Delphi: can you package a mind?
How Delphi is democratizing and preserving knowledge through digital minds
I have been obsessed with digital clones for a while.
It started when I watched Black Mirror’s “Be Right Back”, where a grieving widow orders an AI version of her dead husband, trained on his old messages and voice notes. The show was a warning. I finished it with possibly the wrong takeaway: I want the product.
I have grandparents whose stories I have only half-heard and mentors whose careers I will never fully study. What happens if you can turn a person’s mind into software? Package taste, memory, experience, and reasoning into a product that is “you”? Talk to that digital version of someone and feeling like the real experience? The episode showed with AI, these hypotheticals can be a real product.
The internet has always been about access, but in its current form, it’s only general information access. Access to specialized, personal knowledge is still gated by time, geography, and the mentor deciding you are worth replying to.
Technology has solved similar problems. Web search democratized information, and consumer social democratized publishing; digital minds can be the next evolution: democratized mentorship.
What is Delphi?
Delphi was founded in November 2022 by Dara Ladjevardian and Sam Spelsberg, inspired by a deeply personal experience.
Dara’s grandfather Akbar was one of Iran’s most successful pre-revolution businessmen who built a 30,000-employee empire before landing on the new regime’s hit list and being smuggled out of the country under the belly of a donkey.
Years later, a stroke took away his ability to speak. Dara, struggling as a solo founder, kept reading his grandfather’s memoir and wishing he could ask for advice. Inspired by Ray Kurzweil’s book, How to Create a Mind, he did what any normal and sane human would do: fed everything he could find into GPT-3 and trained it to become a version of his grandfather he could actually talk to.
This seeded the idea for Delphi. Dara later met Sam Spelsberg, formerly on Apple’s crypto services team, during their careers at OpenStore. They started a book club and shared a similar love for learning and mentorship. After a trip to Santo Domingo, the duo decided to pair up and start Delphi, a company that’s not trying to make AI feel more human, but making humans more accessible through AI.
Delphi does this by creating “digital minds.” These are AI replicas of a real person, trained on their writing, podcasts, videos, or a sit-down interview if the archive is thin. Once trained, creators deploy it as chat, voice, or video. They can gate access behind a paywall, capture leads, plug into Slack, Zoom, Zapier, and their CRM. When users chat with someone’s Delphi, the responses will cite the source, ensuring responses do not deviate from what the creator has expressed.
The best proof point is dating coach Matthew Hussey. His Delphi handled over 1 million text conversations and nearly 2 million minutes of voice chat. Delphi says “Matthew AI” generates seven figures at only $39/month, implying at least 2,000 annual subscribers. Hussey told the Wall Street Journal that the ability to give millions of people access to his thinking at 3 am on a Tuesday is a game-changer. Others on the platform include Tony Robbins, Arnold Schwarzenegger, HubSpot’s Brian Halligan, podcaster Lenny Rachitsky, and Khosla’s Keith Rabois.
The pricing structure for Delphi is tailored to its end users. Pricing starts free for general public and runs to $79/month for creators and $299/month for leaders with larger audiences.
The most interesting part of Delphi is probably what Dara described as its end state.
Delphi could be the next version of LinkedIn… There is no way to find people based on their thoughts, their expertise, and their mind. But this is truly how we connect. - Dara Ladjevardian, Founding Journey Podcast
This is a profound idea. If human identity can be captured via a digital mind, how relationships work shifts. If Delphi becomes how people connect, it stops being just a creator tool and transforms into the identity layer for the AI era.
Investors are betting on this vision. Founders Fund led the $2.7M seed in September 2023 with Lux participating. Sequoia led the $16M Series A in 2025, with Anthropic’s Anthology Fund, Proximity Ventures, and angels like Lenny Rachitsky, Gokul Rajaram participating.
A collision of creator economy and AI
Delphi’s market is not AI chatbots. It is the collision of two markets: creator economy and AI persona.
The first is the creator economy. Patreon has 330k+ paid creators, and Substack has more than 3 million subscribers. Forbes sizes the market at $250 billion in 2025, and Goldman estimates it will nearly double to $480 billion by 2027.
Every single one of the 50 million creators globally aims to connect with their audience, and every single one of them hits the same constraint: time. As much as you would like, you cannot be on 10,000 one-on-one calls a week or chat with every single subscribers. If Delphi can meaningfully change how creators can be accessed, it unlocks a brand new vertical on top of an explosive market.
The second is AI personas, a nascent but fascinating category. Character.AI, which raised $150M at a $1B valuation in 2023, is the clearest proof point that AI can create human-like interactions so convincing that users form real emotional bonds with them. This both proves product-market fit but also poses a warning. The platform has faced lawsuits over emotional dependency, including cases tied to real tragedies.
Rather than creating fictional characters for entertainment, Delphi creates real people for utility. Regardless of what form AI persona takes, it will take some time for people to get comfortable to interact with them. However, as Character.AI proved, the adjustment period might be shorter than we think.
What I like about Delphi
1. Mentorship democratized
Mentorship is one of the most consistent predictors of success, but also one of the hardest things to obtain.
The average person who receives dedicated one-on-one instruction outperforms 98% of those who learn through conventional groups, driven by personalized pacing and immediate feedback. A study by Anders Ericsson found elite performers across domains (chess, music, sports, medicine) almost universally had expert coaches giving targeted, individualized feedback.
The issue, however, is that mentorship is extremely difficult to find.
Throughout life, we’re shaped by the people we learn from. But the right guidance is often out of reach. The best coaches are booked. The industry veterans don’t take many meetings. - Jess Lee, Sequoia Capital
With Delphi, mentorship becomes democratized. If you want to learn about the product, go chat with Lenny Rachitsky. Curious about AI? Talk to Zach Kass. Want to test your investment thesis? Deena Shakir will be your thought partner.
The AI-native integration means Delphi is not pulling from a static knowledge base. Instead, the AI clones surface what people are asking and feed those insights back into future content to be adaptive. Not only do users get tailored insights, but the Delphi owner gets direct line of sight into what their audience actually wants.
2. Humans want human content
The internet already ran this experiment once. Wikipedia built a massive repository of the world’s facts, yet we spend way more time listening to opinions on Reddit and TikTok.
Information is cheap and boring. Opinions are engaging and fun.
Ultimately, people want perspective, taste, judgment, and the feeling that there is an actual person on the other side. People want to pick sides and argue with each other with strongly held beliefs. This isn’t just about having opinions. When you create space for people to engage and debate, you create dialogue and differentiated views that shape progress.
Similar to Wikipedia vs Reddit, the world needs general intelligence and also specialized intelligence. As we reach AGI, knowledge and facts will be more readily available than ever. But is having accessible, near-perfect, homogenous information enough? Is that going to satisfy the human itch for opinions, debate, and arguments? No. But Delphi could.
3. Delphi built a great supply-side pitch
The difficulty in building two-sided marketplaces is usually getting the supply side right. For Delphi, this means getting experts.
The pitch starts with knowledge gathering. Every expert has already done the work: podcasts, essays, frameworks, interviews. Most of it ends up scattered across the internet, slowly forgotten. Delphi gives you a permanent, searchable archive of your own thinking. From a user perspective, I cannot count the number of times I’ve tried to find something specific someone said and given up after 15 minutes of Googling. Delphi solves this problem.
Then there’s scale. Matthew Hussey’s clone handled over 1 million conversations, with thousands of users in dozens of languages and available anytime, anywhere. For anyone whose calendar is the bottleneck between them and their audience, that number speaks for itself.
The third pitch is probably the most effective one: legacy. Delphi has a page on its website literally called “Become Immortal.” They are not being subtle. Delphi is how you preserve your legacy and share it with the world forever, even after you are gone. People like Arnold Schwarzenegger and Tony Robbins don’t need another content channel. They signed up because the idea of their thinking outliving them resonates.
What my concerns are
1. Adoption ceiling is limited at current form
While a digital mind is a cool concept, for most creators, novelty might exceed value. From Deloitte’s 50 million creator estimate, perhaps only 1-2% have the voice and scale (100k+ followers) to justify using a digital clone to engage with their audiences. For many smaller creators, building, rather than maintaining an audience, is the challenge.
If we assume around 5% freemium conversion, at $79/monthly cost, Delphi’s estimated ARR at scale would be capped at ~$50M. While Delphi represents a fascinating use case for many of the headliner influencers, the market for customers with enough audience to truly need the service today is limited. If Delphi only works for the top 1% of self-help, B2B, and executive coaches, it becomes a strong niche company, not the identity layer of the AI era.
2. The barrier to adoption and retention is high
Building a good Delphi takes real work. You need to upload your content, run training interviews, review outputs, catch errors, and continuously iterate. For creators whose entire brand is how they think and communicate, a clone that gets things wrong is not just unhelpful, but actively damaging to their brand.
The reputational anxiety compounds this. Gizmodo ran a piece called “Self-Help Ghouls Are Charging People Absurd Prices to Talk to Impersonator Chatbots.” A congressional bill to ban AI chatbots from impersonating licensed professionals was introduced in Congress in early 2026. Even creators who want to sign up need to think about whether having a Delphi actually provides the audience more access or alienates them from their audience.
3. Pressure from platform players and independent users
Delphi has built a name and reputation for itself, but not necessarily the technology. That invites tough competition from platforms with existing distribution. If OpenAI ships an identity layer, if Substack adds a “chat with this author’s AI”, if YouTube Studio rolls out an AI-of-me feature, Delphi faces serious pressure. Any of those moves would be a highly possible product extension for a platform player.
The threat is not just from big platforms. Independent builders are already creating personal digital minds on top of frontier models. Instead of building this layer through Delphi, creators could deploy this through their direct channels.
The harder, longer-term question is: if general AI absorbs everything a creator has ever written publicly, do you need a specialized clone at all? At some point, asking Claude to only cite Lenny Rachitsky starts to look a lot like talking to Lenny’s Delphi.
4. A digital mind is not the same as a digital archive.
Delphi’s core promise is to package a mind. What it currently delivers is a searchable archive of past thoughts. While this is useful, this is not the same thing. Someone’s mind is not just what they have said in the past. It is an encapsulation of who they are: how they interact with others, how they approach unfamiliar problems, and what values they represent.
When I chatted with Deena Shakir, a Lux Capital partner, about a fintech investment thesis, Delphi pulled from her notes. On a topic outside her focus, it had very little to say. A real conversation with Deena would go differently. She would draw on patterns from adjacent domains, connect ideas she had never explicitly written about, and reason through a new problem using frameworks she has applied elsewhere.
This is a fundamental gap between retrieval and reasoning. Real intelligence is the ability to connect things that were never connected before. A digital archive can surface what you said. It cannot think the way you think. Until that gap closes, the most valuable mentorship moments still remain out of reach.
Food for thought
1. Move from RAG to reasoning
Delphi’s current architecture is a sophisticated search. You ask a question, and it finds the most relevant chunks of past content and surfaces them. While this is good at finding what someone said, it is not the same as capturing how they think. The more valuable version extracts principles and patterns. For example, there is a meaningful difference between Buffett saying “buy wonderful companies at fair prices” and “I bought Apple because of strong fundamentals.” The first is a principle; the second, an instance. You can extrapolate principles, but you cannot extrapolate instances.
Delphi could build a personal knowledge graph: individual pieces of knowledge become nodes, connections between them become edges, and principle-level expressions get weighted above specific instances. Studies show this approach achieves 70-80% better performance, specifically for the kind of knowledge extrapolation and reasoning needed for mentorship.
While extrapolating principles may be the next evolution, finding the balance is key. When Delphi draws a connection between two ideas, it needs to show its work (e.g., here is the principle, here is the past content, here is how it got there). The thought process needs to be clean and traceable, and references need to be explicit. Done right, this closes the gap between archive and mind.
2. Own the voice, own the moat
The real unlock is audio and video, and right now, Delphi has a gap. When Dara appeared on Fox with his clone, the result was not convincing. The clone stared and moved awkwardly, kept asking the interviewer questions instead of just answering, and sounded more like a “please stay on the line” bot than a real human. Contrasted with the real Dara sitting next to him, the difference was stark.
For a company mimicking interactions with creators (people known for their energy and personality), the product needs to solve audio and video. The good news is that Delphi does not need to build this from scratch. There are great companies like ElevenLabs and Higgsfield that are actively solving the audio and video problems.
Additionally, multilingualism is where Delphi can build a strong moat. A finance educator in New York reaching São Paulo, Seoul, and Lagos in their own voice, without rebuilding anything, is a fundamentally different value proposition than any existing creator platform offers. The company already supports 40+ languages, meaning a creator who builds their identity layer on Delphi does not just get a clone. They get a global distribution engine.
3. Build a social engine… for clones.
Right now, discovery on Delphi depends entirely on already knowing who you want to talk to. You search for a creator, open their clone, and have a conversation. There is nothing to scroll, nothing to stumble into, and no reason to come back without a specific intent.
A feed built from past content fixes this. Creators’ clones can resurface their best ideas, past interviews, and key beliefs as posts (text or video). You scroll through them the way you scroll through Twitter or Instagram. You follow the ones you find interesting. You build a real discovery engine for influencer content that opens an additional distribution channel for creators and a platform for consumer engagement.
Once the feed exists, the social layer follows naturally. Users can create their own clones based on their beliefs and tastes. Once users have their own clone, the next step writes itself. Your clone follows other clones, engages with their ideas, publishes “UGC”, and builds connections based on shared beliefs and ideas. This extension helps Delphi build out its vision of becoming the knowledge LinkedIn.
Final thoughts
I like to evaluate companies on 4 key criteria: runway for growth, ability to execute, product-market fit, and defensibility.
Runway for growth: 3/5. The creator economy is a $250B market experiecning exponential growth, but the addressable slice of creators with enough audience to need Delphi keeps the ceiling lower than the headline number suggests.
Ability to execute: 4/5. The Delphi team, led by a serial founder with a personal raison d’être, created a fascinating product that caught the attention of the public. The company fought and won hard battles and managed to onboard major creators and investors in the early days of AI clones.
Product-Market Fit: 3/5. While it is difficult to test the PMF of an extremely nascent and controversial market of AI clones, creators like Matthew Hussey’s experience are strong early proof points. The question is whether this is a durable and emerging social pattern or just mania from AI early adopters.
Defensibility: 2/5. Delphi today can be threatened by platform players and independent creators. However, creating a knowledge graph or entrenching the voice/video functionality could create a substantial moat.
Overall rating: 6/10— Invest with caution
Delphi sounds like a Black Mirror episode.
Digital clones, AI replicas of real people, immortalized minds. And to be fair, it looked like one too when Dara’s clone appeared on Fox with the energy of someone reading their own ransom note.
But here is the thing about being early: strange-looking and wrong are not the same thing.
As AI avatars proliferate and human-AI interaction becomes a daily occurance, someone has to define what the best version of this looks like. Delphi’s model is a very reasonable one. Rather than replacing humans, Delphi actually celebrates and preserves human knowledge through AI.
If Delphi can close the gap between archive and mind, it is not just a good business. It becomes the system of how knowledge is preserved and distributed, even when the people who held it are no longer here.