I've been using OpenAI's Codex exclusively for last one week. I used Claude heavily before this.
Claude is great at answering questions. Codex is better at finishing work.
The difference shows up in long running task chains. I can hand Codex a bug, ask it to investigate, implement a fix, write tests, run checks, update docs, and it just keeps going at it until closure.
The best AI coding tools aren't the ones that generate the most code. They're the ones that require the fewest handoffs. 🤟🏻