Let's take the simple concept of "array" or "list". Mathematically speaking, it's a very simple concept: integer numbers are mapped to objects.

But there are so many ways to implement this mathematical abstraction. Do you assume the integer numbers will be used in a sequence, like 1, 2, 3, 4, 5; or completely arbitrarily with possibly large gaps, like 1, 1000, 1001, 5000? Will it be accessed from different threads? Will it have a "create phase" when it is constructed in a single thread, followed by a "use phase" when it is accessed from multiple threads but read only? Will it be modified frequently, or rarely, compared to mere reading? Will you need to make copies of it for further independent modification? Etc.

Now, one possible way is to choose one specific answer, and make the standard "list" in your language mean exactly that. For some purposes it will be okay, for other purposes it will suck, and someone will create a library providing an alternative implementation; and users will complain why they need an extra library for something that should have been part of the language.

Or perhaps you will provide multiple implementation, and users will have to choose. And they will complain about this being too complicated.

You might also spend years trying to find one perfect implementation of "list", that will fare relatively well under all circumstances (never the best one, but also never the worst one).

It would be nice to have a language where the programmer wouldn't have to worry about performance of the underlying data types. Just use them as mathematical abstractions, and everything will work fine. Like, you would still have to worry about algorithmic complexity of your code, but you would not have to worry about accidentally using the existing stuff in a wrong way which is not obviously wrong (and would not be wrong for a different implementation).

This is different from merely allowing people to use both high-level and low-level concepts in the same language. This is about how to implement the language in a way that allows you to do high-level as much as possible, without suffering terrible performance consequences. And I don't mean consequences like "this will be 100 times slower", but rather "this implementation of list, when used in this specific way, will actually have exponential complexity where some other implementation would have been polynomial". Because ultimately each implementation has a weakness, and one must be chosen. And if you treat other pieces of code as black boxes, it means you never know whether you made a good choice.