A retailer is deploying a customer service AI agent that uses multiple tools to look...

The requirement is not just foundation-model evaluation. It is full agent release validation: task completion, correct tool selection, final response quality, golden traces, continuous monitoring, and human audit. Amazon Bedrock AgentCore Evaluations is designed for automated assessment of agent and tool performance before and after deployment. Built-in evaluators can be used for common evaluation scenarios, custom evaluators can test organization-specific criteria, and online evaluations continuously monitor live traffic. That directly supports CI/CD quality gates and post-deployment degradation detection with less custom engineering. Option A creates a custom evaluation pipeline, increasing operational overhead. Option B evaluates only the underlying model, not the agent’s tool behavior. Option D covers online monitoring but misses pre-deployment golden-trace validation and custom evaluator use in the pipeline.