Build a multi-device fitness app by proving the smallest voice-to-structured-data loop first, then use Cursor for scoped planning and implementation while Xcode, physical devices, version control, and privacy constraints provide the feedback the agent cannot infer.
Source episode · 03:00
How I AI: Terry Lin's Vibe Coding Workflow for Building an Apple Watch Fitness Appwith Claire Vo
Terry shows a voice-driven workout logger across Apple Watch and iPhone, traces it back to a voicemail-and-spreadsheet prototype, then explains his Cursor rules for PRD creation, independent review, small committed execution phases, refactoring, and learning the generated code.
A version-controlled iPhone and Apple Watch prototype that captures a spoken workout, shows the parsed record for correction, syncs it reliably, and presents basic history without hiding failures or losing data.
AI-first code editor
Step by step
Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.
7 steps
Prove the core job with the smallest possible loop: record a workout note, transcribe it, parse it into a spreadsheet or local fixture, and correct the result. Use this to define the schema and failure cases before building two apps.
Use Cursor for code navigation and edits, but use Xcode for signing, entitlements, builds, simulators, device logs, and Watch deployment. Start from a clean repository and document the exact build targets and commands.
Write a scoped PRD for one end-to-end milestone. Define the phone and watch roles, data model, account and sync behavior, privacy, offline states, correction flow, analytics, non-goals, and acceptance tests.
Write a PRD for the next milestone of [app]. User job: [job]. Devices: iPhone and Apple Watch. Define the data model, source transcript, parsing and correction, sync states, authentication, offline behavior, privacy, error recovery, non-goals, and physical-device acceptance tests. Label assumptions and open questions.
Have a fresh model review the PRD as if it had no prior context. Ask it to identify missing interfaces, platform constraints, failure states, and ambiguous requirements, then resolve the findings in the document.
Review this PRD with no assumed context. List missing decisions, contradictory requirements, Apple platform or entitlement risks, sync and offline edge cases, privacy gaps, and tests another agent would need to execute it. Do not invent answers. Rate implementation readiness only after listing evidence.
Treat the readiness score as a summary. The cited gaps and unresolved decisions are the useful output.
Turn the reviewed plan into small phases with explicit files, tests, and stop conditions. Commit before the phase, run it, inspect the diff and device behavior, then commit the passing result before continuing.
Implement phase [number] only. Stay within [files and behavior]. Build the affected targets, run [tests], and stop on the first failure. Do not use placeholder data or weaken checks. Summarize the diff and wait before the next phase.
After the flow works, identify large or tangled files, duplicated logic, and high-context modules. Refactor one seam at a time with behavior tests and physical-device builds so cleanup does not silently break sync or recording.
Propose a behavior-preserving refactor for [area]. Explain the current responsibilities, target boundaries, migration phases, tests, build checks, and rollback. Change one phase at a time and stop if behavior or device builds differ.
Ask Cursor to explain each changed component, data path, platform API, and failure state in plain language. Trace one workout from recording through parse, correction, sync, and history, then record that explanation in the pull request or project notes.
Explain this implementation to its owner. Trace one workout from watch recording through transcript, parsing, user correction, phone sync, storage, and history. Cite the relevant files and functions, identify failure handling, and quiz me on the parts I should understand before maintaining it.
Turn an idea into a PRD, user stories, and a plan.
After the steps
How to recover when the loop fails and where human judgment helps.
Recover
Keep building
Design a weekly meta-workflow that analyzes your interactions with your AI assistant to make it smarter. This system learns your writing style from edits, suggests new automations for repetitive tasks, and helps you filter valuable AI techniques from hype.
Create a daily AI workflow that summarizes yesterday's meetings, identifies urgent tasks, and proactively learns internal company terms it doesn't understand. This keeps your AI assistant's knowledge current and highly relevant to your work.
Build a recurring AI task that automatically gathers updates from your calendar, Slack, and notes to suggest weekly priorities and prepare you for upcoming meetings. Start every Monday with a clear, comprehensive plan without the manual effort.
Join 100,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.