Cursor vs. Codex: A Personal, Unscientific Take After Actually Using Both

I want to say upfront: this is not a review, not a benchmark, and definitely not a "which one should you buy" article. I'm not sponsored by either company, I don't have a horse in this race, and I'm not trying to crown a winner. This is just what it felt like, day to day, to use Cursor and Codex as someone who writes code and ships things. Take it as one developer's subjective impression — nothing more.

The First Thing I Noticed: Familiarity

Cursor is built on top of VS Code, and since I'd already spent years living inside VS Code, opening Cursor for the first time felt like walking into a room I already knew. Same layout, same muscle memory, same keyboard shortcuts I didn't even have to think about. There was zero ramp-up cost. I could focus entirely on figuring out what the AI features could do, instead of also having to relearn how to navigate an editor.

Codex, by contrast, felt to me more like a chat application than a development environment. The interface leans heavily into a conversational, message-based experience rather than the file-tree-plus-editor layout I associate with "an IDE." To be fair, I want to flag my own bias here clearly: I'm simply far less experienced with Codex, so some of that unfamiliarity is on me, not necessarily a flaw in the product. First impressions are colored by what you already know, and I already knew VS Code inside and out.

What I Actually Missed When Using Codex

Once I got past the surface-level "which one feels more like home" reaction, a few concrete differences stood out to me in daily use:

Seeing the whole project at a glance. In Cursor, the file tree sits right there on the left side, exactly like VS Code. I can browse the project structure, open any file instantly, and keep a mental map of the codebase while I work. I didn't get that same at-a-glance structural view with Codex. Selecting code and dropping it straight into the conversation. In Cursor, highlighting a chunk of code and sending it directly into the chat panel is completely frictionless — it's a core part of the workflow. I didn't find this as natural in Codex; it felt like more of a detour to get a specific snippet into the conversation. Switching models on the fly. Cursor lets me pick between different underlying models depending on the task — sometimes I want a faster, cheaper model for a trivial fix, sometimes I want to bring in a stronger reasoning model for something gnarly. Having that choice sitting right in the UI, without leaving my workflow, felt genuinely useful. Ask / Agent / Plan modes. Being able to explicitly tell the tool "just answer my question," versus "go ahead and make the changes yourself," versus "lay out a plan before touching anything" gave me a sense of control that matched how I actually think through a task — sometimes I want a second opinion, sometimes I want to fully delegate, and sometimes I want to see the plan before committing.

Put simply: in my day-to-day, interactive coding sessions, Cursor felt like it was built around how I already work, and Codex felt like something I had to adapt myself to.

The Caveat I Have to Be Honest About

Here's where I want to slow down and be fair, because I don't think it's an accurate picture to just say "Codex loses." From what I've read and pieced together since using it, Codex's design isn't really trying to compete on the same turf as Cursor's real-time, in-editor experience. It leans into an asynchronous, delegate-a-task-and-walk-away model — you describe what you want, it goes off and works on it independently (sometimes in a sandboxed environment), and you come back later to review the result. That's a genuinely different philosophy from Cursor's "AI sits beside you while you type" approach.

If that's the intended use case, then judging Codex by how it feels for the kind of moment-to-moment, hands-on coding session I was running it through is probably not entirely fair to the tool. A chat-style interface actually makes more sense if the core interaction model is "hand off a task and check back," rather than "watch and guide every step." My experience says more about the specific way I was using it — mostly interactive, exploratory, one-file-at-a-time work — than it does about which tool is objectively "better."

So Which One Do I Actually Reach For?

Honestly, right now, it's Cursor — almost entirely because of the environment I already knew and the interactive habits I already had going in. But I'd be lying if I said that's a fully objective conclusion. It might just as easily be "the tool that matched my workflow" rather than "the better tool," and those are two very different claims.

If your day-to-day looks like mine — sitting in an editor, going back and forth with the AI line by line, wanting to eyeball every change before it lands — Cursor's approach is likely to feel more natural out of the gate. If your workflow leans toward handing off well-defined, self-contained tasks and reviewing the output later, Codex's model might fit you better than it fit me.

Either way, this isn't really an "X beats Y" story. It's a reminder that the "best" AI coding tool has a lot to do with how you already work — not just which one benchmarks better on paper.