Another (very) small step to understanding AIs for development.
I’m interested in facts instead of hype. Facts can be small, but you can depend on them. Previously I’ve started an experiment with letting an AI write code. The experiment is small to find out what an AI can do, not where an AI fails. For this I did let Claude Code generate a Minesweeper game from a requirements.txt. That document describes the game and the technology that should be used, in this experiment Javascript/HTML/Canvas.
Claude Code was able to create a playable game of Minesweeper. You can read about it in story where I describe more details.
One-Shotting Minesweeper
🧠 MIND BLOWN BY CLAUDE CODE - it wrote a Minesweeper game for me for only $0.24 !
Now I wanted to see if and how Claude Caude can migrate this (mini-)application to a new technology stack. I did ask it to migrate the Javascript/HTML/Canvas to Go/WASM/Canvas.
And it worked!
The generated game works and is the same, except some minor differences in the UI.
Claude Code generated
The game code in Go game.go
Change the HTML to use the web assembly code
Generated JS code for calling the web assembly code
Generated a build.sh file to build the Go project
Generated a run.sh file to start a web server with Python3
Generated a go.mod file for dependencies
Generated a README.md to explain how to start and run the game
It took just a few minutes to migrate - this small - project to a new technology base, and I am impressed. I’m sure in it’s current state Claude Code will fail for larger projects, but it did work for the Minesweeper game example.
If we’d have split the requirements into business requirements and technical requirements, it would have been possible from the beginning to just create the target we want. Generating JS/HTML, or Go/WASM or even Flutter. Perhaps this is the way forward for tech migrations and multi plattform support.