The faulty reasoning is this: If Hacker Hortense does something with powerful language L that Newbie Nathan finds confusing, we assume that language J won’t permit Hortense to do that, and therefore if we standardize on language J, fewer bad things will happen.
This reasoning is faulty. First off, if Hortense and Nathan don’t see eye to eye on how to write programs, no language will solve the problem, because you have a disparity of experience and/or education. If what you want is code Nathan will like, and you don’t want to raise Nathan to Hortense’s level, you have to drag Hortense down to Nathan’s level with coding standards. And you could have done that in Language L just as easily as J. If you ban L, what will you do when Hortense implements Parser Combinators and Monads in J?
The first flaw is the assumption that language features introduce complexity, when in reality it isn’t the features, it’s the solution to the problem that confuses.
This indirectly leads into the second fault with the reasoning. There is a hidden assumption that Hortense is going to write so much code, and there will be so many “problems” Nathan will have with his code, and every problem we eliminate is one fewer problem in the result. This is not how software works.
Let’s say Hortense writes some code in L, and Nathan complains that there are seventeen things he doesn’t understand. Ten of them rely on features of L, seven are independent of the language. If we tell Hortense to rewrite things in J, we cannot assume that the result will have seven things Nathan doesn’t understand. It might have even more as Hortsense works around J’s deficiencies.
For a demonstration of this, look at any modern Java Dependency Injection Framework. These complex and opaque software machines are made up of of XML, interfaces, and classes. They exist in no little part because Java lacks the reflection and meta-programming of languages like Lisp or even Ruby. Are these thing simple by virtue of being written in Java instead of Lisp?
Imagine for a moment that DI frameworks don’t exist, and that Hortense had built something in Lisp with macros for dependency injection. Nathan is nonplussed, so Hortense rewrites it in Java and rolls her own DI infrastructure. Will the result really have fewer things that Nathan doesn’t understand? Or will it be even more inscrutable thanks to the accidental complexity of working around Java's limits??
The “problem” is not that some languages are too powerful, it is that people imagine that teams of programmers can work together on complex problems without communicating, and some people assume that when there is a disparity of experience of knowledge, that people can work together without educating or learning from each other.