Serendipity as a feature, not a bug
Most software treats randomness as a fallback. The best parts of the web treat it as the main course.
Most software is built around predictability. You search for a thing, you get the thing. You sort the list, you see it sorted. The interface promises that pressing the same button twice will produce the same result twice. That's table stakes.
A small slice of the web flips this on its head. The whole point of a "next random" button is that it doesn't promise anything. You don't know what you're going to get. You're trading the comfort of a prediction for the chance of being surprised.
This is harder to build than it sounds, because the surprise has to be calibrated. Pure randomness across the entire web would be terrible, most of the internet at any moment is junk. So the trick is human curation: a hand-picked pool of decent things, randomized within. The randomness is real; the floor is high.
When it works, it produces a feeling that feels almost extinct: stumbling onto something good without having looked for it. That's serendipity. It's not a bug, it's the whole product.