Funnily, I was working on cloning a particular piece of Firebase functionality, and inquired if they could help me with that (or if they had that particular piece in open source), and got more replies from them for that than apparently this paying user ever got.
Yes. They are in the "Embrace" phase of the "Embrace, Extend, Extinguish". Sure they are helpful. Have to be (considering the API drift between the Tensorflow releases and how late they were - Theano is the same exact idea and great implementation, only first released in 2009-ish).
But it's pretty easy to look a bit into the future, and predict what we would see during "Extend and Extinguish" phases of TensorFlow project.
Very typical from Google.