The AWS Machine Learning Blog outlines an approach to building agentic workflows by pairing OpenAI-compatible endpoints on Amazon SageMaker AI with the Amazon Bedrock AgentCore runtime. The described setup is a multi-agent workflow in which each specialized agent selects the model best suited to its specific task.
The post also addresses observability, showing how to obtain token-level insight from SageMaker endpoints. According to the description, this level of instrumentation is not provided by default in Strands Agents.
Who should care
Developers and teams working on multi-agent systems within the AWS ecosystem may find the combination of SageMaker AI and Bedrock AgentCore relevant, particularly those seeking finer-grained observability of model usage.