Building agentic workflows with SageMaker AI and Bedrock AgentCore
AWS shows how to mix Bedrock-hosted and SageMaker-hosted models in a single multi-agent workflow
AWS published a technical walkthrough combining Amazon SageMaker AI (OpenAI-compatible endpoints) with Bedrock AgentCore runtime to run heterogeneous multi-agent systems without rewriting agent frameworks. The reference architecture uses Claude Haiku 4.5 as orchestrator, Claude Sonnet 4.6 for structured budget tasks, and Qwen 3.5 9B on SageMaker for financial analysis — all routed through a single AgentCore container. The pattern matters because it formalizes a cost-flexibility tradeoff: teams can swap in cheaper or domain-specific models on SageMaker without abandoning Bedrock's managed tooling.