The problem with ORMs is that life is complicated, and data describes life, so data is complicated. And abundant. SQL queries are a relatively elegant way to encode the often complicated manner in which data answers a particular question.
<sarcasm>Gah! You know what I hate about building software? The programming. All those data structures are no fun! Lists and arrays and vectors and hash tables and trees, red-black trees, binary trees, b-trees, tries. Why can't we have a programming language that automagically stores all our information in simple variables and just gives me the answer I want?</sarcasm>
You have apparently never used a well-designed ORM. A well-designed ORM isn't to create this magical, leak-proof wrapper for the database. The point is the make common use cases simple and allow the developer to incrementally use more SQL as needed, for more control.
The problem with ORMs is that relational databases are complicated, and you often do wind up having to do manual SQL writing anyway if you want well-performing queries in edge cases. But for simple cases where you have a person table and need to make Person objects in your app... it works.
In fact, Rails used this as one of its big selling points, combined with convention-over-configuration. If you stuck to the naming patterns Rails preferred, it really would Just Work without you having to tell it anything beyond the names of your classes.