I’ve been on a similar train of thought. Just last weekend I built a little experiment, using LLMs to highlight pseudocode syntax:

https://x.com/danielvaughn/status/2011280491287364067?s=46