Pretty much all the LLM libraries I'm seeing are like this. They boil down to a request to the LLM to do something in a certain way. I've noticed under complex

Pretty much all the LLM libraries I'm seeing are like this. They boil down to a request to the LLM to do something in a certain way. I've noticed under complex conditions, they stop listening and start reverting to their 'default' behavior. But that said it still feels like using a library is the right thing to do... so I'm still watching this space to see what matures and emerges as a good-enough approach.