In as much as boxes and nodes systems reflect the complexity of what you're doing, perhaps a lack of scalability goes to show that the complexity is visible whereas we're better at hiding it in a textural environment!

My dogma is howling, telling me I should 👍 this to heaven. But I don't quite agree. I think typical nodes-and-noodles languages are too complex because they aren't actually "visual" — they're just sprawled-out text languages with the demand that you bounce back and forth between keyboard and mouse. Most of this complexity is incidental, just of a different kind than text languages that don't (eg) make control flow so explicit. A true VPL would let you both name and draw nodes, and animate them to reflect state. Then you can build yourself (or use built-in) nice visual abstractions for state machines, remote procedures, query and pattern matching, testing, etc. Yep, that's complex and would take new kinds of skill and effort. But you spend those efforts on things that improve comprehension dramatically. I'm advocating for a Lisp of visual languages, in a sense.

I don't quite agree either — though I wouldn't have written it if there weren't something to the sentiment. For full disclosure, I like NnN (nodes-and-noodles) languages at their best: when the nodes embed control widgets, when edges are dynamically inspectable, when zoom is semantic, when layout is easy, when the interface has nice typed/structured editing. Text has a compactness advantage, is more string search friendly, and in a way doesn't promise as much. Few are surprised if they can write anything sensible in a hello.some-new-language file without having read a tutorial if not a book whereas NnN give more the appearance of immediate access to beginners.

zoom is semantic

Can you elaborate on that point?

@wtaysom I would like to know how much PureData fits the criteria you have described: https://puredata.info/