Tech executives love to promise that artificial intelligence will create as many jobs as it destroys. But when you look closely at what's actually happening, the picture gets a lot more complicated—and more than a little unsettling.
The Job Creation Mystery
We hear the refrain constantly: yes, AI will eliminate certain roles, but it will also spawn entirely new careers. The problem? Almost nobody can tell you what those new careers actually are.
Sure, we've seen AI researchers and prompt engineers emerge. But these specialized roles are hardly widespread. Meanwhile, the World Economic Forum projects a net gain of 78 million jobs between now and 2030. Sounds promising, right? Except when you dig into the details, the fastest-growing positions are farmworkers, delivery drivers, and construction workers—all driven by demographic changes, not some AI-powered productivity revolution.
When Automation Actually Worked
History offers us a blueprint for how automation can create jobs. In the early 20th century, the automobile industry automated manufacturing, displacing skilled craftspeople. But the ripple effects were enormous: more cars meant booms in transportation, retail, and countless related industries.
Could AI trigger something similar? Nobel Prize-winning economist Daron Acemoglu isn't optimistic. In his book Power and Progress, he argues that current AI trends lean heavily toward cost-cutting rather than creating genuinely new capabilities. When AI customer service bots do only marginally better than humans, the productivity gains remain minimal. And new job creation looks increasingly unlikely.
The New Gig Economy: Training Your Replacement
There is one clear example of AI-generated work emerging: professional AI trainers. Companies like Surge AI (valued at $25 billion) and Turing (valued at $2.2 billion) have created an entirely new market. They hire white-collar professionals—lawyers, doctors, financial consultants, even cooks and osteopaths—to train AI models, usually as contractors.
The biggest player might be Mercor.io, founded in 2023 and now valued at $10 billion. The San Francisco startup employs around 30,000 people from diverse fields, paying them hourly to make AI smarter in their respective domains. Their clients? The usual suspects: OpenAI, Anthropic, Google.
Mercor's 22-year-old founder and CEO, Brendan Foody, plans to grow his contractor base "by many orders of magnitude." He keeps meetings to 15 minutes, requires customer-facing staff to work six days a week, and uses AI-powered video interviews to screen new recruits. It's the quintessential startup hustle culture applied to a uniquely 21st-century problem.
The Work Nobody Talks About
The actual work these contractors do is shrouded in non-disclosure agreements. But we know it's technically demanding. They create evaluation frameworks—essentially designing tests for AI models, similar to how the bar exam tests human lawyers. They craft scoring rubrics to teach models what good responses look like.
The pay, by all accounts, is excellent. One contractor told me they'll never make as much money as they're making right now. Which raises an obvious question: how long can this last?
The Uncomfortable Timeline
Tech leaders like Sam Altman and Dario Amodei (CEO of Anthropic) suggest super-intelligent AI is just around the corner. Foody, perhaps predictably, says that threshold is a decade or more away. AI models will be like software, he argues—evolving over decades with constant human input.
It's in Foody's business interest to claim a long runway, of course. But he might have a point. Early large language models were trained on massive amounts of internet text and images, then fine-tuned with help from low-paid data labelers in places like Kenya. (Remember when workers there were filtering harmful content from early ChatGPT?)
Now the industry is shifting. Tech companies want subject-matter experts, not just low-cost labor. Scale AI, which once paid workers in the Philippines and India as little as one cent per task, now recruits contractors with master's degrees and PhDs for complex training work. Meta bought a 49% stake in Scale last summer for $14.3 billion. Surge AI offers up to $1,000 an hour for expertise from startup founders and venture capitalists, and has contracted over 20,000 professionals with doctoral degrees.
The Bigger Picture
Let's say roughly 100,000 professionals are doing expert AI training today, on top of millions more doing traditional low-paid data labeling. Will this be the new form of work that fills the hole AI creates in labor markets?
The transformation is already more subtle than simple job replacement. Wade Foster, CEO of workflow automation company Zapier, puts it this way: "In the past you might have a marketer paired up with an engineer and a designer to work on a project together. Now that might just be one person. You'll see a lot of hybrid jobs and squishing of the job titles."
Mark Zuckerberg recently claimed a single good engineer at Meta can now do work that previously required a large team, thanks to AI. Whether this increased efficiency means less hiring overall remains unclear. British software company Sage saw revenue rise 10% last quarter, thanks partly to internal AI use, then hired more people. Headcount rose 2%.
The Squeeze Play
Here's where the story gets darker for our professional AI trainers. The older market of low-paid data labelers is already being squeezed out. Software handles simpler tasks. Expert trainers take over complex ones. It's not hard to see where this goes.
The lawyers, doctors, and former journalists working for companies like Mercor are likely following a similar arc. One contractor told me they expect the work to last another two years, maybe five if they're lucky.
The irony is almost too perfect: highly skilled professionals, earning excellent money to train AI systems that will eventually replace them. It's the AI job paradox in its purest form—new work created by automation, designed to accelerate automation, ultimately ending in more automation.
In the end, expert AI trainers will probably meet the same fate as every other worker threatened by the technology they're helping to build. The question isn't whether they'll be automated out of existence. It's whether anything meaningful will rise to take their place.