What a great hire, but he should probably start something new rather than work on Tensorflow. Obviously he likes to create new things. Not sure what google is thinking here. Maybe he'll create a higher level language to TF?
There's a lot of things to do in TensorFlow that involve a lot of work (and could even be its own project). Namely support for dynamic computation graphs (a-lá PyTorch). TF as it is is great for production, and even for research, but a lot of people in the latter are starting to look at PyTorch more and more for prototyping.
I think tflearn already occupies that space and it's pretty solid