I'm not sure this is quite true. 99% of early 2000s websites can be made by an unskilled operator automatically, by using something like Squarespace or Wordpress. The other 1% are the hard projects -- desktop quality applications that need to run on 5 platforms and 3 Javascript engines. Most people that do frontend engineering are being paid to work on those hard problems, so the job is going to feel harder than it did 20 years ago. There is no money in the easy things; it's all been automated away.
To some extent, the tooling does make things harder than it needs to be; it's converged to a very strange local maximum that's very far away from the global maximum. The problem, I think, is a complete lack of integration along the entire stack. You write your code in a programming language whose source code is sent to the user to compile, and there are hundreds of minor variants on how the user will interpret that code. You have to defensively handle it all. But, developers want to reuse code, and those runtimes don't really support code reuse, so you have to bolt it on with a fake "compile" stage, where you concatenate all your dependencies together and split it up into chunks to be served to the end user's compiler at just the right time. The language that is used for these tools is a little on the outdated side, so this compile stage takes 30 seconds on one CPU core, leaving your commodity-grade 32 core workstation 96% idle while you sit around waiting. And, people don't even like this language for writing their code, so they write it in a different language that is compiled to that first language. After all that, you have code that can run on users' machines, but that's only like 30% of the problem. You have to serve that code to them, preferably from a datacenter that's physically close to their terminal. You have to serve them ancillary assets, like instructions for how to format the data your app interacts with, and images. Like the author mentions, there are a million different image formats, and you have to pick the right one for the end user, relying on a single line of text like "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/89.0.4389.128 Safari/537.36". That's hard, so there is some service you can buy to do that for you, completely unrelated to that aforementioned "build process". The TL;DR is that there are hacks on top of hacks, and moving parts on top of moving parts, that it's a wonder it works at all. But we're in this state where we can't even fix it, because there is no one party responsible for the end-to-end experience. All you can do is bolt on more and more components and hope that you get closer to a local maximum.
The takeaway is that in a distributed system of glued-together components, no one entity is responsible for the success of your users. And, those that manage to build success for their users will do it all through very careful glue, that can come apart and blow up at any time. That means they have a never-ending job that consists only of unnecessary toil. At some point, the person that decides "FUCK IT" and throws this all away and builds some sort of integrated experience is going to make a lot of money. But there are many lifetimes of work ahead to achieve this goal, and you'll be dead before you finish, so why even try? That's where we are. Good luck.