> It’s why the best engineers try to write as little code as possible. They understand that each additional line is another thing that must be understood, remem

> It’s why the best engineers try to write as little code as possible. They understand that each additional line is another thing that must be understood, remembered, and maintained. As a long time SRE/DevOps, this is the key point that I believe people are overlooking in the "AI will let anyone write code!!" Support does not scale linearly with additional software. If anything, it grows faster due to the "two systems == one connection, three == three connections, four == six connections" etc. Sure, you could argue that LLMs could ALSO do the support but the reason you have people around to begin with is for when the technical systems break in weird and interesting ways. Similar to how planes have pilots even in the age of autopilots (the humans are, of course, still fallible). Another argument might be: well, LLMs will write better software. My counter is that there weren't really excuses for not writing better software before but we still got buggy code with poor documentation. If anything, LLMs will make it easier to crank out barely maintainable and supportable code at even faster rates.