Hey IH!
I just kicked off a pretty ambitious challenge: building 100 iOS and Apple ecosystem apps in 100 weekdays, completely in public.
I think LLMs are still terrible at shipping real iOS apps. Not because they can't write Swift or AppKit, they actually can. But iOS development has hundreds of invisible gotchas that models simply can't see: entitlements, capabilities, background modes, simulator vs device quirks, provisioning hell...
Your code looks correct. Your app still won't build.
So I'm fixing this the only way I know how - by actually building the apps, one per weekday, and documenting every single lesson along the way.
I fed Claude Opus 4.6 over 100 carefully chosen real open-source iOS apps and had it extract every pattern, architecture decision, dependency, gotcha, and framework-specific workaround. That became my starting foundation, the Apple-Knowledge-Base.
Now I'm using Claude Code with that context to build each app from scratch - no templates, must run on a real device, and ships to TestFlight or gets parked with a full explanation of why it failed.
Small utilities. Creative experiments. Framework deep-dives. Cross-device ecosystem apps. Nothing that needs a backend. Pure iOS craziness.
App #1 - JSONCrush: A developer tool for JSON validation with recursive tree parsing, syntax highlighting, and JSONPath extraction. Built in 1h 18min, 21/21 tests passing, shipped.
App #2 - Regex-Forge: An interactive regex tester with real-time match highlighting, color-coded capture groups, and a library of 16 common patterns. Built in 1h 20min, 8/8 tests written, shipped.
The goal isn't 100 perfect apps. It's 100 attempts that surface real lessons about what works and what breaks when you're building iOS apps with AI.
Follow along if you want to watch the mess, the lessons, and the wins in real time.
What iOS framework would you want to see tackled? WidgetKit? CoreML? ARKit? Let me know!
Links to the first two apps: