Did you also use Claude, and you like Codex better, or are you making a more general observation about the leapfrog in creative power agents are bringing to engineering?
I’ll tell you something. I love working with Claude. It’s enthusiastic, it’s nice, it’ll give you suggestions. It’s an all around pleasant experience.
I hate working with codex. It feels like a machine. You tell it to do something, and it just does it. No pretension at being human, or enthusiastic, or anything really.
But codex almost always does it right. And the comments are right, I never run into random usage limits. Codex doesn’t arbitrarily decide to shrink the context window, or start compacting again after 3 messages.
The codex client sucks, claude code is much better. But the codex client is consistent, which is much more important. Claude was amazing 3 months ago. The model is still fine, but the quality of the experience has degraded so far it’s hard to consider using it.
This is my experience as well. Codex is very verbose which is annoying considering the limits. My work flow tends to be have Claude code describe the problem (succinct as it can) based on my mashing of the keyboard description of what I want done then send that to codex. I've tried it the other way around doesn't work nearly as good. Disclaimer: not using the 5 prompts per week opus.
