I think the point here is that going back over the last 50 or 60 years people have worked on CAS, and in the process have explored many paradigms of programming.
They have implemented various ideas including “one language for system builders and users” or different languages at different levels. They have explored expressiveness based on methods, functions, rules, algol-ish control structures, point-and-click menus, etc.
I suppose I would point to Axiom as the most highly developed in the mathematical/ algebraical/ category stream of thought. Probably Mathematica in the pattern-replacement rule-based paradigm.
I don’t know what would be the best exemplar of user-interface-menu-clicking.
Julia does not seem to offer any particular language idea that has not been used previously. So why do a CAS again? Frankly claims that it is 10,000x faster than some system S is pretty likely evidence that system S is broken, ill-suited to the benchmark, or being misused.
Its an interesting claim that if you use Jupiter then you can focus on the actual problem, something you could not do (conveniently?) in any previous CAS. I have not yet seen any evidence of this, i.e. comparing Maxima to Julia. comparing Mathematica to Julia. comparing Axiom to Julia.
RJF