@Chazrox Writing small pieces of code, like functions. Connect it to context7 so it can get docs, but obviously that only works if the docs are complete, and HISE docs are not.
I did once write a full 3d rendering engine in HISE PaintRoutine, with frustum culling and all the jazz. You'd just pass it a "mesh" like a an array of points relative to a center, pass it a location for the the mesh center, pass it camera specs like xyz, pitch, yaw, fov and it would render the whole scene using line calls. Of course it was all running on the CPU so it wasn't very efficient at all.
I did eventually move to WebView and three.js because that's what three is made for and it runs on the GPU, but this is the kind of stuff where even testing it out meant you had to go and learn computer graphics from complete zero just to implement this here, but the toughest question would be where to even look and what to look for. With AI it was much faster to find the relevant information and get it going.
So in a world where a question "is this even viable to try to implement" can be answered in days instead of months, it's like magic wand.