I don't think you're agreeing to what the article is saying.

The article seems to be saying, instead of using CGI which spawns a process per request, to have a single Web server binary in Go/whatever. Which is totally reasonable and per my understanding what everyone already does nowadays (are any greenfield projects still using CGI?)

CGI is a "clever 'Unixy' hack" to add dynamicism to early web servers. They stopped being "relevant" a long time ago IMO.

In fact, I think your diatribe actually contradicts the article.

Basically, the article is saying that they went with the "simple" CGI approach which ended up creating more complexity than using the slightly more complex dedicated binary. The author essentially followed your advice which ended up causing more complexity and hacks.

The morale of the story is, you need to use the right tool for the job, and know when to switch. Sometimes that is the simple path, sometimes that is not.