Claude Tag as an On-Call Responder in Slack

Claude Tag is Anthropic's way of putting Claude into Slack as a team member: it launched with access to selected channels and whatever tools, data and codebases a team connects. In a post, ClaudeDevs shows how its team uses it for on-call. When an alert fires in Slack, Claude pulls metrics, diffs recent deploys and checks feature flags, then proposes a likely fix for people to approve.

The video tells one incident. At 11pm the on-call engineer is paged because the error rate on a payments API has passed 2%, and Claude is already investigating, because days earlier the engineer told it to start whenever an alert fires in that channel. About 15 minutes in, it ties the spike to a feature flag flipped shortly before and a new worker with no limit on concurrent work. It offers two fixes and recommends one. Because the engineer asked it to propose fixes first, it tags the engineer and waits. The engineer replies with one line, to add the cap and keep the feature on, and only then does Claude open the pull request. The tests pass, and the engineer reads and approves the four-line diff.

After the deploy, Claude watches the error rate for ten minutes before calling the problem fixed, then closes the incident and saves a note for next time to check the flags first. The narrator says it acts only within what the channel allows: it can read the team's tools and open pull requests, but nothing merges or deploys without a person approving it, and that setup was done by messaging Claude. By the narrator's account, about an hour of late-night dashboard digging became 15 minutes with one decision for the engineer. That is one team's description of one incident, not a measured comparison.