As is so often the case, IMHO the poster is ignoring the business side of the equation in favour of "correct". A good ORM is fantastic for speeding up development. If you design your db around being well usable with the ORM, the potential development gains (at least in dynamic languages like Python) are HUGE. I left Django in part because of their ORM. But using SQLAlchemy, we develop far, far, faster than if we were using SQL directly. And it's flexible enough to allow me to drop into the SQLAlchemy Query language when I need to hand tune a query, or to SQL itself if I really need to get close to the metal.

Will there be downsides one day? Probably. Will they come even close to the business value of the amount of the coding time we've saved during the critical bootstrap phase? No way in hell. As they say, those are problems I'd love to have.