As AI evolves from simple chatbots to autonomous agents, the challenge shifts from monitoring single inferences to observing complex "agentic loops." The AWS Strands SDK simplifies this by using a model-driven approach where LLMs plan and execute tools autonomously. However, in a serverless environment like AWS Lambda, these multi-step loops can become a "black box," leading to unpredictable latencies, soaring token costs, and difficult-to-trace hallucinations.

In this session, we demonstrate how to instrument AWS Strands agents using OpenTelemetry (OTel) to gain deep visibility into the "thought process" of your AI. We will explore the integration of the AWS Distro for OpenTelemetry (ADOT) with the Strands SDK to capture granular telemetry across agent cycles, tool calls, and LLM reasoning steps.