(I hit the comment character limit.)

...to write books. It is a GUI, I suppose. (In a way that Vim or even Gvim isn't.) But it is heavily text-based. It's a "glorified text editor", not a "graphical editor".

So, I guess what the OP article is going for is something like you write a book about the "why" and "what you're trying to accomplish" and an NLP wizard in your computer turns that "book" into a working program. It's "common knowledge" (not to say "common knowledge" can't be wrong) that human language is not "precise" or "unambiguous" enough to make working software in. But I don't know OP would respond to that. (I think often times, the non-programmers' problem isn't so much in being able to communicate precisely as in being able to think in terms of the specific kinds of questions that need answered when you're doing software engineering. To non-coders, it's not always clear to them that the button on the page is distinct from the action the button takes when you click it. Once the button exists, the task is done, right? And if the button doesn't "work" (whatever that means) before the code to handle the click is written, that's a "bug" (rather than a yet-to-be-implemented feature). This is a big part of the reason why LLMs are so popular. You can make a barely-functional tech-demo/mockup very quickly that chokes the moment you try to use it for anything even slightly interesting, but the news media all picks that up and says "AI just re-coded Photoshop from scratch!!!". (Note the "early alpha" at the top right there.)

But then they shift tactics and say "no, not NLP, actually visual editing". But I don't see how that solves anything. And I'm not sure whether OP sees how that solves anything either beyond just "visual editing" (whatever that even means) "is easier for non-coders". I mean... yeah, if you can figure a reasonable way to make it even remotely feasible.

I linked Glot3, a DSL for doing 3d modelling, earlier. I spent months writing stream-of-consciousness vague thoughts trying to figure out even the basics of the data model that should use, struggling with questions like "what is a vertex really" and "what is a face" before settling on something that I thought was serviceable. After settling and starting to code, I still ran into obstacles -- holes in my prior thinking -- that necessitated relatively significant refinements to what I'd decided previously. And I've been coding for almost 20 years. How is "visual editing" going to make non-coders capable of thinking in the right way to develop software? Let alone describing it sufficiently precisely (even in their native language).

I dunno. Maybe there's some seemingly-magical solution. I think it'd have to be more interactive than just "write a book and run it through a thing". There'd have to be something capable of saying "you've been ambiguous here" and forcing the user to fix it before continuing. (Kinda like a compile-time error.) It would have to be capable of teaching non-coders how to think like coders by stubbornly refusing to work any time they didn't think sufficiently precisely. And even then, it's likely they'd end up with something that, although it followed their instructions, didn't solve their problem. (Just like an amateur building a bridge any more ambitious than a little stone walkway over a creek might end up with a bridge that collapsed under the weight of whatever might pass over it.) In which case, they'd probably have to make pretty significant revisions or start over. I'd worry that the result would end up being organically throwing things at the wall until it kinda-sorta works. Basically the same thing a coder who didn't know what they were doing would end up with if they did the same but with code rather than NLP or graphical editing. A Ball of Mud.

One thing the article says is "The GUI version of a program could merely be one of several human-editable representations". Maybe the idea is that the GUI and human-language and code versions of the project all be translateable between each other at will. But if this whole thing didn't already sound like magic, this makes it sound like straight-up fantasy. Unless they can demonstrate some way to make this work. But if they'd done that, I doubt I'd have to do anywhere near as much research to figure out what it is they even mean.

I dunno. I still don't get it. Literate programming is one thing, but even in literate programming, coders have to be involved, and they have to architect the overall system just like they'd have to the non-literate-programming way to do things.

I kinda want to see what an FAQ would look like. Somewhere where people could push back on the author a bit. Ask the "hardball questions" and see how the author responds. I'm not sure I'll have a picture in mind of anything like what the author is going for that could possibly actually work without seeing something like that.