How this Yelp AI PM works backward from “golden conversations” to create high-quality prototypes using Claude Artifacts and Magic Patterns | Priya Badger

Product management for AI-powered features requires a shift from traditional documentation toward "golden conversations" that define the desired user experience. Priya Badger, a product manager at Yelp, demonstrates a workflow that uses large language models to prototype conversational flows by working backward from example user interactions. This approach involves generating sample dialogues and visual prototypes using tools like Claude and Magic Patterns to refine system instructions and interface design before full-scale development. By treating these AI-generated artifacts as the primary source of truth, product teams can rapidly iterate on user flows, test edge cases, and align design with technical capabilities. This methodology allows for non-linear exploration of solution spaces, enabling teams to visualize and validate complex AI behaviors—such as image-based service requests—before committing significant engineering resources.