We concluded that for serious modeling, 80% or more of what one does is giving names to concepts, leading to the usage of a lot of text any way. And that boxes-and-lines introduces a lot of 'clutter' which is probably meaningful, but not in a strict enough way to do anything useful with it: x/y coordinates of boxes and lines.
That's next to there being a lot of tooling available when going text based (editors, versioning, etc), which is more of an added bonus in the short run.
In the end, we will probably evolve in the direction of using a projectional editor, making it possible to have some boxes-and-lines modeling environment as well.
To me, projectional editing is the most sensible type of programming. Unfortunately projectional editors have not had as warm a reception on HN, because programming is narrowly defined by complex text editing. Anything with a UI and programmers are afraid they will be accused of being users. It's dumb, and most won't realize it or admit it, but that's what it comes down to.
Projectional editors can of course show the model as text and in other formats as well, like boxes-and-lines. You wouldn't be able to permanently place them in any specific spot then, so moving them around would at most be supported during a viewing session.
I don't buy that it's actually better without a UI or that you can't make a UI for configuring complex models (it's hard but has been done many times before). But like I said, programmers seem to accept it more readily without the UI so maybe that's the trick to programmer popularity for these types of systems.