The more natural an analytics interface becomes, the more important I think the underlying data definitions become.
Historically, an analyst could spend time understanding what a metric meant, which table it came from, how it was calculated, and what assumptions were behind it. As interfaces become more natural, more of that complexity gets hidden from the user. That is a great experience when the system is right, but it becomes a serious problem when the underlying definition is ambiguous or inconsistent.
This is why I think semantic layers and governed metrics become more important with AI rather than less. If an AI system can answer a business question in seconds, users will naturally trust and act on the answer. The system therefore needs a reliable understanding of what the metric actually means, how it should be computed, and when it should or should not be used.
The easier it becomes to ask questions, the less visible the complexity becomes, which makes getting the foundation right even more important.