Good points. I just wanted to add that one really needs to think with a focus on business logic. These two suggestions need to be carefully balanced:

1. Business Logic Independence

and Use GraphQL when you need flexible data fetching

With flexibility in data comes the temptation to program business logic in the frontend. Short of using homomorphic computing, this can't be done entirely on the frontend.

Also, it is important to stress that more complicated applications may need a non-trivial domain model (in DDD terms), not just CRUDL. It very much depends on the application, but I'd consider GraphQL as a variant of the CRUDL approach, which might be okay in some cases but not ideal in others due to bringing business logic to the frontend.

Thus, even the choice of GraphQL should not be made just because of tradition but should be carefully considered as one option among others.