← All work signals
Skills signal

The Model Whisperer

A new craft emerges — coaxing exactly what you want from a stubborn machine, part interrogator, part poet.

What is it?

Getting a model to do what you actually want — asking well, correcting mid-stream, pushing back — is becoming a distinct, learnable craft, not a fluke. With generation now cheap and common, the edge shifts to whoever directs it best. Whether that edge survives easier models is the open part.

Why now?

Cognitive output was once scarce, so who asked mattered less than who could supply it at all. Once supply stopped being the bottleneck, the scarce thing became judgement — spotting a blind spot, catching a wrong turn early. Abundance didn’t remove the gap; it relocated it, from making to aiming.

Who’s building it?

Frequent, attentive users build the craft through repetition; course-makers and employers are the ones naming and pricing it.

Power usersPrompt-course makersEnterprise enablement teamsModel providersHiring managers
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 — power users, and the trainers formalising what they do.
Buyers — employers hiring for the skill, and workers racing to acquire it.
Gatekeepers — model providers, who could design the need for the skill away entirely.
What’s at stake
The providers selling the raw capability also control how much steering skill it still requires to use it well.
A credentialing grab — whoever names it first gets to certify.
Wage premiums for a skill nobody agrees how to measure.
What’s unsettled
Whether direction skill is durable, or a training-wheel phase models design away.
Whether it transfers across models, or must be relearned each time.
Who owns the credential — a course, an employer, or the model provider itself.
02 · How the demand matureda progression, not a timeline
Curiosity
Early adopters notice their outputs beat a colleague’s for no obvious reason.
Imitation
The gap gets named — prompting, steering, direction — and others start deliberately copying it.
Formalisation
Courses, playbooks and internal wikis turn tacit habit into something that can be taught and sold.
Hiring signal
Job listings and internal reviews start naming the competency explicitly, and paying for it.
03 · Where the market goesthe forks, not a prediction
→A durable craft, or a training-wheel phase — decided by whether models keep needing a skilled human to aim them, or learn intent unaided.
→Providers could absorb the skill into the product, auto-correcting the steering that used to be scarce.
→A credentialing industry could form before anyone agrees what this competency actually measures, or who owns it.
→The premium may sit with individuals, or get captured by whoever certifies the skill.
Everything here turns on one open question: does directing a model stay a human skill, or does the model eventually learn to read the room itself.
04 · The Taste Economy deskadjacent markets, moving now
The Skills Premium
Whole skill sets are being rewritten
39% of workers’ core skills seen changing by 2030.
World Economic Forum
The Human API
People still supply the judgement machines cannot fake
Meta took a $14B stake in Scale AI (2025).
CNBC
Prompt Literacy
Employers now hire and pay for this
AI-skill job listings jumped from 0.87% to 5.7% (2025).
Forbes
05 · The Wire
01
Trained prompters gained 19 points vs 1 for untrained users.
arXiv (RCT study) ↗
02
Requirement/prompt quality strongly predicts LLM output quality.
arXiv (RCT study) ↗
03
Long-tenured Claude users show 4pp higher success rate.
Anthropic Economic Index ↗
04
Managing AI agents is not the usual focus of employer AI training
HR Dive ↗
The panel on this signal
They agreed on
–The target keeps moving from wording prompts to exercising judgement, so it survives each round of model improvement.
–How unevenly the skill is distributed is the real stratifier.
–The differentiator now sits above raw generation, at the level of direction.
The labour economist disagreed on
–Wage premiums tied to a specific interface skill compress fast once vendors simplify prompting into templates; comparable craft roles have rarely retained pay advantages beyond the tooling’s maturity curve.
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.