Visual programming with connections often just becomes literal spaghetti code, we really lean a lot on linguistic abstractions to manage complexity. I played around with lots of ideas, the latest one being direct manipulation of visually represented abstractions. Fun and somewhat promising (works really well for expressing bubble and quick sort, less well for rebalancing a red black tree after a delete), but I don’t see anything panning out before AI writes all the code for us.
I just don't think quicksort is a good fit for visual programming, (not that I've ever actually implemented quicksort....).
Visual is excellent for things like "At 7PM if the enable switch is on, turn on the sprinkler".
Stuff that's very simple, but you want no chance of it going wrong, and you might want to edit it from a phone.
When you want the least powerful programming model possible, that isn't even turing complete, that's arguably not even programming and just configuration, it's great.
It really depends on your visual rep. For me, I took symbols related to CFG and RegEx representations, and then focused on direct manipulation of those representations. You can find a YouTube video of it probably, and I know I have a paper somewhere, but it’s been so long.
A conversational interface is already going to work well for simple things, but that isn’t very visual. Without abstraction, encapsulation, and generalization, are you even programming?
