The Accidental Perfect Workflow

Why using Claude Code to write and Codex to review actually makes sense

A few months ago I wrote about learning to love the terminal. Since then I've been shipping stuff with Claude Code pretty much daily... Efecto, Remoto, internal tools, even organizing my book.

But somewhere along the way I picked up a weird habit. After Claude Code finishes building something, I run it through OpenAI's Codex CLI for a second pass. Not to rewrite anything... just to review.

Felt silly at first. But it works?

Fresh Eyes

Claude Code is great at building. It holds a ton of context, knows my patterns, generates fast. But it moves on quickly. Eager to hand things back to me.

The problem is when something slips through. A subtle bug. An edge case. Claude built the thing, so it's not great at finding its own blind spots.

Codex approaches the code like a stranger. No history of why decisions were made. Just "does this actually work?"

And it catches stuff. Off-by-one errors. Unhandled null cases. Logic that works for the happy path but breaks elsewhere.

The Workflow

Build with Claude Code. When I think it's done, paste the code into Codex: "review this for bugs and edge cases, don't rewrite it, just tell me what's wrong."

Take those findings back to Claude Code to fix.

Adds maybe ten minutes. Catches stuff that would've taken hours to debug later.

Why It Works

I think there's something to having a second model with no context. It can't assume things. It has to actually read the code and understand it fresh.

Claude knows what I intended. Codex only knows what I wrote.

That gap is where bugs live.

Anyway

I didn't plan this. I just noticed my code quality went up when I ran things through a second model before calling them done.

Your workflow doesn't have to be clever. It just has to work.

I'm @pablostanley on Twitter if u wanna chat.

---

This is a follow-up to How I Stopped Worrying and Learned to Love the Terminal.

also, try a tool I've been working on: https://efecto.app/