Anthropic brings Claude Tag into Slack for shared team workflows

Enterprise and Team customers can delegate work to a channel-based AI agent with approved access to tools, data, and codebases.

Anthropic has introduced Claude Tag in beta, allowing Enterprise and Team customers to add a shared Claude agent to selected Slack channels.

Anthropic has launched Claude Tag in beta for Claude Enterprise and Team customers, allowing organizations to add a shared Claude agent to selected Slack channels and delegate software development, data, support, and other workplace tasks.

Available from June 23, Claude Tag can follow conversations in approved channels, use connected tools and codebases, divide requests into stages, work asynchronously, and return completed work through Slack threads.

The product uses Anthropic’s Opus 4.8 model. It replaces the existing Claude in Slack application, with administrators able to opt into the new version during a 30-day migration period.

System administrators decide which channels, information sources, tools, and repositories Claude Tag can access. They can also set monthly token spending limits for the organization and individual channels and review logs showing each action and the user who requested it.

Anthropic plans to bring Claude Tag to other workplace environments, but it has not named additional platforms or provided a wider rollout timetable. Eligible Enterprise and Team customers will receive an introductory launch credit, although the amount and longer-term pricing have not been disclosed.

Claude becomes a shared participant in Slack channels

Claude Tag operates as a shared agent within each approved Slack channel rather than as a private assistant attached to one employee.

Anyone in the channel can tag @Claude with a task, view its work, respond to its output, or continue a request started by another team member.

Claude Tag can retain relevant context from the channels it has permission to follow, reducing the need for employees to repeat project information with every request. It can also draw on approved data sources and other Slack channels where access has been granted.

Requests can include software development tasks, analysis of product metrics, support tickets, and investigations into technical problems.

Claude Tag breaks a request into stages before working through them and posting its output in Tag breaks a request into stages before working through them and posting its output in the relevant Slack thread. Employees can continue with other work while the task runs.

The agent can also schedule follow-up work over several hours or days. When an optional ambient behavior setting is enabled, Claude Tag can proactively flag information, revisit inactive threads, and follow up on tasks that appear unresolved.

Rahul Patil, Chief Technology Officer at Anthropic, wrote on LinkedIn: "Less synchronous, more async. Less work done alone in a private editor, more done in the open alongside teammates. And increasingly, work that closes its own loop."

Employees can also send Claude Tag direct messages. In those conversations, it uses the personal tools and connectors configured for that user rather than the shared resources attached to a team channel.

Administrators control channel access and spending

Anthropic has designed Claude Tag around separate identities for different teams, functions, and information environments.

An administrator could create one Claude configuration for engineering and another for sales, with each restricted to its own Slack channels, tools, data, memories, and repositories.

Anthropic says information retained by an engineering configuration will not be passed to the sales version or used to give engineering staff access to sales resources.

The product can use information from other approved channels and connected sources, but Anthropic says it will not report content from private Slack channels.

Patil said Claude Tag operates through its own credentials rather than those of individual employees. Administrators can review a record of its actions and identify who initiated each task.

Organizations can also set spending limits before employees begin using the agent. Anthropic recommends testing Claude Tag in a private channel after connecting the Slack workspace and required tools.

These controls determine what Claude Tag can access but do not remove the need for employees to review its work. The launch material does not provide accuracy figures, security audit results, or data on how frequently the internal system completes tasks without human correction.

Anthropic attributes 65 percent of product code to internal version

Anthropic says an internal version of Claude Tag now creates 65 percent of the code produced by its product team.

Patil also claimed Anthropic’s engineers are "shipping on the order of 8x the code per quarter they did over the last few years."

The organization has not published the methodology behind either figure, including how AI-created code is defined, the level of human editing involved, or whether output volume has been compared with defect rates, security findings, or long-term maintenance requirements.

Anthropic says Claude Tag is also used internally to retrieve product metrics, process support tickets, and investigate software bugs.

The external beta does not guarantee the same productivity results for customer organizations. Performance will depend on the tools, data, codebases, instructions, and access permissions connected by each administrator.

Claude Tag is available now in beta for Claude Enterprise and Team customers using Slack. Administrators have 30 days to opt into the replacement for the existing Claude in Slack application.