Amazon Bedrock has introduced Advanced Prompt Optimization, a tool that optimizes prompts for up to 5 models on Bedrock while comparing original and optimized performance. The tool addresses the manual burden of prompt engineering at scale, which affects the generative AI development lifecycle in 4 areas: model lock-in, underperformance, regression blindness, and slow iteration cycles. Customers typically spend days to weeks optimizing prompts and re-evaluating responses when migrating to a new model. The new tool replaces this manual iteration with a guided, metrics-driven workflow, allowing users to migrate and optimize prompts across multiple models in a single job. The system operates in a reinforcement learning-style feedback loop without changing model weights. Users provide a prompt template, example user inputs (text or multimodal), optional ground truth answers, and an evaluation metric. The optimizer then sends the template and inputs to chosen inference models and evaluates responses against the metric, working backward from the evaluation metric to guide prompt and response changes.