Have fully switched to Codex the last few days. Codex mac app is pretty good, probably better than Claude’s native Mac app (product segregation, multi- and sub-agent). The only problem is to judge whether to use GPT-5.4 or Codex 5.3. Not sure why they have made it so complicated. Realizing you can’t have dependency on just one AI provider, and have a cross-check pipeline that keeps auditing your code between different AI models.