Design Playlists & the Home Feed

Open Spotify and Home is a vertical stack of labeled rows: "Made for You", "Jump back in", "Discover Weekly", "Release Radar", "Your top mixes". Spotify calls each row a shelf. The shelves do not all come from one model. A different recommender produces each one, and a ranking layer called BaRT decides which shelves appear, in what order, and what explanation text sits under each card. So the Home request is not "run one model." It is "gather a dozen shelves from a dozen producers and arrange them for this person right now."

That single observation steers the round, because it tells you what kind of system this is. The prompt sounds like a feed, and the word feed pulls people toward the Meta or LinkedIn design where someone posts and the write fans out to every follower's timeline. There is no such write here. Nobody you follow publishes a post that lands in your Home. Home is read-time assembly: when you open the app, the system fans a read out to the recommenders, collects their shelves, ranks them, and ships the page. The hard problems are the fan-out itself, what to do when one recommender is slow, and how fresh each shelf needs to be.

Collaborative playlists are the opposite shape, and that is why they belong in the same lesson. A playlist is a small, shared, mutable, ordered list that several people edit concurrently and expect to see update live. That is much closer to a shared document than to a feed. The two surfaces also feed each other. Finishing a playlist or saving a track is an event, and the recommenders behind Home consume those events to decide what to show you next. Designing them together forces you to connect the read path and the write path through Spotify's event pipeline, which is where a senior answer lives.

The Spotify angle

Reach for a fan-out-on-write timeline here and you burn the round on celebrity hot-spots and an inbox table Spotify does not have.