coco
6.6 KiB
title, source, author, published, created, description, tags
| title | source | author | published | created | description | tags | ||
|---|---|---|---|---|---|---|---|---|
| How I Build Apps with the Codex App and GPT-5.5 | https://x.com/PaulSolt/status/2048226299290828936 |
|
2026-04-26 | 2026-04-26 | My workflow with GPT-5.5 is completely different from my workflow with GPT-5.4 — and this literally changed this week. The GPT-5.5 model is ... |
|
My workflow with GPT-5.5 is completely different from my workflow with GPT-5.4 — and this literally changed this week. The GPT-5.5 model is smarter, the Codex app is faster, and it finally feels like I have the perfect coding partner. Not just for code — for everything around the app. I'm shipping app updates and doing app research in less time than before.
GPT 5.5 + Codex app
Medium reasoning is fast. Surprisingly fast. It feels like I get working code in seconds for small tasks, which means faster iteration loops and more progress.
The Codex app is my go to right now. Codex CLI is still great when I want speed in switching conversations, but the app gives me image previews, diff tools, and a better workspace for bigger problems. Performance keeps getting better week over week.
I just used Codex to revive my 9-year-old Super Easy Timer app, which is pretty much dead. It went from 0.5% conversion rate to now over 7.1% conversion rate.
Super Easy Timer (now free): https://apps.apple.com/us/app/super-easy-timer/id1353137878?mt=12
Just Talk to It
The model is smart enough now that you just describe what you want.
No prompt ceremony. No plan files. Tell GPT-5.5 in Codex what you want, then let it run.
With 5.4, if I asked for too much at once it would go off the rails. With 5.5 I'm just driving it in real time. I describe and steer it interactively with what I want. Then I test it and give it feedback for the next loop.
A fast interaction loop lets me drive the code quickly. GPT-5.5 is smart and fills in a lot of gaps without me asking. Try talking to it.
I use Wispr Flow for voice dictation — though Codex's new global voice dictation is almost a replacement.
Agents Can Create Xcode Projects
Agents need to know if the code compiles and the tests pass. If you're struggling with having agents work on your apps, you need to try this skill.
AppCreator creates new Xcode projects and allows agents to see build failures.
I created App Creator skill to solve my own problems:
- It's hard to spin up new iOS/macOS apps.
- Agents report success when things are broken.
Your agent should always hand back working code, which is what the App Creator skill solves. The App Creator skill lets you create new Xcode projects without opening Xcode. The second part is that it gives your agents the tools to make a repeatable build/test loop. It can't lie about the build being green when it isn't.
Agents should always be able to verify their work before handing it back to you.
Note: OpenAI shipped the Build iOS Apps and Build macOS Apps Plugins. Those work great on top of the foundation from my App Creator skill.
Real Documentation Reduces Hallucinations
Agents have a fuzzy understanding of code, so for critical code that works with Core ML or AVFoundation — You'll want to use the actual APIs.
DocSetQuery for fast, local Apple Documentation
DocSetQuery pulls the local Apple DocSet into clean Markdown and lets the agent search it. Fewer hallucinated APIs and errors.
Research the Market Together
A real partner doesn't just help you write the app. It helps you figure out which app to build.
Shipping an app is a multi-discipline skill.
It requires:
- Coding
- Marketing
- Positioning
- Copywriting
- Design
As a one-person team, it can be hard to do all of these well. So I have been leaning into Codex to help with more and more App Store research.
I created 3 new skills for helping me see what the competition is doing, and to help me position my ideas.
The three tools are:
1. App Store Icon Grid
See what icons look like for any search query. I wanted this ability after using Astro to look up keywords. I love to place my concept app icons on the grid to see how they compare.
Top 16 App Store icons for "photo collages" for competitive analysis.
2. App Store Screenshots
See a grid of the top screenshots. This can help you see what messaging is resonating with users. The top apps use the copy they use because it works.
When you think about your App Store page, consider that on search results users will see 3 screenshots on iOS and 1 screenshot on macOS.
Top 6 "Photo collage" apps on iOS.
3. App Store Reviews
Now you can do more user persona research, figure out features customers love or hate, and position your app.
If you want to build apps, having tools like these three will enable you to build iOS and macOS apps with ease.
App Store reviews from Pic Collage.
Use these tools in Codex to help you research the App Store and find a competitive niche. Take an existing idea and put your own spin on it. Make sure you don't blindly copy an app, or Apple will remove your app.
Download These Skills for Your Next App
Grab all of the tools that I use to make apps:
- Download the App Store Research CLI tools
- Download AppCreator for iOS/macOS App Development
- Get fast local documentation for agents to reference
What's your current Codex setup? Reply below.
Building iOS/macOS apps with AI tools? Follow @PaulSolt for the workflows I'm actually using.
P.S. Need help with App Store Screenshots? Read my other article: The Screenshot Mistake That's Costing You Downloads Every Day