Sapien is a company that is pioneering the next frontier of AI training data through decentralized human intelligence. Pulse 2.0 interviewed Sapien CEO Rowan Stone to gain a deeper understanding of the company.
Rowan Stone’s Background
Can you tell us about your background? Stone said:
“I started my career in the energy industry before my best friend, who I owe my entire career to, introduced me to crypto. That led me to launch a mining farm and take a huge risk by dumping my life savings into equipment from a slew of early Chinese ASIC manufacturers with super opaque (read: terrifying) order processes. Things escalated from there.”
“I helped launch the Horizen protocol (ZEN), then went on to co-found Horizen Labs, an onchain software dev shop. After that I started Launch Code Capital with a few onchain friends. It’s a small fund that backs early-stage teams building in the space.”
“Later, I helped build some early DeFi charting tools and scaled the first B2B DEX aggregation platform, which we sold to Coinbase. I stayed there for three years, leading BD for onchain products like USDC and cbETH. Eventually I helped co-create Base, Coinbase’s Layer 2 network, where I led business and operations across go-to-market, partnerships, and more.”
“I left Coinbase about a year ago to build Sapien. It started as a decentralized alternative to Scale AI and has since evolved into an open protocol for AI training data.”
Formation Of The Company
How did the idea for Sapien come together? Stone shared:
“The idea started with a simple question: could we take the Scale AI model and build it in an open, equitable way? Something permissionless. Something anyone in the world could plug into. That’s where the spark came from.”
“For me, it also tapped into something deeper. I’ve always been into sci-fi, systems, and new technologies. That’s what drew me onchain, and it’s the same reason I leaned into AI. The potential for impact is massive, especially if you give more people a seat at the table.”
“The core problem we had to solve was quality. That’s easy to manage in a centralized facility with full-time staff. It’s a lot harder when people are contributing on the bus, between Uber rides, or patient visits. That’s why our first move was to zoom out and think about how to build a system that could actually scale under those conditions.”
“That’s where the idea for Proof of Quality came from. It’s the secret sauce that makes our whole system work. It’s what moves us from a bottlenecked tasking app to a fully open data foundry.”
Core Services
What does Sapien actually do? Stone explained:
“At the simplest level, we help AI companies get high-quality human sourced training data, and we give people a way to earn by contributing what they know.”
“On the surface, it looks like a tasking app. Under the hood, it’s a protocol built on four systems: Quality Assurance, Reputation, Qualification, and Matching. People start with simple tasks, build reputation based on accuracy, and unlock access to more complex and better-paid work as they go.”
“We’ve gamified parts of the experience to keep it engaging. Not to make it a game, but to help with retention. If you want good results, you need people to actually want to come back.”
Evolution Of The Company Since Launch
How has Sapien evolved since launch? Stone noted:
“We started off focused on simple image annotation. But quickly realized this wasn’t just a labeling protocol. It’s a data foundry, a place where companies can source, structure, refine, test, and fine-tune the human input (knowledge) their models need.”
“It’s still early days, but we’re seeing strong signals. Average session length is pushing 40 minutes and output quality is correlated with reputation and stake size, as we expected.”
Customer Profiles
What kind of people use Sapien? Stone highlighted:
“Everyone. We’ve got contributors who are artists, students, teachers, lawyers, musicians, engineers. Most people think training AI is something only ML experts can do. That’s not true. Everyone holds useful knowledge — cultural, contextual, intuitive — and we’re building the protocol to unlock it.”
Accomplishment
What are you most proud of so far? Stone cited:
“We’ve passed 1.9 million registered contributors across more than 110 countries. They’ve completed over 187 million tasks. That scale gives us reach, but more importantly, we’re starting to see strong session data. People are sticking around and doing meaningful work. Forty-minute average sessions tell me that we’re doing something right.”
“We’ve worked with some major partners. Toyota, Lenovo, Baidu, MidJourney, the United Nations, and Uthana, who we helped with a complex 3D animation workflow. But honestly, the customers I’m most proud of are two major autonomous vehicle manufacturers and operators. That’s the stuff that gets me out of bed.”
“I’m a huge fan of driving and flying, but I’m all for a world where I can hand off control, whether it’s a car, a boat, or a plane, and let AI take the reins so I can focus on whatever else I need to do. It goes way beyond Garmin autopilot, even if that’s already pretty epic. We’re helping build that future. We’re helping build Star Trek. I live for it.”
Challenges Faced
What have been some of the bigger challenges? Stone acknowledged:
“The AI space is moving at breakneck speed. DeepSeek was a recent example. They proved you can get strong results with fewer resources, and suddenly the whole landscape had to re-evaluate. A lot of teams panicked.”
“For us, it reinforced something we already believed. The next wave of AI performance won’t come from just throwing more compute at the problem. It’ll come from better data, and more diverse, specialised, human-validated data. That’s what we’re building for.”
“The other challenge, which I already mentioned, is quality. We’re asking people to contribute from all over the world, often in their downtime; between classes, during commutes, or after work. Keeping quality high in that environment is non-trivial. But it’s also where our model shines. Peer-powered review systems and reputation-based task routing let us operate without needing centralized oversight. That’s the unlock.”
Long-Term Vision
What’s the long-term vision for Sapien? Stone emphasized:
“We want to make it super simple for any company to access verified human knowledge, and for anyone, anywhere, to get paid fairly for sharing what they know. That’s the core loop. If we get that right, Sapien becomes the connective tissue between humans and machines.”
“Over time, we see this evolving into a full-blown data economy. Not just labeling or collection, but the whole stack; from sourcing to structuring to feedback loops and fine-tuning. If you’ve got skill or context or judgment that’s useful to a model, you should be able to plug in and earn.”
“We’re building the infrastructure to make that happen. And we’re doing it in a way that’s open, onchain, and permissionless, so that over time, it doesn’t rely on us at all.”
“I’m not here for a quick exit. This is a long game. There’s real potential to shape how humans and AI learn from each other, and I don’t plan on sitting that out.”
Funding
Can you share anything about funding or market size? Stone revealed:
“We raised a $10.5 million seed round in 2024, led by Variant. That’s given us the runway to keep building out both sides of the protocol: the contributor experience and the enterprise interface. We’ve been able to move fast without compromising on pace or principles.”
“As for the market, it’s massive and only getting bigger. The demand for high-quality training data is compounding fast. We think the addressable market for AI data alone could hit $100 billion over the next decade. But the real shift isn’t just size. It’s how that data gets sourced. Centralized approaches won’t keep up. That’s where Sapien comes in.”
Additional Thoughts
Final word? Stone concluded:
“It turns out teaching machines is a job. We’re making it one anyone can do, and get paid for.”
