AWS Launches Bedrock AgentCore Optimization to Automate Agent Performance Loops

May 5, 2026 · Updated May 13, 2026

AWS launched Amazon Bedrock AgentCore Optimization in preview to automate the "Agent Quality Loop." The service analyzes execution traces and evaluation scores to auto-generate optimized prompts and tool descriptions. This shifts agent development from manual trial-and-error to a data-driven process within the Bedrock managed agent harness.

Optimization inputs: Execution traces and evaluation scores
Validation methods: Batch testing and A/B experiments
Traffic management: AgentCore Gateway target-based routing
Availability: Preview
Primary platform: Amazon Bedrock

Reliability remains a barrier to deploying autonomous agents in production. This release addresses the quality layer by providing a structured way to fix underperforming agents, following SageMaker AI's agent-guided customization. By using real-world failure data, organizations can move beyond "vibe coding" toward repeatable, measurable agentic engineering.

You can now use AgentCore Gateway to run controlled A/B experiments, splitting live traffic between agent versions to compare performance. This extends AWS Strands Agents orchestration to provide the metrics needed to ship agent improvements with confidence. Offline validation also supports managed batch evaluations against curated test sets to catch regressions.

View the full update on aws.amazon.com

AWS AI

@AWSAIMay 4

Stop guessing why your agent underperforms. Amazon Bedrock AgentCore now auto-generates optimized prompts and tool descriptions from real traces and evaluation scores. Validate them via batch tests & A/B experiments. Ship improvements with confidence. Now in preview. https://t.co/ab4AiZByb8

View on X

Still wondering? A few quick answers below.

Amazon Bedrock AgentCore Optimization is a performance-tuning suite designed to improve the reliability of AI agents. It automates the process of refining agent instructions by analyzing real execution traces and evaluation scores. This allows developers to move away from manual prompt engineering and toward a data-driven approach for fixing underperforming agents.

The service uses the Agent Quality Loop to analyze logs of agent actions, known as execution traces, alongside performance scores. Based on this data, it auto-generates optimized prompts and tool descriptions that address specific failure patterns. This systematic approach helps ensure that agents follow instructions more accurately and use external tools more effectively in production.

Developers can validate agent improvements using two primary methods: batch evaluations and A/B experiments. Batch testing allows for offline validation against curated test sets to catch regressions. A/B experiments use the AgentCore Gateway to split live production traffic between different agent versions, providing real-world performance data before a full deployment.

Amazon Bedrock AgentCore Optimization is currently available in preview for customers using the Bedrock platform. It is designed for developers and organizations building autonomous agents who need to move beyond initial prototypes to reliable, production-ready deployments. Users can access these optimization and validation features through the AWS Management Console or Bedrock APIs.

The AgentCore Gateway is a component of the Bedrock infrastructure that enables target-based routing for AI agents. It is essential for running A/B experiments, as it manages the distribution of live traffic between a control version and a treatment version of an agent. This allows teams to compare performance metrics in real time with confidence.