“How do we reduce the friction of adoption? So how do we make the platform way easier? How do we abstract away all this complexity around LLM selection, evaluation, prompt engineering, all the different data providers that have to get orchestrated?”