As enterprises shift to agentic AI systems, the development and operational surface area is fundamentally changing. Unlike deterministic software, AI agents reason, plan, and iteratively execute multi-step workflows to arrive at a complete response. This evolution introduces a new category of operational problems: how do you monitor a system that can “hallucinate,” troubleshoot a reasoning trace, or audit a non-deterministic decision?

This challenge extends to development practices as well, where software engineers struggle to build agents without clear observability into their accuracy, performance, and cost. Traditional observability frameworks focused on “RED” metrics (Rate, Errors, Duration) are no longer sufficient. Today’s engineering leaders must manage agentic health, tracking token consumption, latency across multi-pass reasoning, and the accuracy of automated actions.

In this session, we will explore how the OpenSearch Observability Stack, a unified, vendor-neutral, and Apache 2.0-licensed platform, solves these complex challenges. We will discuss how to move from a “data swamp” to actionable intelligence by correlating logs, traces, and Prometheus metrics to ensure agentic stability at scale.

Join Dotan Horovits and Megha Goyal from AWS as we bridge the gap between raw telemetry and strategic insight, using OpenSearch as the neutral, scalable engine for observing the next generation of AI.