I find just writing/executing the SQL queries with some sort of api (e.g. asyncpg or sqlite3 for Python) much simpler. The only use-case I can think of is easy of transferring between databases (e.g. from SQLite to PostgreSQL) or a specified schema to reference from (so you get type hints and stuff like that).
From then on our team we decided to switch to using an ORM rather than doing messy SQL queries that could lead to that incident.
We are currently still hitting hundreds of thousands of queries on the backend and the team not looked back since.
If you want to model an inheritance hierarchy as SQL tables that is problematic because there are three ways to do it that all have some good points and some bad points but none of which are completely satisfactory.
If you were working in PHP you can make something very simple that looks like an ‘active record’ that writes your INSERTs and UPDATES for you. You could have something like that in Java but you might rather generate some Java Bean stubs to do things that Java way. There’s something to say for “relational-object mapping” that brings relational structures into the OO world of which JooQ is a good example.
When I’ve asked other programmers why they use an ORM the answer usually comes down to “Because I don’t know SQL well enough to write it myself.”
Understanding SQL and the relational model has for me proved the most valuable and durable skill I ever acquired. I’ve avoided rounds of layoffs and got hired just because of that skill.