Amazon Bedrock AgentCore Launches Decision Model Evaluation Support

Amazon Bedrock just rolled out AgentCore Evaluations to score AI agents through their entire lifecycle. The new framework comes with fifteen built-in evaluators. It also supports custom ones like the open-source Strands Decider 2B model. Instead of writing free-form text, these decision models return calibrated probabilities over fixed answers. That setup gives developers a direct confidence score they can threshold and trust. Teams can run them on demand during development or live against production traffic. They also handle batch jobs against stored sessions without breaking a sweat. You can use decision models as a fast screening layer for narrow questions. That leaves deeper investigations to the built-in evaluators. The system lets you connect hosted options like TypeSafe Jev right alongside your own code-based logic in AWS Lambda.