If I understood John's motivation from the video, it's that this enables the Flatten structure to be inspectable, because the innards are not hidden behind an opaque closure (A => F[B]), as is the case in regular Monad.

Btw, this is a problem that the "Selective applicative functor" too aims to alleviate.

You can read more about the inspectability problem (and Selective) at http://eed3si9n.com/selective-functor-in-sbt

The context there is sbt, which is a build tool, but I'm sure the inspectability plays role in many other areas. Note, how he contrasts "Applicative composition" and "Monadic composition".