Taste, machines, and the people who get to decide.

Taste used to be slow. You read what your library had, you watched what the rep theater played that month, you ate at the restaurant down the block, and over years a sensibility settled in — a sediment of what kept working. It was personal because it was constrained by where you lived and who you knew.

Then taste got faster. Streaming, infinite scrolls, recommendation engines. The constraint dissolved, and a new one took its place: whatever the model thinks you'd like next. The library is everything; the rep theater is the algorithm; the restaurant down the block is whatever keeps appearing on your homepage. The sediment doesn't form the same way.

I don't think this is automatically a problem. Tools that surface things I would never have found on my own are some of the best gifts technology gives us. The interesting question isn't whether the algorithm picks for us. The interesting question is who decides what the algorithm picks from, and what it picks for.

The fear is monoculture: a world where the same dozen things get fed to everyone, recursively, until the things that don't fit the pattern stop being made because they don't get surfaced. That's a real risk, and we've watched a version of it happen — in music, in cinema, in the visual language of the internet. The aesthetic of the AI image generator becomes the aesthetic of the brief becomes the aesthetic of the next AI image generator.

But the response can't just be reactive. "Don't use the tools" is not a taste position; it's a refusal. The position I find more interesting is intentional — using the tools, knowing what they flatten, and choosing to build things that widen rather than narrow.

The hardest part of taste, the part nobody can outsource, is the work of disagreeing with yourself. Of noticing what you've come to assume is good and asking whether you actually believe it or whether you've been served it three thousand times. That's a subjective discipline. It's slow. It can't be automated, because the automation is the thing it has to argue with.

So when I build with AI, I think a lot about whether the tool I am making is going to argue with someone or finish their sentences. A tool that finishes your sentences is friendly. A tool that argues with you is generative. I want more of the second.

Subjectivity is the antidote, not the problem. The risk of AI in taste is not that it has opinions — it is that the opinions get washed out into a beige mean. The defense is not to refuse opinions; it is to keep yours sharp. To read poetry the model cannot summarize. To notice the movie the algorithm did not recommend. To insist, without apology, on the thing you actually love.

Who gets to decide what's good? Not the model, and not the discourse. Each of us, slowly, in private, sediment by sediment, with help where it serves and resistance where it doesn't.

The most useful AI tool is the one that makes you more yourself, not less.