Post

So it's useful to talk about the specific ways in which this argument is dishonest because I think a lot of liberal and left-leaning people are very vulnerable to this rhetorical bludgeon—the idea that you should "listen to experts". But let's unpack what exactly Giggies here means by "experts".

For one thing, "software engineers" as a group are not experts. They don't have some form of accreditation, there's no standard of knowledge required to be a software engineer. "Listen to software engineers" isn't the same category of statement as "listen to climatologists".

For another, programmers are extremely poor at evaluating their own productivity, especially in complex projects, especially when factoring long-run code quality and not just "features shipped". This is something that anyone who has worked in software should be very familiar with.

So when I see significant numbers of programmers hopping on to a tech hype cycle, my own very real expertise is telling me to be highly skeptical. I don't know how you can remember, say, the 30-year debate about OOP and think programmers can tell how useful a tool is in a year.

I give a shit about *utility*, about the social and economic value of those tools, and the bottom line on that is deep, deep in the red. You don't have to countenance bullshit "expertise" arguments from data center defenders who are, functionally, climate denialists.

A fundamental problem with "oh it's useful for coding" is that "generating more software faster" has, if anything, negative marginal utility in itself. The world is not suffering from a software shortage. The world is suffering from a software quality crisis, which LLMs seem more likely to worsen.

You probably can use an LLM to make high quality software, but the tools really, really do not bend in that direction and they don't seem to favor that. And "software engineers" entirely lack the safety culture around automation to even attempt this.