A $10 billion startup run by people who've never had real jobs is training AI to do yours.
The Summary
- Mercor, valued at $10 billion, is building AI agents to replicate white-collar work, founded by twentysomethings with no traditional work experience
- The company is actively recruiting professionals on LinkedIn, not to hire them, but to train AI replacements using their expertise
- This represents the most explicit acknowledgment yet that the agent economy isn't adding jobs, it's learning to eliminate them
The Signal
Mercor's pitch is brutally honest in a way most AI companies avoid. They're not selling "augmentation" or "copilots." They're selling replacement. The company approaches professionals on LinkedIn with what looks like recruiting outreach, then asks them to teach AI agents how to do their job. The data becomes training material. The professional becomes obsolete.
The $10 billion valuation matters because it shows investors believe this works at scale. That's not hype money for a chatbot wrapper. That's conviction that professional work can be decomposed, learned, and automated profitably. The bet isn't on making workers 10% more efficient. It's on making them unnecessary.
"The most efficient way to learn a job is from someone who already does it, even if they've never thought about how to teach it."
What makes this darkly fascinating is the founder profile. The team reportedly never held traditional white-collar jobs before building a company to eliminate them. No years grinding through corporate hierarchy. No embodied understanding of what that work actually requires at human scale. They're approaching professional labor as a pure systems problem, which might be exactly the mindset needed to automate it without sentimentality.
The mechanics are straightforward:
- Identify high-value, repeatable professional tasks
- Extract process knowledge from current practitioners
- Convert human judgment into agent decision trees
- Deploy at marginal cost approaching zero
Mercor isn't alone in this approach, but they're unusually direct about it. Most AI companies wrap automation in language about "empowering workers" or "handling the boring stuff." Mercor's model assumes the boring stuff is most of the job, and the parts that aren't boring can be learned too.
The LinkedIn recruitment angle reveals something important about where training data comes from now. It's not just scraped text. It's structured knowledge extraction from people who don't realize they're training their replacement. The professional thinks they're interviewing for a role or consulting gig. Mercor is building a model.
The Implication
If you're getting LinkedIn messages from companies you've never heard of asking detailed questions about your workflow, you're not being recruited. You're being indexed. The smart move isn't to refuse to engage. It's to understand that your professional knowledge has market value as training data, and you should price it accordingly or withhold it strategically.
Watch what happens to Mercor's valuation over the next 18 months. If it holds or grows, that's the market confirming that agent-based replacement of knowledge work isn't speculative anymore. It's the next infrastructure build.