Sourcing the 5% Who Rewrote Their Job Around AI

Robert Half's 2027 Salary Guide, published October 1, 2026, says 72% of managers are raising pay for AI skills and 42% now rate AI expertise above every other technical skill. Four months earlier, the Google/Ipsos "AI Works for America" poll pegged the share of workers who are actually AI-fluent at 5%. Both numbers are true, which is why sourcers in marketing, finance, legal, and ops are suddenly fishing in a pond 1/20th the size they thought.

The gap between "uses ChatGPT" and "AI-fluent" is 35 points

Google and Ipsos define an AI-fluent worker as someone who has redesigned their workflow around AI and uses it weekly across eight or more use cases. Only 5% of workers clear that bar. Another 35% are "AI Explorers" who tinker ad hoc, and the remaining 60% don't use it at work at all.

That gap matters because Robert Half's premium only attaches to the 5%. The 35% can list "Generative AI" on a resume all day; they will not have redesigned a close process, a legal review queue, or a growth loop. The sourcing problem is now a filtering problem: how do you separate the 5 from the 35 before a hiring manager wastes a loop.

• AI-fluent (5%): redesigns workflows, 8+ weekly use cases, usually self-taught, can show a diff between first-draft AI output and shipped artifact.
• AI Explorer (35%): uses ChatGPT a few times a week, cannot name three tools beyond it, lists "AI" as a skill with no artifact.
• Non-user (60%): irrelevant to this search.

The fluent cohort is 4.5x more likely to report higher wages and 4x more likely to report an AI-attributed promotion, per the same Google/Ipsos study. They rarely sit in open-to-work queues. You have to go find them.

5%

51% of AI postings are now outside IT, and the supply didn't follow

Lightcast's 2025 data shows 51% of AI-skill job postings sit outside IT and computer science, up from 39% in 2019. Postings requiring generative AI skills in non-IT roles are up 9x since 2022. The candidate supply did not 9x. That is the entire pricing story.

The surge is concentrated in four function families:

Robert Half quantifies what the squeeze does to specific titles: newly hired financial analysts get a 3.8% pay bump, attorneys with four to nine years get 3.9%, marketing automation specialists get 3.9%, and executive assistants get 3.9%. Staffing Industry Analysts puts the broader non-tech premium at 28%, roughly USD 18,000 a year, with demand for AI skills in non-tech industries up 800% since 2022.

Cole Napper, VP of Research at Lightcast, framed the mechanism bluntly: companies still treating AI as a niche technical skill will compete for talent with firms that have embedded AI literacy across the whole workforce. The second group wins.

What the fluent 5% actually looks like in Refolk's index

Filtering Refolk's index of US non-engineering profiles for real generative AI skills returns a shockingly small pool. Three cuts across marketing, finance, and legal/ops produce roughly 522 candidates nationwide. That is the whole supply, before you have even filtered by location or seniority.

Two things jump out. First, legal/ops is the biggest cluster, which contradicts the usual assumption that marketers lead AI adoption. Second, 522 profiles across three of the hottest function families is a tiny number against tens of millions of US knowledge workers. That is the mechanical reason 72% of managers are raising pay: there is nobody to hire.

522

Chief of Staff is the single best non-engineering fluency signal

The strongest non-obvious finding in Refolk's index: in a 25-profile sample of the legal/ops cohort, 12 held the title Chief of Staff. That is 48%, from a title that represents a tiny fraction of the labor force overall.

The mechanism is clean. Google's definition of fluency, redesigning workflows around AI across 8+ use cases, is literally the Chief of Staff job description. CoS sits next to the CEO, inherits the half-formed processes, and ships the rebuild. If you are sourcing for AI-fluent operators and you are not pulling CoS profiles, you are missing the richest vein in the market.

Fluency is a job description before it is a skill. Chief of Staff fits the definition more cleanly than any title with AI in it.

How to vet AI skills on a resume without being fooled

Ignore the skills section, read the bullets for named tools and workflow diffs, and ask for the shipped artifact. The Ipsos/Groundwork study from August 2026 found 44% of AI users describe their own output as "AI slop," and that users spend four of the six hours AI saves them correcting its output. That is both the problem and the vetting opportunity. The fluent 5% can show the diff between first-draft output and finished work. The 35% cannot.

Six signals that separate the fluent from the explorers:

1. Named-tool density. A MarTech review of 60 marketing postings found 36 named at least one AI product, with Claude in 26 and ChatGPT in 19. Resumes should match that specificity: Claude Projects, Zapier Agents, n8n, Cursor, v0, Granola, Clay. One generic "GenAI" is a red flag; four named tools with a sentence each is the shape you want.
2. Workflow redesign bullets. "Cut monthly close from 11 to 4 days using Claude and an internal RAG tool over our ledger" beats "Leveraged AI for finance workflows." The verb matters.
3. Portfolio artifact. A Loom walkthrough, a Notion build log, a published n8n workflow, a GitHub repo of prompt chains. The fluent 5% have one because they built it for themselves, not for a job.
4. Community receipts. Maven course cohorts, Build Club, Latent Space Discord, Lenny's Community, Every's AI Supremacy group, AI Tinkerers meetups, r/LocalLLaMA. Only 14% of workers have been offered AI training by their employer in the last 12 months, so almost all fluent candidates are self-taught through these venues.
5. A diff story. Ask for the first-draft AI output and the shipped version side by side. Explorers ship the first draft. Fluent operators have the correction trail.
6. Legacy-enterprise employers. In Refolk's index, the companies with the most AI-fluent non-engineering profiles are Accenture, Mastercard, Microsoft, Amazon, Google, Uber, Deloitte (Monitor), and Comcast. Not AI labs. The fluency is being built inside big companies trying not to die.

This is the exact filtering job that Refolk is built for: describe the person in plain English, including named tools and workflow-redesign verbs, and Refolk returns a ranked shortlist across GitHub, LinkedIn, and the open web instead of a keyword-stuffed Boolean dump.

Where to actually hunt the fluent 5%

Maven course rosters, Build Club project pages, Latent Space Discord, Lenny's Community, Every's AI Supremacy group, personal Substacks, and "Show HN" or "I built" posts on Twitter. These are the venues where self-taught fluency gets documented in public, which is the only way to pre-filter around the "AI slop" problem.

Three practical hunting patterns:

• Cohort scraping. Maven cohorts, Every's operator courses, and Build Club's quarterly builds each produce a public graduate list or project showcase. Pull the names, cross-reference the current title on LinkedIn, filter for the four function families, and you have a shortlist with provenance stronger than any skills tag.
• Tool-user communities. Claude community activity is now the highest-signal watering hole, given Claude appeared in 26 of 60 marketing postings versus ChatGPT at 19. The Zapier community and the n8n forum power-users round out the Named-Tool Density signal.
• Legacy-enterprise poaching. The Accenture, Mastercard, Deloitte, Comcast cluster from Refolk's index is counter-intuitive but logical: these firms have budget for formal AI programs, so they produce fluent operators the fastest. A targeted pull of "Senior Manager" and "Chief of Staff" titles out of those eight companies is a quarter's worth of pipeline.

What this means for your 2027 reqs

Price every non-engineering req that lists "AI skills" as if you are hiring from a pool of 522 people nationally, because for the fluent cohort you effectively are. Build the sourcing plan around named tools and workflow-redesign artifacts, not around the word "AI." Prioritize Chief of Staff, FP&A, SEO lead, and senior paralegal as the four titles where the fluency premium is most asymmetric. Treat the self-taught cohort as the baseline, because only 14% of workers have been offered employer AI training in the last year.

The hiring managers who still think they are competing in a normal talent market are the ones who will spend six months on a req and lose the finalist to a 15% counter. Dawn Fay at Robert Half said the premium is for professionals who understand where AI can add value and apply it responsibly. In Refolk's index, that profile is 522 people. Act accordingly.

FAQ

How do I write a Boolean string for AI-fluent non-engineering candidates?

Stop writing Boolean strings for this. The signals are too compositional: named tools, workflow verbs, community receipts, and employer provenance don't collapse into AND/OR cleanly. Use a plain-English sourcing tool, or build a two-stage pipeline where stage one pulls title and function and stage two scores named-tool density and portfolio artifacts separately. Pure Boolean will give you the 35% explorer cohort padded with keyword-stuffers, not the fluent 5%.

Is a Coursera or Google AI certificate a positive signal?

Weakly positive at best. Only 14% of workers have received employer-sponsored AI training in the last year, and the fluent cohort is overwhelmingly self-taught through Maven cohorts, Build Club, and community Discords. A certificate without a shipped artifact is noise. A certificate alongside a public n8n workflow, a Loom demo, or a Substack post about redesigning a close process is a yes.

Why is Chief of Staff a better signal than "AI Specialist"?

Because the Chief of Staff job description already matches Google's definition of AI fluency: redesigning workflows across many use cases next to a CEO. "AI Specialist" is often a rebranded ops or analyst role that signals interest, not fluency. In Refolk's index sample of the legal/ops GenAI cohort, 12 of 25 profiles were Chiefs of Staff, far above the title's baseline rate in the labor force.

What premium should I actually budget for an AI-fluent hire in 2027?

Robert Half's guide gives you specific anchors: 3.8% to 3.9% for newly hired financial analysts, mid-career attorneys, marketing automation specialists, and executive assistants. Staffing Industry Analysts puts the broader non-tech premium at 28%, roughly USD 18,000, when "AI skills" are listed in the posting. Treat the 28% as the ceiling for an explorer who stuck the keyword on their resume, and the Robert Half title-specific numbers plus a competitive counter as the floor for the actually fluent 5%.