The Codex desktop app doesn't change everything - but it's part of a larger trend worth paying attention to. Where IDEs are headed and why specs matter more than code.
No, it doesn’t. The Codex desktop app dropped yesterday. You’ll see breathless Twitter posts and YouTube videos about how it changes everything. It doesn’t. But it is pretty cool, and it’s part of a larger trend worth paying attention to. I’m going to talk briefly about how it’s changing my workflow, and then zoom out to what it means that this app exists at all.
My Workflow (For Now)
I’ll write a longer post on this, but the quick version:
My primary driver is Claude Code in the terminal. I think it has the best features, the most hooks, and the most ability to create a clean development workflow with all the checks I want.
The Codex app is my parallelization layer. The thing that’s cool about it (and Conductor, which is similar) is that it makes Git worktrees easy to use. That means real parallelization.
Here’s how I’m experimenting with it:
I have my main feature or project running in Claude Code in a terminal window Whenever I come up with changes, bug fixes, or investigations outside the scope of that feature, I spin up a worktree in the Codex app. I can chat with it separately. It lets me know when it needs input. It’s totally isolated, and I can merge it back whenever I want.
The TLDR: Codex app is OpenAI’s supported UI for multi-agent parallelized development. In my workflow, I use it to develop small features in parallel while I’m working on the main thing in Claude Code.
The Bigger Picture
... continue reading