Presto's set of functionality ( ANSI-compliant SQL plus so much more) is a pleasure to use as a data scientist who does a bunch of analysis. JSON, arrays, maps, dates, HLL, spatial stuff, rich aggregation functions, etc.
The only real problem with it is that it's easy to get into some thorny-looking transformations because there is so much to work with.
One way I've solved for this in presto (Athena) is creating a table that flattens the JSON, and views on the table to un-nest arrays or show different grains of data for different use cases. Then transformations and joins to other tables become very simple.
This works well for my current area because the JSON has a consistent and well defined schema.
The only real problem with it is that it's easy to get into some thorny-looking transformations because there is so much to work with.