Recipe Overview

Ensuring answer quality can be hard in one pass. The evaluator-optimizer agent addresses this by adding a feedback loop. One LLM generates a response and another evaluates it and suggests improvements. Anthropic describes this loop of generation and critique. For example, a translation agent might produce a draft and then a second agent checks for mistakes. This iterative review solves errors: each cycle refines the output. The pattern is especially valuable for high-stakes tasks where quality matters more than speed, like legal document review or medical diagnosis assistance.