When Claude is tagged into a shared channel, it operates with its own provisioned credentials rather than borrowing anyone else's. Anthropic is calling this access model "agent identity," and the thread walks through how it works in practice.
When multiple people are in one thread, Claude has no clean set of credentials to borrow, so it defaults to acting as itself in channels and as the user in private DMs. This is the behavioral summary of the agent identity model.
In shared channels, Claude operates under its own dedicated service accounts across tools like Linear, GitHub, and data warehouses. DMs stay simpler: there, Claude runs on the user's own connectors under their name, which is where personal email and calendar access live.
Every action Claude takes in a shared workspace appears in each connected system's logs under Claude's own account. Revoking access once cuts it everywhere, and shared channels never surface anyone's private files.
A user singled out Claude Tag as the most polished AI chat integration they have seen, specifically praising the execution rather than raw capability. The team acknowledged the feedback.
The video accompanying the Claude Tag announcement was itself edited by Claude, a small but telling detail about how deep the dogfooding goes on the team.
The team is planning further discussion of Claude Tag alongside Peter Yang and at the AIE event with Cat Wu. They're inviting community questions to shape what gets covered.
A team member clarified that /goal and related features like /loop and routines are useful for general workflows, not just coding. Podcast editing was cited as a concrete non-coding example.
A Claude Code team member replied to the Bun v1.4 announcement with a joke comparing Bun's multi-role identity (runtime, bundler, package manager) to Claude's expanding scope, suggesting people should start connecting the dots.
A planned interview with a Claude Code team member will cover his personal setup, use of /goal and /loop, planning with HTML artifacts, and Claude Tag best practices. The community is invited to suggest additional topics.
Admins provision Claude's identity once and then tune what it can reach on a per-channel basis: broad read access in open spaces, sensitive tooling restricted to cleared users only. A setup walkthrough video is available from the team.
A team member amplified a post framing Claude Tag not as a chatbot or digital coworker but as an org-level harness, suggesting it could run companies at scale over the next year.
After a user argued that agents should proactively do work without explicit tagging, a team member clarified that once Claude is added to a Slack channel, you can instruct it to monitor for work and act autonomously.
