I’m kinda new to Codex CLI and honestly… I think I found a really nice flow 😄🔥

I tried making a few adjustments on
https://github.com/ruiaureliano/sf-symbols-lite

Not sure if this is the “best” way, but it feels super safe and smooth for me.

This is basically how I do it:
☛ Ask for small changes only
☛ Review the suggestion and accept it if it looks right
☛ Build the project in Xcode to make sure everything works
☛ Open the GitHub client and carefully check the diff

Today I used Codex CLI for the first time.
I ran it locally against my PHP API to detect issues and possible improvements.

The first run completely broke everything 🤣🤣🤣

So I changed the approach and asked it to work step by step, showing each change so I could accept or reject it.

Much better experience. I’m actually happy with the issues it detected and the improvements it suggested.

When I first started using AI to help with code, it felt genuinely useful. Answers were clearer, suggestions made sense, and it often saved time.

Lately, though, it feels confusing as hell. Many answers are vague, wrong, or overconfident, and I spend more time double checking than actually moving forward.

That makes me wonder, did AI actually get worse, or did we just get used to it and become more demanding? Are our expectations higher now?