You build gen AI tools to make data more accessible. Beyond efficiency gains, what has surprised you most about teams interacting with these AI assistants?

One I think is adoption. One of the challenges data people have had with especially the traditional SQL codes or dashboards is that we are making certain assumptions on what the end user's comfort or familiarity is for absorption or usage of the same. There are still stakeholders who don't like even dashboards. So the fact that the natural language interface is there automatically drops a certain set of barriers because the person is like, hey, I'm asking the question in a form that I am comfortable with. So that's one of them. But the other one, which is interesting both in a positive and in negative way, is trust. Because of the natural language nature of LLMs and the interaction of agents and all of that, people tend to put a lot more trust in the system than they would normally do. They're a little skeptical if you give them a table of numbers, but if you write them a sentence saying the answer is X, they tend to trust it. Now that's a good thing and a bad thing because even if one doesn't know how an LLM works, one should know what the limitations of an LLM are. It's a probabilistic model. So that means there has to be a human in the loop. There has to be rigorous testing and validation. Just because it's an easy to use interface, you don't just blindly trust it.