Catch-all post: April 8, 2026

08 Apr, 2026

1. There's an uncanny valley that happens in software development: one day you have a promising prototype and the next day you have a bad product. I feel this more frequently now that I can push prototypes into the "good enough to be a bad product" stage more efficiently with AI.
2. My favorite writers on the Internet tend to have a take for everything. If someone's overall output is good enough, take density functions as a sort of proof of work: you have to have thought about a subject quite thoroughly to have a take on everything. One way I've changed as a reader in the last five years is that, although I still dislike "hot takes" that aim at cheap provocation, I appreciate "take machines" (or "intellectual athletes") more.
3. The recent New York Times "code overload" article is quite strange. At one point the security risks of employees having lots of source code on laptops is described as "a crazy risk no one thought of six months ago that they’re trying to solve right now." It is, in fact, an old and extremely well-studied problem in corporate security. Elsewhere we find this sentence: "Not long ago, the process of turning ones and zeroes into computer programs was very different." I'm not sure what that's supposed to mean, but I haven't yet found a candidate interpretation that makes sense. I mention this as yet another data point about the average (low) quality of cognitive debt doomerist writing.¹
4. I'm still finding plenty of cognitive-debt articles that Pangram and I agree are very likely themselves to be LLM outputs. (I don't in general see value in linking to such pieces.)
5. Here is a useful Matt Glassman post about how easy it is to confuse the exceptional and the normal. Many programmers have spent their whole careers in an era when programming has been prestigious and there were huge stables of well-paid software teams everywhere. That is, historically, not normal.

#catch-all #generative AI #software