Great question. We think a key point of difference is that we'll never focus on providing a 'clicky' way of building actual data processing, training, transformation steps.

For example, in Dataiku you can define and transform columns by using a GUI. We never saw that as more productive than writing transformations using i.e. R data frames, Pandas or Koalas (Pandas on Spark). The Python/R scripts that do the actual transformation can be cleanly versioned, re-used, and modified in a much nicer way than anything produced with GUI based transformation/processing.

You also don't ask people to invest their time and skills into a way of doing things that is specific to a particular tool. I.e. if you can write great data transforms in Pandas today you don't have to change anything when you start using Orchest to build your data pipelines.