Amazon Web Services launched an AgentCore system prompt optimizer for Bedrock agents that recommends updated agent configurations using production traces and reward signals and validates them with offline evaluation and online A/B testing
Product launch Provisional 45% confidence first seen
Amazon Bedrock AgentCore launched a system prompt optimizer that uses production agent traces plus a reward signal to propose updated agent configurations. It validates proposed changes using offline batch evaluation and online A/B testing, including safeguards intended to control prompt growth and reuse trace phrases. Separate coverage also described a prompt optimization platform launch for GenAI applications, benchmarking and tuning models and prompts against user data.