Here’s the full initial SPEC Claude generated after the first day of How to Vibe Code an App in 10 Days
# Favorite Places App — Specification
## Overview
A personal iPhone app for saving favorite
locations, organizing them with folders
and tags, and quickly opening them in
any installed maps application.
## Target User
Personal use (single user, no multi-user
or sharing features needed).
## Core Features
### 1. Location Search & Saving
- Search for places by name or address
using Apple's MapKit search
- Save a place with:
- Name (editable, defaults to search
result name)
- Address
- Coordinates
- Personal notes (optional)
- Folder assignment (optional)
- Tags (optional)
### 2. Organization
- **Folders:** User-created folders to
group places (e.g., "Restaurants,"
"Travel," "Work")
- **Tags:** User-created tags for
cross-cutting categorization (e.g.,
"date night," "outdoor,"
"kid-friendly")
- Places can exist outside any folder
(uncategorized)
- Places can have zero or multiple tags
### 3. Browsing & Viewing
- List view of all saved places
- Filter by folder or tag
- Search within saved places
- Tap a place to view details and edit
### 4. Open in Maps App
- Primary action button opens the place
in the user's default maps app
- Secondary option shows a list of other
installed maps apps (Apple Maps,
Google Maps, Waze, Citymapper, etc.)
- Default maps app is configurable in
settings
### 5. Data & Sync
- All data stored in iCloud (CloudKit)
- Automatic sync across user's devices
- Works offline with local cache; syncs
when connectivity returns
## Technical Approach
- **UI Framework:** SwiftUI
- **Minimum iOS:** 26
- **Location Search:** MapKit /
MKLocalSearch
- **Data Persistence:** SwiftData with
CloudKit sync
- **Maps App Integration:** URL schemes
for supported apps
## Out of Scope (for initial version)
- Sharing places with other users
- Map view of all saved places
- Import/export
- Widgets or Shortcuts integration
- Photos attached to places
Spoiler: it was three days
Last Sunday I started building an iPhone app using Cursor and Claude Code.
Back in October I listened to an episode of Syntax called Is Coinbase Really Writing Half Their Code With AI? and the main takeaway was Coinbase primarily uses Cursor and Claude Code for development.
I’ve never built an iPhone app before - I’ve never even written a line of Swift - but I have spent the last five years working as a Quality Engineer for a well known international food delivery app company. I’ve had an iPhone since launch, and a Notes folder full of unrealised app ideas, so this felt like the perfect opportunity to see what is possible in December 2025.
The Idea
An app to bookmark your favourite locations:
- Central storage not tied to a specific map app
- Can open places in any mapping app on your device
- Solves the problem of favourites being locked into specific apps
The Approach
I needed a PM (Product Manager) to define what the app should do, a dev team to build it and a QE (Quality Engineer) to ensure it worked as specified. I would be the PM & QE, with Cursor and Claude filling in as my dev team 🤜🤛.
The Rules
This must not take over my life, and must fit into chunks of time in the evening and lunchtime.
Day One
I signed up to a three month half price offer of Claude Pro (thanks Upgrade!) and started with the following prompt in Opus 4.5:
I want to create an iPhone app with Claude Code. I have setup a repo on Github and want you to ask any questions you need to create an initial spec
We went back and forth for a bit until Claude produced a full SPEC.md file.
I then fired up Claude Code and told it to get working. After about an hour I had something that would run in the iPhone Simulator and I also burned through my usage allowance, which wouldn’t reset for five hours.
At that point I jumped over to Cursor, started a seven day trial and told it to carry on executing the plan. Again, after an hour I’d used up the usage allowance but I had an app you could add places to, sort them into folders and swipe to delete entries.
It was hitting midnight now, which seemed an ideal time to stop (see The Rules) but I wanted to get it running on my actual phone.
I hit a few more snags and as Cursor and Claude refused to talk to me anymore, I hit up ChatGPT and finally I had an app running on my phone.
There was still plenty to do, but Claude and Cursor kept the NEXT_STAGE_PLAN.md doc up to date, so we could pick everything up tomorrow when everyone had had a well earned rest.
*it’s not actually Vibe Coding, it’s Spec Driven Development
(I have no idea if the screen recording will actually post - there’s literally no way to test this without just doing it 😩. If it’s not in the main post, I’ll thread it in a comment)
(Video edited for length, steps removed - turns out Micro.blog has a 60s limit EDIT: using Vimeo instead as Micro.blog munges the video quality)