Making a Game with AI: What I Did as the Designer
The first part of my Jelly Gem development journal: why I made a merge game, what AI handled, and where my work as a designer came in.
My first vibe-coded game: making Jelly Gem, part 1.
Drop, bounce and merge luminous little jellies into a heart.
What AI did
Created the image assets, implemented the drop-and-merge game, and handled everything related to development.
What I did as a working designer
Used references to communicate what I wanted, from colors and shapes to movement and sound, and kept asking for changes until it suited my taste.
Why did a designer make a game?
I'm a brand designer, and I've always liked games. But merge games like Suika Game and 2048? I'm completely hooked.
I'd wanted to try making a game like that for a while. Whenever I tried building something with Claude, though, I wasn't entirely satisfied with the speed or performance. I'd make a little progress, stop, then try again. Finding Astra felt like finally seeing the light.
It made me feel like I could make anything, so I started listing all the things I'd wanted to create. I actually have another finished game and a nearly finished website already. I'll share those in time! A merge game was the first thing that came to mind, and if I was going to make one, I wanted the colors and atmosphere to be entirely to my taste.
What was the concept?
Whenever I saw the squishy toys and slime-like toys that had become popular recently, I thought it would be fun if those soft little things could merge. So I didn't spend long deciding on a concept. I went straight into trying it.
Make them translucent, put little sparkles inside, and let them merge into bigger squishies!
That was the beginning of Jelly Gem.
How did I make it?
I worked through the detailed planning and implementation by talking with Astra in ChatGPT. I'd explain what I wanted, it would make something, and I'd look at the result and ask for changes.
I was made to study coding at one point, but I've forgotten all of it. I have no coding knowledge now, so I didn't write any of the code myself.
Even so, a working game that broadly did what I wanted appeared in a single go. It was genuinely amazing.
But having a game that worked and having a game that felt the way I wanted were two different things.
I tried to explain the tone and style I had in mind as clearly as I could, with references, but Astra's first version felt like it had captured only about 10 percent of what I was imagining.
For the overall mood and the actual background image, it was still faster for me to design what I wanted myself and hand it over. I put those together quickly, and after that Astra understood my intentions much better. We could bring the rest of the game into the right style much faster.
AI was making it, but I still had to keep saying what I liked and what I wanted to change. And for now, making the image I wanted myself and supplying it was faster. I think that was especially true here because there wasn't a single correct answer. It was a matter of my taste.
There were quite a few changes along the way, so I'm going to write about them across a few posts. Seeing the first images alongside the later ones should make the differences much clearer!
I'll also keep steadily sharing my future work on Instagram.
Originally published in Korean on September 28, 2026. English translation prepared October 1, 2026. Read the original on Naver.