The google one looked like it was focused only on a few use cases (language / vision). I've played around with some of the open source stuff (tpot, autosklearn

The google one looked like it was focused only on a few use cases (language / vision). I've played around with some of the open source stuff (tpot, autosklearn, h20's automl). The commercial products seem to do more automated feature engineering. H2o's automl is pretty easy to use with their graphical "flow" notebook interface. Was able to build a reasonable model with no coding aside from prepping the data going in.