--- title: "Inside the Low-Paid Jobs Training the AI | Metaintro" canonical: "https://www.metaintro.com/blog/low-paid-jobs-training-ai-that-could-replace-them" language: "en" author: "drashtigarach" published: "2026-07-13T15:52:45.000Z" modified: "2026-07-13T17:18:35.791Z" --- [Back to Blog](/blog) [AI](/blog/tag/ai)[Jobs](/blog/tag/jobs)[Technology](/blog/tag/technology)[News](/blog/tag/news) # Inside the Low-Paid Jobs Training the AI That Could Replace Them Tech firms spend over $1 billion a year paying laid-off white-collar workers $21.50 to $150 an hour to train the AI meant to replace them. Is it worth taking? [![Drashti Garach](https://cdn.metaintro.com/rs:fill:40:40/q:72/plain/images/5719d740-e510-42bc-8017-e040d145f35f_1766029465094.png)Drashti Garach @DrashtiGarach](/blog/author/drashtigarach) [July 13, 2026](/blog/archive/2026/07)13 min read ![Inside the Low-Paid Jobs Training the AI That Could Replace Them](https://cdn.metaintro.com/rs:fill:1200:675/q:78/plain/images/5418-1.png) [https://x.com/intent/tweet?text=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them](https://x.com/intent/tweet?text=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them&title=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them&title=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them)[mailto:?subject=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them](mailto:?subject=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them) At [Metaintro](https://www.metaintro.com), we track the strange new corners of the labor market so you can make clear-eyed choices, and few corners are stranger than this one. As [The New York Times](https://www.nytimes.com/2026/07/10/business/ai-white-collar-jobs.html) and a widely syndicated [Financial Post report on Yahoo Finance](https://ca.finance.yahoo.com/news/bleak-job-market-pushing-white-110029482.html) have documented, a "bleak" job market is pushing unemployed professionals into a peculiar kind of work: getting paid to teach artificial intelligence to do their old jobs. Major technology companies are now spending more than $1 billion a year to have humans train their machine-learning models, and much of that spending flows to the exact white-collar workers whom AI is displacing. The core question this raises for anyone job hunting right now is simple. Is training AI a genuine career move, or is it a short-term paycheck with a built-in expiration date? ## Why are laid-off professionals training the AI that threatens them? The answer, over and over, is that the alternative is nothing. Take the case of one Columbia University journalism graduate based in Barcelona who is fluent in eight languages and has internships on his resume. He has been job hunting for more than a year, and says only five of his graduating class of 200 are employed in full-time jobs. He described the decision to take AI training work in blunt terms, saying that "sheer desperation" made him do it. That is not an isolated story. It is the emotional engine behind an entire industry. The mechanics are almost cruelly efficient. White-collar workers in North America and Europe say they are now bombarded with AI training job ads on job platforms, and the application process is often fast and automated. When you have watched a year of applications vanish into automated screeners, a role that pays this week feels like rescue. We cover the deeper version of this squeeze in our reporting on [how a bleak market is pushing new grads and the long-term unemployed to the margins](https://www.metaintro.com/blog/2026-hiring-rebound-skipping-new-grads-older-long-term-unemployed), and the pattern is the same everywhere: desperation, not ambition, is the recruiter. At [Metaintro](https://www.metaintro.com), we think naming that dynamic honestly is the first step to navigating it. ## What does AI training work actually pay? This is where the picture gets uncomfortable, because the pay gap is enormous. One trainer in Surrey, British Columbia, spends his days running quality checks on AI-generated videos for a "Magnificent Seven" company, managing a team of eight. For that work, the Maryland-based staffing firm [TekSystems](https://www.teksystems.com/) pays him $30 an hour. Before that, he took a data-labeling role in 2024. San Francisco-based [Invisible Technologies](https://www.invisible.co/), which claims to have trained 80 percent of the world's leading AI models, hired him for $21.50 an hour to identify and label objects in videos, under a non-disclosure agreement that stops him from naming the client. At the other end of the ladder, the numbers jump. According to a [Bloomberg](https://www.bloomberg.com/) report, [OpenAI](https://openai.com/) ran a secret effort, "project Mercury," that hired more than 100 former investment bankers at $150 an hour to train its AI on how to build financial models, explicitly aimed at replacing entry-level work. [xAI](https://x.ai/), Elon Musk's company, has hired AI video-game tutors at up to $100 an hour to train its Grok chatbot. The lesson for job seekers is stark: the more specialized and hard-to-automate your expertise, the more the AI training economy will pay you to hand it over. This mirrors the wider [AI productivity and pay gap we have tracked](https://www.metaintro.com/blog/ai-productivity-pay-gap-workers-wages-2026), where the same technology can both raise output and suppress wages depending on where you sit. ## Which companies are hiring, and how big is this getting? The supply chain behind AI training is bigger than most job seekers realize. Workers report being recruited by [Mercor](https://mercor.com/), [Scale AI](https://scale.com/), Turing Enterprises, and [Appen](https://appen.com/), along with their networks of partners and subsidiaries such as Crossing Hurdles and Outlier.ai. Mercor, a Bay Area startup that trains models for OpenAI and [Anthropic](https://www.anthropic.com/), has rapidly expanded its contractor network to 30,000 people, now including doctors, lawyers, bankers, and writers. Founded in 2023, Mercor was valued at $10 billion in October. We have written before about the [30,000-strong Mercor network training white-collar AI](https://www.metaintro.com/blog/mercor-ai-training-white-collar-jobs-skills-2026) and about [Hollywood writers taking Mercor AI trainer gigs](https://www.metaintro.com/blog/hollywood-writers-mercor-ai-trainers), and the through-line is that this is no longer a fringe activity. It is a fast-growing pillar of the AI build-out. There is a second, less visible layer here too: the annotators and labelers who do the grinding foundational work. Their jobs can evaporate quickly once a model matures, as we documented in our coverage of [the hidden AI workforce and annotator layoffs](https://www.metaintro.com/blog/ai-hidden-workforce-meta-covalen-annotator-layoffs-2026). If you are considering this path, understanding which layer you are entering matters enormously for how long the work will last. ## What is the work actually like day to day? It is quieter and lonelier than the AI boom's headlines suggest. The Surrey trainer's job is to run quality checks on AI-generated videos and, with his team of eight, to identify and describe what the model produces, under a non-disclosure agreement that stops him from even naming the client. His earlier annotation role involved labeling objects in videos frame by frame. This is not glamorous prompt-engineering work. It is meticulous, repetitive judgment work, often done from home on a laptop, feeding corrections back into a system you will never be credited on. The paths into it are as varied as the people taking them. One Vancouver software developer and entrepreneur built a business licensing music on YouTube, sold his music portfolio to a company abroad, and now trains AI freelance, framing it as a way to pivot into the AI world and keep his income going. Another Vancouver worker was hired in June by the platform Toloka, known as Mindrift, as a contract AI tutor to complete coding tasks, hoping the experience moves her toward becoming a software developer. The common thread is that almost everyone treats the work as a means to an end, which is exactly why understanding whether it delivers on that promise matters so much. It also sits inside the broader shift we track in our coverage of [the 2026 entry-level squeeze as AI raises the productivity bar for new hires](https://www.metaintro.com/blog/2026-entry-level-squeeze-ai-raised-productivity-bar-new-hires), where the traditional first rungs of the career ladder are quietly being rebuilt. ## Why are companies paying humans so much to train AI at all? Because the models cannot get smarter without human expertise, and the more advanced the model, the more specialized the humans it needs. Early AI training leaned on simple tagging and labeling. Today's frontier systems need doctors to check medical reasoning, lawyers to vet legal drafts, and bankers to grade financial models, which is why Mercor's 30,000-person network now includes exactly those professions. That demand is what justifies the more than $1 billion a year that major technology companies are spending on human trainers, and why OpenAI reportedly paid former investment bankers $150 an hour for its financial-modeling effort. There is a hard irony baked into this arrangement. The workers most valuable to AI trainers are precisely the high-skill professionals whose roles the AI is being built to automate, which is why so many describe the trade in bleak terms. That displacement is rarely total, though, and understanding the nuance matters for how you plan. Lacey Kaelani, CEO of [Metaintro](https://www.metaintro.com), told [People Managing People](https://peoplemanagingpeople.com/workforce-management/ai-layoffs/) that "AI is not completely eliminating roles, but instead restructuring roles and therefore slowing hiring for some jobs." For job seekers, the strategic read is that expertise still commands a premium right now, even in the act of teaching it away. The people who benefit most are those who use the paycheck to fund a transition rather than treating it as a career, a theme that runs through our reporting on [how AI-displaced workers can face a decade of wage setbacks](https://www.metaintro.com/blog/ai-displaced-workers-decade-wage-setback-goldman-sachs) if they do not plan the next move deliberately. ## Is training AI a good way to break into the AI economy? Many workers tell themselves it is, and that is the pitch: get relevant experience, keep the lights on, and pivot into the AI world before the door closes. Several trainers frame the work as a training ground for new skills in the AI economy. But the data complicates that hope. A 2024 study by the European Microworkers Communication and Outreach Initiative, known as EnCOre, found that only 6 percent of these "microworkers" said they took the jobs to acquire skills to find a new job. The overwhelming majority were there for the money and the ability to work from home, not for a career ladder. That gap between the promise and the reality is the single most important thing for a job seeker to internalize. Training AI can genuinely teach you how these systems think, where they fail, and how to prompt and supervise them, and those are real, portable skills. But the work itself rarely converts into a stable role, which is why so many workers describe it as a bridge rather than a home. We have written about the broader problem that [most workers cannot connect AI training to an actual job](https://www.metaintro.com/blog/85-percent-workers-cannot-connect-ai-training-job-2026), and the microworker research suggests why: the incentives of the work push toward volume, not skill-building. ## How can you tell if a specific AI training gig is worth taking? Not all of these roles are equal, and a little diligence up front saves a lot of regret. Start with the rate and the layer. A basic data-labeling contract in the $21.50 range is different from a specialist role in the $100 to $150 band, and the higher-paid work usually draws on expertise you already have, which makes it both better compensated and more likely to teach you something transferable. Ask how long the project is expected to run, because the foundational annotation work tends to disappear fastest once a model matures. Read the non-disclosure terms carefully too, since the Surrey trainer could not even name the client he worked for, and heavy NDAs can limit how much of the experience you can later put on a resume. Then weigh the trade honestly against your alternatives. For the Barcelona journalism graduate who had applied for more than a year, almost any income was better than none, and that is a legitimate reason to take the work. But go in with clear eyes: the 2024 EnCOre research shows most people in these roles are there for the paycheck, not a career path, so do not count on the gig to become permanent. If the rate is decent, the project has some runway, and the work touches your existing skills, it can be a reasonable bridge. If it is bottom-of-the-market labeling with an ironclad NDA and no learning attached, treat it strictly as short-term cash. Our guide to [the jobs AI cannot replace and what each pays](https://www.metaintro.com/blog/7-jobs-ai-cant-replace-2026-what-each-pays) is a useful companion here, because it points toward the kind of durable work worth aiming your longer-term energy at. ## What does this mean for your career? If you are staring at an AI training offer, treat it like a bridge loan, not a mortgage. Take it if you need the income, but go in with a plan to extract value beyond the paycheck. Document exactly what you learn about how models are built, evaluated, and corrected, because that supervisory knowledge is what employers in the [growing field of AI-adjacent roles](https://www.metaintro.com/blog/ai-companies-hiring-beyond-openai-nvidia-2026) actually want. Contract AI work is part of the same larger shift we covered in [how 76 million Americans moved into freelance and gig work](https://www.metaintro.com/blog/76-million-americans-freelance-gig-economy-takeover-us-labor-market-2026), so budget for the volatility, the missing benefits, and the sudden project endings that come with it. Most of all, keep building the skills that AI struggles to replicate: judgment, relationship-building, complex problem-solving, and the ability to own an outcome end to end. Those are the skills that survive automation, as we lay out in our guide to [staying employable as AI reshapes work](https://www.metaintro.com/blog/2026-career-skills-stay-employable-constant-change) and our look at [the jobs AI cannot replace and what each pays](https://www.metaintro.com/blog/7-jobs-ai-cant-replace-2026-what-each-pays). Training the machine can be a smart temporary move. Betting your future on it is not. At [Metaintro](https://www.metaintro.com), our advice is to use the paycheck, protect your time, and spend the runway building toward something the AI cannot easily take back. --- ## Related Articles - [30,000 Professionals Are Training Their Own Replacements at Mercor](https://www.metaintro.com/blog/mercor-ai-training-white-collar-jobs-skills-2026) - [Hollywood Writers Are Taking Mercor AI Trainer Gigs](https://www.metaintro.com/blog/hollywood-writers-mercor-ai-trainers) - [The Hidden AI Workforce and the Annotator Layoffs Nobody Sees](https://www.metaintro.com/blog/ai-hidden-workforce-meta-covalen-annotator-layoffs-2026) - [Why 85% of Workers Cannot Connect AI Training to a Job](https://www.metaintro.com/blog/85-percent-workers-cannot-connect-ai-training-job-2026) - [The AI Productivity and Pay Gap Hurting Workers' Wages](https://www.metaintro.com/blog/ai-productivity-pay-gap-workers-wages-2026) - [AI Displaced Workers Face a Decade of Wage Setbacks](https://www.metaintro.com/blog/ai-displaced-workers-decade-wage-setback-goldman-sachs) - [76 Million Americans and the Gig Economy Takeover](https://www.metaintro.com/blog/76-million-americans-freelance-gig-economy-takeover-us-labor-market-2026) - [The 2026 Entry-Level Squeeze as AI Raises the Bar for New Hires](https://www.metaintro.com/blog/2026-entry-level-squeeze-ai-raised-productivity-bar-new-hires) - [Career Skills to Stay Employable Through Constant Change](https://www.metaintro.com/blog/2026-career-skills-stay-employable-constant-change) - [7 Jobs AI Cannot Replace in 2026 and What Each Pays](https://www.metaintro.com/blog/7-jobs-ai-cant-replace-2026-what-each-pays) - [AI Companies Hiring Beyond OpenAI and Nvidia in 2026](https://www.metaintro.com/blog/ai-companies-hiring-beyond-openai-nvidia-2026) --- ## People Also Asked ### Q: How much do AI training jobs pay? A: It varies enormously by specialization. Basic data-labeling and annotation contracts can pay around $21.50 to $30 an hour, while highly specialized projects pay far more. OpenAI reportedly paid more than 100 former investment bankers $150 an hour to train its AI on financial modeling, and xAI has hired video-game tutors at up to $100 an hour. The more niche and hard-to-automate your expertise, the higher the rate. ### Q: Are AI training jobs a stable career? A: Generally no. The work is overwhelmingly short-term and contract-based, with no benefits and projects that can end abruptly once a model is trained. A 2024 EnCOre study found only 6 percent of these microworkers took the jobs to build skills for a new career, with most doing it for immediate income. Treat it as a temporary bridge, not a long-term plan. ### Q: Will training AI help me get a job in tech? A: It can teach you how AI systems are built, evaluated, and corrected, which is valuable supervisory knowledge. But the work itself rarely converts directly into a permanent role, and most workers cannot connect it to a job offer. To make it count, document what you learn and keep building durable, AI-resistant skills alongside the contract work. Watching the labor market shift under your feet is exhausting, but you do not have to track it alone. At [Metaintro](https://www.metaintro.com), we cut through the noise on AI, hiring, and pay so you can stay ahead of the market instead of reacting to it. [Sign up for Metaintro](https://www.metaintro.com/signup) to get clear, job-seeker-first analysis delivered straight to you. ### Share this article [https://x.com/intent/tweet?text=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them](https://x.com/intent/tweet?text=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them&title=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them&title=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them)[mailto:?subject=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them](mailto:?subject=Inside%20the%20Low-Paid%20Jobs%20Training%20the%20AI%20That%20Could%20Replace%20Them&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Flow-paid-jobs-training-ai-that-could-replace-them) For job seekers ## The right role, found for you. Built for people who’d rather be good at their job than good at job hunting. [Get Started Free](/signup)[Search matching jobs](/jobs/search) - MatchCompare live roles against your current evidence. - PositionTurn proof projects into role-specific applications. - ImproveUse market feedback to keep the skill plan current. [Return to navigation](#main-navigation)