Getting a database to line up with production code is especially important in python as pep 484 types become more commonplace. If you trust the type annotations but the database/orm is polluting data structures with values that don’t conform to the annotations, you’ve lost a lot of the annotation work’s leverage.
It’s frustrating how almost all scaling advice is focused on stuff like how to manage a fleet of 100s of servers or implement consensus algorithms. Enterprise startups in many cases really should be treating servers like pets, not cattle.