You’re specifically talking about web development and have listed 0 of the things I, am embedded developer, do in an average day. All programming !== web development. There are more difficult problems out there than throwing up a web page. Especially in performance-sensitive applications.

I am specifically talking about the kind of development that employs, at a cursory glance, more than 65% of the development workforce. The lion's share of the remainder works on executables (commercial software) that run on a consumer device (most of which, again, is incredibly similar to web-client development in 2019), and within those specialized domains, I suggest the power-law largely still holds.

If you have some numbers to suggest that embedded engineering is a particularly good field for engineers who want a profession that rewards a high degree of specialization, I'm all ears.

Also: having worked on performance-sensitive backends for most of my career, most applications are performance-sensitive only in the shallowest sense and the marginal returns for improving performance do not in general justify the cost. One can (again, like I said) seek out domains where understanding how to build very low-latency or very energy efficient or highly concurrent (or whatever measure of performance you care about - very durable? very reliable? etc.) applications matters. But that's not 80% of the work out there.

Edit: also I find embedded engineering to be a very strange choice of counter-point. My last job alongside embedded engineers (at TE Connectivity) saw two of them defect to become fungible Java backend engineers so they could find stable work and transition into management. The remainder worked on specialized switches. Their day-to-day was managing a piece of software that polled chip readers and a REST service that hosted that information. I don't recall ever overhearing a conversation about optimizing TCAM usage or packet switching latency under load or whatever, maybe some of them got to get into the weeds on that every once in awhile. I suspect, again, that this is largely akin to having kernel development skills on a team that largely operates in user-space -- not part of the 80% -- but I am absolutely not an expert in the field).

The whole point of the post was that knowledge-intensive domains exist (and are quite common, 20% is a large fraction), and that some domains reward vast amounts of knowledge (hence power-law distributed, not normally/exponentially/etc. distributed), but that you had to seek them out.

You don't always have to go into management to follow this path though. At companies I've worked at, as you become more senior as an IC, your role expands from contributing locally on your team to contributing to the entire org. There is also often a split between the technical leadership and the people leadership on teams, which allows lead engineers and managers respectively to focus on each area.

IMO an underrated and difficult aspect of modern tech is figuring out technical practices that scale with team size, so that you can add more people without sacrificing quality, focus, speed, etc. I'd say microservices and continuous deployment are two examples of technical practices that address this, but there are more.

I've started to propose we s/FANG/Big N/g "FANG" leaves off other obvious big companies, e.g. Uber, and it wouldn't scale to keep trying to add them to the acronym. Here N is used like when one talks about a list "N items long"

Amazon makes a ton of money, and (quite transparently I might add) they plow a lot of their earnings from some of their businesses into investing into their other businesses, which is what most investors want.

This is completely different from Uber and Lyft, where it's not clear that their profit potential supports their current valuations.

Edit:

I just realized what else that stood for...