Rowan Stone, co-founder and CEO of Sapien, welcome. Thank Thanks for having me. Now, I want to start out with a little bit about your background. You spent time at BASE. I think you were part of helping to really grow it in the early stages. What from your experience at BASE can help you to continue your journey in entrepreneurship to build Sapien? What were some things you learned? Great question. Maybe just for context, I actually sold a company to Coinbase. And so I was very lucky to work there for about three years. And while I was there, I ended up kind of owning from a business perspective, a bunch of their kind of more crypto native products. And so that started off with things like stable coins, things like wallet, smart wallet. Basically, we're just trying to figure out how we can make these things more useful, more relevant to people and get that tech stack into the hands of those who could benefit from it. But while at Coinbase, we kind of realized that there wasn't really a real reason for Coinbase to actively be building on chain. And what I mean by that, it might sound kind of weird, Coinbase has like a thousand engineers, some of the smartest builders in the space. They've literally been scooping them up for nearly a decade, a really long time. And those people are building a centralized exchange. They're helping to build a stable coin, a suite of stable coins. They're building wallet infrastructure, but they haven't really meaningfully contributed to the on-chain world by giving themselves like a real home and pivoting everybody's attention to like, how do we actually make this stuff more seamless, more slick? And so myself, Jesse, and a few others, we actually created BASE together. My part of that puzzle was more enablement, kind of thinking from a business perspective, does this make sense? How do we make this make sense? How do we persuade the broader business that this isn't commercial suicide for a big public company to partner with a DAO and deploy a real chain? uh But we ran that from an ops and business perspective. We did all the go-to-market, we did a bunch of the marketing, the legal frameworks, all that kind of stuff. And while deploying, really the big lessons there is that to bring people on chain, It's no use just to make it easy. Like we can remove the commercial barrier if you like to make it cheap. We can remove the technical barrier to make it easy. But unless you actually give people real reason to do something, an incentive of some sort, they're not going to change their behavior. It's really hard to make people change behavior. And so I think for me, that was probably the biggest learning. And we're applying that same learning to our go-to-market for Sapien. The high level here, go for it. Yeah, maybe before we get into Sapien, maybe I could set the stage. Let me read from the homepage and then we can use that as a jumping off point. And if you could help the audience understand kind of the problem that Sapien is trying to solve. And you may need to define some terms because the audience is primarily crypto. But for those that are interested in AI but don't know much about labeling, for example, maybe you can help define some of those terms. But let me start off with reading the homepage. Tell us what this means and who you're speaking to. So train AI with expert human feedback, accuracy, scalability, expertise, custom data collection and labeling services powered by a decentralized workforce and gamified platform for unmatched accuracy and scale. So tell us what that means and what problem is Sapien trying to solve. Yeah. And I'm going to caveat this with I am in the process of nuking our website with the largest bomb I can possibly find because it's like lean, scrappy startup mode. It's free product market fit, and it doesn't say the things I wanted it to say. But let me just take a step back anyway. what you don't like about it too, but go for it, go for it. it's a great V zero. It's absolutely fine. And so our thesis really is simple. In fact, before I even say our thesis, AI, if we're not familiar, is essentially the best analogy I can think of a child. And a child will become a good, useful human if you feed that child with tons of good information. You let it learn through its life, the differences between good and bad and how to be a member of society and how to help people around it. It'll become useful. AI is exactly the same. It's essentially an adolescent, very intelligent adolescent that can do PhD level math and all that kind of stuff. However, it doesn't have all of the context. It doesn't have all of the information. And the way that companies make their models smarter is by throwing industrial scale data at the problem. And that data, most importantly, comes from people, real people. So it's providing real context. And our thesis for Sapien is very simple. We believe that every single one of us, whether we are an artist or a musician or a lawyer, a doctor, an engineer, it doesn't matter. All of us have nuanced understanding within our heads that's super valuable to these companies building AI models. And so we exist and we are building a framework and a system to allow anyone anywhere to properly incentive or properly monetize that nuanced understanding. and transfer that knowledge to the companies that are building AI models. And so this is a series of frameworks in a system format. Most of us are people that have been building in and around the on-chain world for a decent amount of time. And so we're taking a slightly different slant on this problem than most of the incumbents would typically do, deploying both reputation and on-chain reputation system, and a series of incentives and disincentives. much like you would see in Ethereum proof of stake, i.e. stick up some collateral, do some work. In Ethereum's it's verify transactions. In our world, it's provide some knowledge, provide some time, provide some expertise. And if you do good work, you get rewarded, you get more incentive. And if you do bad work, we will slash your collateral. You will lose money. You will also lose your reputation, which is key to the system because reputation is what we use to gatekeep complex, sophisticated work. but also to gatekeep validation privileges and things like this that we can get into in a little bit. So the high level here is it's a system to enable knowledge transfer. Let's get specific. So most people are familiar with Chad GPT at the kind of the consumer at the application level. But if we go backwards in the AI stack, ah where does what you're talking about with providing new ones data or data labeling, where does that fit in? Maybe walk us backwards. Yeah. And so I want to get away from data labeling before we go backwards. And so let me just describe what data labeling is. Data labeling is a way of essentially telling a computer what it's looking at. And so for example, I'm holding up AirPods back when this was first kicking off, even when Sapien was first kicking off, it was actually valuable to AI models to have someone draw a box around the AirPods and say, that is AirPods. This type of work is literally valueless now. computer vision can already recognize what it's looking at and can essentially do the work that people would previously have done. And so the puck is moving towards much more complex types of knowledge, types of information and annotation. But to explain what I'm talking about here, think of building an AI model requiring kind of three components. So the first component is information, data. The second component is an algorithm. And that's essentially like, how are you going to use that information to train the model? And then the third piece it needs, and I'm being very simplistic here, like clearly if anyone's actually listening that understands this, I'm skipping over a bunch of the detail. But the third piece it needs is compute. And so like you've got the data, you've got an understanding of how to apply the data. Now you need to throw massive scale compute in form of GPUs typically, almost exclusively at the issue to train. to run through lots of different simulations and do inference and training to get the model to the point where it's able to function in whatever way you're trying to make it function. And so beta really you can think of as like the fuel. Got it. Now you say you want to get away from data labeling. Is that, you really think computer vision is that good now that it's able to draw boundary boxes over everything and not make mistakes? Or are humans still needed in the loop? Humans are definitely still needed in the loop. it really just depends on what we're looking to annotate, what we're looking to label. And so if it is a vertically specialized model, which is what we typically deal with, we don't generally do like your chat GPTs or your quad fours, like these general purpose models. We typically do a model for autonomous vehicles or a model to teach school children math, like very specialized things that are designed for a specific purpose. And so If you're building, a medical model, the chances are computer vision is not going to be able to figure out what it's looking at, unlike a radiology image or an x-ray. And so we need humans in the loop. We need experts, doctors, like credentialed, certified doctors to help steer and help these models better understand what they're looking at. So they can be useful to augment a doctor in diagnosis or in recovery planning or whatever part of the mission it is. But for all intents and purposes, the vast majority of what was previously labelling work uh is essentially worthless. Computer vision can certainly, around your house and in the outdoor world, most of the time, it's going to be able to capture everything. Where I would kind of throw a caveat is that a big chunk of our work is 3D and 4D annotation. And so it's a little bit different to simple labelling. You're looking at much more complicated data types, typically LIDAR data. And so if you see like the robo taxis in Las Vegas that are run by, by Zooks, a subsidiary of Amazon, these things are driving around with people inside safely. And they're doing that because every single time they encounter a weird scenario or like road works or some sort of random thing that's happening, someone jumps out in front, dog, cat, rabbit, whatever. They have people in the background, they'll kind of default to stopping. So they're not going to do something they don't understand. But then they're capturing gigabytes and gigabytes, petabytes and petabytes of data while driving all around the place. And then they have people in the background that are helping them better understand what they're looking at. And then they figure out how to handle that situation on the back of the human input, which makes them safer and makes them able to navigate that city better. this is a, you're playing in a space that's near and dear to my heart early in my career or in grad school. I actually wrote one of the early neural nets in trying to tackle a problem called the word sense disambiguation problem. And so you're probably familiar with it with a, you know, one example of that is the word bank. know, bank can have different contexts. It could be a financial institution. It could be a type of basketball shot. or it could be the side of a river. And so teaching a machine the context um in which the word bank is used is actually a really hard problem. um fast forward 20 some years later, and the word sense disambiguation problem is actually still a problem. Like machines are still having, they still are unfamiliar with irony, rhetoric, um and really words are. they have machines have a difficult time determining context in which what words are used and which sense. um And then one of the my first job out of grad school was actually at a at a really interesting startup called Inovion where we actually had a very specialized Zeiss camera that took images seven uh seven fields of the eye. And these images were then encrypted and then sent to a reading center at the University of Oklahoma Medical Center where we had ophthalmologist actually look at the eye to determine the presence of cataracts or diabetic retinopathy and diseases. And then it would score them and then return uh the score. And then the primary care physician would then meet with the patient to talk about their eye and like the treatment. And that was a really, really interesting startup, but that was kind of a very, very early. This was like a long time ago. This was like, like AI can do all of that now uh or some of that. But back then, you know, we had actual physicians doing the work and, um but it's, it's work that probably, you know, with computer vision having improved so much, you know, lot of that could be done by, by machines now, but then probably double check by humans. And so Sapien uh is playing in a space that's very interesting to me. And your example with Amazon is super interesting. I'd love to hear more about that. And then maybe we can get into the crypto stuff because crypto is perfect to coordinate all of the behaviors and incentive mechanism and really mechanism design of how to. you know, coordinate all the stakeholders in kind of the system that you're building. But first, tell us about Amazon and you mentioned that and I imagine I'm assuming they're a customer because it's on the website. um What type of work you're doing for them and how did you land that deal because uh that's a great logo to have. I mean, super proud of the customers we've managed to secure thus far. is pretty early days for us, but we're working with 27 customers at this point, including a couple of Fortune 100s, which I think is just showing how much demand there is. Like everybody is in this crazy competitive landscape to try and build either a model that can get towards AGI in the case of ChatGPT and Claude and everyone else. Perhaps AGI is no longer really the goal, but build a model that is generally useful to the vast majority of the population, I think is more likely the goal. And then most enterprise businesses are being squeezed by AI in some shape or form. Large companies, particularly consultants and companies like this, are potentially looking down the barrel of being obsolete. And so rather than admitting defeat, they're doing exactly what they should. They're figuring out how they can augment their workforce with a model that understands their business. that's specialized in whatever remit they might be trying to tackle. And so there's this huge demand and arms race for everyone to kind of build their own thing. And we're doing really standard early days crypto stuff where everyone's like building their own thing, ignoring everyone else, not working together, no shared technical standards, super competitive. And so the demand for data is absurd. And that's really the only reason why we have 27 enterprise customers at this stage. We're only a year and a half old. very unheard of and so grateful to have the opportunity to be learning from these companies and building for them, building alongside them. specifically Amazon, it's 3D, 4D annotation. It's one of our core areas of focus. We actually stumbled upon this completely by accident. I'd love to sit here and say that we're super smart and we did it, but we really didn't. In the early days when it was simply trying to build almost a decentralized scale AI, so they're the of main incumbent in the data labeling space specifically. We realized that having done a bunch of labeling ourselves when we were setting up the product, at least the v0, this is not fun work. Like this sucks. 12 hours a day, data labeling, it's grim. It's really not fun. And so we figured that there's crazy high churn. And the only way we can try and get over a crazy high churn is to just make it 5%, 10 % less mind-numbingly boring. and try and introduce some kind gamification elements and a little bit of fun to try and get people to actually enjoy at least a little bit what they're doing. And so that was part of the early thesis. We actually took investment from companies like YGG and Animoca, and we have Gabby as an advisor behind YGG. And so like adding these elements of gamification, attracting a bunch of people from the gaming space, coincidentally, gives us an ability to do 3D, 4D data way better than the average person. Gamers are just naturally good looking at this type of information. They understand that they literally spend hours every day doing it. so completely coincidentally, we've positioned ourselves in such a way that we can actually be super competitive for 3D, 4D data. we won a number of contracts, including Toyota and GAC and Amazon Zooks. And these are against huge incumbents, multinational companies, and in some cases, their own teams, because we can provide higher quality data. And what that means is just more accurate annotations, typically of LiDAR data, which is like a type of radar. And yeah, that's what we do. So we provide the annotation, and then they use the data that we provide to hone their models, run additional inference and fine tune, and ultimately make the vehicles safer for their passengers. You know, I had Gabby on the show. uh I've known Gabby for a number of years and that makes a ton of sense that uh Sapien is working with Gabby and the YGG community. They have some like, I think 10 million gamers, I think, in their guild. so involving all of them as labelers makes a ton of sense. um I also worked at Amazon's. There's, think, a lot of overlap in kind of what we're discussing. And back in the early days, uh at least in the field that I was in, which was in the supply chain area. We were actually looking at a lot of kind of digital object detection and uh looking at different weights for packages to determine uh very, very early AI stuff. And so that's really cool to see Amazon innovating in this space. Now you mentioned that uh Sapien has beat out lots of other kind of web two incumbents in the labeling space. Why are these Fortune 500s choosing Sapien versus the others? all about quality. I'd love to spin a yarn that we're like super unique in lots of different ways. But the reality is we are able to provide higher quality data. And the reason we can do that is that we have a globally distributed group of people, which means that the data won't have geographic cultural age, sex related biases. It's going to be able to get closer to real truth because it's taking samples from all over the world. Mm. But I think the way in which we approach quality control is really what's setting us apart. Right now, we're kind of doing a hacky version of the future product. We're doing as we should, the things that don't scale. But the plan is to actually move quality control into the community. And so what I mean by that, traditionally, companies like scale will operate in like a hub and spoke model. They'll have people distributed all around the world working in kind of gig work style to provide data. or to annotate existing data. And then they'll send all of that to a hub, giant warehouse, giant office, full of people, could be 500,000 people easily. And they'll do the quality control to make sure that data is accurate. AI models, obviously garbage in, garbage out. So the data needs to be accurate for that model to perform. And then once they're satisfied, they'll ship it off and they'll run the training and they'll get it done. Where we're different is we're aligning the incentives at the very start. And so nobody is doing work. without some form of collateral on the line. That collateral for us is going to be in two different types. There'll be monetary collateral in the form of Sapien tokens, which they earn from doing work. And so we're not asking people to like buy into the system. That would be a big barrier to entry that we don't need. So they're earning, 20 % of their earning is going to come in Sapien tokens. They can then lock those up as collateral against future work. But they're also putting their reputation on the line. And so they need a high reputation to get into the more sophisticated work. For example, higher paying tasks like 3D, 4D data. You can't just rock up as a new user and immediately get access to these. You're going to have to do a little bit of the more simplistic things, demonstrate that you can deliver quality data within whatever areas of expertise you have. And then once you've proven that you can deliver consistently quality data, you have enough reputation to participate more broadly. And that participation could just be more complex work, higher paying work, but it could also be validating the work of other people. And so that's really one of the innovations that we're taking here. We're moving away from hub and spoke QC, which is currently bottlenecking our business massively. We have about 70 people doing QC today. They're all on payroll and they're um in an office in Beijing. And we're going to move away from using these 70 people eventually. They're always going to be there as kind of a backstop, but they're going to end up being a tiny minority of the QC workforce. Because once we allow everybody to participate there, Today we have 1.1 million users on the platform. I don't expect we're going to have anything close to 1.1 million that have high enough quality to do QC. But say we can 10x, like say we can get 700 in the first couple of months, that would be a huge throughput enhancement. And then maybe we can 10x again from there to get 7,000 out of the 1.1 million doing QC. All of a sudden we're able to process vastly higher amounts of data. And that enables us to take bigger chunks of the contracts that are out there for the taking. No, that sounds really fascinating. You mentioned something I want to double click into on the on-chain coordination piece, the mechanism design. You mentioned that 20 % of, so if I'm a worker, if I'm a labeler, 20%, I earn 20 % of my income in tokens and then the other, I'm guessing, fiat. And then I'm staking my reputation as well as those tokens. Is that correct? Maybe walk us through what the coordination looks like. Because that's really what crypto is perfect for. It's designed to help coordinate behaviors and incentives. And maybe walk us through how that looks like. You absolutely nailed it. For me, crypto helps coordinate capital. We're deploying on-chain incentives and mechanisms to help coordinate cognition, essentially. Let's use it to coordinate knowledge and transfer that knowledge where it's needed. Optionality is really important. Not having friction is really important. Everybody listening here, because the vast majority of us are crypto people, we probably don't see the complexity. But to a normal person, onboarding is still not easy. We've done a ton of work. It's easier now. but the UX is still far from ideal. And so the way we onboard, we're using Privy, little plug for them, love their product. Users rock up with an email address. We deploy a wallet in the background using MPC technology and they're able to then very quickly get started without the complexities of keys and things like that. We let the user do a bunch of work without any collateral at all, really simplistic work, work that we know the answer is true, that can help us gauge performance of that person and give them an initial trust score. We then let them graduate on to kind of more important paid work, but still very basic things, things that essentially anyone could do. It could be audio speech type work. There's a lot of companies that are building voice assistants and they need to understand tons of different accents, tons of different dialects, tons of different languages. And so that type of work is something that ultimately anyone can do without credentials. And then they're building reputation over time. Once they hit a certain number of tasks, which we will play tunes with, right now we've set it at 100, but we don't know what the right number is. We're going to have to figure this out as we go into production. They then need to start putting up some collateral. And so the idea is they've had a chance to earn some cash. They're paid a minimum of 20 % in Sapien tokens. And then the balance is actually in stable coins. It's much easier for us to pay a global workforce using on-chain rails rather than us figuring out fiat. He plus countries, it's just not going to happen. And so the user can actually play a tune with what they get, but the minimum they'll receive is 20 % sapien. If they're hyper bullish, they can say, you know what, want 80 % sapien, I want 100 % sapien. It's up to them. But the balance would be in stable coins, they're sapien tokens, they lock them up in a staking contract. And that's the collateral that gives them access to the higher paying tasks, the more sophisticated work. And then again, the idea here is really simplistic. Like if they do good work, we give them a multiplier. So they're staking, they get a little bit of multiplier in their earnings for having the stake because we know they're economically aligned with a good outcome. And over time, if they are consistent, they get more of a multiplier and their earnings go up and up based on their input, their quality, their consistency and their overall stake and stake duration. And they can then progress to quality control if they so wish. And that's really where we start getting huge benefit from the network. And this is the future world that you're describing since Sapien hasn't had a token generation event yet, so there's no Sapien token, is that correct? Got it. are currently bottlenecked on QC, as I mentioned, we have a centralized team, the best we can possibly do with what we have. And we're actually turning away work as a result of this bottleneck. We had a couple of opportunities for very large orders from some pretty awesome companies building some pretty cool things. And unfortunately, we just didn't have the confidence that we could deliver the quantity of data that they needed in the timeline that they needed it. So we took a small piece of that contract instead, but we could have easily 10 to 15x the size of that order had we had this system in production. And so for that reason, and that reason alone, we are charging towards a token as soon as we possibly can. I'm not going to talk timelines because I think it's commercial suicide. Nothing ever goes exactly to plan. But it's certainly the top priority for us to move away from this kind semi-closed system whereby most things are permissioned. into a world where token incentives and reputation enables us to move the vast majority of the work into our front end. And then everybody, once they have enough reputation, can start participating and earning from that work. This is such an interesting space that you're in because there's the Web2 component, AI component, there's the crypto component, and there's actual demand. When you speak about turning away work, it leads me to think, I wonder how much demand there is for this type of service, and how are you gauging your capacity against that demand? Is there a way you can maybe quantify it for the audience just to give us a sense of like how big is this thing? Um. What can and can't I say? That's the real question. I think like, roughly speaking, the big models that we all know of, your ChatGPT, your Claude, companies like this, I don't know specifically about Claude actually. It's probably better that I don't name names. But just generally speaking, you're really big models. These folks are spending 80 to 100 million a month on data for training. And so the amount of money that is in the the AI industry specifically looking for the fuel to power these models is absurd. Like we're talking tens of billions uh and growing exponentially currently. Where we sit is a specific part of that data pipeline. We're not doing the inference, we're not doing the training, we're specifically answering two questions. And those questions are from a customer, I have a data set from my robots, from my autonomous vehicles, whatever it might be. need you to annotate it. I need you to give me the context so I can then use that data to train. Or it could be, I'm building a model, but I'm missing a bunch of my data. I need more context. Help me go and collect the data that I need, structure it in the way that is suitable. And then they'll go and take that for training. And so that's kind of the two parts of the puzzle that we kind of look after ourselves. And the overall market for the data specifically, it's like $30 billion is the kind of estimate that people are saying. 30 billion by the end of 2030. Yeah, that's amazing. I mean, the demand for data that's enhanced and clean and usable by these models is really incredible. And I think Sapien is playing a very important role. How do you view web two uh competitors in this space? You mentioned scale, which is probably the biggest in this space. But there are others also. um I, as I've done some research, you know, prior to this interview, uh it, I thought that rather than competing, I wonder if there's a co-op petition type of relationship perhaps with, you know, Sapien, with the other players in the space. Is that something that you and the team have thought about? Collaboration over competition is an ethos that I've tried to use many, many times in many, many places. We used exactly that ethos when we took Base to Market a couple of years ago. And it's exactly how we are framing and how we are thinking through the problem space for Sapien. And so rather than specializing in a specific part of the market for data, instead, what we're really trying to do here is build the framework, build the system, build the incentives. to allow really the connection of any enterprise demand for data with everyone everywhere that has the context that these models need. And so what we want to do is just build the system that allows that transfer. And by doing so, rather than seeing Scale AI or Appen or Pearl or any number of these companies as potential competitors, they can be customers, they can be collaborators, they can actually help lean in, build with us, create shared technical standards. and ultimately end up with a data pipeline that's way more efficient and much more beneficial for the customers that are building these things. So that's kind of the way we're thinking it through. And in terms of how we service, again, we're doing the things that don't scale. Today, we have like an enterprise sales team. We look like a very standard web two company. We have people, boots on the ground, knocking on doors, pitching customers and closing deals. And like, That's great. It's great that we've managed to get 27 paying customers, but it's horrifically unscalable. And so the right way to do this is to create a system that is somewhat self-service and to have, rather than us going and knocking on the door of a thousand different customers, we build relationships with the rest of the data space who have their own customers. And when they need data annotated, structured, or they need data collected from different people around the world. We have this network, it's ready to go. They can plug into it and they can source exactly what they need. And so that for me is a much more scalable path to market. And that's the route that we're taking here with Sapien. But Sapien's making revenue, which is unheard of for a crypto project. There's very few, mostly in the DeFi space, that's actually making revenue. So that's pretty amazing. How are you thinking about revenue going forward as you approach token generation events? Yeah, I mean, have a slightly more optimistic view of revenue in the crypto space. think over the past couple of years, we've turned the corner. There's a ton and granted it's almost all transaction fee revenue, but like there are a bunch of companies making a bunch of money from people actually using this stuff. And the way I view like valuations for most of these things is like your valuation is what people are willing to pay to use it. And so if you're a chain, your valuation is the sum of your gas basically. But those numbers are going up and to the right. Like they are, there's some decent money being made at the moment. However, you're right. Like we do have real like household name customers that are paying us to provide them with service. And it's early, we're not super scalable yet, but we've found some product market fit and we're doubling down on those areas. And the ultimate intent, particularly in the run up to TGE is to bring as much of the work that we have currently kind of hidden away in private. into the front end so that anyone anywhere can participate. But then really importantly, and this is one that I to be a little bit careful about how I speak about, particularly given that the law is still a little bit gray, we want to find a path. And that path is becoming clearer as regulation unfolds to allow participants to gain from the revenue that the business creates. Now, traditionally, that would be basically impossible. Until recently, was something that everybody wanted to do and nobody could really find a real path to doing. Because if you share revenue with token holders, your token is basically a security. However, the laws are changing. Particularly in the US, we now have people in power that actually see the value of on-chain technology and want to find a way to create regulation that protects consumers. without stifling innovation and without stifling companies that are building. And so I'm starting to become increasingly bullish that we can actually find a path to allowing token holders to accrue some value from the business that's actually able to generate value. I'm not quite ready to explain exactly how. I mean, it's not rocket science. We're definitely not ready to do it yet because the law hasn't changed. But should that start to become clearer, that's exactly the plan. Yeah, yeah. Things are becoming less gray for sure. um But I think what is very promising about Sapien is when you consider this is really a two-sided market of, on the one hand, you've got enterprise customers that need their data labeled, or really anyone that needs their data labeled, but really entities that need their data labeled, and then you've got labelers on the other side. And then Sapien's in my mind anyway, as you're describing it in the research I've done. provides this coordination platform, this mechanism by which it coordinates the work to the labelers. um this mechanism design that's built in helps to ensure that there's high quality and ah promotes good quality, demotes bad quality, and your customers benefit. And then the labelers benefit because then the good ones get promoted. and which is exactly what crypto is meant to do. It's to help to coordinate all the right, the stakeholders and incentivize our behaviors. um I'm curious from a, uh you know, as let's say I'm a Fortune 500 customer and I have work that needs to be done. I work with Sapien. How do I submit work? And do I have uh the say as to which labelers can work on my job or no? Sure. ah Right now, it's partially self-service. You can come into the website, you can upload a data set if you have a data set already. We will estimate what's needed to be done. And then we will essentially match that work with the right people. I'm going to try really hard to get us away from the word labelers. And the reason for this is that, yes, that's how this industry started. But I think when most people hear labelers, it's like not something they want to do. And The key thing here is that we truly believe all of us have information that's useful to AI, and all of us should have the ability to monetize that information. so I think getting away from AI and simply just saying people, like anyone, anywhere, you want to give me your voice because like me, you're from Scotland and the model doesn't understand a Scottish accent? Great, come on in. Or you have experience in some... bizarre language that nobody understands apart from a very small number of people. Well, great. You can monetize that for sure. Or maybe it's some obscure pottery or some really weird thing. It really doesn't matter. Every single person will have some type of nuanced understanding of the world around them that will be beneficial. so for that reason, ah we are called sapient. We're trying to coordinate everyone everywhere. People, humans, sapients. um And I completely lost my chain of thought ranting about labeling. And I apologize because I've been ranting internally too. We are still in the habit of using the word labeling too much. And I'm determined to delete it from our vocabulary so that we can be talking about the real opportunity, which is for everyone. Yeah, and I appreciate that. And I think that that type of repositioning makes a lot of sense because it is much bigger than the actual task itself. in the dirt and sweat of the actual work, it is labeling. But as you zoom out, it's a much bigger job. And so I appreciate you uh repositioning from the specific task to kind of the bigger vision of what it could provide. um No, this is really, really interesting. Can you tell us maybe some customer stories? Not to get anything proprietary, but I think to give the audience a better sense as to um kind of the value Sapient provides and the end outcome, what it could provide either end users or the enterprise. because it feels really abstract, right? Most people are familiar with the applications like Grok and ChadGBT, but people are unfamiliar with all the work it actually has to go through to be able to benefit from a ChadGBT. And in the AI or machine learning kind of pipeline, Sapien plays a really, important part. um And you mentioned earlier, by the way, you said garbage in, garbage out, and Sapien has a really important role in making sure that the data is clean. I guess what can you share for the audience so that they can have a better sense, less abstract sense as to the value Sapien provides? Yeah, for sure. I think my favorite customer story, it's probably going back about a year now. It's one of our earlier customers. Huge Chinese public company. And they existed to provide homeschooling to relatively young kids. Like I think at a pre-high school, in Scotland we call that primary school kids. I don't know what it would be for you. So like young 12, 13 year old kids. And the government decided that it was just not fair because rich kids got homeschooling and everybody else didn't. So they banned homeschooling almost overnight, this whole idea of a home tutor. So this company essentially went to zero. Their company was deleted by a new law. What they realized is there was a pretty obvious loophole in the way the law was crafted. uh So the law was written in such a way that a human tutor was not allowed. there's nothing to prevent an AI tutor from not being allowed. And so they pivoted hard. They obviously had to offload a bunch of teachers and that sucks. But they invested a ton of capital and a ton of time into figuring out how they could deploy kind of kid-friendly AI models specifically designed to teach different subjects. And the first one they did was a math tutor. And they approached us looking for a variety of Mandarin-speaking teachers, essentially, that could teach mathematics at a decent level. They started with the younger kids. And we went out, boots on the ground, in a very traditional recruitment way. We found a bunch of people. It was actually very difficult. We persuaded them that this was a cool idea. This was like very close to day zero for the Sapien business. And we managed to collect the chain of thought reasoning that was needed for them to teach these models, not just the answer to questions, because that's not valuable. Like any calculator can go one plus one equals two. But really what's valuable to the model is like, how did you get that answer? The chain of thought reasoning to get to the outcome is really how you teach, particularly a child, how to tackle the problem. And so they pivoted. We helped them with the math. We helped them with a variety of other kind of GPT style tutors. They now offer a similar service to what they previously provided at a much more competitive price point. And so it's guess much fatter because more children can have access to it. But importantly, their business didn't die. Like they're actually kicking ass at this point. And so I think that's probably one of my favorite stories. The other one that I love, and I'm going to try not to name names because we need to be a little bit careful with some of the customers. There's a couple of autonomous vehicle companies out there that are just building stuff straight out of Star Trek. And I'm a huge sci fi nerd. So I just I love to see it. I love to play a tiny little minuscule part in making it reality. And I look forward to a future in which I love driving. And so I'm not going to give up entirely, but a future in which I can jump into like a completely driverless car that's set up as a little living room, and I can just get a laptop out, do some work, head to whatever I need to be. Like that's a cool future. And it's basically here now, like Zooks, I'm going to name a name there. They've got this. It's in Vegas. There's cool four seat robotaxis. They literally look like something out of June. They're pretty bad ass. And so those are the things that excite me. And I think without going on a rambling ah montage here, the next thing that's really exciting, because we've built up a huge group of people that can do 3D, 4D data to a very high standard, and we're very competitive on delivery. We are actively leaning into another segment that excites me personally, and that's robotics. So you think about, particularly humanoid robotics and manufacturing robotics, they're basically just an autonomous vehicle, except they don't have wheels, they have legs, or they don't have wheels, they have arms. And that means we can turn our hand to it very, very easily. And we've started working with the first couple of companies. And again, playing a tiny part in the rollout of something like that to me is pretty exciting. I was gonna ask about humanoid robots. um And uh so that is quite exciting. Tell me about the, I guess what skill level would you, I guess what qualifies someone to work for Sapien and be able to add their new ones knowledge to digital objects like autonomous driving or humanoid robotics? I'm just trying to think like, practically, what makes someone qualified to be able to do that? Can anyone do that? ah No, is the short answer. I think technically, yes, but in reality, most people really struggle, particularly with the 3D, 4D data. It's really hard. It's difficult to understand what you're looking at. You're not looking at a photo or a video. You're looking at raw LIDAR data. And so it's not something everybody can do. It's actually a big part of the product that we're building over the next six months. And so again, we have like this semi-closed product today where most of the work that's happening is happening in private. And the reason it's happening in private is that we need to onboard particular people to particular tasks. And that might mean I need you to sign a non-disclosure agreement before you can see this dataset. Or it could mean I need you to do a skills assessment to prove to me that you can accurately understand the dataset and provide real valuable inputs before you're allowed to start doing the That's what's typically required for 3D, 4D data. Or if we're doing a medical model or an engineering model or something that requires real credentials, then we need to verify those credentials. And so all of these activities today are handled by our operations team who do a kick-ass job of juggling far too many things, but that's not scalable. That's not going to work. We can't get to tens or hundreds of millions of people with a small group making sure everyone has the right things. And so we're building something akin to like a standard onboarding wizard. And that will allow users who have gained enough reputation to onboard themselves to a specific task and then just follow through by signing the non-disclosure and uploading their credentials, which we'll check with a third party or doing a skills assessment. And then they'll have a record, you like, a sold-down NFT most likely, that tells us that they're suitable and capable of doing that. particular type of work, or they have the expertise that's needed for some subset of work. And so that's how we're planning to handle it. But yeah, today it's it's a bottleneck for sure. It's not easy. Are you also looking at, you know, the maybe potentially training people so that they can upskill and become better at whatever they want to do, but they're not quite good yet at it. Is that also part of the onboarding? isn't yet. it's not yet, not because we don't want to, but simply because it's in the nice to have bucket. We really need the kind of system up and running, ideally humming, before we can start pivoting to the things that we would love to do. And I think education is probably the single most important thing that we need to get done once the system is operational. Because I don't think people understand how valuable the knowledge they hold actually is. Mm-hmm. not everybody wants to participate, clearly. Some people just flat out think AI is net negative or for whatever reason. And so it's not going to be for everyone, but for us, we want to at least give everybody the option. And the best way to do that is to teach them that this exists, teach them that what they hold has value and show them how to monetize that value. Yeah. Rowan, I love what you guys are doing for various reasons. mean, there's clearly demand out there for the type of work that Sapien provides. um And the crypto aspect is also super, super interesting in its ability to coordinate incentives and behaviors and overall mechanism design. And I love the AI aspect because AI needs this. AI is hungry for data, good data, clean data. data with context and you guys are providing that. So I appreciate you sharing with the audience today what Sapien is doing. Maybe I can end with a couple of rapid fire questions that are non-Sapien related if that's okay. Who in the business world do you admire that inspires you to get up and improve the way you do work every day and how you treat people at Sapien? I actually don't get inspired so much by the big splashy figures that we see day in, day out. I actually take inspiration much more frequently from people I work with directly. And so I was at Coinbase for a few years. Brian Armstrong is an absolute beast. Another Coinbase plug, Alicia on the finance team is probably one of the most talented humans I've ever come across. We spoke a little bit earlier about a common friend that we have, Rob Figlione. people like this to actually inspire me more than the kind of like insert any big Silicon Valley personality that you may come across. So yeah. That's good. That's really good. What do you think in either crypto or AI is overrated, but for some reason is getting all of the attention right now? Twitter agents for the AI world if you're like crossing crypto x AI. ah Sure, AI can tweet for you, not convinced it's really adding much value. But it's where we are in the kind of free product market fit grifty part of crypto AI. Mm-hmm. Have you, I don't know if you've been following Kaido recently with all the mind share around uh loud or stay loud, but um there's been a lot of debate recently around mind share and whether or not it actually can command uh demand or price. And so that's kind of been interesting to watch, but that's kind of unrelated to what we're talking about. But yeah, that has to do with crypto and Twitter. I guess on that note, uh You know, most of crypto is primarily on Twitter. How do you think about other kind of distribution channels like TikTok and Instagram, et cetera? For us, because we're trying to onboard consumers and because they're not necessarily onboarding directly to crypto, we actually think that we need to really think hard about things like TikTok. We did some AGC style content. And for anyone that's unfamiliar, this is literally like agentic content creation where there's no real people. We just have AI model spin out some videos, some onboarding with some product placement, talking in periphery about like, hey, Sapien's cool, I can earn some extra income, blah, blah, blah. and you can create a good number of videos in a short space of time. And TikTok algorithm is all about like, consistency and quantity and you can get good exposure. We actually had very good results, probably onboarded 30, 40,000 people in the course of a few days. And so probably going to double down on more traditional social media. Crypto Twitter is somewhere I've spent a long time, probably too much of my life to be honest. But uh It's just not the right fit for Sapien specifically because the vast majority of people that are posting on Twitter, particularly in the crypto realm, probably don't want to do work for five, 10, 15, 20, even 50 bucks an hour. They're probably too busy farming the next big thing or degen swinging all kinds of shitcoins. So that's not our crowd. Traditional social media is really where we're likely to be doubling. Yeah, so you kind of glossed over the fact that the campaign that you did onboarded 20,000 people? Yeah, something like this. onboarded, like, I think it was a little over 30,000 people. ah that's actually huge. Can you share with us more about that? Those kinds of numbers are really, really impressive. Yeah, mean, I think everybody has a different, I guess, version of what good looks like for us. User signups, honestly, to me is a bit of a vanity metric. so great 30,000 people signed up. Much more interested in what did those 30,000 people do? How many of them were we able to meaningfully retain? How many of them actually participated and created quality data? And obviously when you start wiggling in like the real metrics that matter to us, the numbers become much less impressive. But from a raw onboarding perspective, you're absolutely right. Like it's a huge number of people, huge number of eyeballs in a fairly short period of time. Like we're talking days. And that's why we want to really double down. No, that's great. That's really good. um Early in my career, I started a online tutoring marketplace, which I eventually sold to a New York um Stock Exchange-owned, publicly held company. um Yeah, no, was actually a lot of fun. But in that two-sided marketplace, had tutors on one side, and then you've got parents on the other who are really the ones hiring the tutor for their child. and experiment with a bunch of different campaigns, but never had the types of results that you described with 20,000, 30,000 on board, um which is really, really impressive. yeah. think the difference, I'm sorry to cut you off a little bit. I think the thing that I think helps us is that. TikTok has an enormous number of eyeballs and attention spans are pretty short. Like the videos are usually like scroll, scroll, scroll, scroll. However, if you have an opportunity for somebody to make extra money on the side without spending money and without necessarily needing specific qualifications or having like some artificial barrier to entry, you're going to get a decent number of clicks. so that I think is really how we're able to get that number quite quickly. But again, it's TBC what happens 30 days later. Like this is still relatively recent. Maybe we can chat again in a few months. If we get a decent number of these 30K people to actually do meaningful work and contribute, then I'll be high-fiving the team and be like, yeah, that was epic. For now, like, it looks epic, but we'll see. Well, Rowan, thanks so much for taking the time to speak with us. I love what you guys are doing at Sapien. It meets a real need. There's an actual problem that you're trying to solve and you're doing it with both addressing it in the AI space, but also with crypto incentives and helping according to all the stakeholders. And I think that's exactly what crypto is meant to do. So it's beyond just farming. It's beyond just trying to make money, but actually helping to coordinate behaviors and I think that's really amazing what you guys are doing. So thank you. No, thank you. Really enjoyed the chat. Thanks for inviting me.