The Algorithm Wants You Bored: How I Used AI to Widen My Taste, Not Narrow It

As I have mentioned more than once in this very newsletter, I’m bound to the musical loves I developed and nurtured in my teens and 20s. Not by choice, exactly. More by drift. You find the bands that resonate, you build a wall around them, and then you spend the next two decades letting a recommendation engine confirm the wall was a good idea.

For years I told myself Spotify was opening me up. New releases on a Friday. A Discover Weekly that occasionally landed something. The little radio stations that spin off a track you liked. It felt like exploration. It felt curated.

It wasn’t. It was a hall of mirrors, and I was the only thing in it.

Here’s the part nobody at a streaming company will say out loud: the recommendation engine isn’t built to widen your taste. It’s built to keep you listening. Those are not the same goal, and most of the time they actively fight each other. The fastest way to keep someone listening is to give them more of what they already chose; the adjacent, the safe, the 4% deviation. Retention loves a narrow groove. Discovery is risk, and risk is churn, and churn is the one thing the model is paid to kill.

So the thing I’d mistaken for a guide had actually become a fence. It had learned that Mark, 44, Loughton, likes melancholy guitar music and the occasional bit of house, and it had quietly decided that was the whole of me. Why would it gamble on anything else? I was retained.

I only noticed because of the podcasts.

The forty-hour backlog

At some point podcast listening became unmanageable. I’d subscribed to maybe thirty shows. Business, AI, a few football pods, the long-form interview stuff that runs three hours and assumes I have an incredibly long daily commute. The backlog sat there like unread email, except guiltier, because I’d chosen every one of them. Forty hours of audio I genuinely wanted, and no way through it that didn’t involve either bluffing in conversation or giving up.

The streaming app’s answer to this is the same answer it has for everything: here’s a queue, here’s autoplay, here’s a playlist we made you. It will happily play you forty hours. It has no interest in helping me (or you) decide which forty minutes are worth taking the time.

So I built something that does.

It’s not clever, and I’ll spare you the architecture diagram. A script pulls my listening; what I actually played, not what I saved-and-forgot. It pulls the new episodes across every show I follow. Then it runs the lot through Claude with a brief: what’s genuinely new here, what connects to a thing I was chasing last week, what’s adjacent to my taste but outside my obvious lane, and, crucially, what should I skip. It comes back as a single styled digest, in the style of PItchfork/Stereogum. One read. Ranked. Reasoned. On my phone before Gizmo’s had his morning lap of the park.

What I’ve learned

Spotify can’t ask me to be braver, because braver doesn’t retain. But I can write a brief that does. I can tell my own system: assume I’m more curious than my play history suggests, and act accordingly. Because my own model doesn’t have a commercial incentive to keep me up to date; It has exactly one incentive; widen what I listen to, ignore the fluff - I’ve found it to be incredibly helpful in cutting out the noise.

Six weeks in and I’ve gone deep on artists I’d have died never having heard, started podcasts on topics that have nothing to do with my job. And crucially I’ve gone back to records I’d shelved, because the digest kept drawing lines between the new stuff and the old stuff and showing me I’d misunderstood my own collection. The wall around the twenties bands has windows in it now.

The Hack: write your own brief.

Anyone can write their own brief. Not metaphorically, you can literally write your own brief, the tools are sitting right there. You can take the same engine that’s been quietly narrowing you and point it the other way, at almost no cost, in an afternoon.

Most people will spend the next decade being optimised by systems they didn’t configure, towards ends they didn’t choose, and calling it convenience. The few who do better won’t be the ones with the best tools. Everyone has the tools now. They’ll be the ones who bothered to write down what they actually wanted, and then made the machine answer to that, instead of the other way round.

The algorithm wants you bored. Comfortable, retained, and very gently bored. Don’t let it have the last word on who you are.

The Problem Hacker — Smart takes on strategy, growth, and the systems quietly deciding things on your behalf. From Mark Jefford // jefford.