I think ORMs like SQLAlchemy are really useful for many many use cases. I don't think most people who work with ORMs deal with the kind of complexity described by the author, let alone work on such a specific application for 30 months at a time. In that sense, ORMs are super powerful tools that cut down your work, shortens your code and do nifty optimizations once in a while
With that being said, this article totally resounds with me, having resorted to using direct SQL in almost all of my large-ish projects. Using an ORM always starts out nicely, and then eventually it gets messier and messier. Specifically, when your DB data has subtle interactions with a cache, I have found it harder to keep ORM code clean and readable.
SQLAlchemy isn't like most ORMs. It's data model is actually closer to the relational model than OOP objects, and therefore lacks many of the ORM warts.
With that being said, this article totally resounds with me, having resorted to using direct SQL in almost all of my large-ish projects. Using an ORM always starts out nicely, and then eventually it gets messier and messier. Specifically, when your DB data has subtle interactions with a cache, I have found it harder to keep ORM code clean and readable.