At one end would be dynamic and scripted and would progress to the other side of compiled and static. You would start out at one end for speed of development and progressively optimize and compile.
I suppose python and cython are that language. But it's not as popular as it should be.
I first read about multi-level programming back in the 80ies and was hooked.
Objective-C has bit of this, because the C part is static and pretty low level, but the object-oriented part is dynamic and can go quite high level [1][2].
My own project, Objective-Smalltalk[3] definitely has "multi-levelness" as one of its primary goals. The way to get there is to make the elements of the language malleable, similar to Open Implementation[4], but using a software-architectural frameworks instead of an object-oriented one.
[1] http://en.wikipedia.org/wiki/Higher_order_message
[2] http://blog.metaobject.com/2014/05/the-spidy-subset-or-avoid...
https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html
So today we should be able to write 90% of an app in a rapid application development (RAD) language like HyperTalk, MATLAB or Python (or my favorite, the language that doesn't exist that combines all of these) and micro-optimize in places that need it using C/Java/Go/Rust etc. The compiler should automatically generate all glue code for variables and function calls.
My sense of this is that functional programming is gaining popularity because most people are working a few levels too low to be productive. Short imperative code should be able to act as a black box with no access to global variables or sockets and be exercised with known inputs/outputs and be represented in a functional manner. I wish we could deal with logic as truth tables and state transitions rather than the code (declarative rather than imperative).
Do you mean the recent GHC developments that let you defer typechecking to runtime? That seems a little too new, to attribute it to Haskell like that.