I had a short break from real work today, whilst I waited for AWS deployments, to experiment with using AI to build a module interface.
I selected the Joystick module as one of the simplest implementations that exists. I wanted to see whether a stub could be built using only the PRM documentation and simple guidance within the RISC OS Build Environment.
What I got out wasn’t unreasonable. It wasn’t how I would have done it exactly, but it’s quite close, and it’s done some things I’d have not thought of.
It has decided that the author is RISC OS Open, which is clearly wrong. I can probably provide some extra guidance for that in the environment. It decided to use the ZN449 chip, which isn’t a terrible choice have made, and it’s stubbed the routines in a quite reasonable manner.
And the code compiled - it built it to check.
At the end I told it to add CI and git management before committing the code itself.