Gratistellar Devlog: I set out to learn to code. I didn't.
So here's the thing: I'm a designer. I'm kind of a tinkerer. I like to understand things by doing, by trying things out, usually with some kind of a set outcome I'm shooting for. I pick up projects knowing most of them are fun ideas to play with and won't stick around.
I wanted to learn to code in different ways for years. I took classes when I was young. I got the basics (and literally had learned BASIC). I've played with other programming languages and easy coding in games and apps. I understand code conceptually, the logic of it, the structure, that kind of thing. I suppose it's perhaps like the way people might understand a foreign language they studied for a year. I can kinda read it, I follow the grammar, I sometimes can guess what something means. I am not fluent. That's for sure.
When I had the idea for GratiStellar, it seemed like a good project to learn coding. It's simple and (at first) was well-contained. I tried a few things first.
I had a look at Godot. I'd heard good things and it seemed like a reasonable place to start for someone coming from a design background. It's actually great. I may build something in it one day, but as a primarily videogame creation tool it doesn't fit this project.
I tried a couple of the app-builder tools, the ones that promise you can describe what you want and they'll generate something functional. These were great for describing something and getting an immediate prototype. The problem came when you wanted to tweak or refine things. Every prompt broke something or made another thing worse. You couldn't edit code meaningfully. There were too many barriers.
After some research, I landed on Flutter in Android Studio. Flutter made sense for what I needed: I could launch on Android immediately and keep the option of other platforms open for later. I still didn't have the chops to actually code, so I thought, okay, this "AI app-builder" did surprisingly well but was limited. Let's get one of the actual LLMs to help with the coding. Tell it what you're doing and go from there.
Part of the reason was practical, but a big part of it was curiosity. The goddamned push for AI is everywhere. Your phone, your websites, and especially at work. I needed to know what was hype and what is real.
A bit of context here, though, because I think it matters: these tools aren't politically neutral. The environmental cost is real. The labour and copyright questions around training data are real and largely unresolved. If these same tools had been developed under a different kind of system, a lot of the harm would look different. They will almost certainly be used to make wealth inequality worse. But they're also not going away anytime soon, and through this project I've discovered how powerful they can actually be, too.
I'm not going to pretend this is a clean or simple thing.
The first thing is that the ideas are yours. Always. The LLM does not have a vision. It can execute, it can suggest, it can even fill in gaps, but it has no stake in whether the thing you're building makes sense or feels right or is actually good. That's entirely on you. I was the one who knew what I wanted GratiStellar to feel like, what mattered and what didn't, when something technically worked but was wrong anyway or could be more usable. In other words, I drove the design.
The second thing is that these tools are fallible in a specific and absolutely absurd way. They are confidently wrong. They are optimistic to a fault. Toxically positive, even. Hell, they will lie to you sometimes. They can waste a lot of your time if you're not paying attention. I learned to verify everything, push back regularly, and treat enthusiasm with suspicion. These things are the most brilliant fools I've ever experienced and you have to babysit them.
The third thing is the one I'm still sitting with. I wanted to learn to code, but a bit of the way into the project, it became more efficient for me to tell the LLM the specific thing I was doing and then bug test it. Through patches and upgrades to the model, it was being improved in its ability to code and I wasn't nearly keeping up. I was understanding less of the code it was producing, not more. I became something more like a creative director than a builder. I held the vision, I made the calls, I caught the mistakes (hopefully all of them), but I wasn't writing the code and still can't.
I kept going and finished the app. It works. People are using it. And I did not learn to code. Whether that's a good thing depends on your perspective, I think. If my goal was to learn to code, I failed. But I wanted to learn so I could build things, and I did that. The app exists and is out there. If this was about developing a skill, I probably shortchanged myself. If it's about whether the thing exists and does what it was supposed to do, then I got there. I genuinely don't know where I land on this, so it's probably both.
At the end of the day, the app is out there and hopefully helping some people. I'm taking that as a win. The rest is a learning experience. Or lack of learning experience.
You can check it out on the Google Play store: GratiStellar.
kianamaiart
stormborns
annaxmalina
origamiaround
jacobvanloon
andulkaphoto
baratheonpilled
