Food Fight
Eight foods fight it out in rounds of would-you-rather until one is left standing. My concept, art direction and UX, built in about five days by directing Codex.
I always freeze when I open a delivery app: too many options, no idea what I actually want. I also needed a project to get properly fluent in AI-assisted development, so I built a game that fixes the first problem. You pick a category, eight foods fight it out, and whatever is left standing is what you order.
A long list of options is what causes the problem in the first place, so the game never shows more than two at once. Each round is a single either/or choice, and the winner stays in, so you’re always comparing a new option against your current favourite. Eight foods give seven rounds and 128 possible choice paths. Codex ran an automated test across all 128 to make sure every game ends in a consistent final recommendation.
I’m not a React developer, and I didn’t write or review the code. My job was everything around it: I planned the approach in ChatGPT, broke the product into tasks small enough for Codex to build reliably, and set what counted as done. Then I tested every build like a product owner. When something broke or felt off, I worked out what was wrong and described it precisely enough for Codex to fix.
The arena entrance is a good example. The first version worked, but squeezed the whole sequence into 1.27 seconds, so the poster movement and the fighters’ entrance felt rushed. I rewrote the task as specific beats:
The revised sequence runs about 2.35 seconds. Desktop and mobile recordings, plus automated checks for input locking, focus and reduced motion, gave me concrete ways to review the result without reading the code.
Each food fighter started as my own character idea. The risk with generated art is a cast that looks like it came from different games, so I used ChatGPT to keep the prompts consistent in style and structure, then generated the characters in Higgsfield.
The hardest part wasn’t the logic; it was the feel. Beyond the entrance, two refinements made the biggest difference:
From idea to a running V1, in about five days of part-time work
Every possible game checked by an automated test, all ending in a consistent recommendation
A complete experience on its own domain, from category choice to final winner, with one category live and more in progress
The web is full of worlds