According to CNBC, 34-year-old entrepreneur Utkarsh Amitabh started a part-time gig in January 2025 training AI models for the startup micro1, earning $200 an hour. He’s made nearly $300,000 including bonuses since then, working about 3.5 hours each night after his daughter goes to bed. Amitabh, who is also a CEO, author, lecturer, and Oxford PhD student, says “intellectual curiosity drew him in” to the role, which aligns with his background in business strategy and tech. Micro1, founded in 2022 and now valued at $500 million, uses a network of over 2 million experts like him to train models for clients including Microsoft and Fortune 100 companies. The work involves breaking down complex business problems into language machines understand and refining the AI’s responses through a trial-and-error process that can take hours per problem.
The expert economy for AI
Here’s the thing: Amitabh’s story isn’t just about a lucrative side hustle. It’s a window into a massive, quiet shift in how AI is being built. The low-hanging fruit of scraping the public internet is mostly gone. Real progress now, as micro1’s CMO said, comes from domain experts who can “challenge, refine and effectively outthink the model.” This creates a whole new expert economy. We’re not just talking about coding or prompt engineering. We’re talking about lawyers, doctors, engineers, and philosophers—like Amitabh with his master’s in moral philosophy—teaching AI the nuances of their fields. It’s highly paid, deeply specialized, and fundamentally different from the early days of basic data labeling. Companies like micro1 and its competitors are betting billions that this human-in-the-loop expertise is the key to the next leap in AI capability.
What does the job actually look like?
So, what are you doing for $200 an hour? It’s less about feeding data and more about high-stakes tutoring. Amitabh describes getting a complex business problem—something an executive might face—and deconstructing it into perfect, unambiguous steps for the AI. If the model’s answer is off, he has to diagnose where the “subtlety got lost.” Was it the phrasing? A missed assumption? A logical gap? You’re essentially a forensic editor for machine reasoning. It requires, as he says, “immense attention to detail” to catch mistakes both humans and machines might make. And it’s a moving target. The models get better, so you have to level up your own thinking. It’s intellectually exhausting work. For a deeper dive on the general process, this explainer on training AI models outlines the core concepts, though at a much more basic level than what Amitabh is doing.
The trillion-dollar question of obsolescence
Now, the elephant in the room. Isn’t there a risk that by training these AIs so well, experts like Amitabh are building their own replacements? He calls this the “trillion-dollar question.” His take is philosophically interesting. He doesn’t see knowledge as a finite resource where the machine learning something means a human forgets it. Instead, he envisions a symbiotic relationship where human and machine collaboration lifts the output for both. He’s between a “techno-optimist and a techno-realist.” Sure, there will be painful job displacement—that’s already happening. But he, and reports like a 2025 World Economic Forum analysis, believe AI will also create new roles, predicting net job gains by 2030. His view: the fear of AI might actually push us to upskill and ask better questions about our own value. It’s a comforting thought, but is it realistic for everyone, or just for the elite experts at the top of the food chain?
Is this a viable path for others?
Let’s be real. Amitabh’s path isn’t exactly typical. He’s a hyper-qualified, deep generalist with degrees from top schools and a stint at Microsoft. That’s the profile micro1 and its competitors are aggressively recruiting. For them, it’s a talent pipeline problem. But his story does signal a broader trend: the premium on applied, expert knowledge is skyrocketing in the AI age. The barrier isn’t just knowing a topic; it’s being able to deconstruct your own expertise for a machine. That’s a meta-skill in itself. So while not everyone will land a $200/hour side gig, the message is clear. The future of work, especially in knowledge industries, will heavily favor those who can partner with AI, not just use it. The real side-hustle revolution might be in becoming a tutor for the world’s smartest, and hungriest, student.
