Designing AI Products by Starting with Conversations: Priya Badger’s Workflow at Yelp
By: Priya Badger Host: Claire Vo Source: How I AI (ChatPRD / Lenny’s Podcast Network) Type: podcast
Summary
Three workflows demonstrated. Workflow 1: Designing conversational AI by writing “golden conversations” — sample dialogues showing the ideal user experience — then iterating with real images, refining based on patterns observed (image recognition quality, conversation flow, conciseness), and using Claude Artifacts to generate interactive LLM-powered prototypes with auto-generated system prompts. Workflow 2: Using Magic Patterns’ Inspiration Mode to rapidly explore multiple UI design variations side-by-side for new AI features (e.g., photo upload entry points), replacing slow PM-designer back-and-forth with instant visual exploration. Workflow 3: Applying the same techniques to personal projects (community newsletter summarizer via Claude Projects, a “Parent Pal” advisor, a Settlers of Catan timer in Lovable) as low-pressure practice for AI PM skills.
Key Ideas Extracted
- Golden conversations as requirements: Start with example dialogues showing ideal interaction flow rather than traditional wireframes or PRDs — for conversational AI, the conversation IS the prototype
- Inside-out product design: Begin with the end-user experience (the conversation), then work backward to requirements — inverts the traditional linear development process
- Claude Artifacts for interactive prototyping: Generate working chat interfaces that use the same LLM powering Claude, producing both a prototype AND an auto-generated system prompt derived from your example conversations
- Qualitative pattern checking before formal evals: Review generated conversations for image recognition quality, flow, conciseness, and comprehensibility before building rubrics
- Magic Patterns Inspiration Mode: Generate multiple distinct UI options simultaneously for side-by-side comparison — accelerates early-stage design exploration
- Personal projects as AI PM training: Building personal tools (newsletter summarizer, parenting advisor, game timer) provides hands-on AI product experience without needing a formal AI PM role
- Claude Projects for repeatable tasks: Save instructions and context for recurring work (e.g., weekly newsletter formatting from Slack exports)
Notes
- Published on ChatPRD blog Oct 20, 2025 (episode aired ~Jul 7, 2025). 11-min read.
- Context: Yelp’s AI assistant for matching users with service professionals; photo upload feature was the specific problem space
- Priya makes explicit distinction: she manages products WITH AI, not just AI products
- Tools referenced: Claude (conversations + Artifacts), Magic Patterns (UI exploration), Lovable (personal project), Figma (existing workflow comparison)
- Personal projects mentioned: community newsletter (Slack → Claude Projects), Parent Pal (Artifacts), Settlers of Catan timer (Lovable)
- Links to Priya’s Almost Magic Substack for more insights
Raw Content
Re-scraped from ChatPRD 2026-02-15. Full article content captured in Summary and Key Ideas above.