Amazon Bedrock AgentCore Evaluations now uses OpenTelemetry as a framework-neutral contract for scoring AI agents. Teams can keep their preferred development kit as long as the agent emits compatible telemetry, avoiding a separate evaluation integration for every framework.
AWS lists LangGraph, LlamaIndex, the OpenAI Agents SDK, Google ADK, the Claude Agent SDK and Strands Agents among the supported approaches. OpenTelemetry records standardized traces of model calls, tool use and other steps. AgentCore reads those traces and applies evaluations without requiring the application to use an AWS-specific loop or model client.
The design is useful for organizations running several agent stacks or changing frameworks while keeping the same quality checks. It does not make traces automatically complete: developers must instrument the relevant actions and avoid sending sensitive prompt or customer data without appropriate controls. Scores also remain only as useful as the chosen evaluators and test cases. The update primarily removes plumbing that previously tied assessment to one SDK.