How I'd test an AI game builder before trusting it with a project
A polished demo tells me what a tool can produce under favorable conditions. I'd learn more by asking it to build a tiny game, change a rule and fix a deliberate mistake. The second and third steps reveal whether I can keep working with it.
Use a brief you can judge
Ask for one room, a moving player, three collectibles, a visible score and a restart button. State the controls and win condition. Use simple shapes so impressive artwork doesn't hide missing behavior.
Keep the same brief when comparing tools. Record the steps you took and the changes you had to request. Otherwise you're comparing two different tasks and calling the difference a product verdict.
I would define success before generating: the game loads, input works, each collectible counts once, winning is visible and restart restores the round.
Test a change that touches existing behavior
Ask to change the number of required collectibles or add a harmless obstacle. Check whether the tool updates the existing game coherently or replaces working parts unnecessarily.
Then introduce or identify a small bug you can reproduce. Can you inspect the relevant logic? Can you see what changed? Can you undo the fix if it causes another problem?
A tool that generates quickly may still be a poor fit if every revision feels like starting over. I care about the edit loop because most of building a game happens after the first result.
Follow the project out of the tool
Read the current export, hosting, pricing and usage terms before committing. Check what files you receive and whether the project can be opened or maintained elsewhere. Don't assume a downloadable build includes editable source.
Try the shared result on another device and browser. Check loading, controls and any account requirements from the player's side.
Write a short decision note: what worked, what blocked you, and what you still haven't tested. I'd choose the tool whose limitations I can work around for this game. A broader claim that one builder is best for everyone would need much more evidence than this useful little experiment provides.
Sources & further reading
- GDevelop documentation — accessed Sep 9, 2026
- Phaser: making your first game — accessed Sep 9, 2026
Made something playable?
Share your game, world or experiment with people exploring what AI can help create.
Add your space to Velven
