# Brendan Foody (Mercor) - Agentic Data and the Future of AI [Entire Talk]

## Метаданные

- **Канал:** Stanford eCorner
- **YouTube:** https://www.youtube.com/watch?v=NUNyK6P6q8s
- **Дата:** 15.04.2026
- **Длительность:** 47:10
- **Просмотры:** 1,078
- **Источник:** https://ekstraktznaniy.ru/video/51395

## Описание

Brendan Foody co-founded Mercor, a recruiting startup that helps Silicon Valley's top AI labs train their models to do professional-level reasoning by matching skilled workers with enterprise projects. Foody and his co-founders, Surya Midha and Adarsh Hiremath, became the world's three youngest self-made billionaires in October 2025. In this conversation with Adjunct Lecturer Emily Ma, Foody shares how he and his co-founders identified agentic data as the next leap in AI training and benchmarking and built Mercor around it; predicts how the AI revolution will reshape work and the economy; and gives advice to aspiring entrepreneurs looking for a way into the AI market.

Entrepreneurial Thought Leaders is produced by the Stanford Technology Ventures Program (STVP), the entrepreneurship center at the Stanford School of Engineering, and published on eCorner by STVP. STVP empowers aspiring entrepreneurs to become global citizens who create and scale responsible innovations.

CONNECT WITH US

## Транскрипт

### Segment 1 (00:00 - 05:00) []

Okay. Hello everyone. My name is Emily Ma and I am an adjunct lecturer here for the management science and engineering department. You are joining us for the entrepreneurial thought leaders series which is hosted by the Stanford technology ventures program the management science and engineering department at Stanford and basis which is the business association of Stanford entrepreneurial students. I am so excited we are literally starting our second century for the school of engineering with an amazing speaker which I think sets the stage for the next century of engineering and entrepreneurship. So, uh, I'm going to quickly introduce, uh, Brendan Foody, who co-founded Merkore, a recruiting startup that helps Silicon Valley's biggest AI labs train their models to do professional level reasoning by matching skilled workers with on enterprise projects. Earlier this year, Merkor announced that the company had hit 1 billion in annualized revenue, faster than any company, I suspect, that has ever done that in like less than two years or something. Crazy. Um Brendan and his three co-founders or two co-founders Surya Mida and Adar Haimath became the world's three youngest self-made billionaires. Kind of wild. Uh when private investors valued Merkore at 10 billion. Brendan founded a number of earlier ventures including Seros, a cloud consulting and cloud access business. Brendan was also selected as a field fellow in 2024. And uh before that he studied business administration, economics at Georgetown before dropping out to become an entrepreneur. It's kind of wild. I suspect you've known I'm going to dive right in. I suspect you've known you've wanted to be an entrepreneur for much longer than when you were a sophomore. So, uh let's start with a question about donuts. I heard something about donuts and what you did with donuts in high school. Can you tell me a little bit more about that story? — Yeah, of course. Well, first of all, thank you so much for having me. I'm excited to be here and talk with everyone. I think unlike probably a lot of the speakers, I uh am right around the same age as all of you. I uh even recognize some high school friends uh and I just turned 23. And so it's um I think cool to uh be back in a classroom and uh around it all. But one of my favorite entrepreneurial ventures was when I guess ventures is a strong word, but when I was in 8th grade when I saw that Safeway Donuts were selling actually at the Safeway because I grew up uh I was born in Stanford Hospital, grew up not far from here and I saw that the Safeway donuts um on the Safeway on El Camino were selling for $5 a dozen and I felt like all my friends would pay way more than that for buying donuts. And so I started biking to the Safeway every morning. uh on Fridays, buying a couple dozen donuts, bringing them over to my middle school, uh selling them for $2 each, so running call it 60 70% margins, and then scaling up really quickly from there. So, uh my mom would charge me $20 to drive me in her minivan to Safeway. I would buy 10 dozen, 20 dozen donuts, sell them to all my friends. I remember competition popped up uh at my cuz margins invite competition, of course. And so then I uh but they were buying Chuck's donuts which have a higher cost basis of just over a dollar. And so I dropped my costs or my prices to a dollar for per donut for two weeks to drive them out of business. And this was while I was in eighth grade. And so it was before I learned about anti-competitive practices. Um and uh yeah, so I had you know I had a half dozen of those stories growing up. Um but that was one of my favorites. remind me never to buy a donut from you. Got to be careful. — You have to be careful. Yeah. — Um so let's keep moving and building on that story. So after all of these entrepreneurial experiences as a teenager, — um I heard that you actually didn't want to go to college, but you went anyways and then you actually had some you know your co-founders actually were with you during those university days. So what happened during freshman and sophomore year? We have a lot of freshman and sophomores in this class. — Totally. Well, and I'll give a little bit of the background as well for why I didn't want to go to college, which is that I met Adar and Suria, my co-founders, when we were 14 in high school, and co happened our junior year and I started uh getting into sneaker reselling a lot. I saw many of my friends that were reselling sneakers were using AWS to run automated checkout software that would help them to buy and sell items online. And then I started a consulting business where I would help them get AWS credits through this program AWS Activate a lot of people know about. Um, and then I scaled that up to a couple hundred,000 in revenue. And I was telling my mom like, "Why should I go to college to make less money than I'm making right now? "

### Segment 2 (05:00 - 10:00) [5:00]

— What did she say? What did she say in response? — Well, she I don't remember exactly. There wasn't a lot of reasoning around it other than just frustration uh that I needed to go to school. Um, and so I ended up going to college at Georgetown for two years. Siri was my roommate and Adar was at Harvard. And when we were getting started, it was really just that we felt like there were these extraordinaryly talented people internationally, particularly in India, that we wanted to hire to work on projects together. and we were excited about how we could build more efficient processes to make that happen and leverage AI throughout. — That's wild. Okay. So, uh there was a point in time where you hit like a million dollars in run rate and you're like maybe we shouldn't go back to school. Okay. Tell me a little bit more about that time. — Yeah. So we got started um in call it mid to late 2022 just working on projects uh with one another and then merornet's current form was founded in January 1st of 2023 where the premise was this was like after chat had come out and AI was just starting to work the premise was that similar to how humans review resumeums conduct interviews and decide who to hire we could automate all of those processes using LMS to hire people for ourselves and our college friends And so I would hustle sending out like contracts on Panda do to our friends charging them $500 a week taking 20% hiring a college student in India to help them coding up their side projects. And we hustled a lot scaling that to a million- dollar run rate within eight months. And then from there we it was a little bit complicated because I'm thinking again like my mom was really upset with me about going back to school. — How do you talk to your mom? I actually have probably students who are considering this like what do you tell your parents? — Well, so actually Georgetown notified her which was a little bit of a miscommunication on my side and so it didn't land the best. Um but I was very stubborn and so I told her um like no it's going to work just trust me. Um and then from there uh a few months later I actually met Shawn uh who I didn't know would be in the audience but uh Sean was working at OpenAI and what we realized was that there was this enormous transition happening in the human data market where it was shifting away from the crowdsourcing paradigm that companies like Scale AI pioneered of how do you get low and medium skilled people that can write barely grammatically correct sentences for the early versions of chatubt and it was moving super quickly towards the agentic data paradigm of how do we find super highskilled experts the bankers lawyers doctors fang software engineers that can work collaboratively with one another in teams to build frontier evals and oral environments to build the next generation of models and so I it's funny I remember Sean and I met in Barge Gemini ironically which was across the street from the OpenAI office. I was 20 years old at the time, so I was nervous about whether I'd have to use my fake ID because like uh the San Francisco bars are really strict on fakes. But for fortunately, we just got devild eggs and we shared this vision with Shod and um scaled up really fast from there. We became Open's largest data vendor within nine months back when 01 was shaking the world and head and shoulders above everyone else and then quickly became a primary agentic data vendor to all of the top AI labs as well as all the top application layer companies. Um and scaled from the million in revenue run rate to over a billion in 20 months — which is as of now the fastest growth trajectory in history. — Yeah. Wild. You're breaking records left, right, and center. I'm going to go back a little bit to Dar and Surya because you chose to co-found Marore. Talk about that. How did you divide and conquer because co-founding is very different than founding. You're not all just it's you're not calling the shots entirely yourself. — Totally. Well, it started when we were got to know each other really well in high school because we were on the speech and debate team together and Adar and Syria were the winningest speech and debate team in history where there's three big national tournaments in policy debate, the most competitive event. They were the first team ever to win all three of the national tournaments. I'm dyslexic and so I'm not nearly as good as them at debate, but I was always in awe of these, you know, legends that were shaping history and um aspiring to find a way to work together and we were really good friends throughout. And so then Siri was my roommate for two years and so we got to know each other in very close proximity. Um but one thing that we've talked about a lot is that the debate days and being on a team together was sort of like uh a

### Segment 3 (10:00 - 15:00) [10:00]

co-founding partnership um uh in an initial version where we were all bought into one another's success. Um and then I think it was just that I had the most fun working with them. Honestly, that's the most important thing is just being around the people that you're energized by is uh the key to sticking with it when things are difficult and doing your best work um when everything fits together. — What was the most challenging moment you had while being together as friends that then you took to founding together? — Interesting. There were a lot of moments when we would lose tournaments and debate and it felt devastating like where we would — uh put 10 hours a day into prepping for a tournament for weeks on end and all of a sudden in you know one round in single elimination like felt like our debate careers are over. — And so I think that we were always far more bought into debate than we were to anything academic. And so it was generally those experiences. Um and I think that helped to um yeah build more grit throughout the way. I remember every round Adar would always like freak out about whether he'd lost and he'd win like 99% of rounds or something like that, but he always thought he lost. And I always thought that was a really good mindset to develop of like the paranoia of what is everything that we could have done differently and better to um have improved the probability of winning. — Well, it's very clear to me that you are out there to win and that's how you've grown the company so quickly and that you know what to do after that 1% failure, right? When it doesn't work out, it sounds like you pick yourself back up together. — Yeah, absolutely. And I will say like obviously I talk so much more about the things that have gone well. There's been so many periods throughout the company where the future has been very uncertain. — Uh and you know of like when we're dropping out and we only have a couple of our like college friends that are customers and their businesses go under and all that kind of stuff. And I think that especially before strong product market fit, there's this very natural roller coaster of emotions um where the highs feel like the best day on earth and the lows feel like the worst. And so I think having them um by my side was certainly incredibly formative for that. — Okay. How did you guys divide and conquer? You did go to market. Dar was very close to India. Sounds like Syria took care of sort of the internal functions. How did you decide on that how to divide and conquer based on you know all the things that needed to get done? — Well, it's interesting. I actually coded our initial product. Um and I uh or I would code the front ends and then work with some Indian engineers on the backends. Um uh and so it's fun to like spin up the coding tools today. And then eventually once I started getting contracts, we figured a cleaner structure would be that Adar would be CTO at the time to manage engineering since he was studying CS and had spent the most time coding out of all of us. — Uh Surya was is uh very well connected to the IITs in India and so he set up a lot of our initial recruiting functions. Um, and I don't know, it just felt like a natural way that things panned out. — And is it still that way? How has it evolved since? — Not really. It's uh, it's taken many different forms cuz I feel like it's always about working on whatever the most important problem is at any given time. So, it's more dynamic. But the rough split right now is that I focus on the AI labs and really our frontier customers that are um you know building the best agents in the world that everyone uses every day. And then Adorch is focused more so on enterprises. How do we take all of the processes, knowhow, technology, etc. that we've built for the AI labs over the last two years and apply that to the rest of the Fortune 500. — Got it. Okay. You know, we've kind of putted around a little bit about exactly what Merore does and not everybody in this room is a computer scientist or working on AI. We have an incredible like breadth and diversity of students here. Could you maybe start from like square one for somebody who is a layman, a lay person, a student who isn't familiar with, you know, the AI labs like building models like could you tell us a little bit about, you know, what Merkore does and then I want to then tie that to benchmarking, right? So defining benchmarking as the most basic definition. — Totally. Yeah. So actually I'll also give a little bit of history alongside this. The first stages of data creation were behavior cloning data where the two

### Segment 4 (15:00 - 20:00) [15:00]

most common types are supervised fine-tuning data where you would have inputs and outputs for people that have played around with fine-tuning APIs and then RHF data where you would have the preference that you sometimes see in chatbt if the model generates a couple responses and you choose which you prefer. Um what eventually happened as we moved towards agentic data is that people started to create environments where they instead have the humans structure the success criteria similar to how a professor might create a rubric for your exam in a class of you know plus five points if the model does this plus 10 points if they do this minus five points if it makes this mistake and that could be in the form of rubric criteria or unit tests and then you're able to use that data to either measure how well the model is performing to see which model should we be using uh whether it's the newer model or the older model and as researchers experiment that would be a benchmark to measure success evaluate success and you see the benchmark scores whenever labs are announcing their new models. The other way that you can use it is as a reward function for reinforcement learning where you could have the model attempt the problem a 100 times or 50 times. You could score all of those attempts and the model can very quickly learn to hill climb that. And so that's like the more granular technical explanation. But what we do is we build out all of these rubrics, unit tests, and all the context and tools that models use to cover pretty much the distribution of anything you could do in chat GBT, cloud code, cloud co-work, or Gemini. — Amazing. Thank you. — Yeah. — Uh, okay. So, let's talk about Apex. I think that was a huge win. And it's not just one Apex. I know you started with, you know, investment banking and management consulting and, you know, uh, law and whatnot, but that's branched out to include agentic and include SWES and whatnot. Um, why is that important? Where is this all going? Yeah. So it actually ties back to I remember my first conversation with Sean, we talked about IMO medalists and in the early days of models, everyone was talking about very academic benchmarks to set the goals for models of how could we get IMO medalists to measure Olympiad math, how could we get PhDs in GPQA, uh all the way to how could we get, you know, II medalists for um various coding evals. And then there was this big transition in the market that we saw early on moving away from academic capabilities towards economically valuable capabilities of how do we measure the things that professionals do in their real jobs. How do we create the distribution of the prompts, the context, uh the rubrics or success criteria corresponding to that uh and measure that distribution across investment banking, law, medicine, uh software engineering, and all sorts of the most economically valuable domains. And so we built Apex with that purpose and it's quickly gotten adopted by top a lot of the top labs as uh becoming this industry standard for how labs determine whether their models are doing well at the capabilities that enterprises care about and how enterprises make buying decisions for what model they should be using. — Yeah. — How do you find like the authors? right people? — The experts that help to create these data sets, — the most important is making sure a that they're really high caliber, right, of we want to get the top people from uh any given program or uh organization. Uh and B that they cover a representative distribution of what our customers care about. Okay, — so say our customers care about being better at consulting. We would break that down into what are all the different kinds of consulting, whether it's uh management consulting or operational consulting, etc. Uh and then what waiting do we want to have in each of those categories and how do we build out the distribution that corresponds to all of that? Um and so it's really just — like the bottleneck to AI labs automating everything that happens in the economy. Yeah. uh and we can do in all of our jobs is how do they co build this network of human knowledge that covers the distribution of all of the context, all of the prompts and all of the responses that correspond to that. Um and so that's what we focus on doing. — Amazing. Okay, we're going to ask the hard question, — right? Uh once all of this expertise is in essence built into the large language models, what comes next? Let me put it this way. We have a lot of students here who are wondering why should they study CS and how are they going to find their first job or their first adventure out of college. So what would you say? — Well, I think that in every revolution

### Segment 5 (20:00 - 25:00) [20:00]

in productivity, people have underestimated how elastic productivity is. And what I mean by that is I think like as an example as a company if we make all of our software engineers 10 times more productive we'll hire more software engineers. Sure people might not just be coding they might be doing a lot of the things adjacent to software engineering of designing architectures of having taste in the product that we want to use um and setting the right goalpost for the agents. But I think that the investment in human labor to do all the things that models can't do is going to be profound. Um and while I would guess that probably within a few years agents will be able to do a significant portion maybe even a majority of what humans do in our jobs. I think it is a multi-deade process to actually achieve super intelligence where we automate absolutely everything that people can do because models inevitably will just struggle with these super long horizon challenging to verify tasks. — Yeah. Okay. This is a fun one. Uh we talk about artificial general intelligence. I know there's ASI, AGI, there's many ways of going about it. Um when that day comes, what will you be doing? — Well, I think I'll be doing the same thing. Uh but I do think there is a such a large nuance between uh I guess AGI people define usually as models being able to do most economically valuable work versus 100% of economically valuable work. Um and a great proxy for this is that a the Bureau of Labor Statistics analyzed that over the last 225 years we've made the average uh person 20 times more productive. M — so that's equivalent to automating 95% of what you do in your job. Yet at the same time we have more jobs in the economy than ever before. — And I think that will continue where so long as we don't automate 100% and we just automate 95% of what everyone does. We're not going to run out of things to do as a society. We're going to have so much more work. We'll build a thousand times more software. We'll build rockets. We'll cure cancer, go to Mars, etc. Um, and I think there's going to be immense human involvement in all of those processes. — You know, it's funny. Uh, we're experiencing these moments. Uh, the one that comes to mind for me is always the ATM. I think the banking industry was having a moment. Uh, thinking all these tellers would lose their jobs, but actually the financial industry evolved and became much broader and stronger as a result of I mean ATMs were just one thing in innovation in that field. So we're seeing that widespread now and it's helping people through this transition. — I totally agree. Yeah. And I think there's also enormous job creation associated with people building data centers, people training agents for deployed engineers that need to go deep with every Fortune 500 to implement agents. And so with every revolution, there's been some degree of creative destruction that happens. And ultimately humans always have and always will uh play a meaningful part in the economy. — Let me zoom out a little bit. Um what is the role of policy and government in a transition like this? — I think that there can be a lot of value to regulators helping to internalize the externalities to markets. And what I mean by that is sometimes markets don't price in that there are certain you know negative uh things that happen because of business outcomes like the most common example is carbon um from burning fossil fuel and like how could we price in that externality via a carbon tax. I think similarly probably the externality that's like most top of mind for the average American right now is jobs. Yes. — While jobs are like the largest positive thing for everyone's lives in the economy, they're also the largest thing that regulators disincentivize. And so I think figuring out how regulators can best incent through like payroll taxes, income taxes, especially at the low end, I think those are not great. But figuring out how regulators can better incentivize all these positive things that we want more of like new job creation I think is um one of the most important things. — Amazing. Okay. So um in the next year we'll end with this question and then we'll hand it to students. Uh actually we'll have one more question after this. What are you most excited about? You don't have to disclose anything but like what are you excited about in terms of where AI is going? I'm really excited about how it gets applied to the rest of the economy. — Yeah. — Because I think it's easy to be in these like uh echo chambers in Silicon Valley where we think that, you know, AI has 100% adoption when we talk with the

### Segment 6 (25:00 - 30:00) [25:00]

average company in the Midwest. They, you know, just started using chatbt a few months ago. Um, and I think that getting like really permeated throughout every business workflow in addition to every consumer is going to be really exciting. And I think in the context of Merkore that's expanding from just serving the frontier model companies to serving every enterprise in the world and helping them build uh the evals that correspond to their value chain so that they're able to implement and apply agents throughout everything that they do. — All right. So final question which is the classic question we always answer uh ask at the end of these chats is uh um from Tina Celic who started the series about 30 years ago. She wrote a book called What I Wish I Knew When I was 20. I know that wasn't that long ago for you. That was you know you just turned 23. Happy birthday. Um if you could go back to three years ago. Um I've actually I have like two questions here. What would you tell yourself and your mom? — What would I tell my mom? Uh, I I'll think about that one. But I think one of the most important things people don't talk about is focusing on leading indicators when markets are moving really fast. Like if I had to distill what we did strategically into one thing, it would be that. And that we saw, hey, the cutting edge frontier labs see this thing in the markets where we can learn from them about, you know, how the entire economy is going to change shape. And if we just focus on those leading indicators, that gives us a window into the future. And so I would encourage all of you to try to see like what are those leading indicators and how all of these markets whether it's data or compute or modeling are changing extremely quickly and um how do you use that as a window into the future? Um in terms of what I would tell my mom um I'm not sure. I think I would tell her I'm probably something similar. Don't worry. It'll all work out. — Yeah. Truly it has. Okay. So we're going to organize ourselves to do Q& A. — What is something that makes like the human and not cannot be automated through AI? — One thing that I think everyone is over overestimating progress on is the physical world. Like the majority of work that happens uh in terms of dollars is in the physical world whether it's electricians or mechanical engineers um or all of the physical world services. And I think building out the distribution of data for how agents will learn how to do all of that is going to take an extremely long time. And so I think understanding um yeah, what are all the opportunities to accelerate that? What are the things that are going to take a lot longer to do and how that shapes markets is going to be really important. — Hi, thanks for being here Brendan. Um I have two quick questions. Firstly, how does it feel to be um the youngest billionaire in the world? And uh secondly, I was curious what are sort of the blocks of constancy throughout your day. What are the sort of key habits you do every day that has made you into who you are today? — So in the first one, I will say I really don't think about it too much, but if I had to choose two things, it would be first of all very surreal like you know my friends in college just graduated in uh June of last year. So I'm like roughly the same age as most of you and it's obviously beyond my wildest imaginations. And then second of all is very fortunate like so much of what has happened has been a function of having incredible co-founders an incredible team having you know a degree of fortune associated with how we were positioned in the market and um I I'm super grateful for all of those. And then your other question was habits right? Yes. — On habits, maybe if I had to choose one thing, it's that I think people try too hard to be disciplined to do things that they don't necessarily want to do when a lot of the key to success is finding the thing that you can be obsessive about. Like I remember when I was in college, I would have a hard time motivating myself to cover your ears, like turn in homework on time, attend a class, all that kind of stuff. And it was really the key was just like finding the thing that I couldn't stop thinking about where whenever I'm like with my friends or trying to do something else like my mind comes back to Merkore because it's the place where I can have the most impact in the world and make a dent in the universe. And I think that find if you find that thing that you can be deeply obsessive about everything else follows. And that's also what the teal fellowship looks for actually is they look for traits that someone is extremely obsessive because that's one

### Segment 7 (30:00 - 35:00) [30:00]

of the largest predictors of the end outcome of the company. — Thank you. — Yeah, — thank you for being here Brendan. So my question is when you were playing around with all these different projects, what was like the initial ideation process like? Like how did you decide which projects to pursue? And my second question is about the teal fellowship. How did they kind of support you in building what you built today? — Yeah. So on the first one, I would say we were always very customer oriented without with all of them of how do we have our early design partners that can give us feedback on the product um across each of the different initiatives. And that was certainly foundational to being intellectually honest for whether what we were doing was actually useful. Um, and so I'd encourage everyone to always try to spend as much time with your users of any products you're building or the customers of any business because that is by far the most important thing uh to get right in terms of so what was your second question? — Um, the Teal Fellowship. Yeah. — Yeah. It's actually funny because it came later. it. We didn't get the teal fellowship until March of 2024 and that was after things were the company was at like 2 million revenue but we knew that we would be at 50 by the end of the year and so things were really just like on the dawn of starting to work and probably the largest value ad was I think the credibility stamp to investors and hires as meaningful but also the community like I met um some of my best friends my friend Yash who Sean knows uh there who's the CEO of applied compute and was previously our customer at OpenAI. Um and so those are the two things probably the credibility stamp and then the community. — Thank you. — Thank you so much for being here today Brennan. um when you started Meror uh it was around a time when scale already owned a lot of the narrative around AI data and training and I'm curious what gave you and your team the conviction to kind of go for it and assume that the market's not fully taken as opposed to go and work for say Alexander — Wingh — well it's funny because I skipped over part of the story earlier which is before I met Shawn our initial intuition was that we should actually partner with scale and have them become a customer of ours. And so we hired over a thousand software engineers for them using our platform and our technology. And then we got flooded with support tickets that these people weren't being paid and they were like Merkor is a scam because they referred us this opportunity that is scale AI. And what we realized was that there was this gap in the market in part because it was changing really quickly. like scale grew up in autonomous vehicle labeling with people in the Philippines and a lot of the infrastructure was built around that. And if we went directly to the labs, we had the opportunity to build things from the ground up around exceptional people creating this new paradigm of agentic data. And so that was really the moment that it all flipped for us. — Thank you. M — uh hi Brandon, thank you so much for being here. So since you missed uh two years of college, I thought I'd ask you something related to the education system. Um what do you think that education systems around the world can do in order to cope with this wave of agentic AI? And do we need to change the way we actually evaluate human skill at this base level or do you feel like the current system is preparing students to actually take on a world that will look vastly different to the one we inhabit during our schooling? — I think a lot about this because we do I mean we've done over 6 million AI interviews. So I think a lot about how do we effectively assess people um in all sorts of different scenarios. And I think the most important thing for educators to get right is to lean into assessing people using the technology rather than stopping the technology from happening. Like in the initial waves of AI, we saw all of these things of like GPD0 try to stop the use of the models uh which is almost analogous to like stopping people from using calculators or computers. And the better way is how do we encourage everyone to use the technology as much as possible and see what they can do with it. Like now all of you can build a production scale app using cloud code in a week. And so how do we that can be like a oneweek project which is insane and has never been possible before. And so how do we um you know just set these really ambitious targets and goals for students that force the adoption of the technology and

### Segment 8 (35:00 - 40:00) [35:00]

the exploration of what's possible with it. And just another uh question to lead up on that. Was there a moment you had during your founding process that made you realize that the whole agent AI landscape was actually moving in this direction? Because today if we look at it we have you know so many tools like you know open claw that's just a list of them. Um so yeah was there one moment that made you actually realize that okay the whole um AI agent revolution is actually going to be the next kind of change in the whole AI landscape? — Definitely. Actually, I mentioned Yosh and I learned a lot from him um in 2024 and one of the things I guess I only talk about this because this is like over or two years ago at this point and so it's not like you know totally locked down confidential but one of the things we saw with deep research was that the it made very clear that the model is the product. — Yes. And the way to solve these problems is not through stitching together a bunch of API calls with drag and drop processes uh and if then statements. The way to solve the problem uh in almost every knowledgework domain is how do we lay out the end goals for the models or the agents the eval and then how do we just train the model to get really good at those end goals. And uh I think it was work looking at the leading indicator working super closely with him um that gave us that window into the future before everyone else. And I think now since Obus45 everyone is really realizing that because that's like you know great example of the model is the product and the models are just getting so good that uh it really makes sense to lean into that paradigm. — Thank you. Hi, you mentioned that as we automate most of today's tasks, humans will migrate to like other kinds of jobs. So I was wondering what you think those jobs will look like. — I think there's a lot in the physical world. Um but maybe actually I'll break it into two categories. The first category is new jobs that get created and then the second category is things that as an economy we underinvest in. In terms of things that get created, probably the top three that come to mind are data center creation are for deployed engineers or deployment strategists that you probably see at like every AI startup to help enterprises and companies adopt agents. And then the third is training agents where we have you know huge volumes of people paying out many millions of dollars a day to create these de data sets are all environments and eval um for uh helping to improve model capabilities. That's the first bucket of new jobs that get created. And there's also all these things that we just haven't been able to invest enough brain power in as a society of how do we cure cancer and solve climate change, right? Like think about if we could have 40 or 50% of like Stanford's graduates working on stuff like solving climate change in 10 years. That would be so impactful. Um and so I think freeing up people's bandwidth and economic resources to mobilize towards those things will also be another huge benefit. Um another an analogy it made me think about is 200 years ago 75% of Americans were farmers. — That's right. — Right. And as we made farming dramatically more efficient. It just made it so that we could work on all these other things like literacy and medicine etc. that uh we had underinvested in as a society previously. Okay, we will take we're gonna try five more questions. We're gonna try. So, um — I can answer quickly. — Are you can you? Okay. — Hi. Um I'm just curious what are your thoughts on like the ethics behind AI and like recruitment specifically just because it kind of goes based off of past information and then even more specifically for women or other minorities. — Totally. Well, I think it's an incredibly important problem and I have sort of two thoughts on this. first is that instead of using like heruristics for past things, I think grounding everything in performance data of how people actually do at the job uh in the purest sense is the most effective way to make an assessment of like people that have this skill or perform great on this assessment um is the right way to do it rather than whatever arbitrary resume signals people have. The second thought is that the most unfair thing about recruiting processes is that the vast majority, 99% of people can't even engage, right? Like the average one of us, if you submit an application to a hot company or Frontier Model Lab, they don't have the bandwidth to interview. review resumeums, to run you through all of

### Segment 9 (40:00 - 45:00) [40:00]

their on-sites. And the reason is that the entire process is manual, right? people have to manually review resumes, manually conduct interviews, manually figure out who to hire. And so I believe that when you can solve this matching problem at the cost of software, it can enable everyone to be considered for every job and solve the largest inequity in every recruiting process. — Okay. Thank you. — Yeah. — Um I was just curious like you guys went through prod. Um I was wondering like what your experience was with prod not going to like YC or one of the traditional accelerators. It was quite early in tune in prod's history. Right. — I would go so far as to say Meror might not exist without prod. Yeah. — It has been so unbelievably formative in everything like all of our initial customers were fraud teams. my friend Ben who was the founding CEO of Any Sphere after he connected Michael and Aman with uh Arvid and Swale uh founded this company Coofactory where they were an early customer uh Rob and Gavin from Etched and a bunch of others. So I can't recommend them enough. — Thanks. Hey, so um a lot of people are there there's a lot of potential upside coming with AI and we've been talking about some of that today, but there's a lot of people who are also worried about the potential uh downsides of AI. So just to you um what is your greatest fear? What keeps you up at night um over the next 10 years or what's in keeps or scares you the most? And what are you trying to do about it? — I think at a societal level and probably Meror as well, there will be a lot of job displacement. Like I say all these optimistic things about how I think in five or 10 years it's clear that things will be dramatically better and there are going to be lots of jobs that go displaced. I think in a lot of ways the most important problem in the economy is how do we navigate that in a few years we might be able to automate 50% of the work that knowledge workers do and I think that so much of my goal at Meror is how do we help people navigate this problem how do we build benchmarks that give us this window into the future of what are the jobs that AI can actually do and it can't do for a very long time. How do we create these new job categories and how do we work with um enterprises and policy makers to uh create awareness about both of those things? — Great. Thank you. — Hi, thank you so much for being here today. Um, I was wondering like when you're hiring like these top consultants or top like investment bankers, did you ever have to like convince them to like you know like feed data to your models and also like potentially perhaps like lead to the downfall in the future? Like was that ever an issue or was it more just like we'll pay them more? I think that one of the large motivations is people love being at the frontier of the technology using the unreleased models and finding the places where they fail. And the vast majority of people on the project share the mindset that we have that there's not a shortage of things to do in the economy. And when we make everyone twice as productive will accomplish dramatically more and so that tends to not be a concern for the people that we work with. Hi. — Hey. — Uh my question is do you think university is necessary or get big help on your entrepreneurship? — I don't think that it was. I think that there are definitely communities of people that I met through university that were super impactful. like a great example is prod wouldn't have existed without um sort of the Harvard network that Adar was in and um so I think the communities that are formed around universities are super impactful but I don't think that most of the value really came from like the academic structure um necessarily — thank you I have same thought — there are I don't mean to say that like the entire academic structure is useless But uh there are there's a lot of value to it. — Okay. — Awesome. — Uh hi. So I'm just wondering like you mentioned about leading indicators and looking for them. So I'm just wondering like what are some concrete examples and what are some patterns and what's like the difference between something that's an actual leading indicator versus something that might be like uh red herring. So I think I'll say the best leading indicator is associated with dollars being spent because people can say something will happen or say they want something very often but until they actually put their money where their mouth is, you don't know that thing is true. Uh and a

### Segment 10 (45:00 - 47:00) [45:00]

great example of a leading indicator is what's happening in compute right now, right? where there's just like so much demonstration from the most sophisticated labs in the world that they will need to scale up the amount of compute that they have by orders of magnitude. Um, and there's so many granular leading indicators like the fact that everyone is investing in alternatives to Nvidia and we're moving to this multi-chip future is really interesting and there's going to be a lot of technology built around uh routing between different chips. Uh, I think there's so much that's going to happen in the data center buildout, whether it's shortages of electricians or uh, construction and all these different things. And so, you can dig into a lot of the granularity of these, but I think that's one of the most interesting ones to look at. — Okay, we're down to our last minute and uh, before I close and thank you properly, I want to thank all of you for being here, tuning in to another seminar. Uh next week we have uh actually the OGs. So similar to Brendan, you know, founded a company with two of his best friends. We have Kit Rogers, Benjun and Paul Coker come and join us. They started a company called CRI which uh was bought by Rambus in 2014 for a very large sum like 342 million. They were also the first uh Mayfield fellows. So for those of you do I have any Mayfield fellows in the room? Okay. So uh they were part of the first Mayfield Fellows cohort that is hosted by the Stanford Technology Ventures program. So uh we'll have a similar story but looking 30 years back uh and seeing where they are today. Uh Paul invented a lot of the security layers on the internet. So uh be back here next Wednesday. But without further ado, let's give Brendan a huge applause. [applause and music]