AI won't just write code: it will redesign the interfaces we use every day

This weekend, taking advantage of a rainy Sunday, I spent some time doing something that looked small: I redesigned Mercadona's old website.

I have been shopping at Mercadona for more than 15 years, I like their products and I greatly value being able to shop online. For me, saving an hour at the supermarket is not just convenience: it is getting time back that I can spend on other things, like playing with my kids. And in the end, getting time back is one of the clearest ways to earn money, even if that sounds like a cheap mug motivational quote.

The problem is that not every Mercadona customer gets the same online experience. Depending on your area, you can access the new online shopping experience or the classic website. And the classic website is a fairly rough experience. I'm not the only one saying it.Juan Roig publicly acknowledged: “Our website is shit. Customers say it and employees say it”.

Juan Roig sabía bien lo que decía

The quote is good because it is honest, and I have also been suffering it for years. Every 4 or 5 months I search Google to see whether Mercadona plans to update the logistics center assigned to my area. So, tired of this not improving, the following occurred to me:

If a company still does not give me the interface I want, why not build a layer on top of it?

So with that idea I started working with Codex. I did not open a code editor during the entire development process. I spoke with the AI using handy, my favorite STT tool, explaining the problem, reviewing proposals, correcting decisions, and validating the behavior.

The solution I had in mind was a Chrome extension that detects when I am on Mercadona's classic site and generates a new interface on top. This extension visually hides the original DOM and builds a new, much more usable DOM: products with images, clear names, presentation, price, and more comfortable actions... basically a UX closer to 2026 than to 2008.

But here is the important part: the extension does not replace Mercadona's logic. I am not reimplementing the cart. I am not calling internal APIs directly. I am not trying to guess how the backend works.

What I do is simpler and, precisely because of that, more robust. Each new component keeps a reference to the corresponding original component. If my new interface shows an “add to cart” button, that button does not invent a new action: it delegates the click to the original button on the old website.

El punto es crear un adaptador que consuma el DOM original, así no rompemos nada

In other words: I interact with a modern interface, but Mercadona still receives an interaction compatible with its original page.

The interface stops being closed

This is not about building an extension for Mercadona. It is about something much bigger. For years we have assumed that interfaces were the exclusive property of whoever built the product.The company designed an experience and the user had two options: adapt or leave.

But AI is starting to change that relationship. When creating software becomes conversational, the interface no longer has to be a closed surface. It starts to become something an advanced user can adapt, extend, or rebuild on top of existing systems.

Not to hack them. Not to break them. Not to bypass rules. But to adjust the experience to the way they actually work.

Así es como veo ahora la página de compra de Mercadona. ¡Con fotos y todo!

The UX of the future will not live only inside the official product

Part of the user experience may live in personal adapters, created with AI, between the person and the tools that already exist. Today it is a supermarket website. Tomorrow it could be a CRM, a corporate intranet, a medical tool, a financial dashboard, a slow back office, a public administration site, or any software we use because we have to, not because it is well designed.

And this opens up an uncomfortable question for many companies:

What happens when users can fix bad interfaces by themselves?

What happens when the user's real experience no longer matches exactly the interface the company designed?

What happens when the most useful interaction adapter is not created by the software provider, but by the user themselves with the help of AI?

For me, this is one of the important changes coming in the next few months. Maybe we are not quite there yet; besides wanting to solve a problem, having a proposal, and having money for tokens, you also need judgment and knowledge to guide the AI.

AI will not only allow us to create new applications faster. It will also allow us to create small, personal, extremely specific software; software that may not make sense as a startup or need to scale to millions of users, but solves a concrete friction for a concrete person.

Although this sounds less epic than “we are going to change the world”, it is deeply transformative. Until now, many small ideas died because of the cost of implementing them. It was not worth spending days or weeks building something to solve an everyday annoyance. But if you can build it in a morning by talking to an AI, the equation changes.

Así es como cambia la UI gracias a la extensión de Chrome

The barrier is no longer writing every line of code

As I said at the beginning, I did not open the code editor to build this. Therefore, the main barrier to overcome becomes understanding the problem well; designing a sensible architecture, knowing what not to touch, validating that you are not breaking anything, and having the judgment to separate a nice demo from a useful solution.

Codex did not come up with the idea for me or decide the architecture for me. I had to explain that I wanted a Chrome extension, that I did not want to remove the original page's DOM, but simply hide it and connect new components that applied current design patterns to the old components.

In fact, Codex is pretty bad at frontend, so I had to ask for quite a few changes to make everything look good and not break. It also did not propose an optimistic pattern to improve the experience when adding or removing products.

I brought all of that, and AI reduced the distance between having an idea and seeing it working. And that, to me, is the important point.

We are not only entering an era in which AI writes code. We are entering an era in which many people will be able to turn judgment into software. And when that happens, many interfaces will stop being definitive.

Conclusion

The future of UX may not be one perfect interface designed for everyone. It may be something more interesting: stable systems underneath and personal adapters between those systems and each user.

And a new professional skill in the middle: knowing how to use AI to turn everyday friction into a real tool.