As a refugee of Hibernate etc. I definitely agree with this. Python in particular already has data types that map nicely to relational data - dict, list, tuple, iterators - so mixing SQL calls directly in code works great.
Plus you get full access to the particular database's features, direct control & understanding of what's being executed when, easier query & performance tuning, no special 2nd pseudo-sql dialect to learn, no big extra stack of leaky abstractions to troubleshoot, etc.
One of the big pitches of ORMs is "you can change your DBMS mid project!" Which sounds cool and has its occasional applications but is something I've never actually needed in over 20 years of development.
I worked in some projects where we had to support multiple databases. Entity Framework and a few Database Views helped us keep it agnostic. It was very common in the pre-cloud enterprise world.
Uber did it recently, too. They changed from Postgres to MySQL. But I don't know if ORMs helped them or not.
Plus you get full access to the particular database's features, direct control & understanding of what's being executed when, easier query & performance tuning, no special 2nd pseudo-sql dialect to learn, no big extra stack of leaky abstractions to troubleshoot, etc.
One of the big pitches of ORMs is "you can change your DBMS mid project!" Which sounds cool and has its occasional applications but is something I've never actually needed in over 20 years of development.