Why I Didn't Have to Switch From Claude to ChatGPT Codex

I’m a firm believer that when the masses start moving heavily toward one thing, especially in tech, it’s usually a good time for me to question what’s actually happening. I’ve never really been the person to switch tools just because everyone on my timeline is talking about them. That doesn’t mean the popular thing is automatically bad, but I think we have a tendency in tech to hype something up so much that eventually we stop asking whether it’s actually better for the way we work.

For a long time, Claude was that tool.

There was a period where it seemed like everybody was using Claude, especially developers and people heavily involved in AI. The conversation made it seem like Claude was clearly the superior LLM. People loved the writing, the context window, the coding capabilities, and eventually it became one of those things where it felt like if you were serious about using AI, you were supposed to be using Claude.

I personally never saw exactly what everyone else saw.

I used Claude, tested it against my own workflows, and tried to understand the hype, but I never consistently got the deliverables I wanted from it. Whether it was the outputs, memory, context, or just how it fit into everything else I was doing, I always found myself going back to ChatGPT and Codex.

Now we’re starting to see the pendulum swing in the opposite direction. Some of the same tech influencers and builders who were heavily using Claude are talking about moving their workflows over to ChatGPT and Codex because they’re not getting the outcomes they want anymore.

For me, there was never really a switch to make because I was already using Codex.

And I think that says something bigger about how we choose AI tools. I’m becoming less interested in which model the internet has decided is “the best” and more interested in which ecosystem actually allows me to get the most work done.

Part of This Was Always Hype

I don’t think Claude is a bad product, and this isn’t supposed to be an article arguing that Claude sucks and ChatGPT is better. I think Claude went through a massive hype cycle, which happens with almost every major technology product.

Someone discovers something new, a few influential people start talking about it, developers post impressive examples, benchmarks start circulating, and then everyone wants to use it. Eventually, people start treating their preference for a particular model almost like choosing a team.

The problem is that your actual workflow gets lost in that conversation.

I don’t care if a model performs slightly better on a particular benchmark if it doesn’t help me produce what I actually need to produce. I also don’t care if someone with 500,000 followers says they moved their entire development workflow somewhere else. Their workflow isn’t necessarily mine.

That’s really why I stayed with ChatGPT and Codex. It wasn’t some prediction that eventually everyone would come back around. It was simply that the ecosystem worked better for what I was already doing.

The more I’ve used Codex, the more I’ve realized that I don’t necessarily care about the model in isolation anymore. I care about what I can connect to it, what I can build with it, how much context it can have around my work, and how many steps I can eliminate between having an idea and actually getting something done.

How I Actually Use Codex With GitHub

One of the biggest examples is how I build products.

A lot of my projects will actually start in Lovable. I’ll use Lovable to get the initial idea out, start putting together the product, test the interface, figure out what I like and don’t like, and get something functional relatively quickly. Once the product reaches a certain point, I’ll push everything over to GitHub.

From there, I can continue working directly through Codex.

That workflow has become really natural for me because I’m not starting over every time I move between tools. I’m not constantly copying huge blocks of code into a chatbot and trying to explain what the application is supposed to do. Codex can work with the repository itself, which means I can continue making changes, troubleshooting issues, adding functionality, and refining the product from the actual codebase.

So for me, Lovable and Codex don’t necessarily compete with each other. They serve different parts of how I build. Lovable is great for getting me moving quickly, and once the project is in GitHub, Codex gives me another environment where I can continue taking it further.

That flexibility is a much bigger deal to me than arguing about which model technically writes better code.

Then There’s Zo Computer

Another reason Codex has worked well for me is how I use it alongside Zo Computer.

Zo gives me access to cloud computing, so everything doesn’t necessarily have to depend on the resources of my personal computer. When I combine that with Codex, I have an environment where I can build and execute without feeling like everything has to happen locally on my machine.

What I like most about this setup is the amount of friction it removes. I don’t want to spend all day thinking about infrastructure when I’m trying to test a product idea. I’d rather spend that time figuring out whether the product makes sense, fixing what doesn’t work, improving the experience, and getting something into people’s hands.

That’s where I think some of the AI conversation gets too caught up in the intelligence of the model itself. Obviously, I want the model to be good. But once the models are good enough to handle most of what I need, the surrounding experience starts becoming just as important.

Speed matters. Integrations matter. Context matters. Compute matters. The number of steps between asking the AI to do something and actually getting it done matters.

The Integrations Are Probably One of My Favorite Parts

The other part that has made the ChatGPT and Codex ecosystem useful for me is how much I can connect to it.

I can connect Gmail, Calendar, GitHub, and other tools through integrations and MCPs, and a lot of it is becoming increasingly easy to set up. I don’t necessarily need to build some complicated automation every time I want my AI to interact with another part of my workflow.

That’s important because my work doesn’t exist inside one chatbot.

My emails are somewhere else. My calendar is somewhere else. My code is in GitHub. My documents might be somewhere else. The products I’m building have their own environments. If my AI is supposed to actually help me work, then I need it to be able to interact with those environments instead of constantly requiring me to bring everything back into the chat manually.

I think this is ultimately where the AI platforms are heading anyway. We spent the first few years talking about which model could give us the best answer. I think the next phase is going to be much more about which platform can actually understand the environment around your work and help you do something with that information.

The In-App Browser Is Golden

This is probably one of those features that sounds small until you actually start using it consistently.

I love being able to browse while I’m already working inside Codex because it eliminates another place where I would normally have to break my workflow. When I’m building something, I’m constantly researching. I might need documentation, an example, information about a particular API, or just something I want to reference while I’m working.

Normally that means leaving whatever I’m doing, opening another browser tab, searching, finding what I need, coming back, and continuing.

Being able to keep more of that process inside the same environment just makes everything feel more streamlined. It’s not necessarily the flashiest AI feature, but those small reductions in friction start adding up when you use these tools every day.

And Then There’s Hosting

This might actually be the most underrated feature for me.

Being able to build a website through Codex and then host it from the same environment is pretty amazing when you think about how many separate tools used to be involved in that process.

I can have an idea, start building it, work through the code, make changes, test it, and get something online without turning deployment into an entirely separate project.

For someone like me who is constantly building and testing different product ideas, that matters a lot. Not everything I build needs to become a company. Sometimes I just want to see whether an idea works. Sometimes I need a quick internal tool. Sometimes I want to test something with users. Sometimes I want to build something simply because I’m curious about it.

The easier it becomes to move from an idea to something that actually exists on the internet, the more experimentation I can do.

And that is probably what I value most about Codex overall. It reduces the distance between me thinking about something and me actually building it.

Why I Never Really Needed to Switch

When I look at the current conversation around people moving from Claude to ChatGPT or Codex, I don’t really look at it as some victory for one model over another. Claude will continue improving. ChatGPT will continue improving. Gemini will continue improving. Six months from now, there will probably be another model or product that everyone suddenly says you need to use.

I just don’t think constantly switching based on that conversation is particularly useful.

At some point, you have to figure out what works for you and build a workflow around it.

For me, ChatGPT and Codex have worked because they fit into the way I already operate. I build products. I use GitHub. I experiment with tools like Lovable. I work with cloud computing environments like Zo. I connect different systems together. I automate things. I research while I’m building. I want to be able to move quickly without having to recreate the context of what I’m doing every time I open another tool.

That doesn’t make Codex universally better than Claude. It makes it better for the way I work.

And maybe that’s where I think we need to change the conversation around AI tools. Instead of constantly asking which model is better, I think we should be asking which environment actually allows us to produce better work.

Because I don’t really care which AI wins the hype cycle.

I care about what helps me get things done.