SQLAlchemy can accept raw SQL with: result = db.engine.execute("<sql here>") I haven't met an ORM that wasn't more work to implement, especially when trying

SQLAlchemy can accept raw SQL with: result = db.engine.execute("<sql here>") I haven't met an ORM that wasn't more work to implement, especially when trying to keep schema changes in line with existing deployments. That said, I'll take a decent DB client SDK any day; I don't want to spend too many cycles on connection pooling and such. I just want it to handle variable substitution and concurrency primitives at the language level. I don't think there's anything wrong with using tools to help you generate good SQL queries, but you should see them clearly and own what the database is doing. MySQL and Postgres have a myriad of useful functions and extensions that ORMs don't utilize or hide from the developer. That leaves the application to do multiple queries or a lot of logic that isn't needed. It doesn't have to be gnarly; any SQL query that can't be handwritten on an index card is probably too cumbersome to maintain. A 1000 line SQL query is a code smell that means the data model could probably be improved, perhaps with views.