The Exchanges

Q What was the worst product decision you made?

A At one point, uh, Brendan, Siri, and I all thought that chat was the future of all UI at that time, a while back. So one of the initial iterations of the Mercore product was just built around a chat interface. Like there was pretty much no other way to hire people unless you use the Mercore chat bot, because we were so bullish on, on chat. Um, I think we've come around To that. We now, like, mix chat with, with other things, uh, where applicable, or leverage LLMs in other ways, but for a while we thought, like, the concept of a web app tomorrow would be dead, and the way you would interface with all web apps would just exclusively be with chat, so it wouldn't even be, you know, you clicking a button to hire someone, it would be you telling the chatbot to hire the person. I think, you know, it's possible down the line, but we, we may have mistimed a little bit.

AI assessment note: “one of the initial iterations of the Mercore product was just built around a chat interface.”

Q November. I quoted it wrongly. It's, you may be able to correct me, but it's much more now. Um, with 30 people at the time of the fifty million, and I've heard a little rumor on the grapevine that you do nine, nine, six, so nine a.m. to nine p.m., six days a week. Can you unpack, if that's true, why you do it, and how that works in reality?

A Yeah, yeah. It, it's really funny. Um, a lot of people ask me this question about the nine nine six thing. The only reason we actually just floated those numbers out is because we didn't want our team working on Sundays. Um, so I like to think of the nine nine six stuff as more of like a side effect than an objective. We've just really, really carefully selected for working with people who care deeply about the mission. And the side effect about that is They don't want to wait until Monday to, to move the company forward. Um, so people really do it just because they enjoy being in each other's presence. They enjoy what they're working on.

AI assessment note: “I like to think of the nine nine six stuff as more of like a side effect”

Q A lot of young exceptional people are being told today that they shouldn't maybe study CS anymore because actually CS is becoming so automated. Uh, 41% of code is now written by AI in five years time. That'll be extortionately higher. Do you agree with that advice? And how do you think about whether or not young people should learn programming today?

A My take is that programming is actually more important today And it's just going to happen at a different level of abstraction. One could argue that the leap from assembly to Python was actually maybe even a bigger leap than the leap from Python to natural language. So my answer there is that the way we define programming will look very, very different. It may be a person who, you know, has like average skills by today's standards and in computer science, Who's orchestrating thousands of superhuman coding agents to achieve more than we thought, uh, was even possible. Um, but that skill set, which, you know, we can define as programming at a different level of abstraction, programming in English, is gonna be super important.

AI assessment note: “programming is actually more important today And it's just going to happen at a different level”

Q It's 2035, okay? Final one. Where is McCaw then? Paint that picture for me of how big you are? How many people you've placed? Where is McCore?

A I have to work backwards, um, a little bit, right? So how many job seekers are there, right? You know, roughly put it in a couple billions. Um, how many jobs do, does each people, you know, does each person take on, right? People change their roles. So, you know, maybe we factor out all the jobs McCore creates for AI agents and like roughly focus on just the jobs for people, create Couple dozen jobs for each person. Mercor has created a hundred billion jobs and has built the, the unified labor marketplace, meaning that anytime a company wants to hire a person for a specific job or task, They do it through Mercore. And anytime a candidate wants to consider a company for a specific job or task, they do it through Mercore. And Mercore is able to solve the matching problem across every role, across every company in a seamless way.

AI assessment note: “Mercor has created a hundred billion jobs and has built the, the unified labor marketplace”

Q Why then does so many people tell me, including Jonathan at Grok, that actually synthetic data is often more high quality. It doesn't involve the dregs of the internet like Reddit, um, in a lot of cases being included. And actually you'll see this exponential increase in model performance due to actually mostly using high quality synthetic data. Not low quality human data. Why is that wrong?

A So the first thing is that it's not zero sum, right? Even in a world where human data is super important for the next generation of models, it doesn't mean that synthetic data won't also be important. So synthetic data will certainly be a part of the equation, but in a lot of ways, the bottleneck to unlocking and unleashing the next level of intelligence will be expert humans, which brings me back to the question and phrase that you used, low quality human data. Low quality human data certainly won't push the models to be better at anything. Uh, high quality human data will. And again, that's a talent assessment problem. The biggest lever on data quality for creating these post training sets, for example, is finding the right people, um, which again is, is really, really hard to do.

AI assessment note: “the bottleneck to unlocking and unleashing the next level of intelligence will be expert humans”

Q At the time you were like, 1819, and you can correct me if I'm wrong there, but you, Brendon Suria, get interested in labour markets. How does that happen?

A Actually, so Brendon Suria and I started working together, um, without any business ambition. Necessarily. We, we just started a dev shop together. So we were like, cool, you know, let's learn how to build software really, really quickly. Let's go to these startups. Let's figure out things they want built. Let's build it together. And what we ended up doing is recruiting these really, really exceptional folks from India to help us out with our dev shop. And then very, very quickly we realized, you know, the software was one thing, but we had found some really, really exceptional people and it was more about the people than the software. So then we were like, okay, We found these people in a completely manual way. Can we automate this? Uh, and that's how the automated candidates side of the platform was born. And then very quickly we realized Brendan sir, and I couldn't scale well by doing sales manually. So then we had to automate the other side of the platform to the, the company facing platform. And that's how the marketplace was born.

AI assessment note: “we had found some really, really exceptional people and it was more about the people”

Q At the time you were like, 1819, and you can correct me if I'm wrong there, but you, Brendon Suria, get interested in labour markets. How does that happen?

A Actually, so Brendon Suria and I started working together, um, without any business ambition. Necessarily. We, we just started a dev shop together. So we were like, cool, you know, let's learn how to build software really, really quickly. Let's go to these startups. Let's figure out things they want built. Let's build it together. And what we ended up doing is recruiting these really, really exceptional folks from India to help us out with our dev shop. And then very, very quickly we realized, you know, the software was one thing, but we had found some really, really exceptional people and it was more about the people than the software. So then we were like, okay, We found these people in a completely manual way. Can we automate this? Uh, and that's how the automated candidates side of the platform was born. And then very quickly we realized Brendan sir, and I couldn't scale well by doing sales manually. So then we had to automate the other side of the platform to the, the company facing platform. And that's how the marketplace was born.

AI assessment note: “recruiting these really, really exceptional folks... realized it was more about the people than the software”

Q Why then does so many people tell me, including Jonathan at Grok, that actually synthetic data is often more high quality. It doesn't involve the dregs of the internet like Reddit, um, in a lot of cases being included. And actually you'll see this exponential increase in model performance due to actually mostly using high quality synthetic data. Not low quality human data. Why is that wrong?

A So the first thing is that it's not zero sum, right? Even in a world where human data is super important for the next generation of models, it doesn't mean that synthetic data won't also be important. So synthetic data will certainly be a part of the equation, but in a lot of ways, the bottleneck to unlocking and unleashing the next level of intelligence will be expert humans, which brings me back to the question and phrase that you used, low quality human data. Low quality human data certainly won't push the models to be better at anything. Uh, high quality human data will. And again, that's a talent assessment problem. The biggest lever on data quality for creating these post training sets, for example, is finding the right people, um, which again is, is really, really hard to do.

AI assessment note: “the bottleneck to unlocking and unleashing the next level of intelligence will be expert humans”

Q November. I quoted it wrongly. It's, you may be able to correct me, but it's much more now. Um, with 30 people at the time of the fifty million, and I've heard a little rumor on the grapevine that you do nine, nine, six, so nine a.m. to nine p.m., six days a week. Can you unpack, if that's true, why you do it, and how that works in reality?

A Yeah, yeah. It, it's really funny. Um, a lot of people ask me this question about the nine nine six thing. The only reason we actually just floated those numbers out is because we didn't want our team working on Sundays. Um, so I like to think of the nine nine six stuff as more of like a side effect than an objective. We've just really, really carefully selected for working with people who care deeply about the mission. And the side effect about that is They don't want to wait until Monday to, to move the company forward. Um, so people really do it just because they enjoy being in each other's presence. They enjoy what they're working on.

AI assessment note: “I like to think of the nine nine six stuff as more of like a side effect”

Q In terms of like the post-training data side, I'd just love to hear your thoughts on how much will be human data versus how much will be synthetic data moving forward?

A I think a lot of it will be human data going forward, and I think a great example of this is evals, right? Evals for models definitionally have to be outside of model capability, right? In order to see whether model is doing well at a particular task, You need to have an eval set created by humans that is better than the model at that particular task. And humans are going to play a huge role in that, for example. And I think there are a whole set of other use cases, whether it be SFT, RLHF, you know, RL environments, like how the models of tomorrow are being trained that all require these expert humans, uh, to essentially teach the model how to get better.

AI assessment note: “I think a lot of it will be human data going forward”

Q That is very, very kind of you, dude. But I did my stalking beforehand, and everyone told me about your mastery of debating. You and your co-founder, Surya, you were debate champions. How did debate prepare you for founding a company? Let's start there.

A Yeah, well, I mean, one thing about Brendan, Surya, and I is that we actually go quite a ways back. So, I actually first met Surya when I was 10 years old. Um, and the reason we got along so well is because we were pretty much the only elementary schoolers who wanted to compete in high school debate. So at the time, we did Lincoln-Douglas debate, which is sort of like a one-on-one debate format. Suri and I actually even debated each other a couple times. And then we ended up at the same high school, Bellarmine, which is also where I met Brendan. Um, and then all three of us were on the debate team together. Suri and I decided to do policy, so we ended up being debate partners together. And then going on and competing in all of these natural national tournaments. Um, but, but debate is a lot like founding, uh, in a lot of ways, right? Like I like to think of my debate partnership with Surya as sort of my first startup, just because we had fifty-fifty equity in each other's success. If one of us were to mess up, it would tank the odds for both of us. Uh, there's like this constant feedback loop after every debate round about whether you won or lost. Picking the right debate partner is like The most important decision you can make in policy debate, and similarly picking the right founding team is the most important decision you can make while starting a company. So there's that pa…

AI assessment note: “debate partnership with Surya as sort of my first startup”

Q In terms of like the post-training data side, I'd just love to hear your thoughts on how much will be human data versus how much will be synthetic data moving forward?

A I think a lot of it will be human data going forward, and I think a great example of this is evals, right? Evals for models definitionally have to be outside of model capability, right? In order to see whether model is doing well at a particular task, You need to have an eval set created by humans that is better than the model at that particular task. And humans are going to play a huge role in that, for example. And I think there are a whole set of other use cases, whether it be SFT, RLHF, you know, RL environments, like how the models of tomorrow are being trained that all require these expert humans, uh, to essentially teach the model how to get better.

AI assessment note: “I think a lot of it will be human data going forward”

Q For a lot of students who are wanting to start a business, who have a business already, how do you advise them on whether to drop out or whether to stick to the traditional path?

A Oftentimes it's, it's an emotional decision. Like you can try to rationalize dropping out or, ah, starting a company or try to figure out the exact, you know, set of prerequisites that you have to do. But like for me, for example, the moment I knew that I wanted to drop out was, was actually back when we had an office in Palo Alto and the office had exactly three desks, one for Brendan, one for Syria, one for me. And I was like, Surya, man, should we, should we drop out? And then he just looked at me, and he was just like, dude, how hard could this be? Wasn't a logical argument at all, but in that moment, I was just like, let's do this. Let's, let's drop out of school.

AI assessment note: “Oftentimes it's, it's an emotional decision. Like you can try to rationalize”

Q I think when people- I'm so sorry to be like what? Just help me understand. You're teaching me.

A Yeah, yeah, yeah. It could be, you know, well, maybe we could take a step back, you know. And just talk about, you know, labor in general. If we reach the point a couple of years, a hundred years from now where the models are able to do every single job and humans no longer have any work to do, society is going to look really, really different, right? Well, we're all going to be living on a UBI. We're all going to be playing video games all day, whatever it may be. But until that point comes, there's going to be a whole set of tasks that the models cannot do, whether they be specific, you know, economically valuable tasks, like maybe the job that a consultant could do, or maybe a specific category of engineering or even like more niche things, right? Maybe it's making the model better at like some specific hobby, uh, for example, and we're always going to need to fill in the gaps, particularly in that long tail. And Harry, the other thing I'll say is, I think people are really, really in this mindset of like, This, like, unidirectional relationship between humans and AI, right? Where I can't do something, I give it to the AI, the AI takes it to completion. But I think the more realistic breakdown is the AI for a specific use case might be able to get us 6070, 80% of the way there. But for that remaining 40, 30, 20%, you're going to need a human To be able to, to take you all th…

AI assessment note: “AI for a specific use case might be able to get us 6070, 80%”

Q Take me to the seed round, dude. Like, how did it go? Do you remember getting the term sheet? Just take me to that. Cause you were 1819 at the time?

A Yeah, I think we were, we were 19 at the time. So that was, that was just surreal. So what ended up happening is initially we thought we wanted to base the company in New York. So I'll take credit for making the, the wrong, uh, call there. I very, very quickly realized that it was the wrong decision. But what ended up happening is we had moved to New York before Raising the seed round. And for me, actually, the more surreal moment wasn't actually when the money hit for, for the seed round. It was actually when we changed our, uh, like salaries and gusto to 500 dollars a month. I felt like we made it at that time. I was like, amazing. You know, we just moved to New York. We changed our salaries to 500 dollars a month. And then afterwards we closed our seed round. Um, and then when the money was wired, we were just looking at the account, like, you know,

AI assessment note: “we closed our seed round. Um, and then when the money was wired”

Q For a lot of students who are wanting to start a business, who have a business already, how do you advise them on whether to drop out or whether to stick to the traditional path?

A Oftentimes it's, it's an emotional decision. Like you can try to rationalize dropping out or, ah, starting a company or try to figure out the exact, you know, set of prerequisites that you have to do. But like for me, for example, the moment I knew that I wanted to drop out was, was actually back when we had an office in Palo Alto and the office had exactly three desks, one for Brendan, one for Syria, one for me. And I was like, Surya, man, should we, should we drop out? And then he just looked at me, and he was just like, dude, how hard could this be? Wasn't a logical argument at all, but in that moment, I was just like, let's do this. Let's, let's drop out of school.

AI assessment note: “Oftentimes it's, it's an emotional decision. Like you can try to rationalize”

Q Take me to the seed round, dude. Like, how did it go? Do you remember getting the term sheet? Just take me to that. Cause you were 1819 at the time?

A Yeah, I think we were, we were 19 at the time. So that was, that was just surreal. So what ended up happening is initially we thought we wanted to base the company in New York. So I'll take credit for making the, the wrong, uh, call there. I very, very quickly realized that it was the wrong decision. But what ended up happening is we had moved to New York before Raising the seed round. And for me, actually, the more surreal moment wasn't actually when the money hit for, for the seed round. It was actually when we changed our, uh, like salaries and gusto to 500 dollars a month. I felt like we made it at that time. I was like, amazing. You know, we just moved to New York. We changed our salaries to 500 dollars a month. And then afterwards we closed our seed round. Um, and then when the money was wired, we were just looking at the account, like, you know,

AI assessment note: “afterwards we closed our seed round. Um, and then when the money was wired”

Q What role are you best at? What role are you worst at?

A It's an interesting question because we place all kinds of talent. At companies, right? Everything from software engineers, to lawyers, to doctors, to financial analysts, to consultants. So like a huge part of the Mercore platform is actually not like building specifically for any of these roles, but instead building technology that generalizes really, really well, right? You know, one example is the AI interviewer. We, we've built it in such a way that it can immediately pre-process Someone's background and then administer a custom interview to a person, regardless of what role they're, they're trying to take on, um, in a completely automated way. You can literally spin up this interview in under 10 seconds. So, you know, for example, for this, you know, podcast, right? You, you must have spent like a decent amount of time doing research, but imagine like you can just have an agent pull in all the information on someone's profile and put together what would be the superhuman interview or the superhuman podcast. That stuff is possible now, uh, and it's possible for pretty much all roles.

AI assessment note: “It's an interesting question because we place all kinds of talent.”

Q What does no one tell you about scaling that you wish they'd told you?

A Scaling culture is Harder than scaling software. When you're adding people to the team very, very quickly, there, there's this dynamic that the culture that you create with the first, you know, 20 people is in some ways the strongest the culture is ever going to be, and ensuring that that culture stays strong as the company grows, does new things, and new people enter the company is really, really challenging, but in some ways the most important. Actually, our insight about the market is that human data and talent assessment have actually become the same thing, right? Where, you know, I can take you back five years where when we think of this data labeling or human data stuff, it's essentially a crowdsourcing problem, right? Let's say Waymo wants a bunch of their images labeled. You get a bunch of people across the world to draw boxes around stop signs to make the model better at classifying stop signs. But Fast forward to today, and the nature of human data work has changed a lot. Now it's GPT, Foro, or whatever model is not good in a particular domain, so we actually need an expert to make the model better in that domain, and figuring out who that expert should be is 100% a talent assessment problem, uh, and is a perfect application of the platform. With a lot of the labs that we work with, uh, we're able to figure out who are the exceptional people, In very, very specific do…

AI assessment note: “Actually, our insight about the market is that human data and talent assessment”

Q What role are you best at? What role are you worst at?

A It's an interesting question because we place all kinds of talent. At companies, right? Everything from software engineers, to lawyers, to doctors, to financial analysts, to consultants. So like a huge part of the Mercore platform is actually not like building specifically for any of these roles, but instead building technology that generalizes really, really well, right? You know, one example is the AI interviewer. We, we've built it in such a way that it can immediately pre-process Someone's background and then administer a custom interview to a person, regardless of what role they're, they're trying to take on, um, in a completely automated way. You can literally spin up this interview in under 10 seconds. So, you know, for example, for this, you know, podcast, right? You, you must have spent like a decent amount of time doing research, but imagine like you can just have an agent pull in all the information on someone's profile and put together what would be the superhuman interview or the superhuman podcast. That stuff is possible now, uh, and it's possible for pretty much all roles.

AI assessment note: “It's an interesting question because we place all kinds of talent.”

Q When you look at candidate completion rates, how much of that is India versus rest of world today? I know you specialize in finding amazing talent in India specifically.

A So the reason we started with India is because, you know, our parents, uh, immigrated from, from India, Suri and I, so, uh, they went to these amazing schools. So we like started these recruiting campaigns from those schools specifically. And actually like one of the things that got us really, really excited about, you know, labor markets in general and the inefficiencies associated with it was just because like one of the best engineers I've ever worked with on our team, we found through a Facebook ad and I manually interviewed him and actually he didn't pass the interview. Uh, but the reason that we ended up hiring him is he sent me a really, really long message about what exactly he got wrong in the interview and how to Correct it. And I just felt like we got to work with him. It was sort of that that prompted us to start in India. But if you fast forward to today, actually the number one place that, you know, workers on the Mercore platform who, you know, have jobs through, through us, um, are from is actually the United States.

AI assessment note: “number one place that, you know, workers on the Mercore platform... are from is actually the United States.”

Q What does no one tell you about scaling that you wish they'd told you?

A Scaling culture is Harder than scaling software. When you're adding people to the team very, very quickly, there, there's this dynamic that the culture that you create with the first, you know, 20 people is in some ways the strongest the culture is ever going to be, and ensuring that that culture stays strong as the company grows, does new things, and new people enter the company is really, really challenging, but in some ways the most important. Actually, our insight about the market is that human data and talent assessment have actually become the same thing, right? Where, you know, I can take you back five years where when we think of this data labeling or human data stuff, it's essentially a crowdsourcing problem, right? Let's say Waymo wants a bunch of their images labeled. You get a bunch of people across the world to draw boxes around stop signs to make the model better at classifying stop signs. But Fast forward to today, and the nature of human data work has changed a lot. Now it's GPT, Foro, or whatever model is not good in a particular domain, so we actually need an expert to make the model better in that domain, and figuring out who that expert should be is 100% a talent assessment problem, uh, and is a perfect application of the platform. With a lot of the labs that we work with, uh, we're able to figure out who are the exceptional people, In very, very specific do…

AI assessment note: “Actually, our insight about the market is that human data and talent assessment”

Q When you look at candidate completion rates, how much of that is India versus rest of world today? I know you specialize in finding amazing talent in India specifically.

A So the reason we started with India is because, you know, our parents, uh, immigrated from, from India, Suri and I, so, uh, they went to these amazing schools. So we like started these recruiting campaigns from those schools specifically. And actually like one of the things that got us really, really excited about, you know, labor markets in general and the inefficiencies associated with it was just because like one of the best engineers I've ever worked with on our team, we found through a Facebook ad and I manually interviewed him and actually he didn't pass the interview. Uh, but the reason that we ended up hiring him is he sent me a really, really long message about what exactly he got wrong in the interview and how to Correct it. And I just felt like we got to work with him. It was sort of that that prompted us to start in India. But if you fast forward to today, actually the number one place that, you know, workers on the Mercore platform who, you know, have jobs through, through us, um, are from is actually the United States.

AI assessment note: “the number one place that, you know, workers on the Mercore platform... is actually the United States.”