An old coworker explained to me about how his previous company used DBT to create many different projections of messy data to serve many applications, rather than trying to come up with the One Canonical Representation. It truly blew my mind in terms of thinking about how to model data within a business.
The huge limitation with this vision is that it only works in places where you can tolerate some pretty significant staleness. So the promise of this approach excludes most OLTP applications. I simply assumed it wouldn't be reasonable to create something that allows for unconstrained SQL-based transformations in real time, and that no one was working on this. Oh well.
But several months back, I discovered Materialize and it was an "oh shit" moment. Someone was actually doing this, and in a very first principles-driven approach. I'm really excited for how this project evolves.