Around 2000 I took an AI course at the University of Guelph. I don’t think I learned too much. We didn’t talk about neural networks, as far as I can remember. My end of term project was, I think, a pathfinding algorithm wearing an AI costume. There were certainly no discussions of transformers. No CUDA. No PyTorch. None of that existed.
But what I remember doing a lot of was coding in Lisp - a lot of Lisp in the dark University of Guelph CIS lab.
You can implement an agent loop in any language, and modern languages can do everything that lisp can, so there’s nothing magic in lisp. At the end of the day lisp is just another weakly typed functional/procedural language.
Personally I didn’t find lisp that much elegant due to the loose typechecking and the excessive parentheses, but the macro system allows some very terse and smart (and mind-bending, as you said) solutions, which do feel pretty cool when they work fine.
IMHO it’d be probably better to give your agent access to a python interpreter because it’s much more prevalent in the training data than lisp…
tbf high prevalence is a double edged sword, it also means it’s trained on a lot of poorly written code which it will slide into