In this episode of The Sunny Ray Show, Sunny Ray talks with Marcus, co-founder of Prefina, about the future of personal data ownership and AI. Marcus shares his multicultural upbringing across Sweden, Spain, Finland and Kansas, his years as a naval navigation officer in the Finnish Navy, and his path through three tech startups focused on data infrastructure. He explains how GDPR and PSD2 first revealed that personal data could be a liability for institutions, sparking his belief that individuals should own and control their own information. The conversation moves into Prefina's approach of separating the data layer from the application layer, creating a portable personal data cloud that powers AI agents and AI twins. Marcus recounts the moment an AI twin answered questions for a colleague in front of thousands of people, revealing how people interact differently with AI than with humans. He closes with a vision where everyone's business card includes an AI twin that manages their agentic systems, much like email became ubiquitous decades ago.
A former naval officer turned entrepreneur explains why owning your data is the key to trustworthy AI twins and agents.
Is it true you grew up in Finland?
I was born in Stockholm, Sweden to Finnish parents, and we moved around a lot, including Sweden, Spain, and eventually the US, where I went through middle and junior high school in Kansas. Living in different countries shaped how I think about diversity, and it's part of why my family and I have stayed in Silicon Valley, where the filtering process that brings people here creates a unique mix of perspectives.
Before becoming a tech entrepreneur, you were a naval navigation officer. What did that time teach you about discipline and how does it apply to entrepreneurship today?
It reinforced my need for internal control, almost an OCD tendency to always be in the controlling seat. It also taught me that humanity's challenges are the same but the solutions are not, so working with very different people at sea trained me to build teams, since you can't do everything yourself and you have to learn how to talk to different types of people.
Your LinkedIn bio mentions you're a competitive athlete. Tell me about the athlete side and whether physical competition informs how you build a company.
I compete in trail and mountain ultramarathons now, like the Broken Arrow Sky Race in Tahoe. A mentor once told me that since you can't control everything in a startup, you can control your own fitness, and that became a healthy outlet for my need for control. Last year for my birthday I ran over a hundred kilometers to our lakehouse, including twenty miles through a flooded swamp.
At what point did you start to see that the current data model, beyond finance, was fundamentally broken?
It was clear when GDPR and PSD2 came out while I was running an API-first back office company for financial institutions. Clients who had been in a data land grab suddenly realized personal data was a liability, not just an asset. That made me think that if it's our data, we should be the ones benefiting from it, which became the seed for Prefina.
What's your philosophy on learning faster than the challenges unfold?
I believe there's no empirical science to building something, only trial and error, so the question becomes how many at bats you get. A startup shipping daily learns faster than an enterprise shipping yearly. I'd rather throw a good enough guess into the market and iterate than sit theorizing, because the second or third guess is almost always more accurate than the first.
The idea of separating the data layer from the application layer sounds architecturally radical. When did that insight crystallize and how hard was it to convince early investors?
It was much harder to explain before AI agents existed. Now it's clearer: if your personal AI is provided by one LLM company, you likely can't switch models. Separating data from the model, into what we call a personal data cloud, lets you carry your data across different specialized agents instead of feeding the same information redundantly into every application.
When did the AI twin concept first come to you, and what was the reaction when you first described it?
About two years ago, a colleague built an AI trained on his own data and used it at a talk in front of three and a half thousand people via a QR code. He got four hundred signups, nearly ten percent of the audience, and people asked honest questions in every language they wouldn't ask a human. That showed us AI twins solve real availability and comfort gaps, not just novelty.
What does winning look like for you, your company, and individuals everywhere?
In the future, your business card will just have your email, phone number, and your AI twin. That twin becomes your household staff orchestrator, running your agentic systems on your behalf once it knows and trusts you. It's similar to how email started as a niche enterprise idea before becoming something everyone assumed they'd have.
Building something daring? Sunny talks to founders like this every day. Fifteen minutes to see if your story belongs on the stage.
Claim your pre-interview