Bobby Calderwood

Double the number of jobs on a multi-tenant stream processing platform, and you'll more than double the alerts, false positives, and real incidents your team must triage. Signal and noise both grow super-linearly with scale, consuming an ever-larger scale of the platform team’s bandwidth. Applying AI at the right points can bring that support growth curve back toward linear. At JPMorgan Chase, a small platform team supports thousands of user-submitted Flink jobs processing more than 100 billion events per month. In this talk, we share the three-stage agentic support maturity curve that made this possible, and how you can apply it to your own platform: - Amplify the signals that matter. Learn AI/ML techniques can enrich platform telemetry and filter out alert noise so real problems stand out. - Human-initiated, AI-assisted diagnosis and remediation. See how correlating logs, metrics, and historical metadata with point-in-time state provides the context an agent needs to help engineers find root causes and fixes faster. - Agent-initiated diagnosis and remediation. Discover how agents can resolve routine issues on their own, keep the right people informed, and bring in humans only…