The most addictive algorithm ever built

I found an answer to a question I ask every day, usually when I'm about half an hour in to doom scrolling Spurs clips and dogs welcoming their owners back from war: What makes TikTok so addictive?

TikTok isn't playing the same game as everyone else.

The platform's recommendation engine, known internally as Monolith, doesn't care who your friends are. It doesn't need you to follow anyone, like anything, or build a profile. It watches what you actually do - and it learns in real time.

Most recommendation systems work in batches. They collect your behaviour, process it overnight, then serve you updated suggestions the next day. Monolith updates continuously. Every scroll, every pause, every half-second you linger on a video before swiping away - it's all feeding the machine, which adjusts its predictions immediately.

The system tracks micro-behaviours that users aren't even conscious of making. Rewatch a clip? Data point. Speed up your scroll after certain content? Noted. The length of time between opening the app and your first engagement? That matters too.

This is why TikTok can hook new users so quickly. The "cold start" problem - how do you recommend content to someone you know nothing about? - gets solved within minutes. A few swipes and Monolith has enough signal to start serving content that keeps you watching. The result is an algorithm that knows what you want before you do.

🧵 Here's a similar deep dive on the X algorithm. Some dark shit.