Using Codex also has new pain points:
Sluggish architecture migration is a common problem for every AI, especially noticeable on GPT 5.4, with various hacky compatibility layers.
Additionally, Codex adds test coverage every time it makes changes, and the test files are bigger than the implementation! While it looks stable, it's really slow.
The result is that refactoring is extremely, extremely slow! Plus, every time I have to say "continue", I've been sitting at my computer for a full 2 days and Codex is still doing the same thing!
The image below shows Claude's sharp criticism of Codex:
查看原文