Everyone's using AI to build games, thanks to Levels' flight sim.
It looked fun, so I used AI to build a game I've always wanted to have on Indie Hackers — a mini crossword puzzle.
It turns out, not every game is so easy to make with AI. Crosswords have a lot of moving parts that the AI just couldn't one-shot:
- user interface with lots of keyboard shortcuts that need to function intuitively
- algorithm that can actually generate valid crosswords, AI got this wrong over and over
- giant lists of valid words, with ratings on each, so the algorithm will know which words are too uncommon to use in a puzzle
- custom keyboard to render on mobile, bc native mobile keyboards suck for crossword puzzles (blew my mind that this was necessary)
- AI to generate clues for the finished puzzle, and to make sure clues aren't too hard or too easy, and ideally are clever and fun
- UI for you to edit clues when the AI's clues suck
I ended up having to do a lot by hand, but I got it working in the end:
👉 https://www.indiehackers.com/indie-mini/daily
My takeaway is to put a lot more thought into when vibe coding works well vs doesn't. What seems impressive to humans might not take a lot of code (a 3D flying sim on a single map), whereas what seems simple to humans might have a lot of difficult components (a daily mini crossword). If all you have is a few yours or a day, you're better off vibe coding the former rather than the latter, at least for now.
Valuable lesson to learn for sure.