“Make the start button a bit bigger.”
In an ordinary project, that sounds straightforward. In an AI-generated game, I want to check a few things afterwards. Does the clickable area still line up? Does it cover the instructions on a phone? Has anything next to it moved?
I pay quite a lot of attention to the second and third edit when looking at these tools. An impressive first version is useful. Having to check the entire game again after every small change gets tiring pretty quickly.
At KIX, our workflow connects requirements, wireframes, assets and code. A short request can affect several of those. Replacing a character can mean checking image dimensions and collision bounds. Changing the scoring rules also means checking the score shown on the results screen.
I’d like the system to establish that scope before it starts editing. If it’s a button change, keep the work around the button. If the gameplay needs changing, follow the relevant logic. And if the model decides something else needs adjusting, make that visible to me.
It also needs to preserve things we’ve already fixed. A new image shouldn’t undo the mobile layout we just got right.
That’s most of what I want from AI coding tools: less code for me to write, with changes I can understand, check and roll back. I’m happy to read the polished explanation after I’ve tried the result.
There’s more about the generation and editing workflow on the KIX project page.