Matt Watson

Adagio Translate

Español

Recently I’ve been using AI for coding more and more. I’ve never been someone who adopts new technology very early, but at this point it’s impossible to deny the benefits of AI when it comes to programming. One example of this, which I’ve made public, is a document translation app I call Adagio Translate.

It’s a “computer-assisted translation tool.” It’s a program very similar to others I’ve seen in the freelance translation world, designed to make translating documents easier, especially long ones. More or less, I copied a program I liked very much a long time ago called OmegaT, but mine runs in the web browser instead of requiring you to download an application or install dependencies like Java.

I built it entirely with AI, specifically with Codex from OpenAI. I’m a little embarrassed to say it, but I’ve embraced “vibe coding” for certain projects. That is, I don’t really program anything myself, or at least not much. I simply ask Codex to do what I want, and it does it. I blame my brother in part, who inspired me to try this in the first place.

The app has roughly the following features:

But the truth is, there’s still a lot missing. I need to test it more and see if there are bugs. The irony is that I don’t translate much anymore. Adagio Translate would have been very useful to me ten years ago. In fact, I did build an app like this 7 or 8 years ago, but it never worked very well and it took me weeks to program. This one I built in just a few days.

It’s an experimental app, I know, but it’s still impressive. It has helped me better understand AI and what it’s capable of.

Adagio Translate isn’t the only thing I’ve built almost entirely with AI. Several weeks ago, I was working on a friend’s website where I had to create a map with some interactive features as a WordPress plugin. In that case, I relied heavily on GitHub Copilot’s inline suggestions, and it helped a lot. I accomplished in a few hours what in the past would have taken me days, if not weeks.

At work, I recently started a little Laravel app. AI has helped me a great deal because, for some reason, Laravel has always been difficult for me. Many years ago, Laravel came with very simple starter kits, but now they’re quite complex. And as I’ve complained before, Laravel doesn’t provide very explicit instructions on how to set up login sessions and handle authentication manually. With Codex, I created a very basic Laravel app without any of these starter kits or other external dependencies. Then I asked Codex to create a very basic login system, and I gradually added features little by little. Each time I asked it for something, I read the code to make sure I understood it and then committed it to git. I did in four hours what would previously have taken me weeks.

I know there are certain obstacles to using AI extensively for coding. In other cases, I’ve experienced the vicious cycle of asking AI to fix something and it keeps “fixing” it with complete confidence, but the app never actually works. It’s clear there are times when AI is incredibly helpful and other times when it isn’t. It all depends on the circumstances and the particular problems of the project. With AI, everything has to be very clear and the requirements must be well defined.

But when it works, it’s simply incredible. A person can build something with AI without understanding any of the code, as my brother recently did in Swift. However, what I find the most exciting is finding the balance between understanding the code and leveraging AI’s capabilities. If you can find that balance and if the stars align just right, it can feel like you’ve discovered a superpower.