Shyam Sreevalsan spoke at Conf42 DevOps 2025 on January 23 (virtual) about the future of observability. The talk covered three threads: AI-driven anomaly detection, real-time processing at the edge, and predictive analytics as a replacement for reactive alerting.
The central argument was that monitoring is shifting from “tell me when something breaks” to “tell me before something breaks.” Proactive monitoring requires two things that most traditional tools lack: per-second data collection (because you cannot predict what you cannot see) and ML models that run continuously rather than on a query schedule. Shyam walked through how Netdata handles both – distributed agents collecting every second, unsupervised ML training on each metric individually, and anomaly correlation that surfaces coordinated deviations before they become incidents.
For a DevOps audience managing complex distributed systems, the practical takeaway was specific: AI-driven anomaly detection can simplify what would otherwise require dozens of hand-tuned alert rules. The virtual format meant a global audience, with questions coming in from practitioners across time zones dealing with the same fundamental challenge – too many alerts, not enough signal.
