In my experience, novices generally stick with the default. And when they don't, their code breaks because they assumed serializable was somehow magic and they don't handle the inevitable transaction failures (generally on production, when there is load). This is exactly why PostgreSQL drivers that used serializable as a default had to backtrack and default to read committed. And yes, I agree that the blog is encouraging serializable without adequately pointing out the downsides; th extra burden on application developers would be #1 in the caveats in my book, and using serializable when you don't have to will lead to a net increase in bugs.