Helps reduce number of tokens in an output, and skips some chains of the workflow when not needed via code logic ## Example - Context: Chatbot to ask about products and categories in a tech store - Prompt to identify the categories and products (source is JSON) based off of user input, return which ones are relevant - Based on which ones are relevant, add product details into the conversation context - When they ask for details, you can then create a custom output format for it