I skimmed through your article, and I'll address the bolded points
> being able to understand code by reading it is different from being able to do that same modeling yourself starting from a blank screen
That is simply Domain Knowledge. In a specific context, be it a Framework or an App or a Platform or Infrastructure or whatever else, you have to understand it. Of course, if you start from zero and you build it from scratch yourself, you have pretty much the best DK you can have.
That is usually not the case though. If you have external collaborators of any kind, your DK is always challenged. I find myself often dumped into Domains from which I have to gather Knowledge to understand. And, for me personally, having to read code written by an AI, especially now that they've gotten smart enough to not make glaringly stupid errors, is MUCH better than having to read Hand Crafted Code that has been handslopped (yes, I'm going to coin that term now) to meet Deadlines. Shortcuts, hacks, you name it. Terrible codebase sins, accumulated as years, even DECADES of "we'll fix it later".
Of course it's a pain. It always has been. Back in the GPT 4 / Claude 4 days, AI generated code was atrocious, but that is not the case anymore.
The rest of the article, especially these bolded points:
> The weight shifts from the ability to produce directly to the ability to evaluate output and set direction.
> So in the AI era, productivity is increasingly limited by the speed of judgment rather than the speed of writing.
> A manager who steps away from the front line may, over time, lose the ability to perform the work they are managing.
> does being able to evaluate code mean the same thing as owning the mental model of that program?
Are SPOT ON. You can vibeslop anything at the speed of light. But if you don't have the knowledge, taste, and expertise to not lose track of your DK so that it doesn't become an unmanageable mess, it's going to bite you in the proverbial ass.
But, again, that has always been true!
How is the code that was vibeslopped by ChatGPT in 1 minute different from the Junior Dev who spent a week handslopping a new type of wheel? The only difference is quality and speed.
And you need expertise to be able to recognize when something is wrong, and keep your DK updated at scale.