Working on AI game generation at Lobah has made me quite specific about what I mean by a successful result.
A screen, a character and a working start button are enough to get me interested. Then I want to know whether I can finish a round, restart after losing and use the controls on a phone.
Those checks sound small. When you’re playing, though, one broken step can stop you from seeing anything that comes after it.
I used to compete professionally, and I still enjoy games. It’s hard for me to judge one from a screenshot alone. I want to try the controls, see what feedback I get and find out whether I understand the rules. Even a very small game can feel good to play when those details work.
With a generation tool, I also think about what the user does afterwards. They might want a different character, an easier level or a link they can send to a friend. Those are ordinary requests, and the workflow has to account for them too.
That’s part of why I like working on this. AI can get an idea into a playable form quickly, without all the initial setup. There’s still plenty to figure out once someone tries it and says the game isn’t much fun yet.
If you want to show me an AI-generated game, I’d prefer a link I can open. Let me play a round, then we can talk about the models.