I was thinking about this the other way- new languages will always be designed by humans or human/ai collaboration but the task of a certain kind of viability will involve producing sufficient useful training material- and this when humans are intended to be able to also read the code strikes as being a new kind of gate, as such material will have to be automatically produced which may involve arriving at algorithmic definitions of fuzzy human syntactic concepts that refer to taste and readability. Super interesting design problem.