Don Box says "In my mind, the only way this stuff will get traction is if we change the runtimes to work in terms of higher-order expressions.  Otherwise we're stuck in the 1980's CASE world where box-and-arrows and code were like matter and anti-matter, despite the best attempts at roundtripping UML and code."

I think he means we need to be able to reason and debug at the level of the models. In which case I wholeheartedly agree.  But until everything is done using modelling, we need a well-designed transition from the modelling to the code.  I think of this as the "customization cliff" - once you step outside of what is covered by your DSL, you plunge down a cliff onto the rocks of the low-level platform.  Instead, you need to be able to step down easily onto the next level of detail.  "Easy things should be easy, and slightly harder things should be only slightly harder".