Last month, we explored how AI agents are reshaping the way applications work. As agents take on more responsibility, they’re creating new workflows that span models, tools, retrieval, APIs and infrastructure. That shift is creating a new observability challenge: teams need to understand not just whether an application is working, but what an agent did, why it did it and whether the outcome was good.

Traditional observability can tell you when a service is slow or an endpoint is failing, but that only tells part of the story for an agent. As agent activity grows, teams need ways to collect, organize and investigate that telemetry without adding more complexity.

This month, Observe by Snowflake is introducing its agent observability capabilities to help teams tackle that challenge. From instrumenting agents and monitoring production behavior to evaluating quality, investigating unexpected behavior and optimizing performance, the goal is to make agent observability part of the production workflow.

Read on to see what’s new.

Agent Observability helps teams monitor, debug and improve AI agents and LLM applications by connecting agent performance with quality and cost.

New capabilities include the Agent Explorer, online evaluations for production traffic and an optional OpenTelemetry SDK for instrumenting agents and sending traces to Observe. Together, they help teams move from experimentation to more reliable production AI systems.

Read the announcement ›