Market Commentary

The Taste Economy: what human judgement is worth when a machine can recommend anything

Recommendation engines now surface near-infinite, mostly competent options across news, music, retail and fashion. That abundance is quietly repricing the one thing machines still cannot manufacture: a trusted human call on which option is right. This note reads the category — who is rebuilding editorial judgement, who is capturing the value, and where the tension between algorithmic scale and human taste is heading.

Commissioned by · Karin Fischer · Cultural Anthropology

1. The scarcity has moved

For two decades, the operating assumption of digital commerce and media was that access was the bottleneck. Whoever solved discovery — surfaced the right item from an impossibly large catalogue — won. Recommendation systems solved that problem almost too well: Spotify, Amazon, TikTok and a hundred smaller engines now surface options that are, on average, competent. The bottleneck has not disappeared. It has moved one level up, from finding an option to trusting a choice.

When a machine can surface anything, the question stops being what's available and becomes whose judgement you trust to narrow it down.

That shift reframes an entire category of business. Firms that once competed on catalogue depth are starting to compete on the credibility of their selection — a market this note now maps.

2. From tracking clicks to modelling taste

The mechanics of recommendation did not stay still while this happened. Early systems tracked behaviour — what was clicked, watched, bought — and extrapolated forward. The more recent generation of systems model something closer to intent: not just what a person did, but what they are likely to want next, inferred from a denser signal of taste. That shift changed how discovery feels to the person on the receiving end, from a search they conduct to a state of being anticipated.

The newsroom case that anchors the next stage of this read sits at the far end of that timeline — a deliberate reintroduction of the human layer that the earlier stages had spent years automating away.

3. Where the value splits

Not every category is resolving this tension the same way. Two variables decide the shape of the response: how much the category's output can be evaluated objectively (a product spec versus a cultural judgement call), and how much trust in the selector matters to the person receiving it. Plotting the category against those two axes produces four distinct postures, each already visible in a different part of the market.

The newsroom's editorial rebuild sits squarely in verified selection — a category where trust in a named human has become a structural response to an algorithm-created trust problem, not a nostalgic feature.

4. What the newsroom case shows

One case is doing a disproportionate amount of the work in this category's reporting: a major newsroom that holds firm on never letting AI write its published articles, using the technology instead for draft summaries and metadata, with every published word passing through a human editor. Commentary on the case converges on a specific claim — that what survives automation is not general writing ability but original investigative reporting, source relationships built on risk, and judgement calls about what is too graphic to publish.

If verification becomes a saleable product line in news, the same logic — scarcity value rising with the volume of synthetic content around it — has an obvious route into every other category where a machine can generate options faster than a person can evaluate them.

5. Three ways to hear the argument

Ask a platform executive, an independent curator and a sceptical media economist to describe this shift and three different stories emerge, none of them wrong, all of them incomplete on their own. The disagreement is not really about facts — everyone points to the same abundance of algorithmic options. It is about who captures the value that scarcity of trust creates.

All three are describing the same market from different sides of the transaction — and the honest answer is that the category currently rewards whichever party can most convincingly claim the human signature, whether or not the judgement behind it is fully independent.

6. How consistently the pattern holds

The newsroom case is the most fully documented instance of this dynamic, but the underlying pattern — trust migrating toward a disclosed human choice as algorithmic volume rises — shows up with uneven strength elsewhere. Reading across the categories this note has touched, the signal is strongest where the cost of a wrong choice is highest and weakest where the choice is genuinely low-stakes.

That gradient is itself a useful map: it says the curation economy will not arrive everywhere at once, and it will arrive first wherever a wrong recommendation costs something more than mild disappointment.

7. The domains, compared

Zooming into four categories side by side sharpens the point further. Each one has its own answer to who currently holds the deciding vote — the platform's model, a named human, or some blend of the two — and its own account of what a human still adds that the model does not.

The pattern across all four is consistent even where the balance differs: wherever a category keeps a human in the loop, it is because the criteria for a good choice are not fully specifiable in advance — and that is precisely the condition a recommendation engine cannot resolve on its own.

8. What stays unresolved

None of this settles the underlying question the category raises, and it may not settle for some time. A curator's authority rests on the claim of an independent eye — but if that curator is themselves informed, assisted or partly generated by the same algorithmic layer they are meant to be a check on, the distinction the market is paying for starts to blur.

Where that line gets drawn, and by whom, will likely determine which curators retain pricing power and which quietly become another interface on top of the same underlying model.

9. Reading the category honestly

Put together, the evidence supports a specific and narrower claim than the broad idea that taste is having a moment. It supports the claim that verified, accountable human judgement is repricing upward in categories where a wrong choice is costly and the criteria for a right one are not fully specifiable — starting visibly in journalism, and plausibly extending outward wherever the same conditions hold.

The open question is not whether human judgement still matters — the evidence says it does, and increasingly at a price. It is whether the market can tell the difference between a genuinely independent eye and a well-marketed extension of the same machine it is priced against.

Grounded in Folka’s corpus: The Taste Broker · The Taste Graph · The Human Filter

How we read this

This read draws on newsroom-AI industry reporting (GeoBarta, Media Copilot, both January 2026) documenting how one major newsroom structures human editorial judgement against AI-assisted production, supplemented by Folka's own editorial synthesis of recommendation-system and creator-economy dynamics. The newsroom case is the strongest sourced evidence; the broader curation-economy and taste-graph reads are Folka's interpretive extension of that pattern into adjacent categories, not independently measured. The single figure that would most change this read: a reliable estimate of how much consumers already pay, in aggregate, for human-curated selection versus algorithmic discovery — no such number yet exists in citable form.