OpenAI is soooo bad in terms of UX and AX. I've been pushing myself to only develop in cloud environments (as this is a prerequisite for serious autonomous AI development). Claude spins up cloud environments in a few seconds; they work smoothly, and the agents are self-aware of where they are enabling it to improve its environment and settings for future agents through new iterations. Codex, on the other hand, takes a few minutes to spin off a new environment; it sometimes works, it sometimes doesn't, and the agents aren't self-aware of where they are, nor are they aware of the tools to configure the environment.

Subscribe to my weekly newsletter of hot takes: