The AI trainer gig economy just split into two tiers.One side just raised money like a frontier lab. The other side's queues are going empty.
Mercor disclosed annualized revenue past $2 billion this summer and is reportedly in talks to raise at a $20 billion valuation; bootstrapped rival Surge AI is reportedly fielding term sheets near $25 billion. Meanwhile generalist AI-rating queues are thinning out and rates are sliding batch to batch. The gap between the two tiers is now roughly 10x for work that looks similar from the outside — and the ladder up runs through a credential most knowledge workers already hold, not a new AI skill.
No hypeOne gig-economy split, priced in real dollars on both sides
Why nowThe credentialed tier just became a multi-billion-dollar business — while the entry tier is visibly shrinking
If you read nothing else
The "get paid to train AI" gig economy isn't one market anymore — it's two, and they're moving in opposite directions. Mercor, the marketplace connecting credentialed experts to frontier labs like OpenAI, disclosed this summer that its annualized revenue has passed $2 billion, with more than 30,000 verified experts earning over $4 million a day between them, and it's reportedly in talks to raise $500 million at a $20 billion valuation — up from a $10 billion Series C just nine months earlier. Bootstrapped rival Surge AI, run by roughly 110 people with no sales team, reportedly generated $1 billion-plus in 2024 revenue and is said to be fielding its first outside investment near a $25 billion valuation. That's the top tier: credentialed physicians, lawyers, CPAs, and senior engineers reviewing AI outputs in their own field for $75–300+/hr. The bottom tier — generalist raters on platforms like Outlier, DataAnnotation, and Remotasks doing basic response evaluation for $12–30/hr — is showing real strain: task queues going empty, project rates swinging from $35/hr one month to $22/hr the next, and at least one 2026 platform tracker downgrading a formerly top-rated platform specifically over pay instability and thin queues. The mechanism is straightforward: frontier labs training reasoning models need harder, domain-specific judgment calls that only real experts can make, while the easy annotation work generalists have relied on is both oversupplied and increasingly automatable. The ladder up isn't "get better at prompting" — it's bringing a credential you likely already have.
🎙️ Nova's Signal
Nova is FOWL AI's news anchor — she tracks the signals every week.
"When a company most people have never heard of is disclosing $2 billion in annualized revenue and getting valued like a foundation model lab, that's not a gig-economy side hustle anymore. That's a real industry, and it has a floor and a ceiling now — and they're moving apart, not together."
Nova's call: Stop thinking of "AI training work" as one gig. It's two different labor markets sharing a login page — and which one you're in depends entirely on whether you're bringing a credential or just an internet connection.
In Issue 14 we flagged, as one story among five, that domain-expert AI evaluation was paying 6–10x what generalist annotation pays. This week that split is the whole story — because the two sides of it just moved in opposite directions hard enough to be a story on their own.
On the expert side: Mercor disclosed this summer that its annualized revenue has crossed $2 billion, built on more than 30,000 verified professionals earning a combined $4 million-plus per day, and the company is reportedly in talks to raise at a $20 billion valuation barely nine months after its last round priced it at $10 billion. Surge AI — bootstrapped, ~110 employees, no sales team — reportedly pulled in over $1 billion in 2024 revenue alone and is said to be in talks for its first outside round near $25 billion. Handshake, Alignerr, and Turing are running the same playbook: recruit verified domain experts, place them on named projects for frontier labs, and price the work like scarce expertise rather than volume labor.
On the generalist side, the picture is close to the opposite: 2026 platform trackers describe task queues going empty, project rates swinging from $35/hr one month to $22/hr the next on the same platform, and at least one tracker downgrading a formerly top-rated platform specifically over pay instability. It's the clearest, most literal, most dollar-denominated version of a split we keep finding across the AI labor market — and this week it's worth walking through in full: why it's happening now, exactly what it pays on both sides, and the specific, credential-based way to get on the right side of it.
This Week — 5 Developments
01
The Expert Tier Is Now a Venture-Scale Business
One sentence: Mercor, the marketplace that connects credentialed professionals with frontier AI labs for evaluation and training work, disclosed in June 2026 that its annualized revenue has surpassed $2 billion — with more than 30,000 verified experts on the platform earning a combined $4 million or more per day — and is reportedly in talks to raise $500 million at a $20 billion valuation, up from the $10 billion price tag on its own Series C just nine months earlier.
Mercor isn't alone at that scale. Surge AI, bootstrapped since 2020 by founder Edwin Chen with a lean team of roughly 110 people and famously no sales staff, reportedly generated over $1 billion in revenue in 2024 focused specifically on high-quality RLHF and expert-preference data; Bloomberg reported it's now fielding its first outside investment talks at a valuation north of $25 billion. Handshake, Alignerr, and Turing round out a field of marketplaces all competing for the same scarce pool of credentialed reviewers.
Why it matters · This has stopped being a gig-economy curiosity — it's now one of the fastest-growing, highest-paying employers of specialized professional judgment in the entire AI economy, and it's actively recruiting outside of tech.
The signal — Reporting on Mercor's revenue mix puts roughly 90% of it as coming from a small number of frontier labs, OpenAI prominent among them — which means this market's growth is directly tied to how much harder labs' own evaluation needs are getting, not to gig-economy demand in general.
02
The Pay Gap, in Dollars: Roughly 10x for Adjacent Work
One sentence: Generalist AI-trainer work — rating chatbot responses, basic annotation — pays roughly $12–30/hr across major platforms, while credentialed domain-expert review (a licensed physician, attorney, CPA, or senior engineer assessing outputs in their own field) runs $75–300+/hr, with the most senior reward-model-design and complex-reasoning evaluation work commanding $85–200+/hr even before the rarer $300+/hr assignments.
Platform-by-platform, the range holds: DataAnnotation pays roughly $20–25/hr for generalist work and $40–45/hr for coding tasks; Outlier runs $15–30/hr generalist and $25–45/hr coding; Scale AI's specialized contributor work reaches $20–75+/hr. The expert marketplaces (Mercor, Surge AI, Handshake, Alignerr, Turing) sit well above all of that because, as one 2026 market report put it, they're "selling scarce expertise, not volume labour."
Why it matters · The credential is doing the work that "getting good at prompting" used to be pitched as doing — years of professional experience completely outside AI now converts directly and immediately into the top of this pay ladder.
The signal — The gap is entirely explained by existing knowledge, not AI skill: a doctor, lawyer, or senior engineer doing this work reportedly earns on the order of 10x what a beginner earns, for tasks that look similar from the outside.
03
The Generalist Floor Is Visibly Cracking
One sentence: While the expert tier scales into the billions, 2026 trackers describe real strain lower down the ladder — task queues going empty, project rates swinging from $35/hr one month to $22/hr the next batch on the same platform, and at least one tracker downgrading a formerly top-rated generalist platform specifically over empty queues and pay instability.
Two forces are compounding here. First, oversupply: generalist annotation work is crowded, with a large and growing pool of workers chasing the same task queues. Second, automation: the easy, low-judgment annotation tasks generalists rely on are exactly the tasks current models are getting good enough to do without a human rater at all — while the harder, domain-specific judgment calls driving the expert-tier boom in story 01 are, for now, the opposite: too hard for a model to self-check.
Why it matters · Anyone treating a generalist AI-training gig as dependable income should read 2026's data as a clear warning: platform trackers now explicitly advise treating this work as variable side income, not a guaranteed wage.
Flag — This is the "professionalised vs. democratised" split we covered at the whole-economy level in Issue 14, showing up here inside a single job category — same login page, same-looking task, opposite trajectory depending on whether judgment or volume is being priced.
04
The Ladder Up Is Specialization, Not More Hours
One sentence: RLHF (reinforcement learning from human feedback) is now the single largest category of AI training work in 2026, and the highest-paid listings inside it are narrow, verified specialty tracks — one active listing type is literally titled "Conversational Preference Reward Model Evaluator" — where coders assess AI-generated code, physicians assess medical reasoning, lawyers assess legal argumentation, and PhDs assess mathematical proofs, each priced on its own scale rather than one flat generalist rate.
Mercor's most-paid listings sit specifically in this specialist RLHF tier, reportedly running $100–200+/hr. The practical shift this represents: "which platform should I sign up for" matters far less than "which existing professional skill can I get verified on that platform" — these marketplaces are built around narrow, credential-gated specialties, not a single rate card everyone competes on.
Why it matters · This reframes the advice from "learn AI tools" to "your existing non-AI expertise is now a directly listable, separately priced skill in the AI economy" — a far lower bar for most knowledge workers than retraining into something AI-native.
The signal — The practical gate is verification, not application volume: platforms confirming a license number, employer email, or portfolio before unlocking the higher rate is exactly what's kept this tier from flooding the way generalist queues already have.
05
Full-Time and Contract Job Postings Are Confirming the Same Split
One sentence: The split shows up outside the gig platforms too: as of late August 2026, remote "AI Evaluator" roles average $74,153/year (most between $50,400–$90,000), contract AI-evaluator roles average $65,471/year, and postings are increasingly separating a generic "AI Evaluator" title from named domain tracks — Healthcare & Social Assistance Specialist, Physics and Computational Chemistry Expert, Legal Domain Expert — each recruited and priced separately.
That's a meaningful shift from even a year ago, when most listings in this space used one undifferentiated "AI trainer" or "data annotator" title regardless of the actual task. Employers and job boards appear to be formalizing the same split the gig platforms are pricing, once it became economically obvious that a generic listing couldn't attract the specific expertise the highest-value work requires.
Why it matters · A job search in this space shouldn't stop at "AI Evaluator" anymore — pairing it with your actual field ("AI Evaluator + Healthcare," "AI Evaluator + Legal") is now the search the market itself is organized around.
The signal — Watch for this title to keep standardizing over the next few issues, the same consolidation pattern we've been tracking with "AI Orchestrator" (Issue 14) — fragmented job titles pricing the same underlying work tend not to stay fragmented once enough postings exist to sort against.
📡 Signal & Chatter
What the data — and the platforms themselves — are saying this week.
›
Mercor's own disclosure puts more than 30,000 verified experts on the platform, collectively earning over $4 million per day — a scale that puts it among the largest single distributors of expert-priced freelance income in the current AI economy.
›
Surge AI's reported 2024 revenue of $1B+ came from roughly 110 employees and no sales team — an unusually lean structure for a company reportedly in talks near a $25 billion valuation, underscoring how much of this market's value sits in the expert network itself rather than headcount.
›
RLHF is now described as the single largest category of AI training work in 2026, ahead of basic data labeling and generic response rating, per industry trackers following the space.
›
ZipRecruiter data shows contract "AI Evaluator" postings clustering at $44,000–$94,000/year, with remote listings in some metro markets running as high as $50,000–$102,000 — a real full-time-adjacent income band sitting above the per-hour gig rates for the same underlying skill.
›
Platform tier-list trackers now explicitly separate "generalist" from "specialist" tiers, noting specialists in software development, mathematics, physics, medicine, law, and chemistry/biology command 2–4x generalist rates on the same platform.
›
One 2026 platform tracker downgraded a previously top-rated generalist platform to its "B-tier," citing empty task queues and pay instability directly — a concrete instance of the generalist-side strain described in story 03.
👥 Reddit / Community Chatter
What workers doing this work are actually saying.
›
Generalist raters describe rate instability as the norm, not the exception — a project paying $35/hr one batch coming back at $22/hr the next is a commonly reported pattern, with several describing budgeting around this work as unrealistic month to month.
›
A recurring complaint: recruiting ads for "AI training gigs" lead with the $100–300/hr headline rate that's actually reserved for credentialed specialists on expert marketplaces, while the generalist work most applicants actually qualify for pays a fraction of that — several describe feeling misled by the framing.
›
Credentialed applicants report real friction getting into the expert tier despite qualifying on paper — opaque vetting processes, long silence after applying, and rejection without explanation are common threads, even among people with verifiable licenses or degrees.
›
Workers comparing Outlier and DataAnnotation head-to-head in 2026 generally steer generalists toward DataAnnotation for consistency and specialists (coders, mathematicians, rare-language speakers) toward Outlier for higher ceilings — echoing the platform-tier data in Signal & Chatter above.
🔮 FOWL Prediction #15
"Within two to three quarters, expect at least one major expert marketplace (Mercor, Surge AI, or a fast-follower) to launch a publicly listed, standardized rate card by credential type — the same way freelance platforms eventually standardized rates by skill category — because the current opacity in vetting and pay is generating exactly the kind of visible worker frustration that pushes marketplaces toward transparency once they're competing for the same scarce expert pool."
Multiple marketplaces chasing the same finite pool of credentialed experts creates real competitive pressure to reduce friction and be upfront about pay — the platforms that do it first will have a recruiting advantage. We'll check back on this prediction in a future issue.
FOWL AI · August 31, 2026 · We'll score this in a future issue.
✅ 3 Things to Do This Week
One for each type of reader — pick yours
01
If you hold a professional credential sitting idle (MD, JD, CPA, PE, senior engineering background, PhD) — apply directly to the expert marketplaces (Mercor, Surge AI, Handshake AI Fellowship, Alignerr, Turing) rather than generalist gig platforms. Story 02's roughly 10x pay gap is for adjacent-looking work — the credential is the entire difference.
02
If you're already doing generalist AI-rating gig work — treat it as unpredictable side income, not a wage, per story 03. Check whether a niche skill you already have (a rare language, a specific coding stack, a scientific background) qualifies you for a platform's specialist tier at 2–4x the generalist rate.
03
When job-searching this category — stop searching just "AI Evaluator." Pair it with your actual field ("AI Evaluator + Healthcare," "AI Evaluator + Legal," "AI Evaluator + Finance") — per story 05, that's increasingly how the postings themselves are organized.
🚀 From the Lab
Sponsored by FOWL AI
Launch Lab — Building something?
We're not an agency — we don't build it for you. Brand Brain is the AI workspace you use to turn what you know into a website, newsletter, and digital products you actually own. You stay in control of it the whole way.
The specialist end of the RLHF pipeline from story 04, now appearing on expert marketplaces as narrow, verified listing types — one live example is literally titled "Conversational Preference Reward Model Evaluator." The work: assessing which of two or more AI-generated responses better satisfies a nuanced preference or reasoning standard in your domain, then explaining why in a way that can be used to train the model's judgment directly, not just flag an error.
Unlike "AI Orchestrator" (Issue 14), which describes owning a cross-tool workflow, this title describes owning judgment inside one narrow evaluation loop — and it's priced accordingly high specifically because it's narrow: Mercor's most-paid listings sit in this exact tier at a reported $100–200+/hr, well above generic domain-expert review.
How to position for this now: Lead with the specific domain and the specific kind of judgment call you're qualified to make (a legal-argument soundness check, a clinical-reasoning error, a mathematical-proof gap, a code-correctness call) — the narrower and more verifiable your claimed expertise, the higher the tier you clear on these platforms' vetting process.
💼 AI Trainer Platforms / Opportunity Board
Where each tier of this split is actually hiring right now
$$$
Mercor — the largest expert marketplace by disclosed revenue; specialist RLHF and credentialed domain-expert review reportedly $100–250/hr, with some assignments reported up to $300/hr.
$$$
Surge AI — bootstrapped, RLHF- and expert-preference-data focused; recruits credentialed reviewers directly for named lab projects rather than an open task queue.
$$
Handshake AI Fellowship / Alignerr / Turing — university- or portfolio-verified expert programs; reported ranges span roughly $20–125+/hr depending on credential and assignment.
$
Generalist tier (Outlier, DataAnnotation, Remotasks) — baseline roughly $12–30/hr, coding/specialist tasks on the same platforms running $25–75/hr; treat queue availability, not the advertised ceiling, as the real constraint per story 03.
📋 New AI Jobs
Titles actually hiring right now, worth searching this week.
›
Reward Model Evaluator / RLHF Domain Specialist — this week's Emerging Career Title; also search "Conversational Preference Reward Model Evaluator" for the exact listing form.
›
AI Evaluator + [your field] — Healthcare & Social Assistance Specialist, Legal Domain Expert, and Physics/Computational Chemistry Expert are all live, separately posted tracks per story 05.
›
Domain-Expert AI Evaluator (credentialed) — the Mercor/Surge AI/Alignerr tier open to licensed clinicians, lawyers, CPAs, and PhDs at $75–300/hr, distinct from generalist annotation.
›
Contract AI Evaluator — full-time-adjacent contract postings averaging $65,471/year per current listings, worth comparing against per-hour gig rates for the same skill.
The "get paid to train AI" pitch used to describe one gig market. This week's numbers say it's now two — a multi-billion-dollar expert economy that just priced itself like a frontier lab, and a generalist queue visibly running out of both work and pay stability underneath it. Both are real. They're just not the same opportunity anymore, and the platforms themselves have stopped pretending they are.
The useful move isn't "sign up for AI training work" — it's an honest inventory of what you already know that a model can't yet self-check, and whether that knowledge clears the verification bar these marketplaces are gatekeeping behind. For a real subset of knowledge workers, that bar is already cleared. It's just a matter of applying to the right tier.
"Same login page.Two completely different labor markets underneath it."
💬 One question — reply and tell us
Have you tried AI training or evaluation gig work — generalist or credentialed? What did it actually pay, and did the queue hold up over time? Hit reply and tell us which tier you landed in.
Hit reply. We read every one and it shapes what we cover next.
📍 Atlanta · Vibe Code Saturdays
Join us for Vibe Code Saturdays.
A free, hands-on Saturday in Atlanta where you build something that belongs to you using AI — your website, your newsletter, your portfolio, your next idea. No coding background required. Bring a laptop.
We're scheduling the first session now, and we'll notify you the moment the date and venue are set.