Terry Lin’s Vibe Coding Workflow for Building an Apple Watch Fitness App

By: Terry Lin Host: Claire Vo Source: How I AI (ChatPRD) Type: podcast

Summary

How I AI episode with Terry Lin, product manager and vibe coder. Two workflows for building Cooper’s Corner, a voice-powered fitness tracker for iPhone and Apple Watch: (1) Structured vibe coding process — “dual-wielding” Cursor (for AI-assisted coding) + Xcode (for compiling, simulator, debugging); strict three-phase AI cycle with custom Cursor Rules: PRD Create (takes Linear ticket → full PRD with goals, references, Gherkin user stories), PRD Review (second AI agent rates plan out of 10, keeps iterating until 9+), PRD Execute (breaks work into phased checklist, git commit after each phase for easy rollback); “vibe refactoring” rule to keep files under 400 lines by splitting “God mode” files so AI works more efficiently; “Rubber Duck” rule where AI explains new code and creates pop quizzes to ensure learning; (2) Index card prototyping — sketch mobile UI on physical index cards (approximates phone aspect ratio), photograph with phone camera, upload to GPT-4 Vision with “help me upscale this,” bring AI-generated mockup into UX Pilot → Figma with Apple’s official UI Kit. Started with a scrappy V1: Apple Watch voice memos → Python script → GPT-4 transcription → structured Excel data, proving the concept before building a real app.

Key Ideas Extracted

Three-phase AI workflow: Create, Review, Execute: Custom Cursor Rules for each phase — PRD Create builds full spec from ticket, PRD Review has AI score the plan 9+/10 before coding starts, PRD Execute breaks work into phased checklist with git commits
“Dual-wielding” Cursor + Xcode: Cursor for AI-assisted code generation, Xcode for compiling, simulator testing, and catching build-specific errors — necessary for iOS development
Vibe refactoring for AI optimization: Keep files under 400 lines by splitting large files — LLMs waste tokens reading oversized files, smaller files = more efficient AI collaboration
Rubber Duck rule for learning: AI explains its own code and creates pop quizzes — ensures the builder understands what was generated, not just along for the ride
PRD Review as quality gate: Second AI agent reviews the plan and rates it out of 10 — prevents garbage-in/garbage-out by ensuring instructions are clear before execution
Git commits as risk management: AI pauses after each execution phase and commits — creates a trail of small changes for easy rollback, builds trust for more complex tasks
Index card → GPT-4 Vision → Figma pipeline: Physical sketches on index cards → AI upscaling → UX Pilot → Figma with native UI Kit — go from subway idea to polished prototype in minutes
Start with the scrappiest V1 possible: Voice memos → Python → Excel proved the concept before any app development — classic PM discipline applied to vibe coding

Notes

Published Sep 15, 2025 on How I AI (ChatPRD). ~9 min read. (Note: filename says 2025-07-28 but actual publication date is Sep 15, 2025)
Sponsors: Paragon, Miro
Terry Lin background: Product manager, vibe coder, self-described “AI-powered gym bro”
Tools: Cursor, Xcode, Linear, GPT-4 Vision, Figma, UX Pilot, Apple UI Kit, Python, Apple Watch voice memos
App: Cooper’s Corner — voice-powered fitness tracker for iPhone + Apple Watch
Key concept: “Vibe refactoring” — clean up code for AI, not just humans
Two companion workflow guides published Jan 8, 2026
Cross-references: Cursor Rules, structured vibe coding, mobile development with AI

Raw Content

Re-scraped from ChatPRD 2026-02-16. Full article content captured in Summary and Key Ideas above.