Introducing the agent performance loop: AgentCore Optimization now in preview

This article announces AgentCore Optimization, a new preview feature for Amazon Bedrock AgentCore that automates agent performance improvement through a continuous feedback loop.

• Recommendations analyze production traces to optimize system prompts and tool descriptions
• Batch evaluation validates recommendations against predefined test datasets before deployment
• A/B testing compares agent versions on live production traffic with statistical significance
• Configuration bundles enable immutable, versioned snapshots of agent settings
• AgentCore Gateway splits traffic between control and treatment variants for safe testing
• Replaces manual trace reading and prompt guessing with data-driven optimization cycles
• Available in preview across AWS regions where AgentCore Evaluations is supported

AgentCore Optimization enables teams to systematically improve agent quality through automated recommendations and rigorous validation, replacing ad-hoc manual processes with repeatable, evidence-based improvement cycles.