Mercor Hits $10 Billion Valuation by Hiring Experts to Teach AI Office Work

San Francisco's Mercor has built its business on a paradox: it hires skilled professionals to teach AI to do their own jobs. With this approach, the startup, founded by college dropouts, has reached a $10 billion valuation and become a prominent player in the race to automate office work.

How the approach works

To automate white-collar tasks, it is not enough to give a model access to the internet, documents and corporate email. It needs live examples of how specialists make decisions, set priorities, fix mistakes and see a task through to a result.

In this scheme, Mercor acts as a bridge between human and machine: the company pays highly qualified workers to pass on practical skills that previously stayed inside the profession and were rarely formalized.

Essentially, the startup turns human expertise into a training loop for AI. If a model is to do office work in real processes rather than in demos, it needs not only correct answers but correct logic of action. So the value here lies not in abstract "neural network training" but in collecting professional patterns at scale and at high quality: how to analyze a task, which exceptions to notice, where to double-check conclusions, and how to format the result so it is usable inside a company. Such a layer of knowledge can be scaled, updated and fine-tuned on new cases.

Why Mercor took off

For investors, this model looks clear and scalable. Mercor is not just selling yet another AI assistant; it operates at a more expensive and ambitious level, trying to turn expert labor into a repeatable system for automating white-collar roles.

The founders' story adds to the interest: the startup was created by former students who dropped out of college, and in a short time they took the company to a $10 billion valuation. In this market, that reads as a signal: demand for automating knowledge work is already large enough to justify huge bets. In effect, Mercor is selling not a promise of a distant future but infrastructure to speed it up right now.

• Access to professionals becomes the raw material for training models
• Real work scenarios can be turned into data, and data into a product
• A focus on office tasks opens a bigger market than mass-market chatbots
• The more closely AI copies a specialist's expertise, the more visible the economic effect for business

Who this worries

The most uncomfortable part of the story is that people are effectively helping build tools that may, over time, reduce demand for their own skills. For an employer this looks rational: first pay an expert to hand over knowledge, then use the model where a person was once required.

For the specialists themselves, the picture is more complicated. For now, such projects provide income and demand for their expertise, but they also push the market toward making part of intellectual work standardized, measurable and therefore easier to automate.

Not all office work will disappear equally fast, though. Where responsibility, context, client trust, negotiation or non-standard decisions are required, humans keep a strong position. But routine chunks of white-collar work, such as preparing standard materials, initial analysis, structuring information and repetitive operations within processes, are increasingly seen as a suitable zone for AI. Mercor is betting precisely on this transitional layer between the human expert and a machine learning to reproduce their useful actions.

What this means

Mercor shows where the center of the AI race is shifting: from flashy demos to the systematic capture of professional knowledge. If this approach works at scale, the main asset will be not the models themselves but access to people who can turn their work into a set of decisions, rules and actions that AI can understand.