I am a programmer and do not regard this as a breakthrough. Generating some code from a screen design is nothing new and at best controls the user interaction with that one screen, to collect input or display output. No AI program is going to anticipate the background processes needed to unify a number of screens to produce an application of a specific purpose. It would rudimentary and extremely generic at best.
I have examined code produced by an application generator - it is bloated, overly complex and fragile to changes. I can write superior and maintainable code from scratch. There will always be a need for programmers; nothing matches the inventiveness of human imagination. AI cannot design anything of higher intelligence than itself.
Depends on the RAD or CASE tool you use. Some actually generate very robust and efficient code. Granted, some generate crap that is bloated and very fragile, but some do not.
To everyone else, this is hardly a breakthrough. All that this thing does is generate the screens and controls. It does not do 95% of the programming as in what a button does or what a data element is and what to do with it. There are tools that will do 80% of the programming, leaving you 20% to write.