To me, the thing that excites me about this is more efficient ORM querying of data. Where previously you either had to make multiple queries to fill parent and child relationships (whether on demand of explicitly and prefilling a cache) or make a larger join where a lot of data may be duplicated in 1:N relationships making it inefficient (with a few levels of 1:N relationships chained leading to extremely inefficient data transfer), now you could actually just get a JSON data structure back of exactly what each field should be. There are inefficiencies at scale here as well with duplicated key names, but that can possibly be worked around to a degree with arrays and an array type index of some sort.