Equity Podcast
Original podcast
Where to listenCopy share link
Learn more about your ad choices. Visit megaphone.fm/adchoices
4,658 words · 4 speakers
This episode of Equity is brought to you by Schneider Electric. Schneider Electric, your energy technology partner.
Hello and welcome to Equity Tech Hunch's flagship podcast about the business of startups. I'm Anthony Ha and this is our episode where we dive into a tech world trend with the help of industry experts. We're going to be focusing this week on Merckor, a three-year-old startup that's become a $10 billion middleman in AI's data gold rush. The company connects AI labs like OpenAI and Anthropic with former employees of Goldman Sachs, McKinsey, and white shoe law firms, paying them up to $200 an hour to share their industry expertise and train the AI models that could eventually automate their former employers out of business. But where does employee knowledge end and corporate secrets begin? Today, we're bringing you a conversation with CEO Brendan Foodie from this year's
Disrupt, where he explained why companies won't give AI labs their data directly and how Mercor stepped in to fill that gap. Let's take a listen.
Thank you, everyone, for being here. And Brendan Foodie, the CEO of Mercor, thank you for being here.
Thank you for having me, Max.
Of course. So this is your first Disrupt, right?
It is, yeah.
And this is a very interesting moment for Mercor as a business. I've been covering them for a while. I'm really excited to talk with you, Brendan. I wanted to start out with talking a little bit about what Mercor does. My understanding is that Mercor hires these very high-skilled contractors and basically asks them to create trading data for major AI labs. Is that an accurate description?
I think broadly it is, but maybe I can give a little bit of the origin story, so to speak, to give the background of how we got here. It started when we were in our college dorm rooms at 19, where I was working with my best friends from high school, who I met when we were 14, and hiring people internationally that we'd match with our friends and automating all the processes of how we do that. So similar to how A human would review resume, conduct an interview and decide who to hire. We automated all of those processes with LLMs, bootstrapped the company initially. And then a few months after dropping out, we met OpenAI and we realized that there was this enormous transition underway in the human data market where it used to be this crowdsourcing paradigm
that companies like Scale AI pioneered of how do you get low and medium skilled people writing fairly grammatically correct sentences for the early versions of ChatGPT. But as the models were quickly becoming incredibly powerful, it was transitioning away from this low skilled crowdsourcing problem and moving towards this super high skilled sourcing and vetting problem of how do we find the lawyers and doctors and bankers and software engineers that can work to build the most sophisticated agents in the world and the business took off from there.
I was looking through Mercor's job platform before this. I wasn't interested. I'm just, I'm very happy here, but I am just curious.
We'll have to change that.
Well, so I mean, some of the things that they're looking for are like Goldman Sachs investment bankers to kind of submit data or, you know, video game designers or even like CTOs of startups. Like they're looking for people who are not short on cash. Like you have to pay them a lot of money to do this and you're getting very valuable data from them. I'm curious, why do people give Mercor so much data?
Well, I think ultimately what most people care about is having a reliable job that pays them well, where they're doing really interesting work. And we provide both of those things extremely effectively. Our average pay rate right now is $95 an hour. We're paying out $1.5 million a day, just, you know, 18 months after really finding this market. And so it's growing amazingly quickly and we're working on the most cutting edge projects with all of the most sophisticated customers in the world. But I think from the AI labs perspective and maybe more zooming out to the markets more generally, it's structurally more efficient for someone to train an agent how to automate a particular activity than it is for them to do it redundantly instead of.
Having customer support representatives redundantly responding to hundreds of similar tickets. They're going to build an eval once so that the agent can learn to respond to that type of ticket. Instead of having investment bankers redundantly analyzing data rooms of similar companies, they're going to build an eval once and train an agent how to automate those things. And so we're seeing this broader transformation of knowledge work, moving towards training agents how to automate all of these activities and workflows across the economy.
So I want to kind of zero in on some of these examples because I think the actual workflows are fascinating. So say Goldman Sachs investment banker works on Mercurus platform. For example, submits data that helps an AI lab make a financial model. That's very valuable to the AI lab. How do you think Goldman Sachs feels about it?
Well, I think that there is an argument that Goldman Sachs doesn't love the idea of having models that are able to automate their value chain, right? Because it's sort of a competitor, right? I think that over time, ChatGPT will be better than the best consulting firm, better than the best investment bank and better than the best law firm. And that's going to transform the economy radically, which will be a broadly positive force that helps to create abundance for everyone. But it definitely shifts shifts the competitive dynamics. And that's part of the reason that the labs need us is that their customers don't want to give them data to automate large portions of their value chain. So they need to hire contractors that previously worked at those companies
that are willing to understand those workflows and are willing to train models how to automate them.
Do you hire people that currently work for like investment banks or game designers or stuff like that?
Occasionally that would happen, but for the most part, it would be people that previously worked at those companies.
Have you talked to the companies of the places where these people work? Like, do you know how they feel about this?
There's different ends of the spectrum. Like, I definitely think that there are companies that are embracing it and realize that the world's going to change very quickly. And it's better to be on the right side of history and adopt this technology where now that they have the ability To produce so much more, to analyze many more data rooms and do much more work, they can expand their businesses by 10x. I think there's definitely another category of companies that are fearful of that, right? And worry about being disintermediated, having their customers go directly to the AI labs or application layer platforms. My hunch is that the former category is going to turn out to be on the right side of history.
The reason I ask is because, I mean, companies are typically very protective about their data. And when someone leaves a company, you know, there's sometimes like lawsuits about people going to other companies and taking some of that data with them. And I'm not alleging that like you guys are engaging in corporate espionage or anything, but I am just like kind of curious about how you're thinking about that. Are investors concerned about that? Have they asked you about stuff along those lines?
That hasn't been a concern. I think one of the key reasons is that we have contracts with all the experts that say they're not supposed to do that. We give them guidance not to do it. And so
To not like bring documents from their old company.
Precisely. Yeah.
Okay.
So sometimes, obviously, when we're managing these workforces of tens of thousands of people, I'm sure that there are things that happen. But at the same time, when we have contracts with them that they're not supposed to do it, we give them guidance not to do it. Generally, that's doing everything that we can on our side to limit any liability for us, our customers, and really just doing the right thing.
You've clearly cracked something that is very valuable for the AI labs here. It's been reported. I think we reported.
It might have been you reported.
Yeah, we actually, I think we did, that you guys are making, you know, $500 billion in annualized revenue.
500 million.
500 million, sorry. Well, there you go. Confirmation. Okay. And so I'm curious, like, why do you think what you're doing is so valuable and so different from what Scale and Surge and other AI data vendors have done?
Yeah, I think that so much of it starts with who are the people that you want teaching models? Do you want a low skilled college kid that just took their accounting 101 class to teach the model? How to do this problem or someone who might have not even studied the topic internationally? Or do you want the person that was at EY for a couple of years and might be getting their MBA? Or that kind of thing, like the focus on extraordinary people. And really, starting with the DNA as a labor marketplace was the largest differentiator in how it positioned us in the market. And I think there's this other very interesting dynamic that is extremely underrated,
which is that the distribution of value in those that improve models is very power law, where the top 10 to 20% of people drive majority of model improvement. Very similar to how in a lot of companies, like the top 10 to 20% of people might drive majority of the value. And that means that when we're able to build a proprietary advantage around how we source, assess and retain those top 10 to 20% of people, it creates so much value for our customers that's incredibly difficult to compete against. And I think that's the reason that we've been able to go from 1 to 500 million in 17 months and really continue that incredible growth trajectory.
And I think other data vendors have started to catch on and they're starting to pay contractors, high skilled workers lots of money. I guess what's your competitive advantage against them if they start also paying people a hundred bucks an hour?
Totally. I remember when we started doing it, it was like no one had realized the market was about to change. And then obviously, as we started talking to a couple of investors around our series B word got out, around how things were really working and a lot of companies started chasing after us doing similar things. And I think to your point about competitive advantage, it really comes down to the power law, which is that when we're able to be hyper focused on getting the best people through our referral network, where 60% of our new hires are coming through the referral network. And then also learning from the performance reviews of who's doing well on jobs and why to make better predictions. Around who should we hire staff on a given project to be in that top 10 to 20%? It's incredibly valuable and it's really hard to do.
Imagine if when building any of your companies or hiring your teams, your mandate is only hire people that are in the top 10 to 20% of that company. Like, it's a very difficult problem to have the right pattern recognition of what are the characteristics in their interviews, on their resumes and all the assessments that you do on them that translate to that individual actually driving superior returns. And that's what we live and breathe.
I mean, obviously, one of your biggest competitors, previously Scale AI, their CEO was hired to work at Meta Superintelligence Labs and they received a significant investment. And a lot of AI labs like OpenAI, Anthropic, Google, cut off a lot of their work with Scale. How much has that contributed to Mercor's rise in the last year?
Well, I think TechCrunch reported yesterday that we've quadrupled since the acquisition in end of June. And so it's definitely been a catalyst for a lot of continued growth, but accelerating what was already happening really quickly. You know, we'd already scaled from 1 million and run rate to nine figures. As the fastest growing company of all time, we did that in 11 months. And then it just helped to accelerate going from 100 to 500.
And I want to talk about one customer in particular, which is OpenAI. And I know you can't get into too many specifics of, you know, your work with customers. But, you know, you've said on podcasts before that you guys do have some level of customer concentration. I think we reported that actually a lot of your business is from OpenAI. What's the long game with something like this? If you are relying on one customer a lot, what's the long game?
Well, to clarify, we work with all of the top labs in a very meaningful way across all of them, similar to how most of Nvidia's business is concentrated in the hyperscalers. What we're seeing is that every enterprise value chain is needing to transform to teaching agents, where the way that a company like Goldman or other Fortune 500s are going to have a proprietary advantage is going to be leveraging their data and their employees to help improve agents for their specific use cases. And so we're starting to launch all these new initiatives like Apex, the AI productivity index, where we partner with many of these industry leaders to measure how well have we automated
different verticals across law, consulting, investment banking and medicine. That enterprise transformation is just getting started, maybe a fraction of a percent of the way.
How do you convince AI labs to stay? What's important to them right now?
I think the thing they care most about is model improvement. It's a competitive market where everyone's racing to have the best models. And so what they're looking at is for every dollar they spend on us, for every hour that we're able to deliver of people working to create data sets that improve models, what's ROI on that? And so being focused on that power law set of people that drive the top 10 to 20% of model improvement is always the thing that keeps them coming back.
OpenAI launched something a while ago that kind of flew under the radar, which is they asked like, and it's like a LinkedIn competitor sort of thing, which I think I maybe texted you or emailed you at the time and I was like, what is going on with this? And I just wanted to ask, do you have any worry that AI labs are going to start hiring a lot of contractors themselves and maybe doing some of the work that Merck or has traditionally done?
What we found is actually the opposite, where they used to do a lot of this, but then have phased it out because it's gotten much more difficult as the skill of the people has gotten higher. And I think the largest reason is that there's a lot of network effects associated with building all of the base of people that refer all of their friends and then can switch between working at a different labs. And so each of the labs incurring that fixed cost independently causes them to move much more slowly and it becomes much more expensive versus us making the fixed cost investment and building the network, aggregating all the performance data, building out all the referral systems that all of them can then leverage. Every company today is looking for an edge.
Better insights, faster decisions, greater resilience. But as AI drives new growth, it also creates new demands on energy and infrastructure. Schneider Electric is connecting energy, automation, and digital technologies to help organizations make smarter decisions and operate more efficiently. Because the future won't be shaped by AI alone, it will be shaped by energy and industrial intelligence. Schneider Electric, advancing energy tech.
I wanted to ask you just personally, I mean, you're 22 years old. You're in charge of a $10 billion company. You were telling me backstage about kind of a scheme that you ran in high school. You've always been a young savvy person. Can you tell me about the sneaker?
Yeah, yeah, yeah. These are always funny stories. So the one in high school was that a lot of my friends were reselling sneakers, as was relatively common for people my age. And they were buying these bots or automated checkout software to help resell sneakers more quickly. But what I realized was that all of them were running their software on AWS and they were eligible for AWS startup credits, but they weren't claiming those promotions. And so I started a consulting business where I would build them a website quickly. I actually hired my co-founder Adarsh to help build websites when we were 15 or 16. And then I would help them in going through the process of applying to get credits, charging them $995, getting them the credits, doing this hundreds of times.
And so when it got to whether I should go to college, I was telling my mom, like, I'm making so much more money right now than I would ever make in whatever job she wants me to have. After college, what's the point of going to college? And so initially, I refused to apply and they were very unhappy about that until I eventually came around.
But you did go and you dropped out.
Exactly. And so, well, and with that part, it was that I told my parents every semester that I was going to drop out of college because initially I didn't want to go in the first place. And then by the time that I really dropped out, I just didn't go to finals my final semester because I was like, we'll make it work. We'll figure something out. Then the school notified them rather than me notifying them. And so they were all upset that I didn't tell them. And then I told them, like, I've been telling you every semester for the last two years. And so it all all came full circle.
I mean, the reason that story sticks out to me is because I feel like you seem like someone who's been good at identifying loopholes in the system that people aren't taking advantage of. And I do think that Mercor is to some degree, a bit of a loophole in the system that people maybe didn't see in the past.
But to push back on that, like, I think that loopholes are where you're almost finding some like mistake people are making or thinking that you could take advantage of. But what we're doing, I think of as very different in that it's creating huge amounts of job opportunities that are enormously beneficial to the tens of thousands of people that we're paying enormous amounts of money out to and driving the model improvement for the best agents in the world. And so sure, it's like finding an inefficiency in markets, but I think that it is far and away the most value additive thing that I could be doing.
I guess the reason I do think it's a loophole is because I do wonder what the companies are thinking, you know, when their employees are going to give this valuable data to train, like you help train models for another company. I think they might want a piece of that.
I think maybe there is definitely a question of, does the knowledge that lives in a former Goldman banker's head belong to Goldman or does it belong to that individual? I think that it belongs to the individual and we're all about empowering those people to best leverage their skillsets and the way that we can make them the most money. But I can also see the argument against of how the employers may want that knowledge that lives in their heads to belong to them.
Yeah, I mean, you've talked a lot about how Mercor is not just people creating AI data for AI labs, that it's actually the future of work. I want you to kind of expand on that a little bit about like, how is this, is this like a new gig economy?
I think definitely, well, one thing that we're finding really interesting is that the nature of these projects is converging with what you would think a full time work environment looks like, where we add people to a Slack org, they collaborate, they work in teams, they get performance reviews, they have all of these things that feel. Like it's their jobs. And I think that's what full-time employees' jobs are going to converge on, right? Instead of Goldman having all of these, and I don't mean to harp on Goldman, they're not actually a customer, but instead of them having thousands of analysts that are redundantly analyzing data rooms and preparing their analysis, I think they're going to have those people in the Slack org that are building the data sets,
collaborating with one another and training agents for their specific workflow. Because similar to how software is a fixed cost investment, where you do it once and then you can run it many times, training an agent is that exact same thing. And so I think for the same reason that software has been eating the world for the last 30 years, training agents is going to be that next frontier that the vast majority of knowledge work is going to move towards.
I mean, play it forward for me. Like if people are training these AI agents to one day take their jobs, like, is that ultimately in their best interests?
I think it is because the demand for knowledge work and the elasticity of that demand is so large that we're just going to do more, right? Like, you could have said, oh, in late 1990s, software engineers are going to automate themselves away because once they build the software, you just run it. And like, all of a sudden, we've automated that given workflow. But the truth is that software engineering as a profession became incredibly pervasive and like one of the largest opportunities, not only for the individuals employed in that profession, but also for all of the consumers of it. And I think it'll be the same with training agents. Like, we're not going to run out of things to do in the economy. We're going to cure cancer, we're going to go to Mars, we're going to do all of these incredible things
that we otherwise might not have imagined because there's so much elasticity for more productivity in the economy.
I mean, you said before that, you know, you essentially think like the entire world is going to ultimately just be training data for AI models. Can you expand on that? And does that not feel like a little uninspiring?
Well, I think of it as inspiring, right? Like, you're referring to this blog post I wrote about how the entire economy will become an RL environment machine. And it really ties to what I was talking about earlier, which is that in the future, humans will do things once. They'll teach an agent how to do that. Maybe another analogy is similar to the way that founders or managers scale is not just doing the same thing themselves all the time, right? They train an individual that they hire to do that job and then they're able to go on to the next thing and build process. I think it'll be the exact same with agents where you train an agent how to do that job and then you go on to the next thing and get the next agent that you're trained to do that job.
And so imagine if all of us could manage 10 people or 100 people and that's all just inference costs to maintain it. I think that's what the world's going to feel like, where we're able to have so much more leverage in doing this high caliber strategic work, which AI will certainly help with, but build process for agents to automate workflows. And so for anyone that wants to climb in their career to manage a big org and have this leverage, like that's going to be possible for everyone. And I think that is very exciting, both from the standpoint of the impact that an individual can have, but also more importantly, from the impact of what's possible in society, right? Of like all of these things that we might have imagined society doing on a 50-year
time horizon can get compressed to the next decade with the right automations and improvements in productivity that so much of this AI revolution is enabling.
You just raised a massive Series C, we have $350 million, the $10 billion valuation. You announced it yesterday and I'm curious, what is the plan ahead for Mercor and how do you grow into a company? Do you always want to be a company that is just providing data to AI labs or is there something bigger on the horizon?
Yeah, it's a good question. So much of knowledge work, maybe all of knowledge work will converge on helping to train agents. And so that's very much our focus. But in terms of the investment areas of how to spend the $350 million, first of all, we're not rushing to spend it because we are profitable, but in terms of the broad things that I think about, it's inspired in many ways by Amazon, where Jeff Bezos talked about how there were many things that would change about Amazon's business. Over the decades of running it. But he knew that there were three big things that customers would always want through the duration of time, which is that customers would always want more products available. They'd always want better prices. They'd always want faster delivery. He would even say like, he can't imagine a world where customers would say, I want higher
prices, right? Or fewer products. And similarly, we have three big things where we know our customers will always want more candidates on the platform. We know that they'll always want better matching because not exactly price because no two people are the same. And we know that they'll always want faster delivery. Of how do we automate all the processes associated with creating that end value and work product for our customers? And so we structure our product team in each of those three buckets. And how do we best allocate our investments to compound in things that we know will build value for our customers over decades to come?
All right. Well, I think that's our time. Brendan, thank you for being here and thank you all.
Thanks so much for having me, Max. Yeah.
Equity is produced by Teresa Locum-Solo with editing by Kel. We'd also like to thank TechCrunch's audience development team. Thank you so much for listening and we'll talk to you next time.
Page updated October 26, 2026 · transcript and summary are machine-generated