I'm no genius, but I've managed to set up my agents to prefer writing a small python script or whatever to do repetitive/iterative tasks that my language model 'knows' it's not good or efficient at.

You're technically right, which is why I say "good enough".

It's a good strat imo - start simple and get it working with LLMs asap, then you can add heuristics etc based on the data you have and improve with that benchmark

Ah yes, I see the malware authors buying subscriptions to the appropriate software to do that.