I considered leveraging LLMs to help write some parts the scenario, but in the end I prefer doing it myself.

When I said "enemy's AI", I didn't mean machine learning, just how it behaves.