I've recently been noticing a pattern, in editors where you can create collections of objects (for example, in the editor for a multiple choice question when creating the possible answers), where adding a new item is done through clicking on a "ghost" editor for that object, which turns into a fully-usabe, real editor upon the creation of the object.
For example, in Google Forms, adding a choice to a question looks like this:
Comparing it to a more traditional "Add" button, like in this case:

Are there any advantages in using this pattern vs the more traditional "Add" button? Does it lead to better predictability/understanding or to a lower interaction cost?
Does it depend on the complexity of the controls that are generating upon creating the object? For example, in Google Forms it's just a one-line input with two buttons on the right, so it's pretty "lightweight" to the eye; in the second example, there's a taller editor, so maybe it wouldn't be suitable in that case?
