I opened a streaming app the other night and the home screen felt like déjà vu. Not because I'd seen the specific titles before, but because I'd seen this exact version of myself before, rendered in rows. More of the thing I watched last week. A slight variation on the thing I finished last month. A category the system had decided was "me," populated with fifteen near-identical options. It was efficient, frictionless, and faintly claustrophobic, and it took me a minute to name why. The app wasn't showing me the world. It was showing me a reflection, and the reflection kept getting narrower.

We've been sold personalization as a window, a way to cut through the overwhelming vastness of everything and see the small slice that's relevant to us. But a window looks outward onto something you haven't seen. What these systems actually build is a mirror, and a mirror only ever shows you what's already in front of it. The algorithm doesn't know who you could become. It only knows what you've already clicked, and it will hand you more of that until the end of time.

Optimized for Engagement, Not Growth

Here's the thing worth being precise about. A recommendation engine isn't trying to make you wiser, broader, or more interesting. It's trying to keep you engaged, because engagement is what gets measured and monetized. Show me the incentive and I'll show you the outcome. The system's goal isn't your growth; it's your attention, and those two things point in very different directions.

Growth requires the unfamiliar. It requires the book slightly above your level, the idea that irritates you, the genre you assumed you'd hate. Engagement requires the opposite: the reliable hit, the confirmed preference, the thing so aligned with what you already like that clicking is automatic. So the engine learns your patterns and feeds them back to you, tighter and tighter, because certainty converts better than surprise. What you get is more of what you already consumed, never the thing that might have changed you. The whole apparatus is optimized to prevent exactly the friction that growth is made of.

The Feedback Loop Wearing a Discovery Costume

The cruel joke is that all of this markets itself as discovery. The "For You" page, the "Recommended," the "Because you watched." The language is the language of exploration, of a helpful guide introducing you to new things. But it isn't discovery. It's a feedback loop wearing a discovery costume.

Real discovery has a specific quality: it surprises you, sometimes against your own stated preferences. You wander into a section of the bookstore you never visit and something catches your eye. A radio DJ plays a song you'd never have chosen and it lodges in your brain for a decade. A friend hands you a film that sounds terrible and it rearranges something in you. None of those moments were personalized. They worked precisely because they weren't. They came from outside the closed circuit of your existing taste, which is the only place genuinely new things can come from.

The algorithm can't do that, structurally. It can only recommend based on what it already knows about you, which means every recommendation is a slightly modified echo of your past. It calls this "for you," but it would be more honest to call it "from you," because that's where it all originates. You're not being introduced to the world. You're being introduced, over and over, to yourself.

Serendipity Needs the Friction We Deleted

I wrote recently about how modern technology keeps adding layers of interface between us and the thing we're trying to do, how "smart" design so often makes the simple worse. This is the same disease in a different organ. In the name of removing friction, we deleted the friction that mattered.

Serendipity is a friction product. It requires effort, wandering, and the possibility of a wrong turn. The old ways of finding things were full of inefficiency: browsing shelves, flipping past stations, reading a physical newspaper front to back and stumbling into a section you'd never have sought out. That inefficiency wasn't a bug. It was the mechanism by which you encountered things outside your existing preferences, and in smoothing it all away we optimized ourselves into a corner. Frictionless feeds are wonderful at giving you what you want and terrible at giving you what you need, and the gap between those two is where most of a life's growth actually happens.

Identity Freezes When the Inputs Are Filtered

There's a deeper cost, and it's the one I keep circling back to. When every input you receive is filtered to match who you already are, your identity stops being a living thing and starts to calcify. You become the person the algorithm modeled, because the model, by feeding you back to yourself, quietly manufactures the very consistency it claimed to detect.

Eli Pariser called the early version of this the filter bubble, the way personalized results wall each of us into our own private information universe. But I think the bubble metaphor undersells it, because a bubble is something you're merely trapped inside. This is more active than that. The system doesn't just enclose your identity; it reinforces it, rewards it, and gradually convinces you that the narrow band it's shown you is the whole of who you are. A teenager whose feed is tuned to one aesthetic, one politics, one mood, doesn't just see a slice of the world. They start to become the slice, because they've been denied the raw material that would let them become anything else.

We used to worry that technology would make everyone the same. The stranger outcome is that it's making each of us relentlessly, recursively ourselves, sanding away the parts that don't fit the model until the reflection and the person are hard to tell apart.

Choosing the Window

None of this means the tools are useless or that you should throw your devices in a lake. I use these apps like everyone else. But I've started treating the algorithm as what it is, a mirror rather than a guide, and that small reframe changes how you handle it. You stop trusting the feed to expand you and you start deliberately seeking the inputs it will never offer.

That looks like ordinary, slightly inefficient things. Following people who annoy you a little. Reading outside your lane on purpose. Letting a knowledgeable human, a librarian, a friend, a good editor, point you somewhere the engine wouldn't. Keeping a few corners of your life unpersonalized, where chance can still reach you. The goal isn't to reject recommendation entirely. It's to refuse to let a system optimized for your attention quietly define the boundaries of who you get to be.

The algorithm will always show you the mirror, because the mirror is profitable and the window is not. But you were never meant to spend your whole life looking at your own reflection. The things that change you come from outside the frame, and reaching them is going to take the one thing every feed is designed to eliminate: a little friction, chosen on purpose.