Is OpenAI Codex Better Than Claude Code?

Hey, this is Isaac from upstart.co. And in this video, I wanna share a little bit more about my first impressions on OpenAI Codex app. This is the desktop app that's directly competing with Claude Cowork and Claude Code. And I've just had a little bit of an experimentation period here to get a sense how it looks like, how good it is, and how does it compare to Claude Cowork. And I wanna share that with you. The first thing I noticed when I came here and I was trying to like connect my apps, for some reason I'm struggling to do so. I don't know if it's the internet connection or something else, But I did notice that this is like entirely or seems entirely separate from the things you already have connected to your ChatGPT account. So that's the first things I noticed. It seems to be a very somewhat disjointed and disconnected experience, Codex compared to what you already have on your Claude account, which to me strikes me as a little odd. Perhaps there's something I don't know there. Just yes, leave it in the comments if that's the case. But that's one of the first things I noticed and started to realize. I know that, for example, in Claude Cowork, you have, you can switch between Claude Chat, Claude Cowork, Claude Code. That's a little bit more evident. I do notice here that the version of that within OpenAI's Codex is this, where you can go from coding to everyday work. Although I don't seem to understand fully the difference. It seems like you won't be able to see the codes that it's running. Other than that, I couldn't really tell other features are dismissed or not shown. So I'm still unsure of the difference, the exact difference there. I just opened up these side panels. Oh, this is something actually I really liked. It has sort of like a built-in browser, which I think it can have more— they can control the environment a little bit more. At least right now, all the kind of like computer use or browser use from Claude needs to happen like in Cloud and Chrome. You need to install an app. So supposedly— well, I'm out of credit, so unfortunately I may not be able to show it, but I would suspect that I can ask it to say, can you log into this app, do this, do this, do that? And it would kind of like be able to automate that in the browser in its own sort of like isolated environment as opposed to needing my computer to do so. That's something I really liked. I also like the way, like the user experience of how it says awaiting for approval. I do know that Clockcore has something like it where this little icon shows up as either done or in some cases yellow if it's waiting for your approval. That didn't strike me as intuitive as here, just like literally telling me waiting for approval. It does have like a blue one here that I don't exactly know what it means, but I tried a few figure that out. An area of particular interest for us, it's the plugin section that I already, as I already said, it feels a little disjointed from what's already available on OpenAI ChatGPT. I did like that it had some additional things. For example, maybe Claude has this already, but I'm not entirely sure. Vercel, Netlify, for those of you that are a little bit more techie, to be able to deploy web apps. That's something we'd be interested in testing at some point as a development agency. I saw some apps that, for example, I know right now are not available, like HighLevel and Pipedrive. So some certain CRMs seem to be currently available on OpenAI, and I'm sure they'll continue to invest in enabling more and more. One of the things I actually liked about the plugins is actually comes with a set of skills. I found that pretty interesting. They seem to have— OpenAI and Anthropic seem to have different philosophies on like skills and connectors. Anthropic is building plugins or is a collection of skills more like at the department level, a finance set of plugins, an operations set of plugins. I do do like that OpenAI has some skills specific to the app you're connecting. At least it shows some of them out of the box for you to give ideas. So for example, here in SharePoint, share and pull specific shared sites, summarize the sites, edit SharePoint text with some style preservation. Like, I think it really helps people to— especially people that are getting started— to connect a tool and then immediately start to see like what's possible and show you some, some use cases to get their creative juices flowing, which some people sometimes struggle with. Uh, the other area of particular interest for us is the automation section, which we have loved Claude Code Routines. We're migrating some of our existing automations to it and definitely see a future for it in the work we do. I did like— I didn't see this option in the automation section of Codex yet. This can run only on a schedule, seemingly at the moment, while on Claude you have the ability to, you know, have this trigger on a particular event. And the other very, very important thing for security reasons, I do know that I can change allow specific connectors or disable specific connectors from Claude. One of the things it tells you is that it will be able to use some unrestricted tools. So, even if the user, for example, is not able to send a Slack message, as an example, Claude coroutine will be able to because it's meant to run in the background. Although we're currently working and experimenting with adding, creating a repository in GitHub to be able to more granularly set or disable certain permissions. It's a little bit more work to set it up, but I do like that from a security standpoint, we would have that flexibility. It would seem right now that these automations within Claude, within Codex, are very basic and that I cannot— well, maybe I can, and I haven't tested it yet— get this to run on another environment. Local environment unavailable. That's something that we have to test and see if we were actually able to run this remotely, which at least from what I'm hearing, from what I'm looking at here, I wouldn't be entirely sure if that's currently possible. I see some of the previous runs here. For some reason, I had a very hard time trying to connect tools and get this automation to run at all. And also similar to Andropic, I literally tested like 5 or 6 different places and then all of a sudden I'm out of credits. One thing I did test yesterday, it's specifically some plugins they have by default, which are computer use and browser use. I found computer use Relative to Anthropic's computer use, a little bit more user-friendly. Unfortunately, I'm out of credits, so I may not be able to show this live. But one of the things I really like, it takes over my computer and does something in a specific app. I'm out of usage, unfortunately. And while Anthropic did that, I'd like the user experience of Codex a little bit more where it kind of showed the mouse moving through and highlighted it and just felt a little bit more that I could see what the computer is doing and it was trying to do versus Anthropic where Things are happening and similarly clicks are being made, but I can't see the mouse and kind of like get the sense of what the computer is doing so that I could stop it if I wanted to, for example. By and large, very promising app. I just think that, that, you know, OpenAI, it's catching up or working towards that. But overall, I would still give the preference to Claude at this point in time. There seems to be still some catch-up work to do. On Codex's side to improve the experience, to have more availability on the automation side, as well as being able to— for some reason, I just couldn't make certain things work here. And funny enough, I saw this: attach Claude to Codex. No idea what that means. Can I get Claude to work within Codex now? We'll be exploring that in a future video, more than likely. But this is like my first impressions. Of what Codex looks like and how it compares to Claude. I hope that helps. Hope this video is useful. Give it a like if you're seeing this on YouTube, give me a comment, and I hope to see you on the next one.