Primary issue: While dataiku shows the flow graphically, it isn't a graphical designer. Meaning, you can't place the nodes where you want them and instead rely on dataiku's auto layout. These are limitations in the current flow designer:
1. If need to change the input or output of a node, you have to open the node and change the input or output datasets, and sometimes due to display length constraints in the drop down list, it's not easy to see which is the new input dataset you want (unless you've been really good at naming your nodes).
2. Following on from (1), because of the auto layout, the location of your nodes can change on the canvas which can get confusing. In a complex flow, I want certain nodes in certain places as one reads it almost like code. To have them moved causes a lot of confusion.
3. Similarly, it's hard to group components with a visual spacing that makes sense to the author because you have to accept the node placement enforced by Dataiku.
Solution:
It would be great if it followed a little more of a drag and drop design paradigm whereby I could drag the line to a new node (or from a new node). That way I could easily change my input node, not be confused as to which data set I'm using, and not have things rearranged on my screen. This would make it far more of a dynamic graphical design environment. Not that "copying other products" is always the right solution, in this case, I feel solutions like Knime or Alteryx have the edge here and this paradigm has been pretty well established with most low code environments.