The ease in which we can create our own data models for use within applications is very useful as it allows us more flexibility than being tied down by an existing database system's structure or schema design rules (which are often not flexible enough).
This also makes adding new features easier because there isn't much code that needs rewriting when changing from one model type/schema version /data source implementation into another - just change out what kind file format they're using! There really aren’t any downsides so far but I think this could be improved if some things were made clearer about how changes get pushed back up through multiple versions etc., i'm sure others will have other suggestions too though :) We've been able solve specific problems where relational databases weren'r suitable such having user accounts linked across different sites without requiring them all share exactly same tables & fields was.