “Instead of a private back-and-forth, Claude Tag shows up in the open.” The system automatically builds contextual histories from active channels so users do not need to retype project scopes or company data each time work resumes.

Anthropic’s internal product group reports that about 65% of its code was generated through a private Claude Tag instance. External data from Ramp’s May 2026 AI Index shows Anthropic with a 34.4% enterprise footprint compared with 32.3% for OpenAI. The company completed a US$65 billion Series H round that set a post-money valuation at US$965 billion and is preparing a confidential S-1 filing for an initial public offering.

The integration includes administrative controls. System administrators can create scoped Claude identities, confine local memories and tool integrations to approved channels, and use management portals to review query logs and set organisational caps on monthly token costs.

Embedding agents in team channels centralises task information and keeps context in the communication window. At the same time, allowing an automated system to read chat histories, access email and modify code repositories increases internal data-exposure risk if access boundaries are misconfigured. Removing human verification from intermediate workflow stages creates the possibility of systemic errors if the model misinterprets instructions during a multi-step task.

Anthropic reports early customer use cases beyond engineering, including querying database metrics, parsing analytics and processing IT support tickets. Channel-by-channel security settings are available to confine integrations to authorised teams.