> These don’t seem like text2sql problems: Correct, but I would propose two things to add to your analysis: 1. Natural language text is a universal input to L

> These don’t seem like text2sql problems: Correct, but I would propose two things to add to your analysis: 1. Natural language text is a universal input to LLM systems 2. text2sql makes the foundation of retrieving the information that can help answer these higher-level questions And so in my mind, the goals for text2sql might be a copilot (near-term), but the long-term is to have a good foundation for automating text2sql calls, comparing results, and pulling them into a larger workflow precisely to help answer the kinds of questions you're proposing. There's clearly much work needed to achieve that goal.