Agentic AI
AWS engineers to demonstrate observability solutions for agentic AI at upcoming webinar
OpenTelemetry and OpenSearch offer open-source tools to unify monitoring across fragmented AI agent deployments.
The New Stack
Understanding system performance and behavior in production environments grows exponentially harder as organizations deploy AI agents across multiple infrastructure layers. Traditional logging, metrics, and tracing approaches increasingly struggle to keep pace with the volume and complexity that agentic systems introduce—especially when agent behavior can be unpredictable and non-deterministic.
Rather than locking organizations into proprietary monitoring platforms that fragment data across silos, open-source alternatives like OpenTelemetry and OpenSearch provide a unified foundation for observability at scale.
Open-source adoption for AI workloads#
OpenTelemetry has achieved dominant adoption among cloud-native projects, crossing the 95% threshold for new instrumentation initiatives and becoming the standard for greenfield deployments. OpenSearch, an Amazon Web Services–sponsored platform, is increasingly embraced by AI engineers who recognize that effective observability is inseparable from successful AI operations.
The OpenSearch roadmap for this year emphasizes positioning the platform as a primary retrieval layer for AI agents and a critical component of retrieval-augmented generation and agentic AI systems.
Hands-on troubleshooting demonstration#
AWS engineers will showcase these tools in action during a live webinar on July 22. The session will feature a practical troubleshooting simulation using correlated logs, metrics, and traces, alongside a demonstration of how agentic traces move through OpenTelemetry pipelines.
Presenters Dotan Horovits and Rekha Thottan will also introduce Agent Health, an open-source evaluation framework that provides structured pre-production benchmarking to identify unpredictable agent behavior before systems reach production.
Why this matters#
Organizations seeking to adopt these open-source standards without vendor lock-in now have a clear path forward. The webinar addresses how to implement unified observability across both AI agent workloads and traditional infrastructure at scale during the second half of 2025.
