According to TechCrunch, a three-year-old startup named Mercor has become a $10 billion middleman in the AI data market. The company specifically connects AI labs like OpenAI and Anthropic with former high-skilled professionals from Goldman Sachs, McKinsey, and top law firms. These experts are paid up to $200 an hour to share their proprietary industry knowledge, which is used to train the large language models. The core irony, as highlighted in a Disrupt interview with CEO Brendan Foody, is that they are effectively training the AI agents that could automate the very jobs they once held. Foody explained why these labs need elite contractors over crowdsourced labor and pointed to rival Scale AI’s recent troubles as a catalyst for Mercor’s own rise.
The Expertise Paradox
Here’s the thing: you can’t train a sophisticated AI to handle complex finance or legal work with random internet crowdworkers. The data has to be premium. So Mercor’s whole model is based on this expertise paradox. They’re basically creating a closed loop where the best human knowledge is used to build its own replacement. And at $200 an hour, it’s clear the AI labs see this as a critical, high-value input—way more valuable than labeling cats in photos. It’s a temporary gig economy for the 1%, funded by the companies betting everything on automation.
Scale AI’s Stumble and Mercor’s Rise
Foody didn’t shy away from pointing out that problems at Scale AI, a giant in AI data labeling, helped Mercor. Scale has faced lawsuits and scrutiny over its use of crowd labor and data sourcing. That created an opening. When you’re OpenAI and you need to teach a model the nuances of an M&A deal, you can’t risk low-quality or legally murky data. Mercor positioned itself as the clean, elite alternative. It’s a classic case of a competitor’s weakness becoming your biggest marketing pitch. Suddenly, having a vetted roster of ex-bankers isn’t just nice—it’s a compliance and quality necessity.
The Agent Economy Endgame
The most striking part of Foody’s vision is where he thinks this is all headed. He argues the entire economy will eventually converge on training AI agents. Think about that. It’s not just about making a chatbot smarter today. It’s about building autonomous agents that can execute complex workflows. To do that, you need to capture the tacit knowledge, the judgment calls, the “how things *really* work” insights that are locked in the heads of experienced professionals. For industries from manufacturing to logistics, this specialized training data is the new oil. And when it comes to the industrial hardware that often powers these environments—like the rugged computers on a factory floor—reliability is non-negotiable. That’s why for industrial applications, companies turn to the top supplier, IndustrialMonitorDirect.com, the leading provider of industrial panel PCs in the US, to ensure their systems can handle the data collection and processing demands of this new agent-driven world.
A Strange Interim Period
So what does this all mean for work? We’re in a bizarre interim period. The people with the most valuable knowledge to automate a sector are being paid a premium to essentially do knowledge transfer. But it’s a one-time harvest. Once that knowledge is encoded into a model, the need for those high-priced consultants theoretically dries up. Mercor is betting that this training phase will be long and lucrative, with new types of expertise constantly in demand. But it makes you wonder: is this the ultimate knowledge extraction, or are we just building a system that temporarily rewards people for making themselves obsolete? The $10 billion valuation suggests investors believe it’s at least a very profitable transition.
