Work shift

The Human API

People plug their verified judgment into the machine pipeline, sold by the decision and summoned on demand.

What is it?

Behind every capable model sits a layer of human judgement — labelling, correcting, rating outputs a machine can’t yet score itself. As models chase the frontier, that judgement gets scarcer, more expert, priced like any input. Whether it prices by the decision, or the hour, stays open.

Why now?

Cognitive abundance made machine output cheap, but not trustworthy alone — every frontier gain still needs a person to grade, correct or approve it first. That need scales with capability, not against it: the better models get, the more expert the judgement required. Scarcity moves from compute to people.

Who’s building it?

Data-labelling platforms and model labs, buying human judgement wholesale from workers who rarely appear in the product.

Meta · Scale AIFrontier model labsLabelling platformsDomain expert annotatorsLabour regulators

The full read

Why now, who is building it, where the market goes, and what stays unsettled.

Get weekly →

01 · The market

Who’s in it

Builders — labelling platforms, and the expert networks that supply them.

Buyers — model labs and enterprises buying reliability they cannot build in-house yet.

Gatekeepers — platforms aggregating supply, and regulators shaping how the labour gets classified.

What’s at stake

The same expertise that makes judgement valuable also makes it scarce — demand for it grows faster than supply can train.

Wholesale deals like Meta’s concentrate access to reliable supply.

Workers stay invisible even as their judgement becomes essential.

What’s unsettled

Whether judgement gets priced by the decision, or by employment.

Who captures the value — platforms, labs, or the experts themselves.

Whether expert supply can scale with demand, or becomes the real bottleneck.

02 · How the demand matureda progression, not a timeline

Bulk labelling

Generalist raters correct and rank machine outputs at volume, cheaply.

Specialisation

Frontier gaps start needing domain experts, not crowds — cost rises fast.

Pricing shift

Judgement gets priced by the task or the decision, not the hour, as demand concentrates.

Strategic asset

Rights-cleared, verified human input becomes a defensible input companies compete to secure.

03 · Where the market goesthe forks, not a prediction

→Concentrate or fragment — a few aggregators like Scale AI absorb the supply, or a long tail of expert networks competes for it.

→Regulation over gig-style data labour could reclassify the workers before the market settles on a structure.

→Domain-specific judgement — medical, legal, code — may command a real premium over generalist labelling work.

→Model labs may build this capability in-house rather than keep buying it wholesale.

Everything here turns on scarcity: whether expert human judgement stays abundant enough to buy wholesale, or becomes the binding constraint on model progress.

04 · The Taste Economy deskadjacent markets, moving now

Borrowed Brilliance

Authors’ work gets priced after the fact.

Anthropic’s $1.5B author settlement paid $3,000 across 500,000 works.

TechCrunch

The Skills Premium

Whose skills still count, as machines take over.

39% of workers’ core skills seen changing by 2030.

World Economic Forum

The Model Whisperer

Trained prompting skill becomes its own real value.

Trained prompters gained 19 points vs 1 for untrained users.

arXiv (RCT study)

05 · The Wire

Meta was finalising a $14 billion investment in Scale AI

CNBC ↗

AI labs shift human-data contracts after Meta’s Scale stake

Forbes ↗

Mercor’s chief executive says annualized revenue crossed $2 billion

TechCrunch ↗

The panel on this signal

They agreed on

–Human judgement is a scarce, essential input to frontier models, and that scarcity only grows with capability.

–Value concentrates wherever reliable expert judgement gets aggregated.

–The workers supplying that judgement stay largely invisible to the people who benefit.

The labour economist disagreed on

–Current arrangements resemble data-annotation and outsourced-judgment contracting reframed with API terminology; whether a genuine per-decision marketplace with price discovery exists, rather than conventional labour outsourcing under new branding, has not been demonstrated.

Know more about the method →

✦How this signal was built

01 · Panel

A Delphi panel of 4 AI analysts, each reasoning independently.

→

02 · Debate

They challenge and revise each other’s reads over two rounds.

→

03 · Evidence

Every claim is tied to a cited, linked source on The Wire.

→

04 · Synthesis

A moderator writes it into one read, disagreement and all.