Although its workflow is pretty neat, it still requires a lot of transformation coding; especially when it comes to Python and other demanding programming languages. It should be more like other solutions that don't require that much coding. 

DataFlow feels like using a closed environment - you have to keep data in it, and you can't work with virtual data outside of it.

It is not easy to use the R language. Though I don't know if it's possible, I believe it is possible, but it is not the best language for machine learning. This feature could be improved.