How I AI – Episode Summary

Podcast: How I AI
Host: Claire Vo
Episode Title: How this Yelp AI PM works backward from “golden conversations” to create high-quality prototypes using Claude Artifacts and Magic Patterns | Priya Badger
Guest: Priya Matthew Badger, AI Product Manager at Yelp
Date: October 20, 2025

Episode Overview

In this episode, Claire Vo sits down with Priya Matthew Badger, a Product Manager at Yelp, to discuss a practical, workflow-driven approach to designing and prototyping AI-powered features—especially conversational agents. Priya shares Yelp’s strategy of starting with “golden conversations” (idealized sample dialogs), then working backward to product requirements and prototypes, using tools like Claude Artifacts and Magic Patterns. The conversation is full of actionable tips for product managers, hands-on demos, and real-world examples from Priya’s work and life.

Key Discussion Points & Insights

1. What’s Unique About Managing AI Products?

• Dual Focus: Managing AI-powered products is different because you have to design both the user interface and the behind-the-scenes AI/system prompts that generate experiences. (03:30)
• Quality & Consistency: AI products can produce variable results, so a core challenge is ensuring a high-quality, predictable experience for users. (03:30)
• Iterative Experimentation: AI’s unpredictability means continuous testing and refining is necessary.

2. The "Golden Conversation" Approach – Working Backward (04:34–13:47)

• Starting Point: Priya begins by scripting "golden conversations," which are full sample dialogues between a user and the AI assistant, to define what great looks like before building anything. This serves as a wireframe for conversational AI.

  • Quote (06:35):
    “I like to start with thinking about what is that conversation flow going to look like when we add this new functionality. … Write a complete sample conversation between the consumer and the AI assistant…” — Priya

• Workflow in Practice:

  1. Draft a Dialogue: Use AI (Claude, ChatGPT, etc.) to role-play real conversations users might have, including new functionalities (like photo uploads for service requests on Yelp).
  2. Set Prompting Constraints: Specify output format and scenario constraints to guide the AI’s example conversation.
  3. Test with Real Inputs: Upload images of problems (e.g., cracked porch, appliance error, wasp nest) to see how well the AI interprets different scenarios, and practice iterating on the dialogue. (09:57–13:14)
  4. Look for Patterns: Use multiple conversational examples to extract common threads and design requirements.
  • Quote (11:34):
    “You can see it's saying, ‘I can see you've uploaded this photo of a front porch with a significant crack running through the concrete.’ …And then it says, ‘Let me ask a few questions …’”

• Qualitative Review: Analyze conversations for tone, flow, accuracy, conciseness, and ability to drive the user to a solution.

  • Quote (13:47):
    “First I just look at it qualitatively to see, like, does this feel like it sounds like it flows well? Is it concise? Is it easy to understand?” — Priya

3. From Conversations to Prototypes: Claude Artifacts (15:09–19:31)

• Iterative Refinement: AI can rewrite sample conversations based on feedback (e.g., being more opinionated, skipping budget questions).

• Prototype Creation: Claude’s “artifact” feature allows Priya to turn sample conversations into a functional prototype with native LLM integration, system prompts, and UI elements, all within Claude—no manual API wiring.

  • Quote (17:13):
    “Claude has a special functionality built in where it actually can create an artifact that uses the LLM that powers Claude to produce those responses. And that's very unique to Claude.”

• User Experience Testing: Simulate real-world interaction with the prototype to uncover issues that wouldn't be obvious in a static doc (e.g., message length feels different in a chat bubble than in a text file).

4. Visual Exploration: Using Magic Patterns (21:37–28:59)

• Collaborative Process: The initial prototype is a conversation starter across product, design, and engineering teams.

  • Quote (21:37):
    “It's really something that helps me clarify my own thinking and ideas … and also just be a better collaborator because I understand system instructions better…” — Priya

• UI/UX Prototyping: Use Magic Patterns to visualize interface flows and experiment with UI changes through natural language prompts.

  • Inspiration Mode: A favorite feature that lets you rapidly generate and compare multiple design options for a single user flow. (25:33–27:33)
    • Quote (25:33):
      “So here I'm going to show you this really cool feature within Magic Patterns, which is called Inspiration Mode. … you can quickly explore lots of different options.”

• Rapid Iteration: These new AI-powered prototyping tools, says Priya, make product exploration and ideation dramatically faster than static mockups or manual Figma iterations.

  • Quote (24:34):
    “My mind was kind of blown, to be honest, the first time I used these like natural language prompting prototyping tools. … you can just describe what's in my head and actually have it... come to life in a prototype.”

5. Forking and Parallelizing Prototypes (32:55)

• Forking Designs: If an idea seems promising but isn’t the main path, you can “fork” it into a separate workspace and continue exploring it without derailing the main workflow.
  • Quote (32:55):
    “You can fork this design and it will create a totally separate window and chat for you of just that variant … you can just run off with that, maybe on the side…”

6. Non-Work AI Use Cases — Personal Prototyping (34:00–38:51)

• Slack Newsletter Generator: Priya uses Claude to process Slack channel transcripts and generate weekly community newsletters—demonstrating AI for content curation/automation.

• Parent Pal Artifact: Built with Claude for her own family, it simulates parenting advice for discipline scenarios—an example of personalizing AI artifacts for “micro” use cases.

• Settlers of Catan Timer: Priya rapidly prototyped a board game timer and leaderboard app with AI, tailored to her siblings’ requests.

  • Quote (36:11):
    “I think the really fun thing for this is that you can build something that's just really for your own personal use case … And it's a really fun process to do that.”

• Takeaway: Don’t have to be on an AI team at work—AI prototyping tools unlock “build your own tool” possibilities for anyone.

7. Debugging and Working with AI’s Limitations (39:11–40:22)

• AI is Not Human: When stuck, remember that LLMs lose context over long chats, so for large projects or after dozens of turns, start new chats or summarize the context to reset the system.
  • Quote (39:11):
    “I think when AI is not working and you've already tried some of the debug methods… take a step back and be like, this thing is actually not a human. What could be going wrong?”

Notable Quotes

• On Prototyping from the Conversation First
  “You're sort of using an example conversation as your first pass wireframe for building a conversational AI … and you're working backwards from that example conversation, which I have actually not seen anybody do before.” — Claire Vo (07:48)

• On Product Management Fundamentals
  “Work with the artifact that's closest to what the consumer is actually going to experience. And then you can back into all the requirements…” — Claire Vo (09:11)

• On Changing Prototyping Culture
  “This is an extension of that, like wireframing hacky Figma prototype process where … it's easier for someone to understand because they can actually click through and see the flow.” — Priya Badger (28:59)

• On Speed and Creativity
  “It also allows you to approach problems from the front door, the back door, the side door, the window … you can start at the end, go back to the beginning, come to the middle, fork off, go back to the beginning and re prototype. And it's not expensive, it's fast and it's interesting.” — Claire Vo (31:02)

Practical Takeaways & Tips

• Prototype with Conversations: Start with the real user experience—script examples first, then use AI to generate and refine them; work backward to requirements.
• Leverage Integrated Tools: Use tools like Claude Artifacts for rapid, “zero integration” prototyping and play out user flows before writing specs.
• Collaborate Visually: Use tools like Magic Patterns to quickly test multiple UI options, gather inspiration, and compress iteration cycles.
• Fork Early, Fork Often: Don’t limit yourself to one idea—parallelize experimentation.
• Personal Projects are Fair Game: Practice and experiment on personal or family projects to build AI/product skills.
• Debug Mindset: When LLMs get stuck, refresh the context—don’t keep pushing a broken chat.

Timestamps for Key Segments

| Timestamp | Segment Description | |:---------:|----------------------------------------------------------------------------| | 00:00–04:34 | Introduction + What’s different about managing AI products | | 04:51–08:50 | Golden Conversation approach & working backward | | 09:57–13:38 | Generating and iterating example conversations, pattern recognition | | 15:09–19:31 | Using Claude artifacts for rapid prototyping and system prompt generation| | 21:37–28:59 | Visual prototyping and ideation with Magic Patterns | | 32:55–33:36 | Forking prototypes for parallel exploration | | 34:00–38:51 | AI for personal projects and non-work use cases | | 39:11–40:22 | Debugging tactics and understanding AI limitations |

Where to Find Priya

• LinkedIn: Priya Matthew Badger
• Substack: Almost Magic Substack

Final Word

Priya Badger’s “golden conversations” workflow and creative prototyping strategies offer a roadmap for any builder confronting the ambiguity of AI-first products. Whether for work or play, her examples show that with the right tools, anyone can quickly move from an idea to an interactive prototype—learning and collaborating along the way.

(All timestamps in MM:SS format as per transcript.)