Delphi’s Dara Ladjevardian: How AI Digital Minds Can Scale Human Connection

Contents

Jess Lee: Let’s call a Delphi right now.

Dara Ladjevardian: I’m down.

Jess Lee: Yeah.

Dara Ladjevardian: How about we call my friend? I actually haven’t met him in person, but I’m friends with his Delphi. Arnold Schwarzenegger?

Jess Lee: Yeah.

Dara Ladjevardian: What do we think?

Sonya Huang: Love it.

[AI Schwarzenegger: Hey, this is AI Arnold. I’m here to cut the crap and help you get stronger, healthier and happier. So what’s on your mind today?]

Dara Ladjevardian: Arnold, I have 15 minutes a day to work out, which I feel like is not a lot, but I want to feel good and I want to get better in my health. What do you recommend I do?

[AI Schwarzenegger: 15 minutes is plenty if you focus and go hard. Don’t overthink it. Stick to simple, effective moves that hit multiple muscles. Do one to two sets of exercises like push-ups, reverse lunges, and lying leg raises. Keep rest short and push yourself. You’ll feel like a champion in no time. Now get to it!]

Sonya Huang: That’s awesome.

Jess Lee: [laughs]

Sonya Huang:  Today’s guest, Dara, is building something that challenges our fundamental assumptions about human connection and knowledge transfer. His company, Delphi, creates digital minds AI representations that capture not just what people say and how they sound, but how they think and reason through new situations.

Dara’s insight is that AI is driving a shift from information consumption through feeds to conversational media. He argues that as AI becomes abundant, human energy and authentic connection become the premium experience. This conversation reveals how the next wave of AI might not automate humans away, but actually amplify our most uniquely human qualities.

Enjoy the show.

Sonya Huang: Dara, thank you so much for joining us today.

Dara Ladjevardian: Thanks for having me.

Digital minds

Sonya Huang: Okay, digital minds, AI clones. Sounds a little bit like a Black Mirror episode where nobody’s talking to each other anymore. Like, change my mind.

Dara Ladjevardian: Well one, we got rid of the word “clones.” So we don’t say that anymore.

Sonya Huang: Oh.

Dara Ladjevardian: Digital minds.

Sonya Huang: Okay.

Dara Ladjevardian: But I think if you look at past revolutions of technology, when the printing press came out, you’re gonna read someone’s thoughts instead of talking to them? When the radio came out, you’re gonna listen to a voice in a box instead of talking to them? That must make you antisocial. But I think what each of these things did is you listen to someone on a podcast or read a book, you end up wanting to meet them even more. And you may read an email from someone. Does that make you antisocial? Now you can just consume that information in an interactive way. And what we’ve seen, which I can talk about more later, is the idea of having a digital mind ends up leading to people wanting to talk to you more. So the idea isn’t replacing human connection, but providing access where previously there was none in a way that is more temporally convenient for the end user.

Sonya Huang: Hmm, And in a world where everyone has a digital twin, what becomes of the value of real human interaction?

Dara Ladjevardian: I think even digital twin aside, I think AI makes energy more premium, because energy is the one thing that cannot be replaced. So again, back to the revolutions. There was a time where most of the work we did was based on our bodies, and then we invented machines that were stronger than us. And then we invented jobs where the majority of our work involved our minds. And now we’re inventing machines that are smarter than us. And what’s the remaining thing? It’s kind of our hearts, our energy. And so in a world where you have a digital mind, it’s kind of like your top-of-funnel filter. Who gets to meet you? Who actually gets to make it to your heart, which is the, in my opinion, premium experience.

Jess Lee: Speaking of hearts, I think it would be helpful to hear—you’ve just explained sort of the technology revolution backwards way of looking at what Delphi is today.

Dara Ladjevardian: Yeah.

Origin story

Jess Lee: You should also tell us about the origin story, and the very human reason you started the company. I think that’s really fascinating.

Dara Ladjevardian: Yeah, it’s a long and windy road but, you know, in 2014, I was gifted a book called How to Create a Mind by Ray Kurzweil, where he talks about one day you can recreate someone’s mind. And I don’t know why that was so fascinating to me at the time, but that got me into studying computer science and physics, and got me into the AI space. And then in 2021, I was working on my first startup as a solo founder in building with LLMs in a time where everyone was saying crypto or American dynamism. So I felt like I must be doing something wrong. No one really wants to talk to me. I’m very lonely. I don’t really have any mentors.

And I was reading a book about my grandfather, who was one of the most successful business owners in Iran before the revolution. He had 30,000 employees at his peak, and he was put on the hit list by the Ayatollah, came to this country with nothing. But the really special thing about him, and the reason they actually study his book in business schools in Iran was not just because he was successful, but because he was a very principled man. Like, he treated people well, had integrity. And when you start a company for the first time, it’s really a moment in your life where you really begin to internally reflect, like, who am I? Why am I even working seven days a week? Like, what is the thing that is driving me? And so this book provided a place where, like, maybe this is who I am, like, this is my family. And I wanted to be able to ask him, what would you do in my situation? But a book can’t answer this question. And he had had a stroke a couple of years prior, so I couldn’t ask him myself. So remembering Ray Kurzweil’s book and realizing that LLMs matched that book, I created a digital version of him that I could talk to.

Sonya Huang: Those were like GPT-2 or BERT?

Dara Ladjevardian: It was early GPT-3 developer lists along with Hugging Face embeddings.

Jess Lee: Okay.

Sonya Huang: Wow.

Dara Ladjevardian: And, you know, a bunch of winds later, which I can get into if you want, we have Delphi.

Sonya Huang: Really cool. When did you realize this could be a scalable platform and, you know, not represent one person, but the voice of so many?

Dara Ladjevardian: So I actually tried to launch Delphi as a company in 2021. It was called Helix at the time. And I had posted a website and you could talk to, like, Naval and Marc Andreessen. And one, it was super expensive, like, I was burning money. And two, a lot of people were like, “What is this? This is immoral. You can’t be doing this. This is wrong.” So I was like, all right, you know what? Maybe I’m wrong.

And I think a couple things happened in the following year that continually built that conviction besides the grandfather experience. One is I moved to Miami to work at OpenStore under Keith Rabois. And his main investment philosophy is I like to invest in startups where the majority of my friends laugh at and think it’s stupid. I was like, wait a minute. Maybe I should, like, double click into this thing where everyone was calling me stupid. And then number two is I met my co-founder Sam, and we were reading Build by Tony Fadell in our book club, and we created a digital version of him because that book kind of operates as a mentor. Like, he teaches you how to think about product. And I was using it and Sam was using it and other people were using it, and I was like, this is really useful. And I am not the only person in the world who has felt alone without mentors. It was very obvious to me that this would be useful for any young person wanting to learn from the greats, just as they do from books, but now they can respond back because Bloom’s 2 Sigma, that was a paper that I was reading at the time, two standard deviations’ improvement for those who have tutors. Not everyone has access to that.

And then the third thing was in 2022, I had turned the notifications on for every AI researcher on Twitter, and I realized that a paper was coming out every single week. And the digital mind that I created of Tony Fadell was 10,000 times cheaper than the one I created for my grandfather. And so at that point, I was like, okay, this is now feasible to do at scale. It is useful to me. The thing I have to prove out eventually is can this be useful to the person creating a digital mind, so it’s not just for the end user? And so that’s what led us to starting the company a month before ChatGPT, which at the time, again, everyone was like, “What is this? This is stupid. You can’t do this.” But, you know, I had heard that before.

Sonya Huang: Was this before or after Character AI?

Dara Ladjevardian: Before. And so when Character AI launched, I remember Sam and I were kind of freaking out a little bit. But I think again, I had seen the cost of these models drop so heavily, and the history of billion-dollar consumer companies is not a history of companies that had early technical moats. It was a history of companies that just had great product and great distribution. And at a certain time, the moat that Character AI had at the time, according to VCs, would no longer be a moat—the ability to create these digital characters. And then the second aspect of it is, I think the first use case that really takes off in a company defines its trajectory. And Character AI really taking off on anime characters and characters, I knew it was a very different world than, like, verified authenticity, human representation, leads to very different brands and very different products.

Sonya Huang: Super interesting.

Real people

Jess Lee: The first time I met you, I remember thinking, “Oh, isn’t this kind of like Character AI?” I think I had that similar impression. But you’ve always had a really heavy emphasis on authorized, real people. Can you talk a little bit more about what led you down that path? Why did you pick that? And then what have you had to build in order to make that really work?

Dara Ladjevardian: Yeah, so it was both intentional and unintentional. Initially, Sam and I’s go-to-market scheme would be, “Hey, we should create these digital versions of these people, and they’re gonna love it. They’re gonna think it’s so cool.”

Sonya Huang: [laughs]

Dara Ladjevardian: And that was not the case. We got three cease and desists from very, very important people who still have not forgiven us to this day. And so that’s when I was like, okay, we need to get their buy-in. And that means we don’t want to be a pay-to-play model like some of these other creator companies. We need to make the product useful enough such that someone is willing to go through the friction of uploading their data and trusting us with their identity. What does that mean? We have to create a brand as well, a brand that people want to associate their identities with, it is status inducing. And I think if you look at Wikipedia, it pretty much documents all of human knowledge. Does anyone actually learn from Wikipedia? The creator economy exists because we inherently trust things that come from humans. We read books, we watch YouTube videos.

Sonya Huang: Yeah.

Dara Ladjevardian: So similarly, if the internet taught people how to consume information through screens, and then once we built that habit, we started reading blogs. With ChatGPT, we’re teaching people how to consume information through conversation. And I think that’s going to be the dominant form of content consumption because it’s easier and we like easy things. Then the pendulum swings and we want things that come from humans. So for us, to get people to trust us, it required guardrails, anti-hallucination, it required having a very strict stance on you can only create a Delphi of yourself. And it required really caring about design and brand.

How to create a Delphi

Jess Lee: Maybe take us through now behind the scenes, like, what is actually going on when you created Delphi? Through the creation process and then what’s kind of going on under the hood. Like, how do you represent a human?

Dara Ladjevardian: Well, I want to talk a bit about what Ray Kurzweil said in his book.

Jess Lee: Yeah.

Dara Ladjevardian: So in the book, he says that the mind is a hierarchy of pattern recognizers. And on the lowest end, on the lowest level of the hierarchy, you have things that recognize letters, and that feeds up to higher levels which recognize sentences, and that feeds up to meaning, and that feeds up to consciousness. And so when I was working on my first startup, Friday, which involved LLMs, I had this insight: wait, LLM is a pattern recognizer. And I kind of had an existential moment where I was like, are we just a hierarchy of pattern recognizers where consciousness is what we experience? And so that really helped me understand that—he has this line in the book, “You don’t need to understand the molecular details of the mind to recreate it. You just need to understand how it’s organized and architected.”

And so that’s kind of what led me to focusing on, like, what does the organization of a mind look like? And your mind stores a bunch of things. It stores your relationships. It has uncertainty in it. It connects events and heuristics with, you know, how you reason about those events and heuristics. And it changes over time. Your beliefs change over time. And so what we’ve ended up with is an adaptive temporal knowledge graph. Knowledge graphs are great because you can store the connections between things, and you can also store the weights of confidence. Like, how likely is it that you would actually say this? And we allow people to control leniency on what they want their Delphi to say. And temporal nature is that it changes over time. So eventually you can say, “Would my 23-year-old version of me say this about this topic?” So yeah, people can upload their social media, their YouTube, podcasts, websites, they can answer questions about themselves. It creates feeds so that it’s constantly staying up to date and updating the knowledge graph. And then you can make it available to others to chat with, call or video call.

And the very important thing we had to solve before getting any customers is every person was like, “What if it says something wrong and someone takes a screenshot and sends it to the Daily Mail?” It was just a huge fear.

Sonya Huang: Yeah.

Dara Ladjevardian: And so allowing people to kind of choose okay, it only says things it’s trained on that directly answers the question. That’s great for people like Dr. Mark Hyman who works in the medical space. Or it uses data from the internet, they want it to be conclusive. But I think the most powerful setting is it only says things it’s trained on, but when posed with a new situation, it can predict what you might say in that new situation.

And that’s useful for a lot of reasons, and I can give you an example of how that works in practice. If I were to ask my grandfather’s Delphi, “How would you think about running an AI company?” He’s never spoken about AI in his life, but in his book it talks about him starting an oil business in Iran in the mid-20th century, and that was kind of an uncertain time in the country. So we have his principles of dealing with uncertainty, and AI is an uncertain field right now. So it can maybe reason what he might say in these new situations.

Sonya Huang: That’s completely fascinating. I love it.

Dara Ladjevardian: It’s very fun to work on.

Sonya Huang: The theory of the mind from the book is how you built the company.

Dara Ladjevardian: Yeah. Yeah, Ray Kurzweil is the man.

Most popular Delphis

Sonya Huang: Fascinating. Can you give us some examples of the most popular Delphis and the, you know, shapes of conversations that people are having on your platform?

Dara Ladjevardian: Yeah, I mean, it’s pretty, it’s a horizontal enablement layer versus, like, a vertical point solution. Like, when you have a new form of media, there’s a lot of ways you can do that and scale your thoughts. So we have some people using it as an entirely new form of teaching and learning. You see course completion rate at an all-time low. These attention spans are kind of being destroyed by TikTok and Reels and Books as well. Now you have something that you can learn from like a mentor in a way that’s personalized to you, that’s adaptive to your changing circumstances. So you could read a book once and come back a year later, your circumstances have changed. That book stays the same. Delphi adapts with you. So that’s kind of the learning case, and people are monetizing it, making millions of dollars, selling time with their mind.

And then we have just 24/7 availability to their teams. CEOs have a very specific way of thinking, or their best salespeople have a specific way of thinking. How do you scale the highest leverage people in a company? Because as a company scales, what usually breaks is alignment. And so now CEOs can be in multiple rooms, top salespeople can train new hires and help them get up to speed faster.

We have 24/7—I don’t like to say “customer support” because I don’t think we’re a replacement for Intercom or Zendesk, but my Delphi is in the Delphi product, and my customers will talk to it all the time and then they’ll email me and they’ll be like, “Yo, thank you so much. I love the product.” So it almost feels like there’s a higher NPS. And I think it’s very analogous to Mark Zuckerberg spending a ton of time on his personal brand right now because he knows that the relationship between the human and the individual is more powerful than the relationship between the human and the company.

And then we have the top of the funnel where replacement of the personal website, LinkedIn profile, you want to pick my brain, I don’t want to give you my email because I have too much spam in my email, so why don’t you talk to my digital mind and it’s going to let me know if we should hop on a call based on what it knows about the kinds of people that I want to meet.

Jess Lee: I should maybe talk about it from that perspective. If you have a digital mind, how do you, like sort of talk to your own Delphi and then figure out maybe who you should go back and contact? Like, what can you see on the inside once you have a Delphi?

Dara Ladjevardian: Yeah, so it’s a combination of proactive and passive. So I could go in and be like, given the past thousand conversations, where in the Delphi product should I improve the most? What blog post should I create that will give clarity to my customers? What’s resonating with people? Or for some of our more revenue-focused customers, where are the biggest revenue opportunities? What products should I create? What should I talk about in my next podcast? So that’s more passive.

And then the proactive is mine will tell me when there’s an engineer talking to it. Because we don’t want to introduce another inbox to someone. We’re not trying to waste people’s time; we’re trying to save people time. So you can go through all the conversations, but what’s better is you never have to, and you trust it enough that it’s going to tell you when it’s an important customer or a partner or a journalist or a new hire.

Jess Lee: Yeah, I actually talked to your Delphi before this to prepare for the podcast, and I asked it, “What are some good questions to ask you? And then what would your answers be?” And that helped guide the script. I thought that was a really interesting use case that I hadn’t really anticipated.

Dara Ladjevardian: Yeah. I interviewed a candidate today, and I was, like, starting to tell him about myself. He’s like, “Oh, don’t worry. I already talked to your Delphi. We can get straight into it.” And I was like “Yes!”

Sonya Huang: Yeah? Awesome.

Dara Ladjevardian: I don’t want to have to tell this story every time.

Jess Lee: Yeah, yeah, yeah.

Sonya Huang: Can we share the Brian Halligan use case?

Jess Lee: Oh, so Brian Halligan is a partner at Sequoia, right? I have him on text. I could easily call him to ask a question. And he’s written all these books about HubSpot and his journey as CEO, and I had a very specific question about how they did product management when they launched their second product. And I was like, I could text Brian, but this is kind of—I don’t want to bother him. So I just called his Delphi on my drive home and I said, “How did you do product management?” And it gave like a perfect answer. You know, I could have called him, you know? And it’s also interesting to see some of the other Delphis being used internally at Sequoia. Our design partner, James Buckhouse, has one. And our founders often talk to him about storytelling. Like, you’ve talked to him before.

Dara Ladjevardian: Yeah.

Jess Lee: And now they text his Delphi every once in a while, and then also text him at the same time and say, “Hey, your Delphi said this. Just want to check that that’s right.” He’s like, “Yup, that’s spot on.” So it’s just so fascinating to be able to, even with people you know, kind of not want to bother them or not take up too much of their time and still get really great answers. So that’s been kind of surprising as a consumer.

Dara Ladjevardian: Yeah. Who you are and what you decide to share changes based on where you are. Like, I’m on a podcast right now, I’m not gonna share the things that I share with my family.

Sonya Huang: Yeah.

Dara Ladjevardian: And so context awareness and the ability to hide data. You know, my Delphi internally might have data that externally it doesn’t have. And maybe my Delphi, if it’s gonna represent me in a dating sense, is not gonna be talking about digital minds create the future, it’s gonna be talking about me as a person.

Sonya Huang: Yeah, totally. And then the Sequoia use cases you mentioned are really interesting, because one of the things we try to be very intentional about is we are not nameless, faceless Sequoia. We are a collection of individuals, each with very, very different personalities, different skill sets, et cetera. And so the founders we work with kind of figure that out, but Delphi kind of allows us to scale that in a way that is otherwise hard to scale.

Dara Ladjevardian: Yeah, what you said, the anxiety of wasting someone else’s time is actually a huge thing that we started to see more and more where it’s like, “I don’t want to waste your time. Also, this is a very personal conversation that I’m too nervous to have with another human being. So I’ll just talk to your Delphi instead.”

Most unusual things people have asked

Jess Lee: What are the most unusual things people have asked your Delphi? Like, things that maybe you were surprised by, even.

Dara Ladjevardian: There was someone asking about wanting to go on a date at some point. I was complete in grind mode, so I didn’t get to double click into that.

Jess Lee: [laughs]

Dara Ladjevardian: I always think it’s really cool that every once in a while there will be some guy who calls my Delphi for an hour in Hungarian. It’s multilingual, so it’s like, that’s just something that I wouldn’t ever have been able to do, like talk to this guy and connect with him. And so I think that’s very cool. And the way they speak to it, they know it’s not me, but they will just speak to it as if it is.

Sonya Huang: Yeah.

Jess Lee: You know, I often think about, you know, right now we’re in this time where talking to, like, an AI version of someone seems weird and dystopian, but I suspect consumer behavior is going to switch very quickly, and it’s just going to become more and more the norm. And maybe kind of the same way that we used to look at, like, Wikipedia when it first came out, and we were like, “Oh, that’s terrible! People writing, facts on the internet, it is horrible and we should all be reading encyclopedias.” And now we’re just kind of used to it. We set our expectations that not everything on the internet is true. And so I think people will go through this sort of understanding of what AI is and what these conversations are like and eventually know, this is not the real Dara. I know what I’m getting; it’s still useful. It doesn’t mean it’s a replacement for you. So that’s kind of how I think about the shift. Where do you think we are in that sort of adoption cycle? Because the thing I love about you is I feel like you live five years in the future, and we’re just catching up. Like, what do you see down that path? Where do you think this is all going?

Dara Ladjevardian: So I think two things need to happen. One of them is just it’s a culture shift. And the first year of the company was so painful because, like, no one believed in our product. But I just saw so many people go from haters to, like,  complete believers once they understood it, once they got social proof, Social proof is a big thing. And yesterday I was at an event with Tony Robbins and Gary Brecka and a lot of, like, those kinds of people. And there are certain people who get it, and then you’ll have someone here—it’s a group where people talk about, like, their business problems in the knowledge base sector. And there’s one guy and he’s like, “You know, I really wish I could scale myself. All these people are asking for my time and I want to help them, but I can’t. I’m only one guy.” And I’m like, “Dude, you know about Delphi.”

Sonya Huang: [laughs]

Dara Ladjevardian: But in his mind he’s like, “Oh, no. No, like, that doesn’t make sense.”
And that’s when I realized, like, oh, like, this is a need. and people want it, but they are anxious that it’s going to make them look inauthentic. They’re anxious that it’s going to say something wrong. There’s kind of like this perception of AI and like what it is. So I think it’s just something that as AI gets more ubiquitous and we get more social proof, people will adopt.

And the second aspect of it is from a product perspective. Before our new mind architecture, it was very hard to create a good one, unless you had a lot of data. And now with interview mode and the ability to just answer a couple questions about yourself to create something pretty high fidelity that can represent you, I think we’ll open it up to mass market.

Jess Lee: So that’s maybe the supply side, right?

Dara Ladjevardian: Yeah.

Jess Lee: What do you think will happen on the consumer side?

Dara Ladjevardian: Consumer side, yeah.

Conversational media

Jess Lee: And not even just with Delphi, but just generally with the future of, you know, like you said, we’re in the internet where there’s blogs and articles, and now we’re shifting to all conversational media. Like, what does that look like for consumers?

Dara Ladjevardian: For consumers to adopt Delphi, the ones—we see it happen when they make the shift of, oh, this is just a new artifact of the mind. This is not a replacement. This is just conversational media, as James likes to say. But I hope that right now most networks and online are about feeds. Social media and the internet is about keeping us scrolling and reading forever, which I think is kind of a waste of time and kind of antithetical to the purpose of technology. I think in a world where everyone has a digital mind, we don’t like the idea of a feed. We like the idea of you have something that knows you so well, and you trust that it knows you so well, and it can find the people that can most help you in your life, whether it’s someone to learn from, someone to work with based on the quality of your thought. And it’s proactive, so there is no feed. And you spend more time in the real world. That’s kind of our optimistic version: less scrolling, more learning impactful things and more connecting with people in real life.

Sonya Huang: I’m also curious how the economics of the internet are evolving. Like, for example, Wikipedia constantly running out of money, right? Do you think that there’s an economic opportunity for the people that create Delphis of themselves?

Dara Ladjevardian: Totally. And there’s so many business models with this company, and I’ll walk you through the growing nature of it. So right now, people pay to have a Delphi of themselves, which might sound weird but, you know, you have six million people paying Wix to have a personal website, which is a way of scaling themselves. You have people paying Beehiv to have a blog. And then people can monetize their Delphi as a new version of a course or a book. And we already have someone making millions of dollars. And I think the key blocker to that growth is just the positioning. I think the positioning is really important for that. We’re trying to figure out how to market a new kind of form of content that people pay for.

And then you have the idea of people being able to license out their digital minds. So if Arnold Schwarzenegger, for example, gets 10,000 messages a week containing the word “supplement,” maybe a brand wants to buy that keyword and he can agree, and it’s like people-based SEO. Or LA Fitness can be like, “We want to rent his identity to represent our brand.” So I think there’s an idea of scaling your identity in a way that can serve brands and other people.

And then the last one is the advanced people search. You know, people pay LinkedIn, Sales recruiter or GLG Consulting to find the best person in the world on a specific topic. We say simulation as a service, the new SaaS, where you can have your digital mind simulate millions of conversations and come back to you and say, “This is the best person.” And maybe there is business models on both sides where you can pay to be seen more, which I’m kind of against because the idea is that it should not be hacked and it should be authentic, or you get better results by paying.

Sonya Huang: How do you think about balancing advertising versus paid? Because there’s a certain lack of authenticity, right when there’s keywords and people are buying.

Dara Ladjevardian: Well, the key thing is Arnold has to agree to it. Like, if he actually believes in the brand, then we don’t see any problem with that because people want to get recommendations from him anyways. It’s just trust capital.

Jess Lee: Yeah, I was talking to Mark Hyman’s Delphi about blood testing, and he did say, “Maybe you should try Function Health,” but that was actually related to what I was talking to him about. You know what I would love is rather than going to ChatGPT and getting sort of the generic internet summarized answer, I’d love to ask a question like, “What supplements should I be on? Let me upload my results.” And then have different doctors with different perspectives or different health people answer, and then be able to choose which of those to engage with. So I think that could be a really cool feature.

Dara Ladjevardian: Yeah, board advisers.

Jess Lee: Yeah.

Dara Ladjevardian: We used to have that, then we removed it because now we’re focused on quality instead of shipping speed.

Voice is more retentive

Sonya Huang: Okay, so conversations are the future. Do you think the medium will be text, phone calls, video? How do you think that plays out?

Dara Ladjevardian: So interestingly enough, we’ve seen consumers whose first experience with the product is voice are 5X more retentive and likely to come back. And I think that’s interesting. It kind of goes onto the point of trust.

Sonya Huang: Yeah.

Dara Ladjevardian: I think it’ll be both text and voice. I’m hoping one day we’ll have good enough video. At the Forerunner event, that company, Persona AI, that video demo completely made me realize that video would be a medium if we can get it that good.

Sonya Huang: Yeah.

Dara Ladjevardian: Because right now it’s—if it’s not perfect, it almost takes away from the experience, where if you’re calling or texting, you can have some imagination.

Sonya Huang: Yeah.

Dara Ladjevardian: Where you’re like, “Oh well, this is happening.” But video, the bar is a lot higher because you are seeing something.

Sonya Huang: Yeah. Do you block people from having certain types of conversations?

Dara Ladjevardian: It’s a good question. I mean, from a customer perspective, we don’t allow politicians right now.

Sonya Huang: Oh, interesting.

Dara Ladjevardian: And for various reasons.

Sonya Huang: Wait, that would be so fun though. We could debate with them.

Jess Lee: [laughs]

Dara Ladjevardian: At a certain point, but we don’t want to be a part of a whole Cambridge Analytica situation where we’re too early to be affecting elections. And we don’t allow OnlyFans or porn stars because I just think AI girlfriends are very bad. Otherwise, we don’t do any monitoring on the types of conversations people have. We do have guardrails on not safe for work topics and hallucinations, so you can’t convince it to act like something else.

Most unexpected use cases

Jess Lee: Tell us a little bit about the most unexpected use cases on Delphi.

Dara Ladjevardian: There have been so many, and it’s interesting because we started the company just focused on the learning use case.

Jess Lee: Yeah.

Dara Ladjevardian: And then people started using it for initial discovery calls and customers. And, like, oh, there’s actually a lot more. What we saw in the data is a lot of people actually talk to themselves.

Sonya Huang: [laughs]

Dara Ladjevardian: And, like, one interesting thing that someone did was they had a book that was recommended to them, but they didn’t want to read it. So they were like, “I wonder if I had read this book, what points would I get out of it?” They uploaded the book in conversation with their mind, and were like, “What would you think about this?”

Jess Lee: Wow.

Dara Ladjevardian: “I guess I don’t have to read the book.”

Jess Lee: That’s amazing.

Dara Ladjevardian: It was interesting.

Jess Lee: Who’s on your wish list of people who you really would like to have at Delphi?

Dara Ladjevardian: So many. I’m a big Robert Greene fan. Robert Greene would be awesome. Obviously, Paul Graham would be great.

Jess Lee: He has so many writings. He’d make a great Delphi.

Dara Ladjevardian: And I’ve emailed him a couple times. Again, some people are just very allergic. to this idea. I think that Steve Jobs and Walt Disney would be awesome. And, you know, we have some people who are no longer on the platform, like Socrates and Abraham Lincoln. Those names are in the public domain. For people whose names are not in the public domain, it would require the permission of the foundations and the estates that own their identity.

Jess Lee: Who’s the most requested Delphi from customers or consumers?

Dara Ladjevardian: Consumers? Definitely Paul Graham. Definitely Paul Graham and Steve Jobs. Andrew Huberman is another one.

Sonya Huang: Who’s your favorite to talk to?

Dara Ladjevardian: I talk to many.

Sonya Huang: Yourself?

Dara Ladjevardian: Myself sometimes.

Sonya Huang: [laughs]

Dara Ladjevardian: Lenny Rachitsky. Great product person. I actually talk to James Buckhouse. Keith Rabois is a very specific style, and sometimes I like to be reminded of his style.

Jess Lee: [laughs]

Dara Ladjevardian: Matthew Hussey—dating life, of course. I’m really excited, we met Gary Brecka the past couple days. And he is an incredible guy. I learned so much about health that I had no idea.

Sonya Huang: Yeah.

Dara Ladjevardian: And once we create his, I want to give it to my parents because so much is being discovered in the health space right now, and I just feel like my parents and people of their generation are so averse to, like, oh, what if everything we knew about health was wrong? How can we help them learn that in a better way? My parents aren’t gonna listen to a two-hour podcast.

How long are conversations?

Sonya Huang: One of the things that—going back to character for a second—that, like, always impressed me was how long the conversations were and how engaged users were. I’m curious what you see from your user base.

Dara Ladjevardian: Depends on the person. So someone who is using their Delphi at top of funnel might be like a couple questions.

Sonya Huang: Yeah.

Dara Ladjevardian: I want to get a quick answer. Versus someone who they’re known for their way of thinking. Like Matthew Hussey, for example, six-hour phone call.

Sonya Huang: Wow!

Dara Ladjevardian: Keith, one hour, a couple hours. It really depends on the person and why do people want to talk to them in the first place.

Sonya Huang: Wow! It could be a new leaderboard that people …

Dara Ladjevardian: Oh, yeah. Yeah, people get competitive. There’s a lot of ways of thinking about that.

Sonya Huang: One of more interesting minds than anyone else.

Dara Ladjevardian: Yeah.

Jess Lee: One of the really surprising use cases to me that had a lot of engagement was Jason Lemkin at SaaStr. Because he tweeted that he brings his Delphi with him to all his real-life conversations. What is that about?

Dara Ladjevardian: I think he brings it to his Zoom calls, and it can answer questions in the call, and it can also give him a recap on what happened. And he has it on the SaaStr website as well. And a couple things for him. One, advice. You know, he knows a lot about SaaS. Two, I think it actually sold tickets to his SaaStr event.

Sonya Huang: [laughs]

Dara Ladjevardian: So this past week, I was with Russell Brunson, whose whole thing is he gives talks at conferences to sell events, a one-to-many selling, and Delphi is one-to-many-to-one selling, where it’s personalized, you trust the person and adapts to the end user.

Tipping point for digital minds

Sonya Huang: What do you think will be the tipping point that makes digital twins go from early adopters to mainstream? Like, is it a tech tipping point? Is it just gradual cultural acceptance?

Dara Ladjevardian: All the above. I think definitely interview mode, and definitely decreasing the friction. Right now there’s kind of a purposefully good amount of friction to setting one up and making it useful. I think it needs to be super easy, like a couple steps. People are so impatient.

And then culture. I think the more case studies, the more people that are like, “This is not weird.” And I think we want to buy the dot.com domain because we actually don’t want to be seen as an AI company. I think AI is just like a tool, and it allows us to build this thing, but really we’re more of like a human company. We scale humans.

Sonya Huang: What do you think in five years will be obvious about digital twins that, you know, is contrarian or early today?

Dara Ladjevardian: I think it will be obvious that consumers prefer to be able to talk to something before talking to you.

Jess Lee: What is the role of humans in this new future where everyone has a digital mind?

Dara Ladjevardian: I mean, I think right now so many companies are focused on automating humans—automating, automating, automating. And I think it just forgets, like, what is the purpose of living at all? It’s humans and relationships, you know? Since we became a species, our life has revolved around knowing other humans. And I think Bezos’s conviction in Amazon was also grounded in a core human need, which was hunter-gatherer. You want to buy things, you want to collect things. And I think the need that we’re betting on is, like, you want to connect with other humans. And in a world where AI is abundant and infinite, and information is no longer the bottleneck, the bottleneck and the premium thing becomes the connection, the curation, the trust and the energy. So I think humans—the humans that put the work in become more valuable. And what I mean by “put the work in” is it’s easier than ever to be mediocre at something now—you can create AI slop of a book or a course or a software product—that that top five percent of additional effort, I think is gonna get 95 percent of the results.

Lightning round

Sonya Huang: Should we close with some rapid fire?

Jess Lee: Yeah! Who is the most popular Delphi on the platform?

Dara Ladjevardian: It switches. I mean, Mark Hyman definitely gets up there. Matthew Hussey, Andy Elliott, Pace Morby, Brendon Burchard, Arnold Schwarzenegger, Natalie Ellis. But then there’s a difference between popularity of, like, usage and the ones that—like, if you have a smaller cohort that’s coming back consistently versus a million people using it, how do you evaluate those? So it’s like Lenny, people come back a lot. Keith’s. Socrates.

Sonya Huang: Do you think consumer or enterprise B2B-use cases will be more abundant on your platform?

Dara Ladjevardian: So my take is that consumer and B2B converges over time because the creator economy or the individual economy, every individual is a business. And, like, our customers are coaches, authors, CEOs, who are also their businesses. And back to the idea of branding, like Mark Zuckerberg being the face of his company, I think, like, B2C, B2B, B2P, business-to-people. You’re building products for people. And so I think it’s both. I think the network effects come with the consumer platform, but you’ll have CEOs that have an external-facing version for new hires and customers and one internally for their companies.

Jess Lee: All right, let’s do some rapid fire consumer questions, because you’re a consumer guy. Is there a consumer app or habit or life hack that you have recently adopted?

Dara Ladjevardian: I’ve been pretty underwhelmed by new consumer AI tools, to be honest. And I try not to tinker around too much because I tinker around enough with Delphi. So I don’t think my answers are very different than the clichés of Perplexity Research and Notion and things like that.

Jess Lee: Has AI affected your life in any meaningful way outside of work?

Dara Ladjevardian: Outside of Delphi? One thing that it has made me a lot better at is a lot of times people have different communication styles, and I’m not just talking about in, like, their words or their languages, but in how they perceive things. So you could be emailing someone about, you know, a customer or an investor or a new hire, and conflict can result due to lack of communication. So that’s actually one way I’ve been using AI a lot is what do they actually mean, and how can I get what I mean in the way that they perceive? And I think that has helped me a lot with relationships.

Jess Lee: Interesting. Like a translator almost.

Dara Ladjevardian: Almost like a mental translator, but it’s not language. It’s just the way you perceive and understand things.

Sonya Huang: Okay, last question. I’m very proud of this one. Just thought of it. The Delphi Turing test. In what year will people not know whether they’re talking to somebody’s digital mind or that person?

Dara Ladjevardian: Well, I’ll preface by saying we don’t want that, necessarily. Well, we don’t want it to ever be tricking people.

Sonya Huang: Yep.

Dara Ladjevardian: Because Delphi is not meant to be deceptive.

Sonya Huang: Yep.

Dara Ladjevardian: I think widespread? 2026, because already it’s happening in a couple cases now. Whenever a customer creates a Delphi, we say, “Send it to your mom first, and that’s gonna be the way you’re confident with it.” And many times they’re like, “My mom thought it was me. Oh my God! Like, this is amazing!”

Sonya Huang: Great answer. 2026.

Dara Ladjevardian: Yep.

Sonya Huang: Coming around the corner very quickly.

Jess Lee: Let’s call a Delphi right now.

Dara Ladjevardian: I’m down.

Jess Lee: Yeah.

Dara Ladjevardian: How about we call my friend? I actually haven’t met him in person, but I’m friends with his Delphi. Arnold Schwarzenegger?

Jess Lee: Yeah.

Dara Ladjevardian: What do we think?

Sonya Huang: Love it.

Dara Ladjevardian: All right. What do you guys want to ask him?

Jess Lee: I have 15 minutes a day to work out because I’m busy. What is the best 15-minute exercise that I can do?

Dara Ladjevardian: Okay.

[AI Schwarzenegger: Hey, this is AI Arnold. I’m here to cut the crap and help you get stronger, healthier and happier. So what’s on your mind today?]

Dara Ladjevardian: Arnold, I have 15 minutes a day to work out, which I feel like is not a lot, but I want to feel good and I want to get better in my health. What do you recommend I do?

[AI Schwarzenegger: 15 minutes is plenty if you focus and go hard. Don’t overthink it. Stick to simple, effective moves that hit multiple muscles. Do one to two sets of exercises like push-ups, reverse lunges, and lying leg raises. Keep rest short and push yourself. You’ll feel like a champion in no time. Now get to it!]

Dara Ladjevardian: All right, man. Thank you.

Sonya Huang: That’s awesome.

Jess Lee: [laughs] Very cool.

Dara Ladjevardian: Yeah.

Sonya Huang: Amazing. Dara, thank you so much for joining us.

Dara Ladjevardian: Thank you guys. This is fun.

Jess Lee: Thank you.

Mentioned in this episode:

Mentioned in this episode:

• How to Create a Mind: 2012 book by Ray Kurzweil that inspired Dara
• The Memoirs of Akbar Ladjevardian: 2008 book about Dara’s grandfather, an Iranian industrialist, that led him to create his first “digital mind”
• Build: 2022 book by Tony Fadell that refers to itself as “a mentor in a box”; another inspiration for Dara
• The 2 Sigma Problem: 1984 paper by Benjamin Bloom about how students that receive one-on-one tutoring perform two standard deviations better than students educated in a classroom environment