Inside a Personal AI That Acts as You, Not Just Talks Like You | Sam Kececi, Founder @Sentience

My Sentience has closed deals during the conversation, which is insane. Imagine if 90% of the tweets you saw were just written by the same guy. You would still get annoyed.

I still want to talk with the real you. I don't want to talk with the digital twins.

Both my grandparents passed away a couple of years ago. Remnants of their knowledge was in a couple of places. It was in text messages I had had with them, some voicemails they had left me. The content of them that lives in my brain is pretty fallible. You touched on this. The human brain is not meant to remember thousands and thousands of people. This is a small business owner named Pat. He's a jeweler. He's thinking, I'm gonna have to pass this business on to my daughter. He would have now this building source of knowledge. His daughter wants to ask, how do you think about advertising? Or how do you think about working with tough customers? He will be able to pull from actual concrete examples.

What would that look like if everybody has their sentience digitized?

No one can compete with you at being you. And I think that's true now more than ever. Like, you can't really compete with an AI. This Phase 7 is labeled unknown, meaning I don't know if we can do this.

Today I have Sam Kocetsi. He's the founder and CEO of Sentience Company, and he's building a personal AI that doesn't just talk like you, it actually takes actions for you. And he just launched about 6 weeks ago. Today he's going to teach us how he did it. Welcome, Sam.

Thank you. Super excited to chat today.

I was just telling everyone the version I found and researched about you from LinkedIn. Is there a non-LinkedIn version of you you would like to share with us?

Yeah, absolutely. LinkedIn doesn't capture much. I mean, it captures There's the horrible details. But yeah, I grew up in New Jersey. I was always interested in 2 things, I guess, like computers and the human mind growing up. I thought I wanted to be a neurosurgeon when I was a kid. Thankfully, I did not go down that path. I don't think I would have liked the 8 years plus of schooling. But yeah, I have always been fascinated by the sort of hybrid between the 2. I went to Columbia, studied computer science and economics. Was previous to Sentience, a CTO of a company called Macro. Started Sentience in sort of late 2025. And yeah, the sort of quick headline, which you touched on, is we create hyper-accurate simulations of real human beings that can both communicate as them and also operate on their behalf. So excited to dig in more.

Your company has a very ambitious name, Sentience Company. Can you tell us what is sentience in your mind?

When people ask me what I'm building, I can just say I'm building sentience. It is accurate from both the company perspective and also from the perspective of what we are actually trying to pursue. I've always been fascinated by what are the sort of emergent qualities of technology that can actually start to converge upon something beyond just output of text and towards something that actually feels alive and feels sentient. I mean, there's this prevailing wisdom, I guess, of the model creators, like the big labs, that we kind of need to be very sort of anti-sentient. And our goal is actually to kind of pursue this direction of AI that not only can be useful and productive, but can actually carry your taste, your judgment, your wisdom, things that make you who you are. And so for me, if we fast forward 5, 10 years, I hope we will achieve some sort of small sliver of an AI that feels more sentient and more human than kind of the status quo. So that's why I named the company that. Our vision in one line or one word, that's the vision. Um, and everyone at the company can say I'm building sentience and that can be factually correct.

Why are you doing that? What do you think as humans we're looking for? Are we looking for something like Jarvis?

Yeah, it's a really interesting question. I mean, I think there's like kind of 2 bimodal ways that AI will go. I think there is like the use cases of AI that are centered around sort of purely productivity, right? So this is like coding is kind of the obvious example of this. AI is amazing at writing code. And by AI, of course, I'm generally referring to language models and then the agentic harnesses. Surrounding language models, whereas obviously AI can mean a lot of different things as well. But, you know, just to kind of get that definition, it's quite good at things that sort of like don't require your unique humanity. So like coding is an obvious example. I don't really care if the Python code that Claude Code writes has sort of my unique experience and wisdom behind it. Yes, sure. There is some aspect, maybe I write Python code in a little different way than others, but broadly, like, I'm okay sort of outsourcing that to the weights of the model. And thankfully there's like tons and tons of Python code that these things have been trained on. And so I'm like very happy to let my Claude code agent write code. But on the flip side, you know, something like this podcast or interviewing a candidate for my company or speaking to an investor about the vision for the product or talking to a customer about the use cases, that's like an area where I actually will not want Claude to sort of use the default weights. I don't want to outsource my thinking to just like the generic off-the-shelf one-size-fits-all model because inherently that is like trained on sort of the overall corpus of the internet. And like, I have my own unique views and wisdom and ways of doing things, right? The way that I write an email or the way that I talk to my team is different than a different founder, and they're going to build a different business as a result. So yeah, I think there's like areas where AI and specifically like kind of the off-the-shelf agentic harnesses will be really, really powerful. Coding, data analysis, you know, you kind of see now like AI for finance and these are really great use cases that don't require my unique context and my unique knowledge. But for me, like running my business, that like doesn't account for a lot of the taste and the wisdom that I have. So that's where Sentience is really like focused on. It's that other side that I think like A lot of people are kind of ignoring. It's that taste. It's that wisdom that we do a really good job of capturing. One of our goals actually with building this, I tell our team like, we're building first and foremost for me. So this is great because I am like the strongest power user of Sentience. And, you know, if we at least can build something great for myself, if we can build a great Sentience for me, then we have succeeded. I don't know, there's some amount of, I guess, healthy narcissism in there. But I totally agree. Like, the interaction of you and I as human beings is quite fundamental. But what I will say is I'm actually able to be fully present on this conversation because my sentience is actually interacting with my team right now. I can tell you that I have Slack open on the side of my computer, and in the course of this roughly 5 to 10 minutes that we've been chatting. They've asked MySentience 3 questions and it's given 3 answers. So, I mean, you all know sort of the— I think everyone has kind of been on those calls where it's clear that the person on the other end is like multitasking or like sort of not fully present and engaged. And yeah, I find like there's some real beauty in like being able to free us as human beings up to like Be present and yet like the company can continue to run. You know, I obviously like hopefully won't get hit by a bus or something, but yeah, the company can actually kind of continue forward with my sentience and then it leaves me to actually do the things that require my unique humanity. Right. And those are like things like connecting with people like you and meeting people and actually being face-to-face interviewing people, stuff like that. We were all promised this sort of, I think, world where we can do the things that are like uniquely spark joy for us as human beings. And I, and I think what sort of the status quo of agents has led to is actually you end up just kind of micromanaging more agents and you're very productive. You're maybe 10x more productive, but there's sort of still this like missing horizontal layer across all of the context and all of the knowledge of your life. And, you know, we're not focused on, again, things like coding and things like data analysis, but specifically for me answering questions with my team, booking calendar invites, prepping for calls, sharing what I care about and what I value in like an engineering candidate that we talk to. These are things that like, I don't need to repeat myself 10 times anymore because I have this, this proxy for myself. And yeah, it's, it's quite exciting for us.

Was there a moment that you decided this, this thing that you're going to build? Because you left, you left your previous company, right? And you raised $21 million. You just mentioned. So it sounds like a promising business, but you left.

Yeah, I think like, I don't know, it's hard to say. Maybe subconsciously there was something. I can tell you 2 things and they kind of approach it from different angles. The first is that I lived this pain viscerally as CTO of my last company. I was managing a team of around 15 engineers. And what I found is I basically became like a glorified information router. I don't know if you like, There's this analogy of like the people at call centers that used to have like the switchboard operators that used to have to just sit there and route calls from one person to the next. And I think what a lot of people, once you kind of get beyond that initial flat team structure, what a lot of people realize is like, yeah, really what you're doing is just relaying information and kind of like regurgitating your own Decision-making. So, you know, taking a piece of information from sales, giving it to engineering, taking a piece of information from engineering, giving it to sales, right? You're kind of just this like, you know, toll booth in the loop. And so I lived this pain like very viscerally and the more like people I talked to that had been in a similar position to me, I think the more people I realized like, oh, this is actually a universal problem that some people have switched on to. And then some people kind of have just accepted as like, a painful way of life, unfortunately. So that I think was kind of piece number one. I think piece number 2 is maybe a little bit more profound and maybe speaks a little bit towards the longer-term ambitions of the company. But both my grandparents passed away a couple of years ago, and one thing I was struck by is the sort of remnants of their knowledge was in a couple of places. It was in text messages I had had with them, some voicemails they had left And then broadly, it was kind of just in here, right? In my own brain and the brain of my family members and so on. They didn't have a massive Twitter presence or hundreds of podcast episodes to pull on. But what struck me is like, okay, the content of them that lives in my brain is pretty fallible. You touched on this. The human brain is not meant to remember thousands and thousands of pieces of information. So I think a lot about my grandfather, who was actually an entrepreneur. He started a company, and actually both my grandparents were. My grandmother started a dance studio. My grandfather started a real estate appraisal business. And I would love to be able to tap into their wisdom of like, oh, what was it like in sort of the early days? Okay, how did you get over these struggles? And really, unfortunately, like you can't. So I think that was the second piece. And It's been super interesting. A ton of people that we've now started working with over the last couple months have sort of fallen into this category of people that are thinking about their own sort of legacy within their business. So I'll give you one concrete example and there's probably, I don't know, 20 to 40 people in this category, but this is a small business owner named Pat. He's a jeweler and he runs a, I think, 20-ish person jewelry store, kind of high-end jewelry store. And he's kind of thinking about, okay, I'm going to have to pass this business on to my daughter. And he uses a product every day from the standpoint of, okay, he can connect all of his sources. He can have this deep understanding of how he runs the business, how he makes decisions. He was telling me the other day, like, he used it to kind of like have all of his employees screen their outreach. So like they do outreach and He's like, before you do that, run it by my sentience to make sure my sentience approves. But as he's thinking about kind of giving this business over, handing over the reins to his daughter, he's like, I actually have now this building source of my knowledge that if my daughter wants to ask, hey, how do you think about advertising? Or how do you think about working with a tough employee or a tough customer? Like, it will be able to pull from actual concrete examples. So We're really excited about both the immediate term, like aggregating knowledge and like you said, kind of having this entity, this agent with perfect memory of everything from your life. But then we're also excited about kind of what this looks like and where this goes into kind of the longer term. So you asked me about kind of the inspiration and the origins of the company. I wake up every day like super pumped about what the world can look like and you know, even 2, 3 years if we can, if we can build fast enough.

What would that look like if, let's say one day everybody has their sentience digitized?

And yeah, I mean, our goal is to have a billion people on earth create a sentience of themselves. And I think in that world, you know, a lot of this information transfer changes and it frees people up to actually like skip to the part of interactions that matter. So I'll give you like maybe a couple concrete examples. I interview on average about like 10 to 15 candidates a week. I want to interview everyone. Recruiting is like sort of the largest portion of my time. A lot of those conversations kind of follow the same script. It's like introduction, go over resume, and then sometimes you kind of get into some really interesting stuff, right?

Yeah.

So in this world, imagine if like every single one of those interviews could actually happen between 2 sentences. And then what the 2 human beings get is kind of a really rich digest, a really rich report of all of the information that was extracted from there. Then when we actually go to meet, because again, like our goal is not to sort of replace the human interaction, but to sort of augment it. We actually have like a list of really interesting level-down topics that we can go into right away. And so maybe instead of a 30-minute call, that's like 15 minutes of like I don't know, small talk and kind of feeling out the resume. You just get into like, oh, wow, here are the really interesting overlaps that you should focus on right away. And then if you think about the way organizations and teams, in my opinion, should operate, all of the time spent on just knowledge transfer, like basically getting knowledge from my brain into your brain. And if we work together, there's all of this time, all of this scaffolding.

It's bottleneck.

Yeah, 100%. Even just like something like engineering standup, which is like ubiquitous at every organization. And if you add up the total amount of time of engineering standup in the global economy, it's like, you know, trillions of dollars potentially, right? All that is doing basically is just like sharing knowledge and making sure people are on the same page. We run engineering standup internally just through Sentience. We have like a skill basically that's just called /standup. And it's actually quite simple. It just iterates. Uh, through every single person. And just like a real engineering standup, you give your update, it pulls from context like Linear, it pulls from GitHub conversations. Um, you can connect your Claude code and stuff like that. So it actually knows kind of what you're working on. And then it'll give a, an update. And the really cool part where this goes from kind of like, you know, basic agent, right? Because you can do this with kind of like Claude or a basic agent is like sometimes the sentences will. Disagree, or sometimes the sentience will be like, oh, actually, I'm also working on that. Like, you know, we're overlapping. Or sometimes it'll crack a joke in the way that like an engineer on the team will crack a joke. And so these are kind of the glimmers of like, oh wow, this is actually kind of operating and simulating how I would communicate with my team. And so what we can do as human beings is just, okay, we simulated that in the morning. We can all review it as a team still, right? It's not going to totally replace like— Teamwork, but then we have a really clear sense of like, oh, here are like the 2 or 3 like sticking points. Everything else looks good. Let's kind of move on. So we use it for standup. Um, you know, we've, we've set up other teams. We set up a handful of like recruiting agencies who use this for like similar, like going over the candidates they're talking to. But yeah, in, in this future where, where everyone has this, where a billion people on earth have this, this product, we're just really excited about what that kind of multiplayer interaction and multiplayer simulation can look like. So yeah, our goal is to, you know, actually get into the sort of simulated reality world.

What's your sentiment about, in general or on average, engineer teams prefer to have a daily standup? Or is it because it's kind of a discipline that, you know, the engineer teams have been carrying through for so many years that it's just, It's just people are used to it, but, but do you feel the sentiment that people don't really think it's very productive or it's not time worth spent?

Yeah, I have never been, I mean, I've built several engineering teams. I've worked on other engineering teams. I don't think I've ever been on a team or worked with engineers that are like looking forward to engineering standup as like a core part of their day. And I think people view this as like a necessary evil. I can tell you, like, even at Macro, we tried to kind of minimize the amount of time. And there's nothing worse than like an engineering standup that's like, On the calendar for 15 minutes and like bloats to 45 minutes. And then like, you know, 95% of that time the team is zoned out because it's not relevant to them or they're multitasking. Yeah, they're multitasking. They're, they're vibe coding. Like they're actually kicking off code. This is like a good thing, by the way. You should, you should, you should be like, you know, actually getting work done. Engineering standup is just like this, this sort of, uh, Like, like necessary evil to get around the pain of knowledge transfer.

But what about relationship building? Like, you know, if you, if we outsource all these, you know, meetups to sentience, and when is the time that the real people are going to meet with each other and build a relationship?

Yeah, I mean, I think this is a great question and even speaks kind of more strongly to this idea, right? What if we spent 15 minutes in the morning every day just getting to know each other while our sentiences handle kind of the like boring Knowledge transfer piece of standup. And, and yes, I understand there's like actually some interesting overlap between knowledge transfer and, yeah, yeah, between, between knowledge transfer and, and, um, the sort of bonding, but I can tell you for a fact, like the way we use the product and we use this every day, I think we average around like 150 to 200 messages sent and received. And then a handful of that is actually agentic operations, which we can. we can get into as well. All right. Like that frees us up to like, we play Mafia at the end of every week. Yeah.

If, if you're the interviewer, I'm the interviewee, and I send my sentience to talk with your sentience, there's a risk of what if I just put a bunch of like, you know, I'm not a system engineer. I'll say I'll put a bunch of system engineering book in my sentience and pretend that I know the knowledge that I don't actually know. Would that be a risk for like, you know, having these kinds of conversations? Because I think the dynamic of interview could be different. However, I think if I'm a real person talking with your Sentience, that works. I think that there's no problem with that.

Yeah, it's a really interesting question. Yeah, almost like, can you kind of game the Sentience such that it scores higher or something? The way I can tell you I use Sentience right now, I send it to everyone at the end of the call. And it's interesting because we do, as I mentioned, I'm on a ton of these calls and I actually prefer to meet in person generally, and we're in New York and Williamsburg, so we'll have people come by. But generally at the end, people have questions, right? Maybe have 3 minutes at the end of the conversation for a handful of questions. And we never cover— everyone always has more questions than I have time for. So I'm okay. Unfortunately, I have to wrap up, but I'll give you my sentience. And I would say actually around 2/3 of the candidates I interview, end up having conversations anywhere from like 10 to like 50 messages. We've had candidates that have spent like hours after the conference. They will spend hours actually like sort of diving deeper, asking about what they could work on, whatever. Some people just like chat with it. It's funny, right? And if I think about like, hey, 2 things are amazing about that. One is I absolutely don't have time to do that. So I've already like been able to scale myself beyond the physical capacity of And 2, I think there's actually a second really interesting thing, which is those candidates actually probably would not have a super long-winded conversation. They actually would not share all of this information that they would with the human version of me face to face. But because I think they have this kind of more sort of safe and like—

So they want to ask smarter questions.

Yeah, they can ask questions. And of course I can go and review. I mean, we, we, we, we You know, I can, I can see that. And it's often sparked conversation. I can tell you, MySentience has closed deals, like during the conversation, which is insane. It's like closed contracts. It's like closed investors, like during the convo, who they literally say like, okay, sounds good. And the other really cool thing is then like we have this escalation feature. So anything that MySentience will flag as like, this is a really important piece, I want to like Elevate that to Sam, I can get a notification that says, oh, this conversation is going on right now. Here's like 2 interesting nuggets that actually are worth like diving deeper and then I can follow up with them in person. So really like the amazing part of this is it really does multiply my time and what I've found, I mean, yeah, there's totally like a lot of people kind of, I think it's a valid pushback of like, you know, Is this just going to replace your interaction with all the engineers, or is this going to replace interaction with all the people you're interviewing? And really, like, it's every single instance, it's just strengthened that. And it's allowed me to kind of, kind of connect deeper and, and, and have more conversations than I would otherwise be able to. I can tell you for a fact. So, so I, this is a great point. Uh, you know, the way we see sentience now, actually, uh, I'll, I'll, I'll use coding again as an example. Um, we, we're not building a, coding agent, right? There's a lot of great companies, Claude, obviously, et cetera, even startups that are building really great coding agents. But what we can do is we can actually call these agents just that, you know, we can prompt these agents through Sentience. So what Sentience is really, really good at is kind of being that umbrella layer over all of your knowledge, over all of your context. And yeah, I totally imagine like as, as these agents become sort of more, um, employee, more skewed towards actually being employees within companies. Your sentience will sort of be that umbrella layer across managing all of that. And then again, with the purpose of the human is the driver, right? I'm still the driver of this, but we've almost like, we've replaced the cognitive load of managing people now with the cognitive load of managing 15 agents. I used to manage 15 humans. Now I manage 9 humans and I manage however many. We've, we've like, we've, we've sort of like, you know, evolved in what we've taken one step forward and we've sort of also still had that same fundamental problem. What we're really good at is capturing the way you make decisions. And we can get into how we do this if you're curious, but the way you make decisions, the taste, the things that make you, you. The sort of hot takes and the way that you sort of think about things differently than other people. And I mean, to me, that's like the most amazing thing about human species is like, we are all different. And I mean, actually, if you look back in— I don't know, this is not a history podcast and I'm probably out of my depth now. From what I understand, actually, one of the kind of reasons we are the dominant species on Earth is not because of our raw IQ, although obviously are smart, but I think like monkeys have better like road memory than us. It's because of like the social fabric of like, you know, sort of that like cognitive diversity around, okay, you know, one person is a fisherman, one person is, you know, managing whatever. And, you know, I think we kind of want to preserve that. I think there's like a risk that with AI, we kind of lose the uniqueness between us.

That's still one LLMs do right for me.

Yeah. I mean, it's funny, like, There's so much of this, um, what I would call sort of like anti-AI slop sentiment, which I think is totally valid.

Yeah. Have you seen the recent commencement speeches? A lot of—

Yeah.

Yeah.

I've seen a lot of them and 100%, right. And I think what that's a pushback against is actually not the sort of like pure technology, because I think the pure technology is really exciting. And, but I think it's a pushback against like the uniformity of what this writing creates and You know, you can see it. This is not like a hypothetical, by the way. You can feel it. It has a certain smell. It has certain patterns. And that is actually not like just a fact of life about AI. That is a fact of life about the AI models that everyone uses, namely like 3 of them, right? This is Claude, GPT, and Gemini. They're generally all sort of like post-trained and trained in a very similar way. Like, I think it's almost like even if you remove AI from the equation, like people get upset about tweets written by AI. Imagine if like 90% of the tweets you saw were just written by the same guy. You would still get annoyed, right? Because you'd be like, I want something different. I want something new.

Definitely.

And so my sort of hot take here is actually what people are complaining about is not necessarily that it's AI written, but they're complaining about the fact that It all sounds the same. It all has the same patterns. It's all one. You know, you're like, everyone goes to the same, everyone uses Claude, everyone uses, and yes, they are 3 different tools, but like actually fundamentally, if you kind of unpack like the way they construct an output, like it's all basically doing the same thing and it sounds the same. Right. So yeah, like if I went on Twitter and my whole Twitter was just like one guy, I would be like annoyed at that guy too.

You would quit Twitter.

I would quit Twitter. I would block that guy. You know, a friend of mine is building an AI detector company and, you know, it basically can go on Twitter and say, is this written by AI or not? And it will flag it. And, you know, a lot of people have like, there's a very polarizing, right? Some people are, you know, love this. Some people hate it. But like the people that love it, I think are like, hey, I'm tired of seeing the same The same guy writing all the content that I see from everyone. But our sort of vision for the future is what if every single human being had their own very hyper-personalized model?

Do you really think it actually can have high fidelity representation of a specific human? Is that even possible?

Yeah, I mean, there's, there's, so 100% I do.

With current technology or there's something still missing?

So I think the purely prompt engineering approach can get you sort of there, but not to the degree That like we would consider our mission a success. And by that, I specifically mean like take a frontier model, let's say Claude, take Claude, give it a system prompt that says talk like this, blah, blah, blah. And then like we're done. I don't think that's like that. I would be very surprised unless there are like potentially like massive advancements that the labs make about customizability. And generally, I don't think any of them are going to open up the weights like for the frontier models. Now there's a second category. You can do things with like open source models, like LoRA and fine tuning, which we do. So, so we do those for the actual like way that we construct outputs. So every person actually has their own, we actually create a model that's unique to every person given enough context. That said, I do think that there is probably like a step function advancement needed, maybe created by us, maybe created by somebody else that we kind of Leverage.

You talk about your technical differentiators.

So this is our master plan.

Um, I really think I would love to know is phase 1, which is perception. It's talking about basically data input, right? In order to make a version of me or version of you, there's a dataset that you need to put in the engine. And what is that dataset? And how are you collecting the data?

Yeah. Yeah. Good question. Yeah. So first off on this, on this master plan, I was like deeply inspired by Elon published the Tesla master plan, I think in like 2007 or something. And I thought 2 things were kind of cool about it. One, I thought it was really, really sort of admirable that he was putting the master plan public and it It's kind of like, you know, hey, like, here's what we're going to do. Good luck trying to copy it. And we feel very similarly. The second thing is, like, you know, Tesla is like a remarkably complicated company to build. But like, if you read that master plan from 2006, it's like really simple. It's like, you know, a middle schooler could like read that and be like, okay, I get it. Like, I get what Tesla is going to do. And I'm a big fan of Richard Feynman, and I think he often advocates for this idea of, yeah, you can take something very complex and distill it down. So anyway, yeah, just jumping in, our sort of first goal is, can we capture in as high fidelity as possible all of the sensory input from the things that hold context for your life? So what does that mean? There's kind of 3 categories. There was what I would call like latent data. This is things that already live in systems that you use. So your Notion, for example, if you use Notion, we use Notion. That's where a lot of the like company repository information lives, linear things like email, things like messaging, Slack, these kind of fall into that category. These are like systems that you use where latent data exists. Now, That is interesting to a degree because they're already kind of siloed. But if I asked you, hey, is your soul properly captured by just what's in your email? You'd be like, well, you're kind of missing something, right? So the second category then is, can we capture what I would call sensory input? So this is like Capturing audio. So we have an Apple Watch app, we have an iPhone app, and we have a desktop app. So if you're like, you know, working on something and you want to like snapshot or kind of like bookmark it, we're able to do that for you. And then I guess if you kind of fast forward into the end state, like we sort of see this as like a horizontal layer, like almost like a USB thumb drive, you can like plug in and out of every system. But what we do better than basically any other product that I have seen is we are able to synthesize across all of these different modalities into like a unified understanding of you and all the topics and people and threads in your life. So like what that means sort of at a simple level, and I can go into how we do this technically, but that means like, let's say actually this conversation, right? We planned this podcast. We connected a couple of weeks ago. Some of that information lives in email, right? But then some of that lives in the content of that first call we had. Some of that lives like right now as we're having this conversation. Some of that maybe like lives in like me looking up your podcast on YouTube and like browsing around and finding that information. So if I actually want the like holistic synthesis of what I know about you and your podcast, it's not really enough to just give an agent like an email tool call and say like, Okay, go like search for Angelina. It actually needs to like build this representative understanding of, of like the entities in, in my life. Um, and it also needs to update that, right? Because like a day ago, we didn't do this podcast and then tomorrow we will have done this podcast. And so that's actually like a very different world. So, um, this is kind of like now getting into number 2, phase 2. Um, but it's like how we go from perception, which is like this fire hose of raw sensory data. into memory. Because like, imagine you as a human being, um, and, and by the way, a lot of this we've kind of modeled off of human mind, um, which goes back to, I guess, I don't know, all the way to my childhood, which is my kind of passion.

I wanted to learn about neuroscience after I, you know, listened to some of the podcasts, the YouTube podcasts talking about it as well. So I'm really intrigued by how you solve this problem.

I mean, same. Like, I, I also will caveat with, uh, My knowledge on a lot of these things is Claude deep research, podcasts, and just like an abnormally obsessive amount of time reading on my own. So I have like very little, like actually rigorous, what I would call like formal education on this. But it's just been, yeah, it's been something. I mean, Teddy, who's our chief of staff, actually studied neuroscience. So he's a Kind of more of the, uh, oh, he is the real scientist. Yeah. Yeah. Um, he, he was a, he was a neuroscience at MIT and, um, I think has had a passion on this as well. But, um, yeah, I mean, again, if you, if you think about like how the human mind is processing information, like even as we have this conversation, you know, there's a bombardment of sensory input, visual, audio, you know, we don't like. literally capture every single, like, iota of sensory input, but as much as we can from a digital perspective, that's like goal number one. And the problem is if you just like throw all of that in a RAG vector DB, which is kind of like what a lot of other sort of—

I would do it that way, to be honest.

Yeah. So like, so we can talk about that. I mean, that sort of gets you, that gets you a working version of this. Um, now that has like several problems. Uh, first is just like the, the raw volume of content, especially if you're capturing really granular information, like capturing screenshots and then capturing like a bunch of audio convos and then like every single email.

Can I ask if the data capturing, is it continuous? Because for instance, I'm sharing a screen with you right now. If, if actually your app on my desktop actually knows I'm sharing this, this, this view with you for like an hour, then is it going to be recording all these information?

So it will only capture things that you explicitly want it to capture. So it's not taking like a raw video feed. Um, you know, we did think about, there's sort of the extreme version of this, which is, okay, what if we actually like, you know, even you could plug into glasses or something and like do like a raw video feed of every single thing in your life.

Um, it's kind of scary, to be honest. Like if my watch is constantly listening to me and then It's a little scary, isn't it?

Yeah. Yeah. So we, we sort of opted away from, from that like pure always-on capture everything, uh, approach in, in favor of like a little bit more of like a, um, curated, uh, set of information. Um, but of course it's a trade-off, right? Because let's say, you know, something crazy happens in a conversation and you, you didn't capture it. Like you, you're sort of, um, something to be lost. But in any case, sort of the flaw with like the pure embed everything, throw it in VectorDB is a couple of things. One is you're dealing with like a bunch of different modalities, but that can be solved. You could like normalize it down to a JSON format or something. But the second thing is like what we have actually like done a really good job of solving is the like temporality and the updating over time and sort of like the The idea that like you are actually an entity and your podcast actually is an entity that consists of like many raw memories. So we basically build this like very, very close to a knowledge graph, but actually like kind of one more layer up from a knowledge graph. And we update this every night. So you know, we call it sleep, uh, the sleep phase. But, uh, every night we run through and we'll actually synthesize all of the memories from that day into like a bunch of topics. And it kind of ends up looking like Wikipedia, um, where you have like a bunch of these markdown files, they're all linking to each other. And we actually expose this. So one really cool part about our product is like, you just get this like auto-generating self-wiki of your life.

Every day?

Every day. So it updates every day. Everything that happens to you that day gets categorized, gets filed away. We generally do a good job of like stripping out the noise and then keeping the things that matter. But the cool thing is like, because we're actually capturing everything, if you do want to like look up some really, you touched on this right at the beginning, but if you do want to look up some like really granular piece of information, like, you know, what was the email address that this one person I met 5 years ago, You know, had, um, we're also able to do that, right? So you kind of want like best of both worlds. You want like raw retrieval, but then you also want like, um, synthesized, like, and temporally updating understanding. Um, that's what we're like quite good at. Uh, it's funny, we have, uh, an employee who, uh, for like, I think similar amount of time has just keep— been keeping one Apple Note per day where, uh, he just writes like, every single thing that happened in that day, like super, super low-level granular. And Sentience now can actually ingest Apple Notes, so we can connect to Apple Notes. And one of the, the like sort of frustrations I've had for a long time with a lot of these like second brain tools is, yeah, it requires so much manual investment and the reward is there, like generally, right? You can kind of see these insights, or at least there's the, the promise of a reward, but maybe that's part of the pursuit. But it's just like people are spending hours and hours a day curating their personal knowledge base. You can go on Reddit on the personal knowledge management system subreddit, and there are these people that are spending, oh, here's how I spend 4 hours and I have to curate this every day. But yeah, our view is like, We have the technology now where this can happen automatically for you and you actually shouldn't need to set this up yourself. Like we talked to a lot of people who have like kind of jerry-rigged janky setups that do similar things, but it's like, you know, this is kind of our bread and butter. We can, we can set this up for you and then you can use this however, however you want. Like, again, we integrate with other agents, right? So we're not trying to like replace Claude Code or replace Claude. Like we're kind of just being this.

One of the things that I find very interesting, executive function, that comes to me as the, you know, the prefrontal cortex, that's the chief commander of our brain. And nobody in the large language model space is talking about how we can, you know, have that in the attention mechanism because it doesn't come with it, not, you know, for free.

Yeah.

And emotion is also very interesting because I think a lot of decisions when we are making, emotion is part of the decision-making function, right?

Yeah, this phase 7 is labeled unknown, meaning I don't know if we can do this, to be honest. I think 1 through 6, I'm very confident. I think we've labeled a couple of them as like there will actually need to be net new research breakthroughs to achieve these things. And we may make those breakthroughs, but also potentially other folks may make those breakthroughs. The, you know, yeah, phase 1 through 4, we actually kind of consider it to just be a matter of engineering, meaning like, actually, we think we have the raw ingredients and the raw tools, but it's a matter of building it. On that last phase, yeah, I mean, I mean, this is, this is also maybe just more of a Philosophical question. Like, if there is a version of you that thinks like you, acts like you, have we achieved something? I don't know. I would argue yes. Other people would argue no. My sort of personal belief is that consciousness is not purely limited to like our physical form. I think like silicon can have consciousness, but that kind of gets out of the technical and out of the product. And that's where I'm less, I guess, less well-versed. So maybe we can leave that someday to philosophers or something to figure out.

And do you have any suggestions for people who are maybe confused, puzzled and confused about what's going on? I think they need some guidance and me too. I mean, I have, I have years of industry experience. I still feel confused.

Yeah. I mean, I'll take, I would say, take everything I, I say. I need guidance too. I think everyone does. But I can offer kind of what I, I can offer kind of what I think is my unique mark on, on hopefully the world. I think there's a lot of like, I just had a friend text me. He just got into Harvard Law School and he was like, is there even a point of doing this? And, you know, like, are lawyers just going to be automated in the future? And then there's, I think, versions of that, you know, ranging from like, you know, how should I actually apply myself or how should I build skillset? I think it's inevitable that like the raw act of writing code, for example, like deep knowledge of Python syntax. is actually going to be less and less useful over time to the point where it'll basically be no longer useful at all. Similar to how operating a punch card machine or something is just like functionally no one needs to do that, or that's like a useless skill. I guess here's what I would say. I think there is a sentiment that like, yeah, we can just outsource our thinking to AI. I would strongly, strongly reject that. And I would say that Who you are uniquely matters now more than ever, actually, because of like this increasing uniformity. So this is like your opinions, your hot takes, the things that you can debate and disagree with or have a contrarian opinion on. This is your experiences, your knowledge, your memories. I think like these things actually matter a lot. And I'll quote sort of like, this is not my original idea, but I think it's like Naval On Twitter, I read this, but it's like, no one can compete with you at being you. And I think that's like now true more than true now more than ever. Like you can't really compete with an AI at coding. Yeah. Like a year ago you could, you could be like, oh, you know, there was actually arguments. Oh, I'm actually better at coding than an AI. I can do it faster. And, and maybe now there's like maybe 1% of people that's true. And then it'll be 0.5 and—

No way in terms of speed, I think.

No way in terms of speed. Absolutely not. And some people I think can make a valid argument like, hey, I actually write higher quality code than an AI. And I think that's still true. But I think the point being is like, you're not going to compete with AI at writing code. But what you can sort of like, and by AI, by the way, I specifically again mean the frontier language models and the frontier agent harnesses.

Yeah.

You can't compete with those. And so like, The real question is like, yeah, what can nobody compete with you at being? And that's you, who you are with your experiences, your strengths, weaknesses, et cetera. And so I would just, I guess, encourage people to lean into that rather than shy away from it. I think people and humans are diverse and different and lean into the things that make you quote unquote weird or unique rather than shy away from it.

I know how you feel because I also I mean, on the positive notes, I agree with you. If, you know, if AI is working in such a way so that each one of us can just be ourselves, then maybe it's a positive result.

Exactly.

If we can spend more time on creations and experiments, right, that we never had the time to do, maybe the world will become a better place. But I don't— if you were to teach me one thing to help me become a better founder, what would that be?

That's a great question. I, it feels like a cop-out. I honestly would say the same thing. I think it would be lean into the things that make you unique. And I think like a lot of founders will try to emulate like another successful founder. This makes sense, right? And I think it's not a terrible thing. Like if you're like, hey, I want to be like Sam Altman, okay, XYZ, how does he operate? Or I want to be like Peter Thiel. What I would argue is like, I don't think any of, like, I don't think Zuck was like trying to be like anyone. I think he just was himself and he built the company the way that he wanted to build it. I think the best founders aren't actually following a playbook. They're just charting their own course. And at every point in time, you can't really look to the past to create the future necessarily. So I would argue, I don't know if Bill Gates tried to build an AI company today, he may fail because he's a different— he maybe could only build Microsoft or whatever. So I would say just to This sort of TLDR is like, yeah, rather than try to just emulate what you read or try to copy someone else, just be yourself. That's like super cliché, I guess.

But it's very hard to do because we are always comparing ourselves with other people and it's always harder to be ourselves because most people, I would say, maybe have a hard time finding ourselves, right? Because we don't spend time on it.

Yeah, I agree. I meditate. I would recommend anyone meditate. That's a more practical Meditate 10 times a day, or sorry, 10 minutes a day. Um, it doesn't have to be crazy. Sit there with your eyes closed and see what happens.

Awesome. Thank you for your tip. And before you go, do you want to share a little bit about who you are hiring? I know you have roles open and can you share about how has AI changed how you interview people or how you measure performance or, you know, what type of person, what kind of traits are you looking for?

We're hiring sort of 3 molds right now. The first is obviously engineering. There's actually 2 kinds of molds within engineering. One is sort of the AI/ML research angle. So a lot of the problems we talked about of actually making models personalized, actually making them sound like you, building an agentic retrieval system that works across a massive amount of data and can sort of truly emulate the mind. That's like the ML/AI researcher side. We're also hiring full-stack engineers to build the product. Second category, we are hiring a growth lead. So someone who will basically, we've built a product, it's in the hands, a lot of people love it. We want to kind of scale that. So someone who could own all things growth at the company. And then third is a product designer. So someone who would lead product design at the company. On the, you know, what we, I think, look for and how things have changed. Obviously I talked about how we use our own product in the interview process. I think that I place a really high emphasis on EQ plus IQ. emotional quotient plus intelligence quotient. We are super flat. We all work together. We generally don't— I'm allergic to bureaucracy and unnecessary hierarchy. And then I think the second thing is we generally look for people who are a little more holistically minded. So people that actually kind of care more about the vision. I mean, you heard me talk about our impact on the world. I look for that. People who actually want to make a positive impact in addition to build Really amazing technology. So that's true for all the roles. And then our interview process is actually, we don't do LeetCode. We just basically work with you. So.

That's important.

Yeah. Yeah. So, I mean, like for us, like we just want to work with you. So we'll, we'll structure the interview process to be very much just hands-on, mutually kind of get a sense of working together. Absolutely.

I have one more question. To talk with Zero Sentience, I have to get approval. Why is that? And do you, are you actually approving it manually?

Yeah, I saw your request. I actually also saw you asked MySentience that question and it gave you the correct answer. It gave you a perfectly accurate answer. We generally are focused on not necessarily like the purely external use case. I mean, you, you heard me describe a bunch of different use cases across both our own company and a bunch of our users and, and where we find sort of the strongest kind of magic and the strongest killer use cases is actually within more internal or within sort of more intimate groups rather than like a, you know, put it in your Twitter bio. Uh, that said, my Sentience, obviously I, I give it to most people. So right now we're gating it. I imagine like in the future we may have a version that is, uh, purely external. So.

Okay.

Awesome.

Thank you, Sam, so much for the chat today.

This was great. I, uh, you know, great questions and really great chat. So thank you.