Brendan Foody (Mercor Founder/CEO) | Stanford ETL Summary

1. Key Themes

The Shift from Crowdsourced to Agentic Data is the Core Market Transition

The defining insight of Mercor's success is identifying a structural shift in AI training data. The old paradigm—crowdsourced, low-skill labor writing simple text—is being replaced by a fundamentally different model requiring elite human expertise.

"There was an enormous transition happening in the human data market, shifting away from the crowdsourcing paradigm that companies like Scale AI pioneered — low-and-medium-skilled people writing barely grammatically correct sentences for early ChatGPT — toward the agentic data paradigm: finding super high-skilled experts — bankers, lawyers, doctors, FAANG software engineers — who can work collaboratively in teams to build frontier evals and RL environments for the next generation of models." 00:07:03

Evals Are the New Moat — The Model Is the Product

A profound strategic realization has emerged across the AI industry: building evals (success criteria frameworks) is more important than assembling software pipelines. The eval defines what intelligence means for a given domain.

"With deep research it became very clear that the model is the product — the way to solve problems in almost every knowledge-work domain is not stitching together API calls with drag-and-drop, but laying out the end goals (the evals) and training the model to get really good at those end goals. Since Opus 4.5 everyone is realizing the model is the product." 00:35:43

Reinforcement Learning from Human Experts at Scale Is the Current Frontier

Mercor's technical contribution is building the scaffolding—rubrics, unit tests, contexts, tools—that allows RL to function across economically relevant domains, not just academic benchmarks.

"You can use that data to measure how well the model is performing (a benchmark), or as a reward function for reinforcement learning — have the model attempt the problem 50-100 times, score all attempts, and the model hill-climbs. We build out all these rubrics, unit tests, contexts, and tools covering the distribution of anything you could do in ChatGPT, Claude Code, Claude Cowork, or Gemini." 00:15:38

Leading Indicators Over Lagging Signals as the Core Strategic Discipline

Brendan's single most distilled strategic lesson is orienting toward frontier customers whose behavior reveals how the broader economy will reorganize, before that reorganization is visible to others.

"If I had to distill what we did strategically into one thing, it's focusing on leading indicators when markets move really fast. The cutting-edge frontier labs see things in the markets where we can learn how the entire economy will change shape. Focus on those leading indicators and they give you a window into the future." 00:25:31

The Physical World Remains Deeply Underestimated and Underserved by AI

Despite Silicon Valley's confidence in AI progress, Brendan flags that the majority of economic value sits in physical-world work that AI cannot yet touch—and the data to train it there barely exists.

"One thing everyone is overestimating progress on is the physical world. The majority of work in terms of dollars is in the physical world — electricians, mechanical engineers, physical-world services. Building out the distribution of data for how agents learn to do all of that is going to take an extremely long time." 00:27:06

Enterprise AI Adoption Is Years Behind, Creating a Massive Second Wave Opportunity

The Valley's assumption that AI is widely adopted is deeply wrong. The real opportunity—which Mercor is now pivoting to pursue—is helping every non-frontier company build its own evals.

"It's easy to be in Silicon Valley echo chambers thinking AI has 100% adoption. The average company in the Midwest just started using ChatGPT a few months ago. In the context of Mercor, expanding from serving frontier model companies to serving every enterprise and helping them build the evals that correspond to their value chain." 00:24:08

The Network of Human Knowledge Is the Actual Bottleneck for AI Progress

Contrary to narratives focused on compute or algorithms, Brendan identifies the coverage of human expertise as the binding constraint on AI capability expansion.

"The bottleneck to AI labs automating everything in the economy is how they build this network of human knowledge covering the distribution of all the context, prompts, and responses." 00:18:02

Obsessiveness, Not Discipline, Is the Real Predictor of Startup Success

A counterintuitive operating philosophy: forcing yourself to be disciplined is the wrong approach; finding a domain you are genuinely obsessive about is the actual lever.

"People try too hard to be disciplined to do things they don't want to do, when the key is finding the thing you can be obsessive about. The Thiel Fellowship looks for obsessiveness because it's one of the largest predictors of company outcome." 00:28:25

Dollars Spent Are the Only Reliable Leading Indicator—Everything Else Is Noise

A clean heuristic for filtering signal from stated preferences, applied specifically to where AI investment is flowing right now.

"The best leading indicator is associated with dollars being spent — people say they want things, but until they put money where their mouth is you don't know it's true. A great example is compute right now: the most sophisticated labs demonstrating they'll scale compute by orders of magnitude, everyone investing in alternatives to NVIDIA (a multi-chip future), shortages of electricians and construction in the data-center build-out." 00:44:32

2. Contrarian Perspectives

AI Will Create More Jobs Than It Destroys, Even at 95% Automation

Against the dominant narrative of mass unemployment, Brendan offers a historical economic argument that is easy to dismiss but hard to refute.

"The Bureau of Labor Statistics found that over the last 225 years we've made the average person 20x more productive — equivalent to automating 95% of jobs — yet we have more jobs than ever. So long as we automate 95% and not 100%, we won't run out of things to do. We'll build a thousand times more software, build rockets, cure cancer, go to Mars." 00:21:08

Scale AI's Infrastructure Is Actually a Liability, Not an Asset, for the Agentic Era

The conventional view is that Scale AI is the dominant data company with an entrenched moat. Brendan argues Scale's infrastructure was built for an obsolete paradigm and structurally disadvantages it in the new one.

"Scale grew up in autonomous-vehicle labeling with people in the Philippines, and infrastructure was built around that. Going directly to the labs, we could build from the ground up around exceptional people creating the new agentic-data paradigm." 00:32:12

University Communities Matter, But Academic Structure Mostly Doesn't

A nuanced but genuinely contrarian distinction: the value of college is social capital formation, not curriculum—which has direct implications for how founders should evaluate staying versus dropping out.

"I don't think it was, for me. The communities formed around universities are super impactful (Prod wouldn't have existed without Adarsh's Harvard network), but most of the value didn't come from the academic structure itself." 00:43:37

Jobs Are the Economy's Biggest Positive, But Also What Government Taxes Most Aggressively

A rarely stated policy contradiction: the government simultaneously treats job creation as paramount and penalizes it more heavily than almost any other economic activity.

"Jobs are the largest positive thing in the economy, but also the largest thing regulators disincentivize via payroll and income taxes, especially at the low end." 00:23:02

Recruiting Can Be Fundamentally Fairer by Eliminating Heuristics Entirely

The premise that human judgment in hiring reduces bias is inverted: grounding everything in actual performance data removes the largest source of inequity.

"Instead of using heuristics for past things, ground everything in performance data of how people actually do at the job. When you solve the matching problem at the cost of software, everyone can be considered for every job — solving the largest inequity in recruiting." 00:38:50

3. Companies Identified

Mercor

AI-powered recruiting platform and the dominant agentic data vendor to frontier AI labs. Reached $1B annualized revenue in ~20 months, currently valued at $10B, conducted over 6 million AI interviews.

"We scaled from a million in revenue run rate to over a billion in 20 months — as of now the fastest growth trajectory in history." 00:08:28

OpenAI

Frontier AI lab; Mercor's breakthrough anchor customer, becoming their largest data vendor within nine months during the o1 era.

"We became OpenAI's largest data vendor within nine months back when o1 was head and shoulders above everyone else, then quickly became a primary agentic data vendor to all of the top AI labs as well as the top application-layer companies." 00:08:00

Scale AI

AI data labeling company; mentioned as having pioneered the crowdsourcing paradigm but now structurally disadvantaged for the agentic era due to infrastructure built around low-skill, Philippines-based labor.

"Scale grew up in autonomous-vehicle labeling with people in the Philippines, and infrastructure was built around that." 00:32:12

Anysphere (Cursor)

AI coding tool company. Ben, founding CEO of Anysphere, was one of Mercor's earliest customers through the Prod accelerator.

"All our initial customers were Prod teams — Ben (founding CEO of Anysphere), Cofactory, Rob and Gavin from Etched, and others." 00:40:34

Applied Compute

Compute infrastructure company; Yash is the CEO, previously a customer at OpenAI, met Brendan through the Thiel Fellowship community.

"The Thiel Fellowship came in March 2024 when the company was at ~$2M revenue... Largest value-add: the credibility stamp to investors and hires, and the community (met Yash, CEO of Applied Compute, previously our customer at OpenAI)." 00:29:58

Etched

Custom ASIC chip company building transformer-specific silicon; Rob and Gavin were early Mercor customers through Prod.

"All our initial customers were Prod teams — Ben (founding CEO of Anysphere), Cofactory, Rob and Gavin from Etched, and others." 00:40:34

Cofactory

AI company; early Mercor customer through the Prod accelerator network.

"All our initial customers were Prod teams — Ben (founding CEO of Anysphere), Cofactory, Rob and Gavin from Etched, and others." 00:40:34

AWS (Amazon Web Services)

Cloud provider; Brendan built his first real business at age 16 helping sneaker resellers access AWS credits through AWS Activate, scaling to hundreds of thousands in revenue.

"I started a consulting business helping them get AWS credits through AWS Activate, and scaled that to a couple hundred thousand dollars in revenue." 00:04:25

NVIDIA

Semiconductor company; mentioned as the current dominant compute provider but with the leading indicator pointing toward a multi-chip future as alternatives are funded aggressively.

"Everyone investing in alternatives to NVIDIA (a multi-chip future), shortages of electricians and construction in the data-center build-out." 00:44:32

4. People Identified

Brendan Foody

Co-founder and CEO of Mercor. At 23, one of the world's youngest self-made billionaires. Thiel Fellow 2024. Former Georgetown student. Serial entrepreneur from age 13 (donut arbitrage, sneaker resale consulting, Seros cloud consulting). Drove Mercor from $1M to $1B+ ARR in 20 months.

"We scaled from a million in revenue run rate to over a billion in 20 months — as of now the fastest growth trajectory in history." 00:08:28

Adarsh Hiremath

Co-founder and CTO of Mercor. Former Harvard student. Won all three national policy debate tournaments with Surya — first team in history to do so. Manages engineering and enterprise expansion.

"Adarsh became CTO to manage engineering since he'd spent the most time coding. Adarsh is focused more on enterprises — how do we take all the processes, know-how, and technology we've built for the AI labs over the last two years and apply that to the rest of the Fortune 500." 00:13:43

Surya Midha

Co-founder of Mercor. Brendan's college roommate at Georgetown. Co-winner of all three national policy debate tournaments. Leads recruiting functions, leveraging deep IIT connections.

"Surya is very well connected to the IITs and set up a lot of our initial recruiting functions." 00:12:49

Sean (OpenAI)

Early OpenAI employee who became Mercor's key customer relationship and shared the vision for the agentic data transition. Met Brendan when Brendan was 20, at Bar Gemini across from the OpenAI office. Instrumental in Mercor becoming OpenAI's largest data vendor.

"I met Sean, who was working at OpenAI... We shared this vision with Sean and scaled up really fast." 00:07:03

Yash (Applied Compute CEO)

CEO of Applied Compute, formerly a customer at OpenAI. Met Brendan through the Thiel Fellowship; cited as a key learning relationship on the direction toward agentic AI and evals.

"I learned a lot from Yash in 2024. With deep research it became very clear that the model is the product." 00:35:43

Ben (Anysphere Founding CEO)

Founding CEO of Anysphere (makers of Cursor). One of Mercor's very first customers through the Prod accelerator. Represents the quality of early design partners Mercor cultivated.

"All our initial customers were Prod teams — Ben (founding CEO of Anysphere)..." 00:40:34

Rob and Gavin (Etched)

Co-founders of Etched, a custom chip startup. Early Mercor customers through Prod.

"All our initial customers were Prod teams — Ben (founding CEO of Anysphere), Cofactory, Rob and Gavin from Etched, and others." 00:40:34

Peter Thiel

Founder of the Thiel Fellowship; cited as recognizing obsessiveness as the key predictor of entrepreneurial success.

"The Thiel Fellowship looks for obsessiveness because it's one of the largest predictors of company outcome." 00:28:25

5. Operating Insights

Design Partners With Real Money on the Line Are the Only Valid Signal

Brendan's ideation and product validation process is ruthlessly commercial: a customer relationship only counts if dollars are exchanged. Stated interest is worthless without payment.

"We were always very customer-oriented — having early design partners give feedback. Spend as much time with your users as possible." 00:29:58 "The best leading indicator is associated with dollars being spent — people say they want things, but until they put money where their mouth is you don't know it's true." 00:44:32

Use a Price War Tactically and Briefly to Destroy Competitors With Higher Cost Structures

In eighth grade, Brendan intuited and executed a predatory pricing strategy: drop to cost-parity with a higher-cost competitor for exactly as long as needed to drive them out, then return to normal margins. This is a clean, repeatable playbook for commodity or near-commodity markets.

"Competition popped up because margins invite competition. They were buying Chuck's Donuts, which have a higher cost basis of just over a dollar. And so I dropped my prices to a dollar per donut for two weeks to drive them out of business." 00:03:03

Credibility Stamps Unlock Compounding Hiring and Investment Cycles

The Thiel Fellowship's greatest concrete value was not the money—it was the credibility signal that unlocked better hires and investor trust at a moment when revenue alone wasn't enough proof.

"Largest value-add: the credibility stamp to investors and hires, and the community." 00:28:25

Build From the Ground Up for the New Paradigm Rather Than Retrofitting Legacy Infrastructure

When Mercor attempted to partner with Scale and route customers through their platform, it exposed Scale's structural inability to serve elite workers. The operational lesson: legacy infrastructure built for one paradigm is a ceiling, not a floor, when the paradigm changes—build clean.

"We realized there was a gap in the market: Scale grew up in autonomous-vehicle labeling with people in the Philippines, and infrastructure was built around that. Going directly to the labs, we could build from the ground up around exceptional people creating the new agentic-data paradigm." 00:32:12

6. Overlooked Insights

The Apex Benchmark Is Quietly Becoming the Industry Standard for Enterprise AI Capability

Brendan mentions Apex almost in passing, but its implications are enormous. Whoever defines the benchmark by which enterprises evaluate AI models effectively defines the procurement conversation for enterprise AI—a position with extraordinary strategic leverage, analogous to how S&P indices shape investment flows.

"We built Apex with that purpose and it's become an industry standard for how labs determine whether their models do well at capabilities enterprises care about." 00:16:34

If Apex is genuinely the industry standard measure of economically relevant AI capability, Mercor doesn't just supply training data—it also sets the goalposts every lab is training toward. That is a compounding strategic position that receives almost no attention in the conversation.

The Prod Accelerator Is a Hidden Kingmaker Behind Multiple Breakout AI Companies

Prod is mentioned only once, almost incidentally, but it surfaces as the common origin point for Mercor, Anysphere (Cursor), Etched, and Cofactory—a remarkable concentration of outcome for a single community. It is named as potentially the reason Mercor exists at all.

"Mercor might not exist without Prod. All our initial customers were Prod teams — Ben (founding CEO of Anysphere), Cofactory, Rob and Gavin from Etched, and others." 00:40:34

For an investor or founder, Prod deserves far more scrutiny as a talent and deal-flow network than its near-zero public profile would suggest. If it seeded multiple billion-dollar companies in a single cohort, it is likely one of the highest-signal early-stage communities in AI—and almost no one outside the network is paying attention to it.