Building agentic workflows with SageMaker AI and Bedrock AgentCore

This article presents a technical solution for building agentic workflows by combining Amazon SageMaker AI with Amazon Bedrock AgentCore runtime, enabling the integration of managed foundation models with custom models. The architecture connects three model-hosting paths through a single Amazon Bedrock AgentCore container, utilizing models such as Qwen 3.5 9B on SageMaker AI and Claude Haiku 4.5 on Bedrock. This integration provides cost optimization, data residency, and model flexibility in a single production-ready architecture. The practical implication for engineers building AI systems is the ability to deploy specialized agents that collaborate on complex tasks while using the most suitable models for each task.

⚡ Key Takeaways

• The Qwen 3.5 9B model is deployed on SageMaker AI using the vLLM Deep Learning Container (DLC) with image vllm:0.22.1-gpu-py312-cu130 on ml.g6e.2xlarge instances.
• The architecture utilizes the OpenAI-compatible API to invoke models on SageMaker AI, such as Qwen 3.5 9B, and Amazon Bedrock models like Claude Sonnet 4.6.
• The integration of SageMaker AI with Amazon Bedrock AgentCore runtime provides token-level observability, which is not available by default in Strands Agents.
• The solution requires an AWS account with permissions for Amazon SageMaker AI, Amazon Bedrock, and AgentCore, as well as a Python 3.12+ environment with specific libraries installed, including sagemaker-core, openai, and bedrock-agentcore.

This solution enables engineers to build agentic workflows that combine the strengths of managed foundation models with custom models, providing a flexible and cost-optimized architecture for complex AI tasks. By integrating SageMaker AI with Amazon Bedrock AgentCore runtime, engineers can deploy specialized agents that collaborate on tasks while using the most suitable models, resulting in improv

✅ Practical Steps

1. Deploy Qwen 3.5 9B on SageMaker AI using the vLLM Deep Learning Container (DLC) with image vllm:0.22.1-gpu-py312-cu130 on ml.g6e.2xlarge instances.
2. Integrate the deployed model with Amazon Bedrock AgentCore runtime using the OpenAI-compatible API.
3. Utilize the Strands Agents multi-agent system to route user requests to specialized agents, such as the budget agent and financial analysis agent, which invoke their respective models on Amazon Bedrock and SageMaker AI.

Want the full story? Read the original article.