As a long-time developer advocate, one constant struggle of mine has been building good-looking demo applications. I have little-to-no design talent, and although I’ve worked with many talented designers over the years, I’ve always been hesitant to “waste” their time on the trivial sample apps I build.

But over the last few weeks I’ve increasingly realized that this will never be a problem for me again because of AI.

For example, last month I wrote a sample app showing Blues users how to change their Notecard’s Wi-Fi network using our API. The initial functioning version of that sample looked like this:

The initial app was a simple <form> that worked, but obviously needed a bit of design before I’d be comfortable sharing it with the world.

Previously, at this point in the process I would’ve reached for a design library like shadcn to help clean up my unstyled HTML. And although tools like that are still useful, I found an even better trick: I asked GitHub Copilot to “make this look professional”, and it turned out to look surprisingly decent, first try.

The implementation was a little verbose as the AI used a lot of Tailwind class names, but I still found the sample easy enough to understand. (The focus of the sample was on using the API anyways.) The full source code is available on GitHub if you want to check it out for yourself.

Since then I’ve started to use this approach for every small thing I build, on both work and personal projects. For example, I’ve long maintained a Pokémon GO checklist app that I’ve always thought looked pretty ugly.

To improve the design I again asked Copilot to simply “make this look professional”, and I was once again blown away.

Are there broader societal implications of AI trivializing a task like this? Maybe. But am I ever going to build an ugly demo app again? No, I will not.

And for those curious—I’m using GitHub Copilot with Claude 3.5 Sonnet as a model. In my limited testing Claude 3.7 in “extended thinking” mode produces even-better-looking sites, but it’s very pricey to use on its own and not available in GitHub Copilot quite yet.