Pedantic
Legend
I think it's actually possible to have the cake and eat it here, particularly in a world of digital tools. The key is just expressing your NPC/monster archetypes (skirmisher, brute, etc.) in the same mechanical terms that PCs use, adjusted appropriately for level. I imagine the approach would be some combination of prepacked feat selections, templates, maybe even some NPC classes.In traditional D&D, Humanoid monsters started out with a base set of stats. Then you'd tack on PC class levels (!) to make them more powerful.
This inherently made the Humanoid monster's upgrade feel like it was part of the world - it was casting spells that the PCs cast and recognize, it was doing moves that PCs did (or could do), etc. A goblin rogue backstabbed like a PC did, a goblin fighter got extra attacks like a PC fighters.
Now, a problem develops in that the "ideal" PC is full of bells and whistles for a single human being to spend time tweaking. The player also tweaks the PC between fights and sessions, and PCs last for a long time. Monsters, on the other hand, are "ideally" often run by the half dozen or more by a DM, and rarely survive an encounter.
This means that the detail level of a PC should be higher, to keep the player entertained, while the detail level of a foe of a PC should be lower.
This became exceedingly clear in 3e, especially with spellcasting foes. Building a single spellcasting foe could take hours or days of a DM's time, and running them was also complex.
In 4e, they reacted to this. The monsters where built as an emulation layer; the idea is that you build what the player experiences and you leave the other details up to the DM. This means your monster abilities and PC abilities no longer pull from similar pools, and indeed could use utterly different mechancis!
You also identified the problem with this approach:
Doing it honestly is simply a bigger design ask. You can't jump straight to outputs, you have to shoot for targets and then constrain yourself to a specific set of tools, and a lot of effort needs to be put in to making this all easy to use for the GM. That probably looks like publishing lots of sensible sample NPC statblocks to use, creating a digital tools, committing a lot of time/page space to both your NPC/monster creation methodology and then also creating a separate but equally approachable NPC/monster evaluation methodology.You can bring all this back to the generic "CR 4 spellcaster", but the laziness of "do we haaaavvee tooooo?" kicks in. And you get sub-par cardboard cutout spellcasting monsters who don't feel grounded that cast Arcane Blast with acid damage that is mechanically identical to the cold damage you got last fight.
The ultimate call is clearly that this wasn't worth it.