Recipe Overview

Tasks often vary by type (e.g., translation vs. coding vs. QA). A routing agent first classifies input and then sends it to the specialized sub-agent best suited for that task. The problem it solves is that one-size-fits-all prompts fail across diverse requests. Anthropic describes routing as directing inputs to specialized models or prompts based on content. For instance, the agent might detect a math question and use a calculation model, or recognize a creative prompt and switch to a more imaginative model. This solves accuracy by matching task types to optimal handling, improving both speed and quality across diverse workflows.