Join 5,901+ iOS/macOS developers using Codex and agents to build and ship apps. Expect practical tutorials, repeatable workflows, and hard-earned lessons from 7 shipped apps and time at Apple, Microsoft, and GoPro.
I Tried to Break GPT-5.2 — It Kept Working (SwiftUI + macOS)
|
New Codex models are incredibly good for Swift code for iOS/macOS apps. You can easily install codex via here Just open Terminal/Ghostty and type: npm i -g @openai/codex
Using codex, you can use simple language to talk to your agent. See the first video if you want a breakdown on the tools I am using, and some of the rules in my agents.md rule file. 1. How I use Codex GPT 5.2 with Xcode (My complete workflow) 2. ❌ How I Stopped Codex From Breaking My Builds 3. I Tried to Break GPT-5.2 — It kept Working 4. 🔴 LIVE Recording: GPT-5.2 Codex vs Real-World ML App (Spider Detection) |
AI Code Brief - Paul Solt
Join 5,901+ iOS/macOS developers using Codex and agents to build and ship apps. Expect practical tutorials, repeatable workflows, and hard-earned lessons from 7 shipped apps and time at Apple, Microsoft, and GoPro.
Hey Reader, My last app launch was a flop. I wrote the App Store copy after I’d already submitted for App Review — and I knew it. If you’re in that same boat, that’s okay. But don’t stop there. This week’s article breaks down the research workflow I should have used from the start: using AI and competitor reviews to write copy before you design a single screenshot. Read: The Screenshot Mistake That’s Costing You Downloads Every Day Grab the free 100 Best Practices for App Store Screenshots...
Hey Reader, Before you build iOS app screens, pick colors, or hire an artist, you need a visual direction. I put together an article on four ways I explore the style of an iOS app or game so I can move from vague idea to something real. Read it here: 4 Ways to Explore the Style of Your iOS App or Game Dangerous Snake App UI Concepts using Pencil Learn about: Mood Boards Hiring an artist Sketch prototypes ChatGPT visual design Agentic UI design AI is now good enough to help with a big part of...
Hey Reader, Agents write better code when they can read docs instead of guessing from memory. A 2026 paper found up to a 20% improvement in first-try code accuracy when models were given the right programming context. So what do you do with that? Don’t rely on the agent to know everything from their training set. Instead, provide tools or local docs that they can reference. Your local docs should include sample code so agents can use the code correctly (the first time). If you’ve ever read...