It's a nice post, just like the countless others saying the same thing over the decades. But it brings to my mind the infamous The POSIWID principle (the Purpose of a System is What it Does) - while readability is doubtless important, in practice none of us (except maybe tech evangelists and educators) write code with the intent of "how do I best communicate these ideas to my fellow human". At the end of the day, code is judged first and foremost on how well it actually executes its intended logic on a given computer architecture.
Everyone keeps saying that there's a good reason that work on 4GL and 5GL failed, but I do think that there's a missing layer here, and that perhaps if we want to get AI SWE agents to work properly, we'll have to finally discover it - a layer in which we actually communicate system design decisions to fellow intelligent beings (be they humans or AI), coupled with a way to verify that the code matches those design decisions.
Everyone keeps saying that there's a good reason that work on 4GL and 5GL failed, but I do think that there's a missing layer here, and that perhaps if we want to get AI SWE agents to work properly, we'll have to finally discover it - a layer in which we actually communicate system design decisions to fellow intelligent beings (be they humans or AI), coupled with a way to verify that the code matches those design decisions.