I was done with Claude. Not because it was bad—but because it was too much.
Every time I asked it to write a Ruby method, it came back with a novel: three comments explaining the obvious, a safety check for a null pointer that couldn’t exist, and a paragraph of apology. Codex, on the other hand, just… gave me the code. Fast. Clean. No editorializing.
And I realized something: We’ve been having the wrong debate about AI coding assistants.
Everyone is obsessed with which model scores higher on HumanEval. Which one writes more accurate Python. Which one is “smarter.” But after a week of using Codex as my primary tool—and switching back to Claude for comparison—I found the real differentiator isn’t intelligence. It’s personality.
Claude is the overachiever who writes you a 10-page report when you asked for a bullet point. Codex is the senior dev who gives you the answer and walks away.
One developer in the comments nailed it: “Codex seems tuned to not spit out word vomit, both in its chat interactions, and its code. Claude is obsessed with mass.”
That’s not a bug. That’s a design philosophy—and it has massive implications for how you should choose your AI assistant.
Let me give you a concrete example. I was building a scraping and data processing pipeline. I gave both Claude and Codex the same prompt. Codex produced 30 lines of clean, executable Ruby in under 2 seconds. No comments, no fluff—just the logic. Claude came back with 60 lines, including a provenance tracking system, validation for three different edge cases, and a comment on every single block.
One tool respects your flow. The other is terrified of being wrong.
And here’s the twist: the developer who prefers Claude’s approach isn’t wrong. If you’re building a safety-critical system, you want overengineering. You want comments. You want the model to second-guess itself. But if you’re iterating fast, prototyping, or just trying to unblock yourself—Codex is faster, because it trusts you.
One commenter put it bluntly: “While everyone is still stuck on and fascinated by Claude, I’ve been using Codex every day. It’s just faster. Period.”
This is the real war nobody’s talking about. It’s not Claude vs. Codex on accuracy. It’s speed vs. safety. Concision vs. completeness. Trust vs. paranoia.
And the choice isn’t universal. It depends on your workflow. For Rails developers who hate reading comments that restate the obvious? Codex is a revelation. For a team shipping medical software? Claude might be the safer bet.
But here’s my hot take: most developers are overvaluing “safety” and undervaluing speed.
We’re conditioned to think more code = more thorough. More comments = more professional. But every extra line is a cognitive load. Every bloated response is a tax on your attention. The best code isn’t the most documented—it’s the code that solves the problem and gets out of your way.
I’ve been using Codex for a week. I’ve gotten more done in that time than I did in the previous month with Claude. Not because Codex is smarter. But because it respects my time.
The best AI is the one that knows when to shut up.
FAQ
Q: Isn't this just a matter of personal preference? How can you claim one is objectively better?
A: It's not about objective better—it's about fit. Codex is better for speed and iteration; Claude is better for safety and auditability. The article argues most developers overvalue safety in their day-to-day work, which is a trap.
Q: So should I switch from Claude to Codex right now?
A: Not necessarily. If you're doing prototyping, scripting, or Rails work, yes. If you're building medical software, financial systems, or anything where a bug costs lives, stick with Claude. Know your domain.
Q: What about the commenter who said fewer comments is a bad thing? Doesn't that hurt readability?
A: That's an old-school mindset. Good code is self-documenting. Comments that just restate what the code does are noise. The real question is whether the code is readable on its own—not whether it has a comment on every line.