Anthropic's Slack Agent Claude Tag Becomes an Automated First Responder for On-Call Engineering Incidents

Anthropic says the team that builds its own developer tooling now lets Claude Tag, its Slack-based AI agent, act as the first responder to production incidents — pulling metrics, diffing recent deploys and checking feature flags the moment an alert fires, then proposing a fix that an engineer can review and merge. The workflow is laid out in an Anthropic engineering post describing how the company's own Continuous Integration team uses the tool, and it replaces what used to be a largely manual, often after-hours routine: an on-call engineer waking up, opening a laptop, and spending roughly an hour tracing a CI/CD failure back to its cause. In one example Anthropic cites, 44 tests on a new service silently stopped firing; in cases like that, Claude Tag typically posts an initial, evidence-grounded analysis within about 15 minutes of an incident opening, with a reported median of 14 minutes and a fastest recorded case of 4 minutes to name a root cause. Behind that first message, an orchestration agent spins up smaller executor subagents that query Grafana dashboards, log stores, GitHub history, Kubernetes state and PagerDuty alerts, then keeps refining its hypothesis or handing context to the next shift as the incident develops. For issues that call for a rollback rather than a code change, it manages feature-flag adjustments under canary-traffic monitoring instead of merging directly, and it logs outcomes to a running internal record plus a recurring "ci-weather" report so later incidents can draw on precedent.