Hey Reader,

My iOS and macOS app development workflow has changed since my last GPT-5.2 Workflow video.

GPT-5.4 lets me iterate on bug fixes, features, and ideas much faster than before.

Watch: How I Build Apps with Codex and GPT-5.4​

​

If you want to build apps with agents, use my app-creator skill. It can scaffold a new Xcode project and teach your agent how to create a Makefile for an existing Xcode project.

2 Tactics You Can Steal

#1 Use a Learnings.md File with Your Agent

Self-improvement is one of the holy grails for agents. Give them a place to capture lessons so they can avoid repeated mistakes.

Ask your agent:

Periodically, ask the agent to review the conversation and append new learnings. Over time, ask it to reorganize the file for future agents.

#2 Just Talk to It

GPT-5.4 and Codex 5.3 work well with a full plan, but they also handle small fixes with very little detail.

I prefer a hands-on workflow: request a small change, test it, and steer from there. Skip the big plan. Take a screenshot, draw a sketch, or ask for one small task.

You don’t need to be comprehensive. Let Codex plan just-in-time.

When the request is small, you can test the result in minutes and change direction before you fully know what you want.

My Workflow:

1. Ask for a small feature using Wispr Flow
2. While it works, run the app
3. Record your next feature request
4. Queue or steer the agent with the new work
5. Repeat

This loop is much faster than spending 30 to 60 minutes planning and re-planning a feature. Play testing is more fun than creating a master document.

Reserve “Plan mode” for initial app concepts or larger features it can build while you sleep.

Pro Tip

Enable the new Fast mode with the latest Codex CLI (and Codex app).

Type /fast

It runs about 1.5x faster (and uses 2x tokens).

5 Resources and Links

1. OpenAI launched a brand new all-in-one thinking + coding model. What GPT-5.4 Actually Means for Developers.
2. Need more intelligence? Dive into the details behind ChatGPT 5.4 Pro and Thinking Models.
3. Up-to-date documentation makes agents write better code. Try sosumi.ai for better docs from NSHipster.
4. GPT-5.4’s 1M Context Window sounds great on paper, but there are 2 flaws. Read all about when to use a 1M Context Window here.​
5. SwiftUI is hard for agents, even in GPT-5.4. Grab this skill from Paul Hudson, and your agent will thank you.

Community Spotlight

1. Smarter agents write better code. Steal Cameron’s Self-Improving Agents.md file.
2. Visualizing task management can give you insights. This winning hackathon project does just that.
3. 4 minutes 27 seconds to build an iOS workout app. Try the app-creator skill today.​

Quick Question

What model do you prefer for coding? GPT-5.4 or GPT-5.3-Codex?​

Talk soon,

Paul Solt
Paul@SuperEasyApps.com

P.S. Next week, we’ll talk about how to improve your documentation.
​
Do you have any documentation tips or MCPs you use?