i agree they know what they are doing, but lets be blunt its nowhere near enough or fast enough.
Not sure they could really do anything quicker (or certainly better in the short term).
Navi is supposed to be a clean slate architecture, but obviously that takes time to design/sample/debut/produce. In the mean time what else could they do?
Pretty much the options were something like:
A) Carry on and evolve another Vega based product (but 4096 shaders etc is basically the limit) and is a dead end / wasted development resources
B) Fix/Develop Polaris - but to go much further with that it needs a wider memory bus (384-bit or 512-bit ala 280X or 290X) but that then makes the chips/boards costly to produce (and also would be publicly admitting Vega/HBM was a failure)
C) Stop all existing development and "reset", focus resources on new architectures with intention that Navi will be competitive new product. In the interim, use newer fab tech (12nm/7nm) for quick/easy respins of existing products, that don't need much development resource.
Option "C" is where we seem to be at.