AWS – Amazon CloudWatch adds generative AI observability (Preview)

Amazon CloudWatch now helps you observe generative AI applications and workloads, including agents deployed and operated with Amazon Bedrock AgentCore (Preview), providing insights into AI performance, health, and accuracy. You get an out-of-the-box view into latency, usage, and errors of your AI workloads to detect issues faster in components like model invocations and agents. You can also find issues faster using end-to-end prompt tracing of components like knowledge bases, tools and models. This feature is compatible with popular generative AI orchestration frameworks such as Strands Agents, LangChain, and LangGraph, offering flexibility with your choice of framework.

With this new feature, Amazon CloudWatch analyzes telemetry data across components of a generative AI application, helping quickly identify the source of errors. For example, you can pinpoint the source of inaccurate responses — whether from gaps in your VectorDB or incomplete RAG system retrials — using end-to-end prompt tracing, curated metrics and logs. This connected view of component interactions helps developers optimize workloads faster to deliver high levels of availability, accuracy, reliability, and quality. Developers can keep AI agents running smoothly by monitoring and assessing the fleet of agents in one place. The agent-curated view is available in the “AgentCore” tab in the CloudWatch console for genAI observability. Generative AI observability is integrated with other CloudWatch capabilities such as Application Signals, Alarms, Dashboards, Sensitive Data Protection, and Logs Insights, helping you seamlessly extend existing observability tools to monitor generative AI workloads.

This feature is available in Preview in 4 regions: US East (N.Virginia), US West (Oregon), Europe (Frankfurt), and Asia Pacific (Sydney). To learn more, visit documentation. CloudWatch pricing applies for collected and stored telemetry data.

Read More for the details.