I agree, CRUD on the backend becomes trivial and it scales pretty well .

I had very pleasant experience bolting Hasura on top of existing Postgres db with couple of hundred tables and about 1tb of data. The query performance was fine and sometimes actually better in few places .

Across the board, there was a huge boost in render times by avoiding verbose round trips of HATEOS style discovery.

Only annoyance I had is Hasura not having a proper caching layer or replica support in the OSS version, even on their cloud it is limited to 100kb and does not support session variables - almost all queries are user specific in a app so without session variables it is not useful at all .

However it can be handled by putting in proxy with cache capabilities in front of Postgres.

—-

Having said that, I am not yet confident that it will scale well for public APIs with arbitrary query patterns unless more limits on nesting etc can be enforced, but for most internal APIs it should be great .

Hasura has also support for REST endpoints in 2.x so that is an option

if selling /delivering API is the primary value of the team/business perhaps no generated backend will solve out of the box all the challenges