ZICQ
中 Log in / Sign up
Newsroom Agentic #AWS #SageMaker #Bedrock #Multi-Agent #OpenTelemetry

AWS Introduces SageMaker AI and Bedrock AgentCore Integration for Multi-Agent Workflows with Token-Level Observability

Avatar of Mr.Xu

By Mr.Xu Compiled & Reviewed by Editorial

Published: · 7 views

中文阅读 (Chinese) English Version
Co-reading Select any sentence in the article to highlight, annotate, or copy a quote card. Join discussion ↓

Summary:AWS has introduced a new integration that combines Amazon SageMaker AI's OpenAI-compatible endpoints with Amazon Bedrock AgentCore runtime, enabling developers to build multi-agent workflows where each specialized agent uses the model best suited to its task. This setup offers cost optimization, data residency, and model flexibility in a single production-ready architecture. Additionally, the post details how to achieve token-level observability for SageMaker endpoints using custom OpenTelemetry


AWS Introduces SageMaker AI and Bedrock AgentCore Integration for Multi-Agent Workflows with Token-Level Observability

Developers often face the challenge of integrating managed foundation models (FMs) with their own cost-optimized or domain-specific models without rewriting their agent framework. AWS addresses this by integrating Amazon SageMaker AI's OpenAI-compatible endpoints with Amazon Bedrock AgentCore runtime.

Key Features and Benefits

  • Multi-Agent Collaboration: Specialized agents can use the most suitable models for their tasks. For example, the budget agent uses Claude Sonnet 4.6, while the financial analysis agent uses Qwen 3.5 9B.
  • Cost Optimization and Flexibility: The integration offers cost optimization, data residency, and model flexibility within a single production-ready architecture.
  • Token-Level Observability: Custom OpenTelemetry spans enable token-level monitoring of SageMaker endpoints, addressing the default limitation of Strands Agents.

Technical Implementation Details

  1. Deploy Qwen 3.5 9B Model: Use the vLLM Deep Learning Container (DLC) to deploy Qwen 3.5 9B on Amazon SageMaker AI.
  2. Build Multi-Agent System: Utilize the 'agents as tools' pattern from Strands Agents to instantiate and configure each agent with its respective model.
  3. Deploy to Amazon Bedrock AgentCore Runtime: Use the Bedrock AgentCore starter toolkit to deploy the entire workflow to the AgentCore runtime.
  4. Achieve Token-Level Observability: Manually emit a gen_ai.chat span and extract token usage data from Strands' internal AgentResult.metrics.accumulated_usage.

Key Configuration Steps

  • Enable Amazon CloudWatch Transaction Search: Ensure X-Ray sampling rate is set to 100% to capture all traces.
  • Install Strands Agents[otel] Package: Provides OpenTelemetry support.
  • Set Environment Variables: Enable agent observability.
  • Use opentelemetry-instrument as Container Command: Ensure OpenTelemetry auto-injection.

Industry Impact and Developer Recommendations

  • Flexibility in Multi-Model Collaboration: Developers can more flexibly select and integrate different models to meet specific needs.
  • Importance of Observability: Implementing token usage monitoring helps with cost control and performance optimization.
  • Automation and Integration: Leverage AWS-provided tools and APIs to simplify the construction and deployment of multi-agent workflows.

Conclusion

AWS' integration of SageMaker AI with Bedrock AgentCore provides a powerful solution for building multi-agent workflows. Additionally, by using custom OpenTelemetry spans, developers can achieve token-level observability for SageMaker endpoints, enabling better cost control and performance optimization.

Related Resources


Source: AWS Machine Learning Blog (2026-08-14)

— END —

Tags: #AWS #SageMaker #Bedrock #Multi-Agent #OpenTelemetry

Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.

What's your take on this story?
🚀 Explore & Submit Works

Community Comments

Loading live comments and annotations…