A friend recommends a book. The friend is not a professional critic. The friend is not a public figure. The friend is a person whose taste, over the course of years, has proved reliable. The recommendation arrives in a single sentence. It cuts through three months of algorithmic suggestions, three weeks of public discussions about what to read next, and the open browser tab where another book is currently being reviewed by a thousand voices that are not the friend’s voice. The reader buys the book that evening. The book turns out to be one of the better books they will read that year.
The economy of taste, which several years ago was widely assumed to have been displaced by algorithmic recommendation, has not collapsed. It has migrated. The places where taste now lives, and the people who now carry it, are different from the places and people of the previous era. The value of taste, against most observers’ expectations, has gone up rather than down.
##### Why the algorithms did not win
The algorithms did not win for reasons that, in retrospect, were predictable.
The algorithms are good at producing more of what the audience has already responded to. They are not good at producing what the audience would respond to but has not yet encountered. The two functions are different. The first function is satisfying. The second function is what taste was always for.
The algorithms also operate at a scale that flattens the texture of the recommendations they produce. The reader who has been served algorithmic recommendations for any length of time eventually notices that the recommendations begin to feel similar to each other, even when the individual items are different. The flattening produces a kind of fatigue that algorithms have not, so far, been able to design around.
The algorithms also have no skin in the recommendation. The friend who recommends a book is staking some part of their friendship on the recommendation. The algorithm is not staking anything. The reader can tell, on close attention, the difference between recommendations that are vouched for and recommendations that are produced. The vouching turns out to matter.
##### Where taste now lives
Taste now lives in several places, none of them the public surfaces that the previous era assumed would be its home.
It lives in private group conversations among people who trust each other. The text thread of three friends who read in the same category. The Slack channel of a small group of professionals who exchange recommendations. The dinner among colleagues where one of them mentions a film. These conversations have, in many categories, become the most useful sources of recommendation that exist.
It lives in newsletter writers whose taste the reader has come to know over time. A newsletter that arrives weekly, written by a person whose voice the reader has read for a year, accumulates into a relationship in which the reader trusts the writer’s recommendations the way they would trust a friend’s. The newsletter, over time, becomes a taste source whose value to the reader exceeds what the public surfaces can provide.
It lives in small physical institutions whose curation the audience has chosen to trust. The bookstore whose front table is curated by a person whose choices the reader has learned to follow. The cinema whose schedule reflects a programmer’s actual sensibility. The record store whose staff picks are read carefully. Each of these is a small operation. Each of them, in many cities, is doing better, in real terms, than they were five years ago, against the prediction that they would not survive.
It lives in podcasts and long-form interview formats where a host’s taste is the reason the audience listens. The host who recommends a book on their show is making a recommendation the listener can evaluate against the host’s other recommendations. The accumulation produces a relationship the algorithmic surfaces cannot replicate.
##### Who now carries it
The carriers of taste in the new conditions are, in many cases, not the carriers of taste in the previous era.
The major critics still exist. They still have their audiences. The audiences are smaller and more specific than they used to be. A reader who follows a major critic does so deliberately, often in addition to other taste sources, rather than as the default that the previous era assumed.
The new carriers tend to be specific people whose work the audience has chosen to follow over time. A writer with a few thousand subscribers. A bookstore owner whose recommendations show up in their monthly newsletter. A film programmer whose schedule the audience reads. A music writer who runs a small site or podcast. Each of them carries a fraction of the audience that a major critic of the previous era held. Cumulatively, across many such carriers, the audience for personal taste recommendations is large and growing.
The audience also carries taste, increasingly, in the form of word of mouth. The friend recommendation. The colleague mention. The book passed across a dinner table. These have always existed. They are now, in many categories, the most useful taste mechanism the audience has access to.
##### What this changes for creators
For creators of any work that depends on being found by the right audience, the working position is that the algorithmic distribution has lost some of the priority it held five years ago, and the personal-taste distribution has gained some.
The creator whose work has been picked up by a small newsletter writer with a loyal audience is, in many categories, in a better position than the creator whose work has been featured by a much larger algorithmic surface. The first audience will read the work carefully. The second will mostly scroll past it.
The implication is not that creators should ignore the algorithmic surfaces. They are still useful for some forms of distribution. The implication is that the creator who has cultivated relationships with specific personal-taste carriers in their category has built a distribution asset that the creator who has only cultivated platform presence does not have.
##### What this changes for audiences
For audiences, the implication is that the work of finding what is actually worth attention has shifted from the algorithmic surfaces to the human ones.
A reader who wants to find good books is, in our reading, better served by identifying a few specific people whose taste has proven reliable and following their recommendations than by scrolling any of the larger surfaces designed for the same purpose. The same is true across categories. Good films, good music, good restaurants, good writing in any genre, are now reached most effectively through the small set of carriers the audience has come to trust.
The work of identifying those carriers is itself a form of effort. The effort is not large. The return on it is significant. A reader who has assembled a small taste committee whose recommendations they read carefully tends to read better books, watch better films, and discover better work, than a reader who has not.
##### The honest framing
The honest framing is that human taste did not lose its standing when algorithmic abundance arrived. It gained ground, because the abundance produced a new shortage. The shortage is the judgment that decides what is actually worth attention. The audience that has solved the shortage by relying on specific human carriers of taste is, in our observation, getting better recommendations than the audience that has not.
The friend recommended a book. The reader read it. The reader, in turn, recommended it to two other people. None of the algorithms had any role in this. The economy of taste has, in many categories, returned to a shape it held for a long time, and the shape is doing well in conditions that everyone assumed would have replaced it.
