Tyler Cowen interviews Brendan Fudi, 22-year-old CEO and co-founder of Mercor, an AI company founded in early 2023. Fudi is reportedly the youngest unicorn founder ever, leading what may be the fastest-growing company in history by some metrics. Mercor operates as a marketplace connecting AI companies with domain experts who help train and evaluate AI models.

The conversation explores Mercor's unique business model of hiring specialists like poets, economists, and legal experts to create evaluation frameworks for AI systems. Notable advisors include Larry Summers for finance, Cass Sunstein for law, and Eric Topol for medicine. They discuss the AI Productivity Index (Apex), which measures model performance on economically valuable tasks rather than academic benchmarks.

Fudi shares insights on talent assessment, drawing connections to Talent How to Identify Energizers, Creatives, and Winners Around the World by Tyler Cowen, and discusses philosophical challenges in AI evaluation, referencing Kant's Critique of Judgment regarding the difficulty of systematizing aesthetic taste in domains like poetry.

Why Mercor Pays Poets $150 Per Hour

Mercor hires domain experts to create evaluation rubrics for AI models, similar to how "a professor in English class would create a rubric to grade an essay or poem" - Brendan

The challenge with subjective domains like poetry is balancing expert consensus with diversity: "you want some degree of consensus but you probably don't want too much consensus because you also want to get all perspectives"

Cowen raises Kant's philosophical objection from Critique of Judgment that "taste is that which cannot be captured in a rubric," questioning whether systematic evaluation can truly capture aesthetic judgment

AI Productivity Index Shows Rapid Economic Progress

The Apex project measures AI performance on economically valuable tasks rather than academic benchmarks, showing "25-30% improvement per year" with GPT-5 scoring 64%

Methodology involves surveying hundreds of experts from top firms (McKinsey, Bain, BCG) to understand how they spend time, then creating corresponding prompts and evaluation rubrics

"The largest disconnect in AI research is that everyone was focused on academic evals like GPQA for PhD-level reasoning, wholly disconnected from outcomes customers actually care about" - Brendan

Models excel at individual tasks but struggle with "longer horizon tasks" and "integrating multiple tools," though Brendan expects major progress within months

The Future of Human Expertise in an AI World

Brendan predicts that within 2-3 years, experts like Cass Sunstein will struggle to find mistakes in AI legal responses, though taste-based domains will remain challenging longer

"Models are advancing very quickly in being able to automate 50% or 75% of what humans and experts are able to do, but will really struggle with that last 25%" - Brendan

Future knowledge workers will primarily train AI models rather than perform tasks directly: "Instead of investment bankers doing analysis, they'll teach models how to do it once and use it many times"

Workers will only need domain expertise, not technical AI knowledge: "So long as they can find where the model makes a mistake and understand the frontier of model capabilities, it's relatively easy to create criteria"

Labor Market Efficiency and Hiring Innovation

Mercor has grown to "just over 300 people" by using its own AI technology to automate resume screening and interview processes

Drawing from Talent How to Identify Energizers, Creatives, and Winners Around the World, Brendan criticizes traditional interviews for focusing on "vibe space conversations" rather than measuring actual job skills

"The largest inefficiency in labor markets is that everything is disaggregated" - companies consider only a fraction of available candidates while job seekers apply to dozens of positions

Brendan advocates for project-based assessments: "Just give them a project and grade them, in essence. That's the cleanest way to do it"

Economic Elasticity and Industry Transformation

"Software demand is extremely price elastic - if we make software engineers 10 times more efficient, we'll have even more software engineers, maybe 10 times as many building 100 times as much software" - Brendan

Other elastic domains include business building, product distribution, and capital allocation, especially early-stage investing where "enormous inefficiency" exists

Education will see "much better access to information, much better ways of explaining that information" but may be less elastic due to time constraints and the importance of personal relationships

Personal Background and Entrepreneurial Origins

Brendan's eighth-grade "Donut Dynasty" involved buying Safeway donuts for $5/dozen and selling them for $2 each at school until administrators shut him down

He moved operations "50 feet off campus" and paid his mother "$20 a week to drive me in her minivan" to transport inventory

Brendan attributes entrepreneurial success partly to being dyslexic: "quite early on, you have to learn how to delegate" and focus on strengths rather than weaknesses

He and his co-founders met at age 14 on their high school speech and debate team, with the company name "Mercor" meaning "marketplace" in Latin from their Jesuit school education