Spotify shipped an assistant that can reach back through everything you’ve played and change the song that’s on right now. That’s the whole feature. You ask for something, it searches years of your listening, and it quietly rewrites the queue. On paper it’s a skip button with a memory — a small, well-made convenience, and it genuinely is convenient. Anyone who has fought a shuffle algorithm at 11pm knows the appeal of just saying the thing and having the room adjust.
The interesting part isn’t the button. It’s the sentence we’re going to use to describe it.
We’ll say the AI picked the song. The AI knew what you were in the mood for. The AI decided to swap the track. Every one of those sentences does the same quiet work: it hides the actual chain of events behind a single mysterious actor. A product team chose a default. An incentive chose the product team. A model trained on your own past behavior chose the most likely next thing. “The AI decided” is a costume thrown over a series of very ordinary human decisions, most of them made months ago by people you will never meet.
The costume is comfortable for a reason
Naming a machine as the one who “wants” or “decided” is convenient in a way that has nothing to do with accuracy. It lets everyone off the hook at the same moment. If the assistant surfaces the same three artists all week, that’s not a choice anyone made — it’s just what the AI does. If it nudges you toward the tracks the platform happens to pay less to stream, well, the algorithm is a black box, who can say. The grammar of “the AI wants” launders responsibility. It turns a set of authored decisions into weather.
There’s a cleaner way to talk about it, and it costs nothing. Say what actually happened: the system was tuned to do this. Someone set the objective. The model optimizes for it. That sentence is longer and less magical, and it points at something you can actually hold accountable. Mystery is the enemy of accountability — the more we let software be an inscrutable agent with moods, the less we ask who built the mood in.
I keep coming back to how much of the current AI conversation runs on this small grammatical trick. Not lies, exactly. Just a pronoun problem. We keep giving the machine a “wants” it doesn’t have, and in doing so we take the “chose” away from the people who do.
Behavior follows the room
There’s an older idea sitting underneath all this, and it has nothing to do with software. The broken-windows observation — that a cracked pane and some litter change how people behave in a space more than any lecture about character ever could — was really a claim about context. People aren’t fixed quantities acting out their nature. They respond, minute to minute, to the environment they’re standing in. Fix the window, sweep the platform, and behavior shifts without anyone’s personality changing at all.
A recommendation surface is an environment. That’s the thing the “AI decided” framing keeps us from seeing. The assistant doesn’t just answer your request — it shapes the room you make the request in. It decides what feels available, what surfaces first, what a reasonable next choice even looks like. You still feel like you’re choosing. You are, technically. But the menu was set before you walked up to it, and the menu is where the real power lives.
This is why the feature is more interesting than it looks. An assistant that can change your music on command is, underneath, an environment that has learned to rearrange itself around each person in real time. Not a fixed room you learn to navigate — a room that reshapes its walls based on what you did last Tuesday. That’s a genuine advance in what software can do. It’s also a genuine transfer of authorship, from you to whoever tuned the walls.
The honest question
None of this is cause for alarm, and I don’t think it’s a trap. A tool that finds the song you half-remember is a good tool. The point isn’t to be suspicious of the feature — it’s to be precise about it, because precision is the only thing that keeps the room honest.
So the question worth carrying isn’t “is the AI deciding for me.” The AI isn’t deciding anything; it doesn’t want your evenings to sound a particular way. The question is who set the objective the model is quietly optimizing, and whether that objective is yours or someone else’s dressed up as yours. When the queue bends, ask what it’s bending toward. When the assistant “knows what you want,” ask who defined want.
The comfortable version of this story is that we’re handing small decisions to a helpful machine. The truer version is that we’re moving into rooms that rearrange themselves around us, built by people who’d rather we thought the furniture moved on its own. The furniture never moves on its own. Someone always set the room — and the moment we stop saying so is the moment we stop being able to ask them why.
Leave a Reply