Mercor, a recruiting startup matching skilled professionals with AI labs, was valued at $10bn by private investors and reached over $1bn in run rate in twenty months.
Brendan Foody says Mercor shifted from hiring tool to agentic data after a chance meeting with Sean at OpenAI, recruiting bankers, lawyers and doctors to write rubrics that score model behavior.
Foody argues productivity gains tend to create more work, expects humans to stay central for a multi-decade period, and says regulators should reconsider payroll and income taxes that discourage hiring.
Mercor turns professional expertise into the data AI labs need
Mercor is betting that expert human judgment is the scarce input for improving AI, and that it can recruit and score that judgment at scale. Labs now train models against rubrics written by professionals, so whoever supplies the experts shapes what models can do. Foody argues AI will still expand work for decades, because models struggle with long tasks that are hard to verify.
Takeaways
Rubrics written by experts now train models. Older supervised fine-tuning and preference data are giving way to rubrics and unit tests written by bankers, lawyers, doctors and software engineers. The same rubrics feed both benchmarks and reinforcement learning rewards.
Apex benchmarks economically valuable professional work. Apex measures capability in investment banking, law, medicine and software engineering. Foody says it is becoming an industry standard for labs and enterprises, and Mercor wants benchmarks showing which jobs AI can and cannot do.
Productivity gains have historically created more work. Foody cites a Bureau of Labor Statistics analysis showing average productivity rose about twentyfold over 225 years while jobs grew. He compares AI to the ATM and the farming shift.
Humans stay central for decades. Foody expects a multi-decade period of human involvement because models will struggle with long, hard-to-verify tasks. Automating the physical world will take longer still. He sees displacement as a major issue for society and Mercor.
Taxes on hiring deserve regulators' attention. Foody argues policymakers should look at payroll and income taxes, especially at the low end, because they discourage hiring. He is most excited about AI reaching average companies across the economy, including in the Midwest.
Performance data beats resumes in hiring. Foody says performance data is a better signal than resume proxies. Solving matching at software cost would let far more candidates be considered for every job. Mercor has run more than six million AI interviews.
Thiel Fellowship and Prod mattered for community. The Thiel Fellowship arrived in March 2024 at about $2m in revenue, and Foody says its credibility and community mattered most. Prod's teams were Mercor's first customers. He values university communities over coursework.
How Mercor reached agentic data
Foody met cofounders Surya Midna and Adarsh Haymath in high school speech and debate. He left Georgetown after two years, and their first business consulted for sneaker resellers on AWS credits.
In early 2023 they built a hiring tool placing college students in India on side projects for friends. It hit a million dollars in run rate within eight months.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
They first tried to partner with Scale AI by supplying engineers, then saw a gap for expert agentic data.
A chance meeting with Sean, then at OpenAI, pushed them to recruit top professionals to write rubrics and tests.
Mercor became OpenAI's largest data vendor within nine months and passed a billion dollars in run rate in twenty months.