The tool is powerful, but the quota disappears far too quickly for serious coding work.
With terminal access, repo analysis, web research, and high reasoning enabled, you can hit the limit in under an hour of real usage. At that point, the pricing starts to feel disconnected from the actual amount of productive work you can get done.
The result is frustrating: just when the workflow becomes useful, the session gets cut off.
For this kind of product to make sense for developers, the heavy-use limits need to be much higher, or the pricing needs to change.
I just wanted to mention that you can use your regular access to ChatGPT to help you plan and reason.
Explore the models available in your codex as well, as each is tuned for different aspects…
While you can certainly reason within Codex, it sounds like chatting a bit with ChatGPT can help you understand how to make tactical agent commands… Prompt it for tips on using it effectively if that’s the concern.
Generally, having very specific goals and a prompt that is crafted for those goals, before you even start coding up… is very helpful in keeping the agents focused.
I agree that the tool is powerful and feel that the Codex can hit limits fast. My personal workflow is to use ChatGPT for the “thinking part” and then hop over to Codex once I have all the requirements, research, and acceptance criteria all figured out. It’s even helpful to have ChatGPT create a prompt for you which you can guide it through and then paste that prompt over to Codex.
I started on Pro, but I ended up moving to Business because even coding solo I was hitting the limit within a few hours.
That said, I was doing fairly heavy work: a complex system and a CJS server of about 12k lines, so this wasn’t exactly light usage.
With Business, it’s been better so far. I haven’t pushed it to the max yet, but I did get a warning once, and about 30 minutes later it cleared by itself.
So for now, Business seems more usable than Pro for my workflow.