Yea I largely agree with you on that point. I think when discussing Jon, Casey (and to add another, Mike Acton), there's actually a series of advice that they give that get lumped into a whole, and people don't really see the parts of what they're saying and instead focus on the part that sounds most critical to their work.
I do agree that if you take from their "teachings" that every dev needs to optimize every thing, and never use any other language than system languages, that advice is myopic for most devs. However, I don't really see them arguing for that, at least not entirely.
From following their teaching for a while, they mostly preech about the following things which I agree with, even when talking about higher-level languages, technologies.
- Get the clowns out of the car: Don't make things needlessly expensive. Write simple procedural code that maps cleanly to what the hardware is doing. This is essentially stating OOP, large message passing, and other paradigms that abstract the problem away from the simple computations that are happening on your computer is actually adding complexity that isn't needed. This isn't about tuning your program to get the highest amount of performance, but rather, just write basic code, that is easy to follow and debug, that acts as a data-pipeline as much as possible. Using simple constructs to do the things you want, e.g. an if-statement versus inheritence for dynamic dispatch.
- Understand your problem domain, including the hardware, so you can reason about it. Don't abstract away the hardware your code is actually running on too much where you lose vital information on how to make it work well. I've seen this many times in my professional career, where devs don't know what hardware the code will be running on, and this inevitably makes their code slower, less responsive to the user and often drives up cost. There are many times in my early career (backend engineering), that just simplifying the code, designing the code so it works well for the hardware we expect, greatly lowered cost. The hardware is the platform and it shouldn't be ignored. Similarly, limitations that are imposed by your solution should be documented and understood. If you don't expect a TPS greater than some value, write that down, check for it, profile and make sure you know what your specturm of hardware can handle, and how much software utilization of that hardware you're getting.
- Focus on writing code, and don't get bogged down my fad methodologies (TDD, OPP, etc). Writing simple code, understanding the problem more deeply as you write, and not placing artifical constraints on yourself.
Now each of these points can be debated, but their harder to argue against IMO then the strawmany idea of them proposing that you must optimize as much as possible. And they argue that you will actually be more productive this way, and produce better software.
FWIW, you may have some datapoints showing that they do propose what I called a strawmany version of their ideas, but I have seen them advocating for the above points more so than anything else.
---
I do want to add, for Jon Blow, I don't think he has a problem with people using engines. From what I've seen he's played, and loved games that used engines in the past, and had no problem with their output in terms of gameplay or performance. From his talk about civilization ending relating to game dev, he's more concern that if no one tries to develop without an engine, we as a civilization will lose that ability.