Builders Unscripted: Ep. 2 - Ashe Magalhaes, Founder of Hearth AI

Hey, Ashe. So great to see you. Thank you for doing this. Thank you for having me it’s awesome to be here. I actually remember the very first time we met was in this OpenAI office. And at the time, you were building Hearth, right? Can you tell us a bit more about what it was? Yes. Hearth AI was founded in 2022, and it was the first agentic CRM. So 2023, we're using GPT 3.5, and we were trying to kick off all these agentic workflows when the underlying systems were still pretty fragile. So I was trying to wrangle you and other people. And I remember, you had great feedback for the API at the time. And one of the things that you were trying to get done was like structured outputs before we even had them. You were trying to do like retrieval with JSON schemas. You were so early. What was it at the time to like build in AI? Yeah. So, I mean, my history was as an ML engineer. So I had been studying and I really rode the wave of deep learning and the commercialization of deep learning. And then when kind of agentic infrastructure, and LLMs came out for the first time, it was totally different, right? It was non-deterministic, and it was seeing what we could stabilize within a product experience because our users didn't care that, like LLMs were new, they wanted to see the flexibility of what was just then possible. Yeah. That's amazing. And speaking of your ML background, I think what I find fascinating about you is the wide range of like, life technical experiences you've had. Like you were an ML engineer at Airbnb, you worked on satellites at NASA, but you also started by building a solar powered racing car. Like, I want to hear that story. Yeah. So I really try to stay within frontier tech and Stanford has a solar car team with strong ties to like Tesla and SpaceX. So for two years I was building, heads down, a solar powered car with that team. I was on telemetry, CAD modeling, meshing, and then autocross. I was one of the drivers, so I was training within autocross clinics throughout California. That's crazy. And so did you actually, like, race in that car that you built? So we shipped the car to Australia. Every two years Australia has an international solar car competition, the Bridgestone World Solar Challenge, and we spent three months there prepping. We were camping in the Australian Outback, and then I would drive for six hours at a time through the outback, and then we did the race, which was five days from Darwin to Adelaide. Wow, was that scary? Yeah, we were going 90 Kph, which is around 60 miles an hour, and we built the car end to end, so it wasn't like a company came in and did it. And teams were crashing, Cambridge's team had crashed. And so I was very aware that I had spent two years building something with a team that I would put my body into and that level of rigor, like really played into later experiences. Right? Because like that level of risk and like rigor that you need to have is extremely high. Yeah, but it was so amazing because I think that was a form of like frontier tech to really be in the weeds with a team where you know, it is your body on the line, it's your teammates body on the line. How do you make safe decisions and also really push the bounds of what's technically possible? Amazing. And so what got you into building in the first place, starting on that project all the way to like being a founder? Yeah, I think just a lot of curiosity and really interested in, bringing things to life. So I view the builder archetype as quite close to the artist archetype. I think I'm happiest viewing myself as like a conduit for creativity. And I think engineering for me has always felt like a tool for creating and incarnating and to be able to do it with satellites and cars and code. But also like, you know, poetry and videos and other mediums. That is such an energizing way to spend every day. Incredible. And so when you were so early building like an AI product, like three years ago, when you look back now, like, how do you think about that experience? Like what were the hardest parts or the things you did not expect. Yeah, I think there were so many norms that were being disrupted. So the fact that our team was tearing down our like agentic stack every couple of weeks, that was really hard to maybe explain to folks around us because the history of tech had been, okay an ML you maybe spend a year laying the data pipes, and then and only then can you do something predictive and interesting with the ML product. And for us, it was like, no we want to surface an experience to users that is flexible, but it was so open world that there wasn't consumer behavior yet around querying an agent. So for example, like folks didn't know in 2023 that in Slack, the Slack bot could respond. They just thought that it was an outbound ping so there was a lot of new behavior from the eng. side, but also from the user side to educate them on. And I remember, on my side, I was trying to build agents too, before joining OpenAI. And, you know, this was the era of NanoGPT, BabyGPT, people were like trying to find like the early incarnation of what an agent would be. But most people outside of the space would not really understand and grasp, what an agent could do. And I think for me it underlined the importance of having relationships with other builders. You know, we were really all in the same plane of learning this stuff for the first time and to have friends that were like open and transparent and we were sharing tricks and also friends within AI labs that really made the difference with creating products that worked in some capacity. And so you've worked on very deep tech and very deep systems, right? What got you to think about like this human connections, human relationship which is kind of like the core thesis of Hearth right. Yeah. There is this amazing book, “When Breath Becomes Air”, written by Paul Kalanithi. And I didn't realize it, but he had sat in on a neuroscience class that I was in, and he was presented as this guest that had a terminal illness. And he was like, you get to ask this person anything. And, you know, he has this terminal illness and how is he viewing life? And, what he talks about is all achieving all of these accolades and making it to the top of the mountain and essentially looking around and being like, and now what? And I think from a young age I had all of this access to like really interesting frontier tech, but I was realizing I kind of missed that like spiritual footing. And for me, after some soul searching, I think like for so many people, what do I think I'll care about on my deathbed? You know, the satellites are amazing, the solar cars are amazing, all the ML stuff is cool. But I think what I'm going to care about is how connected I was to myself and to other people in my life, and how much presence and love I exchanged. And then it became obvious I want to push the needle on that, on connection. And I think, I say this a lot, but the arc of human progress bends towards love and connection and innovation. And AI I believe, should push the needle on that. And do you think now that like what you were exploring a few years ago with like this like AI agentic CRM was the first kind of manifestation of something like broader that you were gonna explore basically now? Yeah. And, you know, even before that I had been working on teams where they had an external network of people, and the task was to mine that network to match people with other people or opportunities or capital. And so it's kind of this age old problem of as modern humans, we are connected to so many more people than ever before, and maybe feeling lonelier and less able to act on it, both individually and as part of any cooperative team. And so I've been fixated on this for so many years. And the agentic CRM was a multiplayer pass at it. But, you know, my thesis is on relational intelligence. Its this belief that AI should augment the human experience and allow us to feel more connected in the time that we have, and that AI should extend our brain's ability to reason on who we’re connected to and why. And speaking of brain, I see you mentioned a few times like, on your post, like the idea of a second brain. Can you tell us more about this and how you think about it? I love the phrase second brain because it's a little bit less technical. So I think anyone in the world can maybe visualize what a second brain looks like. And it's this idea that we will have a symbiotic relationship with our AI products, and have a relationship with the product is the key word. So it will be this continuous feedback loop, on our stream of consciousness in which the AI is trusted as maybe this bird's eye intelligence to guide us, to see farther than we could with our own cognitive limitations. So one thing that I find very interesting, as you're exploring these topics now, and maybe compared to what you were building a few years ago, is that the tools have changed drastically. And I'd love to hear more about, like, what's your workflow today? Like the way you approach, like new ideas, following your own creativity, like what's Ashe's workflow? So a big thing I'm leaning into this year is really building while connected. And I think for so many years as a founder I built in more of a silo and I'm very excited to be building in public to be connecting with folks on these different agentic workflows or products and leaning into the variance that you get when you build in public and you allow all of these folks to comment on what you're putting out there. And so something I've enjoyed doing is I have my Ashe.ai public facing website, which you could go on, but there's a whole back-end system that is behind a secrets page. Where I will, if I have a product idea, I'll kick it off. Now I'm using Codex 5.4 but, you know, I'm driving really fast to like test what I should be experimenting with, and then I'll have it within ashe.ai. I'll put it on on X, and if people are reacting to it, I'll have 5.4 extract it into a standalone product quite quickly, which is cool, or open source a part of it, because a lot of folks are asking for like an open source version to pull into their own personalized software. That's super cool. So you are exploring different ideas in parallel. You're putting them on your own personal website, using Codex to build many of these ideas. But then as you see them kind of getting traction, you're extracting them into complete products. Yeah. So the name of the game is speed. And I think the key thing is you could build so much software nowadays, but what will you build that you yourself use every day that you find to be a beautiful and elegant experience, and that you can surface to users in a way that is sustainable and isn't just breaking with the fragility of maybe a vibe coded product. And so to be able to drive 5.4 and Codex really fast, but still have kind of a reliable, foundation, which is new, I think this is new as of the past two months. Has let me put out, like aesthetic video as a platform where I'm doing a lot of video editing. And I'll have guests, but I want them to have access to everything they say in case any part is sensitive. And I was able to really quickly put out that platform. And I use it every day and guests use it every time I film something for collaborating for agentic clips. And that's just one example. That's awesome. And is there anything like surprising to you or in your workflow, like the way you use Codex, anything that has been like, fascinating to you? I think the speed is really important where I want to be able to drive the agent very fast in productizing, because I need to still validate it with a user base. Like just because you can build many things nowadays doesn't mean it will be a product that users understand or could come on, or that you yourself will want to use every day. So being able to have that rapid feedback loop building in public has been very important. And the other layers instrumentation. So I have my Ashe.ai system on Slack and then I have all my agentic workflows broken up by channels. For every product that I put out, I have a layer of instrumentation and observability so I can validate are people using the tool in the way that I expect, let alone like any errors or things that are worth me, like quickly trying to patch. That's awesome. I would love to see it. Do you want to show it to me? Okay, so this is my Ashe.ai Slack workspace. And you can see with all of my channels, these are all agentic workflows that I have here. And I'll get alerts and up manage throughout the day. And I have all of these errors for aesthetic.video. So I should probably kick off something because I do have people on this. And I think I know what this is. “Kick off a task to fix the infinite load on the drafts page.” So I trust 5.4 will figure that out. And yeah, so this is Ashe.ai. This is public. This has kind of my portfolio and products and also goals. But then I have a secrets page, and this is where I have everything from different templates. So I'll reuse the same Stripe template. I have these animations, to prototyping potential products. So I have my Oura integration here. Videos was the start of aesthetic.video. And then I got enough validation where I went ahead and built this out as a standalone product. So here I built it out and then told 5.4 “hey, like break off this chunk.” But obviously we have so much happening in Ashe.ai and put it on aesthetic.video. And then from here, this is how I organize the content that I'm putting out. I have an automatic X integration. And then when I want to collaborate with my interview guests I will have the video here. They can add comments and it'll be time stamped. I have an auto generated video guide. I even have an agentic clipper to get those fun reels. Wow, that's really cool. I loved, like seeing your video hub on X did you, simply just ask, Codex to extract all of the prototype and make it a standalone product, like in one shot? Yeah. So actually it did really well with one shot. So I had my template here and I need to use this for myself. So I tried it out with Ashe.ai. I posted it, I had enough people be like “hey can you open source this, can you put this out?” And after getting that validation, I told it to to break off a chunk. I let it, let it go. And then I had a good bit here and then I've added things to it. So I have like Google storage for all my videos. I pulled in, the Gemini embeddings too, to do some smart things with searching on videos. So I have like search over here and I could search for people laughing. Yeah. And then I play with it, like I have my chat here. Sometimes I'll use Opus 4.6 for the chat, sometimes I'll use 5.4 it's really fun in this moment to pull in all of these different agentic tools and model providers. Cool. I guess, like I need to, ship more video content and use the hub for myself too. We can put this video on here. Exactly. We should. Amazing. Wow, thank you so much for showing me that. That's really, really cool. And what I love too is that you were a solo founder before, and, like, you were able to build so much by yourself one more time. Like, this is incredible. Like, how have the tools, like, changed to help you build all of this? Yeah. So I think what's interesting is, as an engineer, kind of one of my defining attributes was speed. And I've always really valued running at a problem. And then instead of living in my head for so long, getting real world feedback loops to kind of coalesce my vision with reality and the magic of this moment is being able to drive like 5.4 incredibly fast. And obviously I also have that like autocross training. I want to go fast and trust that it's going to kind of have a foundation that isn't so fragile that I’ll to spend a ton of time redoing it, and then having that, building in public kind of feedback loop where I can put things out into the world and see the variants and creativity that comes back in. So, for example, maybe last week I had my Oura ring and I pulled the Oura API, and on my Ashe.aI goals page, I put up my wakeup times because I really love waking up at 5 a.m. and someone commented like, ‘hey, you should create a social version of this where we can all have a level of accountability for waking up at 5 a.m’. and the magic of this is now I can quickly sprint at this without it taking up so much time. And I think what's been cool about 5.4 specifically around productizing is I'll have my whole Ashe.ai system, I'll build the prototype, I'll see if I actually use it, because that's important. And then if it's validated in some way, I'll tell 5.4 hey go split this off into an open source repo, or go split this off into its own standalone product. And I'm seeing it do it quite reliably. Like I think that's kind of a huge difference, even, going back like six weeks ago. That's incredible. I have an Oura ring too. So I guess I should look at the leaderboard, but I'm not sure if I'm if I'm waking up early enough to be up there. That's really, really cool. And I love the fact that now that, like, we can build so much, it's really down to the creativity and the process you have around capturing your ideas and bringing them to life. And I love how you're describing the fact that you have this kind of secret space where you can experiment freely, and then you're waiting for a sign or a signal that, like things are picking up to then extract them, do well, is that right? Yeah. And I think, you know, our time is is zero sum. And I think especially as the builder archetype moves closer and closer to the creative archetype or to the artist archetype, You want to channel kind of your full self into what it is that you're building. Otherwise maybe why would people resonate with it? I think there's a real moment right now in which folks are inundated with AI generated content and have like maybe a repulsion to, to that video or to that product. And so I'm asking myself a lot, hey, what does it look like to step into this new world in which I'm building with my full self and being vulnerable enough to share that with other builders or potential users? And one of the things like, speaking of builders, that I love about your process is how much you're building in public, right? Like you're learning as you go it looks like. And you are also talking to a lot of builders making your own videos. What have you learned in that process? Has that helped you, like, build better products and get better ideas? Yeah. So a big thing for me is I want to live a very aligned life in day to day. And, the thing I'm trying to improve upon is, for a while I built feeling quite disconnected. And so if I'm building for relational intelligence, I want to live relational intelligence. And I have found that having these conversations where I'm really present and really curious, I mean it feels amazing. I think we're wired as humans to be connected and to kind of bring that into conversations around frontier tech feels like I'm stepping into the Renaissance moment that exists right now, and I feel very excited and I feel very abundant. And I know that there's a lot of fear in the world which is validated, which is which is valid. But, I think we have so many reasons to feel excited with the new wave of creativity being unlocked. I agree, I think like there's so much like exciting, so many exciting things we can build now. So for us engineers and for, like, all the builders watching, like you're mentioning these like Renaissance moment, how do you think they should grab it? Like there is so much potential. There's also some grief in some ways about like the work changing so much because now there's models that like GPT 5.4 are so good at writing very high quality production code. The role of being an engineer is changing. Like, what's your view on all of this? Yeah, I think this is the moment to lean into creativity as the North Star. I mean, if we go back to the Italian sculptor Bernini, he worked very diligently for like four hours and spent the rest of the day strolling or going to drinks or dancing or like really being having fun. And I know, like Peter talked a lot about the competitive edge, being having fun. And I think the, the exciting part of this moment is, you are incentivized to bring your full self to the table as a builder. And I think in part that's because there's so much noise with so many people trying to maybe sell you something when you go on X or, vibe code their own internal tool, or create AI generated content or AI generated movies. I think the premium now is on risking something by being vulnerable, by sharing and creating kind of a trust layer, or more relationships with folks for why they should engage with you not only as a builder to use your tool, but like as a human that they trust with, their data and whatever else. Yeah. And attention really because, I think this was in the news, but this question of putting out an article that's like AI written, I think a lot of users felt a lot of readers felt disrespected. And I think the premium is on feeling authentically connected to the person on the other end in this moment. And that's a call for all of us to share more, I think, about ourselves. And when you think about engineers and builders, do you think that naturally they should have more taste and interest for the customers and like, kind of like the role starting to blend when you work on the team, right. Because those tools are giving super powers to everyone. Like what do you see kind of the interplay between different roles and different teammates? Yeah, I think this is a moment to really attune to yourself individually first and to ask yourself, what is beautiful to me, what is impactful to me? What am I here to put out into the world? And have that individuality and creativity and from that place, then cooperate and build. I think we had a previous era in which, we were building for so many people in the world that we're in ourselves, maybe addressing markets first. I think the era of personalized software is now a mirror back to us to really figure out, okay, what is it that I want to bring to life? Absolutely. And you were mentioning Peter from OpenClaw that we had for this first episode. And it sounds like when you're working on your projects, you're no longer just working in projects. You're also working in systems. Right? That's something very interesting. I'd love to hear more about your workflow. Sounds like you're also using OpenClaw to build a system around your distribution. I would love to hear how you're thinking through that. So I'm very quick to pull out whatever state of the art thing comes out into my full agentic stack and all of the different workflows that exist. So I went deep into OpenClaw. And for me, you know, I don't have a marketing background, let alone like distribution across all of these different platforms. I will say the area that OpenClaw kind of shined for me was to figure out distribution for, you know maybe I'm really present on X and put out a post there to be able to then lean on OpenClaw and agents to re-share that on Reddit and what other platforms. Not in a way that feels spammy. And this is a really fine line, but in a way that allows me to get that idea out there in the language that is specific to that platform, because they are all different. So it was able to do this pretty impressively on Reddit, with seemingly human responses back pretty quickly. So I think for anyone that's looking to figure out distribution, which I certainly am, it's an interesting moment of like, okay, maybe I don't need to spend all this time on Reddit, on LinkedIn, on Instagram, whatever else. Maybe I can have my agents help me out so that I can be present with the idea. That's very cool. Any favorite way to talk to your OpenClaw? Where do you have it set up? So I will say I'm a big Slack person and OpenClaw is one piece in my full agentic system. So I do love Slack for naturally breaking up my agentic workflows by channel. So I have like a news channel, a birthdays channel, a goals channel, a money channel, basically anything in my life that I'm doing repeatedly there's a channel on slack with agents tied to it, and the important thing is the agents are up managing me to try to like, reduce my screen time and let me be really present with people. That's incredible. I love that. Really, really cool. I need to like, upgrade my setup. It gives me some ideas now. Yeah, I'll show you. I showed someone at a bar once and they were like, are you the only person on this Slack workspace? I was like, yes. And they started laughing. But it's the future! It’s you and your agents. Using all my agents. I'm not alone. You have a whole team with you. That's really cool. One thing that I love, that we don't, I don't think we have talked about yet, is the fact that, like, as you think through these systems for yourself and as you're going to pursue this thread around connections across humans in this AI era, you also thinking about wearables and hardware a lot, right? And you're working on one. Tell me more about like why that matters. Yeah. So I am very driven by I think, a life that feels present and that feels connected. And I want to be across from someone looking at them, fully present with them. And I really do want to minimize my screen time, my time on my phone and my time on my desktop. And so I'm excited. First about the social etiquette layer. I don't like ambient recording. I do think we will need a stream of consciousness that is systematized for our second brain. When I think a lot about a second brain for my people, I think there will be some social etiquette around you saying something really interesting. You seeing me touch a part of my body and agree that that will be put into my second brain, or pull me back and say, actually, hey, I'm not comfortable with that. So first I want to play with what might the future social etiquette be in a way that feels elegant and not gauche? And then second, just an analog experience that lets me have that second brain for my people so I can focus on really listening to what the other person's saying. So being fully present, but also thinking through the social norms of what that means to have such ambient intelligence almost. Yeah. And I think when it comes to our relational landscape, the human brain is naturally pretty limited. So if you think about each person in your life, there are so many facts and dimensions to who they are. Because a relationship is, who am I? Who is this person and who are they to me, over time? And relationships are mirrors back to ourselves. And so I'm excited for an intelligence layer that guides me on the abundance of who I'm connected to. That's really great. And so in this era of like building being easier and you exploring the frontier of the intersection of like software, hardware and wearables, very, very exciting. How do you foresee like, you own part in this journey and like, like in some ways influencing other builders to follow, that journey and that same like, that same chain of thought almost like, do you think engineers will also have to explore a lot more like that software to hardware frontier, or do you think they need to bring their own taste and their own vantage point to the solutions, like how you thinking through that? Well, first I want to lean into my own sense of self. And I think what's nice about this moment is the way it's empowering people to build their own personalized software layers, but hopefully present more authentically, with that building in public layer. And so I think a lot about aesthetic for example, and visuals and beauty. And I think anything I build in the future, I want to interact with it in a way that feels beautiful to me, even down to managing my agents. And we've talked about this, but part of what was cool about Codex is it feels like a really beautiful experience to start to manage all of these different agents and sub agents that are being kicked off. And I think what's exciting looking ahead is that you can imagine every touchpoint with agents in the future, or these systems that you're building or hardware or hopefully board design feels like a piece of art, feels like something that feels resonant and inspiring to interact with. Yeah, and it's amazing how fast we've transitioned from, you know, AI systems that were able to, like, write a bit of code all the way to like AI systems that can reliably take on, like a huge chunk of work, in your case like extracting an entire prototype and turning it in one shot into a complete product. Is there anything that has like, surprised you or delighted you along the way as you've been using like Codex? I assume you're using the Codex app now? I'm using the app. I switch between all of these different systems, all the time. So still using Cursor, still using Claude Code, using Codex, using all of these different systems. And I think what's defining about this moment is no two days are the same, like the whole ecosystem is moving too fast, that I really feel like a creative operating in bursts. So you know like, I'll have a product like the aesthetic.video and right away saw 5.4 was like the right partner for that because it was moving really fast but reliably with understanding okay, there was going to be like a user authentication layer, like this was going to be a product I put out. There was going to be a Stripe layer to this. and then having that like sprint for however many hours and then pausing and putting it out into the world and doing videos and doing something different and having a lot of variance in my day and workflows. And that feels really new to this moment. I think in the past, as an engineer, you know, I would be doing really similar things for like 11 hours or like day in and day out. And now I know I keep mentioning the artist archetype, but I'm best served by, okay, when am I most creative versus like just sprint, just like a marathon? so now builders can truly build anything with these tools, like Codex. Do you think like there’s any, quality, like maybe like thinking through taste or the curiosity that matters more these days? Like what? What would be like, top of mind? I think the call is for folks to be really present within themselves, within their bodies. I do think there is a pull in the direction of what you want to create and what you want, what is resonant to you, whether that's film, whether that's photography, whether that's like products, whatever that pull is, I think aggressively pursue it and curate things that are similar. And, for me, the advice I would give my younger self is to be uncomfortable, you know, like there is discomfort in growth or in putting yourself out there. You know, anytime I've put something out there that has a little piece of me in it, it is scary and there is discomfort and I'm risking something by being seen. But it's so worth it to build that reflex and that muscle, because that's what's meaningful. And that's also what lets you pull in the variance of real world feedback and get more creative and keep, you know, building. Just to talk about Peter and OpenClaw one more time. I think that was an amazing example of putting something out into the world in which so many people poured energy back into it, and it had this emergent quality of growing beyond what anyone would have guessed. Right? And I think that's kind of the magic of what's possible now, when you have these relationships and feedback loops with all of these other builders. Yeah. And I think what has like, really, amazed me with OpenClaw was not just the unique insight that Peter had to put something amazing out there. It's also the community around it, like the number of people now, like wearing lobster hats and like really passionate about making their own personal agents. That's quite special. Yeah, yeah. So back to your question about taste. I think that's really a question on self attunement and individuality. And if something is resonant to you that's worth passionately pursuing and sharing and let it like go out into the world, let it breathe and allow there to be a surprise in what could happen. Right? And maybe as we wrap, you were talking about advice for your younger self. You've had like all these experiences from being an ML engineer, like building satellites, those racing cars that we talked about. You've been the founder a couple times and you're now kind of embracing these tools every day. What would be your advice for builders around the world like that maybe like earlier in their journey, just starting to explore these AI tools like how should they approach this brave new world? First, curiosity, you know, and having fun, I think is great advice. And really honoring that curiosity, you know, if we have these desires, I think it's for a reason. And to let those pulls guide us and to to take that risk because you want to know what happens. You know, I think curiosity is such a magical part of the history of being an engineer. And I think as everything is changing, that is the same pillar. It's to be driven by curiosity, to bring something to life that is newly possible. And I love how it's a very similar advice to what Peter shared. Also, trying to have fun along the way and trying to follow your passion, your curiosity. That's awesome. Well, I think it's a great way to end. Thank you so, so much for spending time together, and I can't wait to see what you build next with all of these great ideas that you're putting on X, every day. And, can't wait to see the device too, that you're working on. Thank you so much. This was so fun.