Most ORMs are heavily inspired by object-oriented programming which fosters the object-relational impedance mismatch. In addition to logic for storing and retrieving data, models often also implement business logic (ActiveRecord is a good example here). This makes for bloated and complex objects that are difficult to work with in the application. A solution can be to lower the level of abstraction and use a more lightweight query builder (like knex.js for Node.js). These kind of tools give you more control in constructing your queries as well as the ability to optimize them.
These tools still require you to understand quite a bit of SQL though and the productivity leap compared to writing manual SQL isn't as high. I believe that query builders are the best compromise we have today for accessing a database from an application.
Regarding the dual schema dangers that are mentioned by the author, I strongly believe that these can be alleviated using code generation tooling that helps to keep your database in sync with your application models (approaches like SQLBoiler in Go where application code is generated based on the database schema are an example here).