Instacart's Blueberry AI: Because Your 3 AM Pager Deserved a Sidekick

Instacart has built Blueberry, an AI-assisted incident response system that uses multiple AI agents, operational data, and 14+ years of historical incident knowledge to help on-call engineers investigate production issues faster. When an alert fires, Blueberry launches roughly 10 subagents in parallel and posts a grounded root-cause hypothesis directly in the relevant Slack thread within about three minutes. The system processed approximately 25,000 diagnostic passes in April across 270+ Slack channels, with diagnostic accuracy improving from the mid-60% to the high-90% range after grounding with historical data. Engineers retain full responsibility for diagnosis, mitigation, and remediation — Blueberry gathers context and generates hypotheses but does not make production changes autonomously.