What's the opinion about keras from google research team?

One of my co-workers asked this question but I’ve no clue to answer. It is assumed that after tensroflow 2 (keras), the research space might change in terms of API.

But gradually, Jax becomes popular and to research scientis from gogle are more inclied to use it. So, as a research scientis at google, what is your opinion about keras for research tools? Will you recommened it to other? or Jax? or PyTorch.