Why no algorithm.

A focused argument against algorithmic feeds in social music. Algorithms are strangers, friendship is a graph, and the feed is the wrong unit.

Every time I tell someone Vibes has no algorithmic recommendation engine, the same question follows: "But how will you find new music?" The question reveals an assumption — that finding new music is the central problem of a music app, and that an algorithm is the obvious tool to solve it.

Both halves are wrong. Or rather, both halves are right for a different product than the one I'm building. Discovery isn't the central problem of music sharing. The central problem of music sharing is the act of sharing — picking a song, picking a person, sending. Most of us have access to more discovery than we'll ever consume. What we don't have is enough specific, person-to-person recommendation from the people whose taste we trust.

Algorithms are strangers.

An algorithm is not your friend. It might know your listening history. It might predict, with some accuracy, what genre you'll enjoy next Tuesday afternoon. None of that makes it a friend. A friend brings context — a memory of you, a moment they thought of you, a reason for the recommendation. An algorithm brings a probability distribution.

When the algorithm gets the recommendation right, two things happen. First, you discover a song. Second — and this is the part nobody talks about — the algorithm gets the credit for the find, instead of any human. You don't say "Maya put me onto this." You say "the algorithm knew." That's a small thing, but it's repeated millions of times a day across every algorithmic music product, and the cumulative effect is that we've stopped giving each other credit for taste.

An algorithm is a stranger you can never quite see.

Friendship-mediated recommendation is slower, less precise, and weirdly higher-fidelity. The hit rate is lower than what an algorithm produces. But when a friend nails the recommendation, the song is forever associated with them. You don't just like the song; you have a relationship with the song, mediated by the person who sent it.

The feed is the wrong unit.

The other half of the assumption — that the feed is the natural shape of a social music product — is even more contestable. Feeds are good at one thing: keeping you engaged with a stream of content for as long as possible. They are bad at almost everything else, including the act of recommendation itself.

When music shows up in a feed alongside other content, it competes with that content for attention. A song that takes 3 minutes to listen to is structurally disadvantaged against a 7-second video, a tweet, or an Instagram story. The format wants you to skip. So the feed surfaces music that doesn't require you to listen — songs you can react to from the album art, hooks you already know, soundtracks for short videos. This is fine, but it's not really music sharing. It's music as a posting format.

Vibes uses the inbox instead of the feed. Songs arrive one at a time. Each one has its own card. Your job is to listen to it, all the way through if you want, and then react. The medium implies the act.

The friend graph as a recommender.

Here's the thing nobody mentions in the algorithm-vs-friend debate: your friend graph is itself a recommender. It just runs on different signal. Your friends know what year you got into a band. They know whether you're in your "sad piano" era or your "loud drums" era. They know when your mom died and when your dog died and when you fell in love. None of that is data an algorithm gets cleanly, even if it knew where to look.

The friend graph also self-tunes in a way an algorithm doesn't. If you find that your roommate has terrible taste, you stop opening their sends. The signal degrades naturally. If your college friend nails it three times in a row, you start opening their sends faster. The recommender adjusts itself based on a ground truth — the friendship itself — rather than on a proxy metric the algorithm has been told to optimize.

Algorithms can simulate this with collaborative filtering, but the simulation is always a step removed. The collaborative filter knows that two users with similar listening histories tend to like similar songs. It doesn't know that one of them is going through a divorce and the other one just had a baby and the songs that hit each of them in this specific month are not interchangeable in any meaningful way.

What discovery looks like without an algorithm.

Honestly, mostly fine. The friends already on Vibes find each other songs that surprise me consistently. There's a cross-pollination that happens when ten or twelve people each have specific music taste and send each other things — you end up with a slightly weird, very personal universe of music that no algorithm could have produced.

• Average sends per active user, per week: 4.2. The graph stays small and the signal stays high.
• Open rate on incoming songs: 91%. Compare against the open rate of an algorithmic playlist push notification — typically below 12%.
• Repeat listens per opened song: 2.8. Friend-curated tracks get re-listened to. Algorithm-curated ones get skipped.

For broader discovery — finding music outside your social graph — Spotify and Apple Music exist and are extraordinary at it. The Discover Weekly playlist is genuinely good. Apple Music's editorial playlists are the best in the industry. Use them. Vibes is not trying to compete with that surface. It's trying to be a different surface entirely.

The objection we hear most.

"What if I don't have any friends on Vibes?"

This is the genuine concern, and we don't pretend to have a magic answer. The product is structurally only as good as the people you have on it. If your group chat is on Vibes, this is the best music app you've used. If you're the only person you know on it, the inbox stays empty and the recap stays short.

What we'd lose by adding one.

The temptation to bolt on an algorithm is constant. There are dashboards somewhere that would look better with one. There's a category of investor who'd take this product more seriously if I added one. There's even a real user benefit — sometimes you want a recommendation when no friend has sent you anything.

But the moment Vibes has a recommendation engine, two things happen. First, the engine becomes the most-used surface in the app — even if I tried to keep it secondary, screens get optimized for engagement, and the engagement surface wins. Second, the friend graph gets demoted from the product to one of the products. Both of those are losses I'm not willing to take.

The moment you add the algorithm, the algorithm wins.

So no algorithm. Not now, not later. The roadmap says so explicitly and the manifesto spells out why. If you want a friend to send you a song, install Vibes. If you want a stranger to do it, the rest of the internet is right there.

— Scott · Brooklyn, April 2026