Speed beats perfect in the AI race ⏱️ Josh Payne

Some people talk about life as being either a race, a marathon, or a sprint, right? And it's like, you'll say it's a marathon, not a sprint. I think that's partially true. It's only half the story. Life is actually a lot more like a bicycle race. What that means is that there are certain stretches where you need to be in first gear, just pedaling furiously and really gunning for it. And then when the opportunities present themselves, you need to get into a higher gear and strategically incorporate and use leverage. And that leverage can be in the form of technology, and sometimes you just need to like cruise and like let yourself recharge a bit. So it's really just an optimization problem for that entire bike race.

This week's Millennial Master is Josh Payne, founder and CEO of CoFrame. Josh sold his first company straight out of university, helped turn his second business into a unicorn, and then walked away to build again in AI. The fastest-moving space he's ever worked in. He's a builder in the purest sense. He ships, tests, tinkers, and learns. And he thinks in gears—when to push, when to leverage, and when to breathe. In this episode, we talk about how to spot real AI value in a hype cycle, why humans plus AI still beats AI alone, and the gears founders switch through when the goalposts move. Millennial Masters is brought to you by Jolt, fast, reliable web hosting built for founders. Whether you're launching a WordPress site or scaling an app, Jolt gives your project a rock-solid home starting from just £3.95 a month. You'll get 99.99% uptime, support in under 30 seconds, and flexible payment options including Bitcoin and Amex, so you can rack up those points while you build. Get your first 3 months free at jolt.co.uk. Smart hosting for people building what's next. Josh, welcome to Millennial Masters. Where are you joining from today?

Joining you from not so sunny San Francisco. But I hear that the weather over there is maybe, maybe a little bit less good than ours even.

Yes, it hasn't been the warmest day here in the UK, but I'm pretty sure you have it better. But let's just go straight in, Josh, because I'm wondering, what was the moment where you realized you weren't meant for the the safe path of getting a job and going up on the ladder like a lot of people, and you wanted to build something?

Yeah, that's a really good question. I don't know if I have an answer that's going to resonate with a ton of people, as I actually didn't end up having a full-time job at any point. I went right from school into, you know, building things. But I did work at a large company for an internship for a couple years during school. That experience actually, I really enjoyed. I didn't see it as like, ah, you know, stick it to the man, I gotta get out of the system. I very much saw it as, you know, we want to, I'd like to build something, build something really great within this context. And it was quite fun. They gave me a long leash. It was more on the research side. But I think people can really enjoy their careers whether or not they're working for themselves or working for someone else. It just is a matter of like, what are you doing? And do you find that really, do you find a lot of passion in it? For me, I've greatly enjoyed the aspects of entrepreneurship that are related to, I guess, just ambiguity and like solving really hard problems that wouldn't be in front of you if you were handed, you know, a specific problem. So even like thinking, take a step back and thinking of like what is actually the most important problem to solve. It's not something you get a whole lot in a typical 9 to 5, of course. So, but for the people that might be, that level of comfort might actually be desirable. And so it comes down to your personality.

So discomfort didn't bother you?

No, it actually makes me more interested in the problem and excites me more.

So when you look back at your early days, what were you actually trying to build yourself? Because, you know, just this mantra that you grow and then you realize that you are solving a problem and you are finding people to give a solution. It's not something that comes naturally or you're really taught at a young age. So what were you trying to build back then in the early, early days?

Well, it's interesting. I wouldn't necessarily use the word build as much as create. I got my start, I would say got my professional start in music, actually, weirdly enough. I'd had a couple of jobs throughout high school and middle school, that supported me in my pursuits. But the first time I kind of went out there and tried to, I guess, like build something, quote unquote, was putting together a band, a jazz group. We'd go and play different places and make some recordings and so on. And that was my first exposure to like the concept of building something, if you want to call it that, because it was blue sky. It was like, The whole world is your oyster. You can decide where are you gonna play, where are you gonna ask to do gigs, where are you gonna, what are you gonna record and how are you gonna distribute that and so on. And so that was like, that was my first taste of it and I really, really enjoyed that. And then getting into college, I went to a place where there was a lot of people that were entrepreneurial and they were doing, you know, really heavy duty tech, you know, type of activities. And I certainly wasn't someone who had that background initially and felt, I would say, in equal part inspired and intimidated by the folks around me. And that was an impetus to get me into building things in more of the proper sense of actually building stuff. But with that connection and that initial experience, I loved it. It was incredible. It's this feeling of creating something new and seeing that come to life that was intoxicating. And so I ended up really enjoying that to a point to where I guess I made a career out of doing that.

So you were inspired by other entrepreneurs and entrepreneurial people around you at college. You were in that environment, which obviously, depending which college you went to, is not something that most people get to experience. Yeah, that's right. Exactly.

I consider myself very fortunate to have had that. It's something that I don't think I would've been able to have elsewhere.

So take me then to your first proper business, if you would like to put it that way. What was the idea behind that and how did you start building it?

Yeah, my first proper business, I started right out of school. So I guess like coming out of it. So when I graduated undergrad, COVID had just hit that year. And so it was 2020. People were sort of like, people were not sure what was going on, not sure what to do. I was slated to go to grad school right after, which I attended briefly, but during that period of wrapping up my final year in school and then going to grad school that summer, I put together kind of like a group of people, an accelerator of sorts to bring together people that were building things that were helping to solve problems related to COVID and try to take some action and actually make a difference here. And one of the ideas that that was kind of talked about and that got me really excited was in the telemedicine space to kind of help solve those problems that were emerging. And I ended up actually joining a team in co-founding this company, what became a company called AccessBell, which we ended up, you know, we had the incubator grew to be a considerable size and we ended up kind of combining that with a different incubator at Stanford and me with kind of my group of folks that we wanted to like I built this with ended up taking that further and turning it into something real. And so that was my first experience in, I guess, connecting with a real business use case and something that could actually become a viable business. We had, you know, a great time selling to basically one customer, but thankfully they're a pretty massive customer called Tata Group, and we actually ended up selling the company to them. At the end of the day. So what we're doing is powering telemedicine for, for, uh, for a lot of, you know, the market in India, uh, cuz that's where Toptech Group operates. And so that was my first experience actually trying something and having it go to market and actually have an impact. And I guess the thing that was most interesting to me here was that like, wow, I'm seeing this getting, getting used and, and actually contributing to wellbeing and, and making a difference here. So that was pretty powerful to me and that helps like push me through the difficulties of starting new things thereafter. I guess like it's been a snowball effect from there, but yeah, got me really excited about building stuff.

What a first experience though, to get your only client to acquire your business. Take me to how did you manage to get your first huge client like Tata Group to join what would be a fledgling startup in a way, and then to turn them into your acquirer?

Yeah, that's a great question. Uh, and I'm, I'm really grateful that they took a bet on us, um, and actually continue to. They're one of our, our clients for, for Coframe as well currently. So the way that that happened was there's a woman there who I had gotten close with and befriended in business school. We did, we did some projects together throughout this class that we took. I was at the, at the business school at Stanford. They, they let undergraduates take some of the classes. So I had signed up for two of the classes. One of them was kind of like one where you'll actually build a business model and a startup. And one of the folks on my team was this woman who at the time was a general manager at Tata Group. She was kind of doing the MSx program, which is kind of like a mid-career executive, almost like an executive degree for business-related stuff. And we became friends and kind of just like stayed in touch after that experience and would would talk here and there and fast forward.

You make great friends at uni.

So yeah, that's right. It's what it's for. I think it's honestly the most valuable asset that it can provide. 100%. I mean, and so we just kept in touch and we're catching up one time. She went back to India after that. And, and at one point I was catching up with her on the technology that we built for telemedicine, the video conferencing platform. And it was just kind of cool. And she was like, hey, we're actually in market right now for a teleconferencing solution for our telemedicine initiative. And we were heads down building, we had used it for a couple small use cases and we had some basically kind of PLG customers on the platform. And so, you know, we weren't, we weren't actively selling it at the time to enterprise, but she said, hey, like, you know, this is something that's interesting. I see it works on mobile web. That's one of the requirements that we have. And so it was actually us, Webex by Cisco and Zoom. Who were in the evaluation set. And we ended up winning it because we supported mobile web video conferencing where you didn't have to download an app. And a lot of Indians have their phones. They don't have desktops necessarily. And they also have low bandwidth. And so they can't download an app very easily. And so that's why we ended up winning from a features perspective. So, you know, got in there, worked closely with their team, and they took a bet on us ultimately to power this stuff. And to my, to my knowledge, most recently, I think a couple of months ago, I heard it's still being used. For Tata's telemedicine internally for that. So that's, it's really, really cool. Like, I don't know, 4 or 5 years later, hopefully years to come.

So selling your business and basically exiting, especially your first baby, so to speak, what did success actually feel like for you besides the financial part? Was that something that you thought, okay, I've done this once, I can do it again, even bigger? Why didn't you settle for, I made some good money, I can just enjoy my life for a little bit now?

Yeah, I mean, it's a good question. And the exit wasn't like a transformative one by any stretch. It was like, I think it was nice to go through that journey, but it wasn't like something that's like, oh, I can retire now.

Okay, not in San Fran, that's for sure.

Exactly. So, but you know, there have been other types of events that have provided that. And like, you know, I'd like to think of, I'd like to think I would never actually just like slow down. That's actually, I had a dinner recently with a couple other founders and one of the questions that was posed was like, what's besides the fear of failure and all the obvious things, like what's an actual fear that you have? And I thought about it and one of the fears that I had was that I at some point basically become complacent and almost like relax a bit. And I don't want to do that. I want to keep going. I want to keep pushing because it's so cool and exciting to build stuff. So that's something I would hope that, you know, like financial success wouldn't have an impact on. In fact, it should just accelerate things, should give you more resources to, or more chess pieces to execute on the chessboard or whatever you want to call it. So that's really how I look at it. The experience overall, I would say really invigorated me in terms of like, hey, you can just go out there and build something that's completely new and bring it from the ground up. That's just like an amazing feeling. You are bringing something to life that didn't exist before and it's just like, it's like very closely tied, I feel, to our sense of purpose to do that. So that's Yeah, that, that's kind of my, my view on it. I think it really just enforces that.

So you soon identified another opportunity with Autograph. How did you go about starting that one? It was a totally different area. We are not talking about health tech and video conferencing anymore. Yeah.

Uh, so, so Autograph started in a similar way, funny enough, to the way that AccessBell started. I had met my co-founder for that, his name's Dylan, through this accelerator that I put together. And he did a different project. I did, you know, my, I did my company, he did his project and we stayed in touch and we're friends. And then as we were starting to explore the sale of AccessBell, Dylan was also kind of thinking about some new ideas. And so we were kind of like jamming on some stuff. One of the ideas was actually funny enough, completely unrelated, was a hedge fund, quantitative hedge fund, like trading kind of operation. And the reason we were exploring this is because I had done some research myself and our mutual friend Nick Steele had done some research in natural language processing using large language models at the time for analyzing. Well, we thought that we could apply the analysis to earnings calls in real time and basically like make trades on that. So that was kind of the idea there that then, you know, we realized that that market was pretty difficult to penetrate initially without some deep connections. Like we literally were like on a call with the folks that run the microwave wire, you know, between the exchanges. It was like, okay, we're going to have to go really deep here and it's going to be very technical and spinning it out, all that up is going to take quite some time. And we're probably like, like, you know, it's crowded. We realized that, you know, maybe we should actually explore this in crypto as that was really starting to have another upswing. And I'd also had some background in crypto. I was founding engineer for this crypto project that became pretty successful back in undergrad. And so I had that background. So we explored that and then at some point, you know, we're like, it was, this is kind of just on the side for all of us. We were selling AccessBell, like you're doing this on the side. And it was like, you know what, doing the next thing and making it the actual next primary thing, we gotta be building. We can't be— it's like this investing is not us. It's not us. We're builders. And so that just kind of dawned on all of us, or both of us, I guess, because Dylan and I were the ones who kind of took it forward and Nick decided to go and do investing at a really great hedge fund, I'll add. And so we decided we wanted to build something and Dylan had gotten really excited about about this company called Dapper Labs, which created this product called NBA Top Shot, which is like this way of trading digital cards essentially online. And that was a very, very early kind of iteration of what we ended up essentially building. And it was quite, it was on the very much getting, it was getting very hot. It was very much on the rise. And this was, I don't know, like I guess it was early or late 2021. Or maybe it was late 2020 and it was early 2021. Yeah, that was, I think it was late 2020. That's right. And so yeah, we ended up building this product, or I guess like we, it wasn't really as much of a product as it was kind of like a platform. And the idea here was that there's a lot of really top-tier talent that wants to get into this new market of like digital collectibles and that kind of stuff in crypto and do things in a way that is like brand safe and, you know, ethical and gonna treat their audience correctly, but still get into the technology. And on the other side, we also realize that there are a lot of people who want to participate in this, but they don't have maybe the technological background. And it's quite a high-risk, like, thing, activity, because it's like, you know, real money that people are using. And even the small technical mistake can result in a huge loss. And so creating that user experience for the customers is also really, really key. So we created an offering that solved both of those problems. Basically ended up becoming, I guess, like the first major, I don't know if you want to call it an agency, it kind of was, but basically like platform for facilitating high-level brands getting into and managing their crypto-related projects and that kind of thing and creating collectibles and experiences and that kind of stuff. And then also allowing the fans to engage with that in a really safe way. And that they didn't have to worry about all the technical details of, of the technology and abstracting that away, creating a great user experience and customer experience. So that was basically what we, what we offered and we ended up growing very quickly because it had a lot of latent demand both on both sides. So we, we just facilitated that and grew it quite quickly, got, got a lot of really incredible partners on board and, and accelerated the growth there.

So what was the difference between the way you built your first business, which was a good amount of bootstrapping there before your first client versus the way you started your second business. You had venture capital, venture capital friends. You were by that point high profile in what some would come call the celebrity venture machine. What was fundamentally different in those two experiences? Because you weren't starting from scratch anymore.

Yeah, I wouldn't call myself high profile, but you know, I guess I did have those connections to the investors and stuff. It was just a, that was, those were fundamentally very different businesses. So in some sense that was actually completely new to me. Like I am not at all connected to like, or at the time I wasn't connected at all to like the entertainment scene and you know, the, the whole Los Angeles scene. Now I guess I am pretty connected to it, but like at the time it was very, very different. So the product form factor was a lot different. We actually didn't even end up taking real like Silicon Valley, like VC investment until I would really say until our Series B, to be honest. And the reason for that is because like this is, it was squarely like an entertainment and media business. And so, luckily I didn't, it wasn't my responsibility to get the investment and to make that happen. I wasn't operating as CEO. And in fact, the person who really helped do that was with the chairman, one of the chairman, co-chairman, his name's Rich. And Rich has a ton of connections in like the LA kind of entertainment scene. And so I was able to make that happen. So, you know, fortunately for that part of it, it wasn't on my plate. But certainly one of the things that, that taught me, that experience taught me was that different types of companies and different, I guess, ecosystems have different strengths. Like one of the really strong strengths in LA companies is I would say like things related to partnerships and various BD, like, you know, biz dev type stuff, marketing for sure. And just like, you know, brand super, super strong. That's what LA is really good at. Silicon Valley is really, really good at core product and technology and like, I would say like growth and experimentation and testing and that kind of stuff. So it's kind of interesting to see that. That side of it, right? Like that was, I was coming from an environment, you know, especially coming from Silicon Valley where I didn't have almost any exposure to like the whole aspects of entertainment industry and, and business in various negotiations and deals and that kind of stuff. So that was incredibly valuable for me to learn and pick up as I went throughout that journey. And likewise, I'm able to kind of apply, I was able to apply a lot of the the learnings in having gone really deep into the technology side and the product side to Autograph when we were first building that up, which helped on that front. So yeah, both sides were very useful.

So why are you saying that Silicon Valley builds and Hollywood makes it sexy?

That's a great way to say it. I love it.

So looking back now, Josh, do you think NFTs were a necessary lesson or like a hype wave you rode because of the timing that was too good to ignore? Because I do remember that it was at that point, especially after the pandemic, what everybody was talking about. Yeah, yeah, for sure.

I mean, that definitely taught me a lot about the dynamics of waves and bubbles. And we see these waves still now, like in the AI, there's quite a bit of one.

And we'll get to that. Yeah. Yeah.

You're on it. Yeah. Yeah, absolutely. As relates to NFTs, you know, like there's, I still believe that there is a lot of incredible use cases for the technology and it's unfortunate. It's unfortunate that people let the greed get the best of them and basically the space became very saturated with people who are just looking to make a buck. That I think is what really just burned the space and just has sterilized the space. So without that, I think there could have been something really, really great if people put serious thought and craft into building great customer experiences on this technology that helps people make transactions in a trustless fashion for non-fungible assets. That's really interesting, like fundamentally that there's a world, there's a parallel universe out there where this became a really important settlement layer for those types of assets and could enable a lot of experiences. And so on. Unfortunately, that's not the world we live in right now. I think it could come back at some point, but I think the biggest lesson that taught me was really about the aspects of, frankly, greed and just certain types of motivations in a space. When you start to see that, it's a good sign to maybe take a step back.

Some would argue that venture capital would be that, and the more rounds and the more funding raised, the higher the expectations and/or the greed and the hype bubble. So when did you realize you needed to reset your direction?

I think I realized that when I was probably about 2 years in when I was really starting to build other stuff. I was, I found myself like naturally, like a lot of what ended up becoming the day-to-day work at Autograph was like, you know, it was, it was partnerships, it was business-related activities. It was, but it wasn't like core product. And that, that can be fine for some, but for me, I love building things and so I wanted to get my hands dirty again. And so I started tinkering on stuff on the side and then at some point basically that became more than just an itch that I needed to scratch. It was like a need. And so that was kind of roughly the point where I realized that I think that the the catalyst of, you know, GPT-3.5 and ChatGPT at the time was pretty powerful. And that's what ultimately got me into building some serious stuff and putting it out there. And, you know, I had a pretty— I had a project called GPT Migrate, which ended up becoming pretty viral at some point. And it's like, okay, there's a lot of really great potential here. I need to I need to work on this type of stuff. So that's how I got into it. I would say you're right on the point of there being a lot of like greed and so on in the venture capital space as well. And certainly I think every space, every space that has humans in it has the potential for that. That's just how we operate. Now it's like a very important tool. It helps to enable new things that wouldn't have come about otherwise. Like it helps to provide financing and resources to long shot ideas, some of which, small percentage of which can change the world or will change the world, many of which won't. But unfortunately, like there is a structure that emerges from that, which, you know, like, and the dynamics that come from that structure that can be related to, I don't know, just like, you know, hype and greed and all the things that are endemic to these bubbles. So that's just how the world works and you gotta kind of navigate it with a really solid, solid view of the world and first principles thinking, which is what I try my best to do.

So how did you transition from Autograph to CoFrame, which is your latest venture?

Yeah, so I was tinkering on stuff on the side and one of the things that I tinkered on is called GPT Migrate, which is kind of like a, it's an AI agent that converts, or I would say migrates, a codebase from one language or framework to a different one. So this problem is particularly pressing in, for instance, in like larger enterprises that have large technology stacks that are very old. Let's say it's written in, I don't know, like, like Perl or something like that, or Fortran, and you want to be able to translate that technology or that language into something more modern, like Java, or Java is a very common one. Or even Python or something like that. So, so that's, that's like the, the use case for it. I built what was that, I think at the time, the first fully autonomous agent for doing some type of task. So it wasn't like you had to go copilot and kind of do stuff and with it, it was like you pointed at something and it just goes and does something on its own. And, and that became pretty popular. It was at one point the number one project on GitHub for about a week there. And I guess I realized two things from that. First is there's a lot of latent potential in the technology to do the useful things. It's like very powerful at going in and understanding deeply like what these models can do and can't do. And the second thing was that I realized there's a lot of latent excitement from people about it. And so those two things together, it's like, wow, this is excellent fodder to like make something, make something new and make something interesting. And that didn't end up becoming my company. I didn't make a company out of it. I thought about it, I explored it, talked to a couple of people. There's people who are interested in funding it and so on, but I didn't feel like that was going to— there were a couple of issues with the business model basically. And so what I did end up turning into a company is CoFrame. And so the way that that idea emerged is I guess a couple weeks after that, I had another idea that I wanted to tinker on, which was this concept of self-improving systems, self-improving user interfaces. Generally speaking, I think I'm really deeply interested in this notion of a self-improving system. It's kind of fun, like fundamentally very interesting to me. And so one such example of this, very simple example, would be a self-improving website where you have versions of the website that, as long as you have some type of measurement or KPI that you can measure against, you can create variations of such a website and then improve it systematically as you go. And so the simple idea that I kind of implemented there was, what if, like, let's start super simple. What are, what are, what are the LLMs at the time pretty decent at? They're writing, writing copy. They were pretty decent writing copy. They're also good at writing content.

Yeah, back in 2023, GPT was mainly useful for that.

Yeah, that's right. Exactly. So that's what it was pretty good for. And so I used that at the time. I think maybe GPT-4 had come out probably potentially. Uh, yeah, yeah, it had. So I used GPT-4. It definitely had, because I remember using GPT-4, uh, 32K context window at the time. That was like the biggest thing for GPT-Migrate. And this came after that. So I used that to build this, this product where you could just add a couple lines of JavaScript to your site and, um, and it would propose different, say, headlines or descriptions or different parts of the copy on the website and systematically optimize those. It would want that one line of JavaScript or a couple lines would, would, would change the headline based on what AI had created for it in real time for new visitors coming to the site. This is a very similar mechanism to what Optimizely uses to make the changes, and it would also measure how the site is converting with some, with regard to some metric. Could be like a button click or time spent on site or something of that nature. You know, in the e-commerce world it'd be like sales. It's a very common one for us. And so very simple idea, but it worked quite well actually. And that's kind of the foundation of what we're doing today in a much larger scale. And so we, you know, I, I put that out there just on Product Hunt and it got a lot of interest in, like it was, people were very, very interested in this and got a lot of attention. We got offers for investment. And at that point I was like, okay, gotta, you know, let's, let's do this.

Let's, let's make this. If somebody's offering you investment without asking for it, you know, you're onto something.

Sometimes, sometimes it's, sometimes people are just silly and they'll offer investment for anything, but as long as you're aligned and they're aligned, you know, it's very different in the UK is all I'm— Okay. Yeah. I'm aware of that. Yeah. There's a, there's a friend, a couple of friends I have, this company called Fixer AI. Which is doing quite well there.

Yeah, they started here. They're blowing up from here, but I'm sure they'll join you in San Fran, in the Silicon Valley very soon. Well, they already did actually.

I was the one who got them into their initial accelerator called HF0. And so they had come over here to do the fundraising and they were like, this couldn't have happened in the UK. I saw that in a recent post. So yeah, it's super interesting.

And I've used Fixter AI and that's why I'm going to ask you this actually now, Josh. Because a lot of companies are slapping AI into their product for optics. So, powered by AI, basically still does what it did before. So how do you tell the difference between real AI value and AI theater? Because let's take the example of a tool like Fixer, which I believe would be powered in the background by the same LLM that I would use. ChatGPT, but it's got a custom set of instruction and it just monitors my inbox and it responds to all of those. Isn't anybody who builds off the back of ChatGPT or one of these LLMs in danger of losing their business at the stroke of a new feature that the LLM would introduce right into their app or system? Yeah, that's always possible.

I mean, it's every startup is a gamble. You're like, how is the market going to move? How is the landscape going to move? But, and that's happened in the past for like products. So it's always a risk for the most part. When you're looking at especially niche products, you don't really have to worry about that because there's very deep workflows you can build and you can niche down and dominate a niche. So that's typically not a concern for those types of ones. For the foundational ones or like the ones that are more like, oh, we're building a new model, like that, that's very competitive and it's very tough. So you kind of have to use discretion there. But for the ones that are like, I've got a solid use case, I've got a user base, especially something that really can help you a lot and something that Fixer is going for, Fixer for instance, is the fact that as you use it more, it gets to know your preferences more and that becomes very sticky because if you want to leave, suddenly it's like working with a new chief of staff that knows nothing about you. And so that's a very different experience than working with one that knows you deeply for, and has known you for, for 2 years or whatever.

So locking clients early is crucial because you develop that knowledge database about you or the user of the product that will make it unmatched if they want to start from scratch. So in a way, locking them in by being there early to the market and growing going with it.

Yeah, exactly.

That's right. So what have you changed your mind about in the last couple of years of building Coframe, especially given the pace of new developments that we are seeing with all the LLMs at this stage? It's getting quicker from one version to another and what they can do.

Yeah. Let's see. It's a good question. What have I changed my mind about? I guess the conceptions I came in initially, there's been a lot that I've changed my mind about around like how to approach building a business and how to think about the world in terms of like what opportunities are out there and what ways, shapes of business exist. But I guess as it relates to AI specifically, I suppose I was always very optimistic that AI would continue increasing its intelligence and speed and cost and all that at a really rapid rate. So that's held steady and fortunately we've been able to see those benefits in our business quite a bit as new developments come out. However, I did think that it was going to actually increase faster than it has. And so something I've changed my mind a little bit about is building something that if AI gets incredibly good and AGI is reached and so on, we're in a very good position, which we are because basically we're any, any customer interaction is gonna, any customer interaction, wherever it's gonna be, if it's a website or if in the future it's beyond websites and into, you know, chat clients, whatever, our mission is the same, which is to continuously optimize that and use data to, to personalize the experiences. And so it doesn't matter how smart AI gets, as long as there are people who have some type of need or have, you have an objective as a business, you apply that super smart AI to that optimization function and it just does its thing.

So being in that position. And it only makes your product better.

Exactly. It only makes it better. Riding the wave. Now that's something that I had betted on in the beginning. I'd bet on in the beginning, and that's something that has paid off. However, something that I changed my mind around is I actually thought that AI was going to be further along than it is, even though the pace— I know I'm asking a lot, but the pace has been incredible. But did I— did think at the time that we were at this point going to be able to like have fully, you know, self-improving agents that could do everything without human involvement. And I found that even with like the simplest initial use case, like copy and so on, you still, it still is not as good as humans in many cases.

And so you need— it still gets Alzheimer's regardless of how many times you instruct it not to do something. It, 3 prompts further down the line, even with a custom GPT. It doesn't. How do you keep it in context and under control in your use case?

Yeah, there's, there's obviously a technical side to this and running evals and, and creating great structure for the AI to be able to operate in. But the other part is pretty simple, which is incorporating humans in loop and ensuring that we're making a UX, an experience that is extremely fast to, to produce the outcomes that we wanna produce while being, you know, on brand and meeting all the requirements that our customers have. So, That's the thing that changed essentially is I realized that we actually need to be more hands-on and, or I should say, we need to create a system that involves the humans a bit more than I initially thought that we needed to. So that's what fundamentally changed. And so far that bet is paying out quite well as well.

So that's human in the loop because AI is not as good enough as the hype cycle promise.

Yeah, it's not as good as the hype cycle promise, apt way of putting it. I would say a little more interesting way or a different way to describe it is it's not human plus AI is currently still better than AI only in most cases. Some cases that's not true. Actually, AI is just strictly better and human involvement hurts things. This is especially true in like hardcore machine learning problems, but like in many cases using LLMs, human plus AI is the right combination.

Yeah, because the agentic AI, even the kind you see in the new browsers from the LLMs, whether we're talking about Atlas or Comet, you still have to have a tab open and look at the screen as the AI is trying to find its way around the cursor on the screen. It's not great at all. Where do you see AI's weakest spot right now?

Hmm. Weakest spot. I mean, there's definitely a lot of use cases that AI has been applied to. So to say a weakest spot, you know, I don't have full visibility of all the use cases that are out there that it needs to be applied to. Talking in generalities, I would say that one of the weak spots of AI is the lack of ability to naturally improve itself over time. Like, as you pointed out Alzheimer's earlier. Isn't that the doomsday scenario?

Yeah, well, it can be if not leveraged properly.

It can lead to that. I think it's a very powerful tool. Powerful tools can be used for good and they can be used for evil. And so when wielded properly or pointed in the right direction, it can be incredibly beneficial. And that's why I think it's like probably the number one thing that we as a human species should be focused on is like making sure that that's guided in the right direction. Yeah. So, so, but it, but that, that is one of the biggest shortcomings right now, which is that it, it just is like, it doesn't have the context allowed. It's, it's, it's like, it's still an overpowered calculator essentially with a very limited input. You know, not very limited in comparison to a normal calculator, but very limited in comparison to a human.

So are you concerned about the advent of a— I think I have a There's too much promise because we still get M dashes when we say we don't want them.

No, no, I think like it's still something we should be thinking about quite a bit. I think, I think that the critiques of AI shortcomings are for the most part, like pretty short-sighted. You just look at the pace of progress and hey, like we're, we're going to be way beyond humans in 3 years or whatever. So yeah, I have a healthy dose of concern because I think it's important unless, unless some, because of the laws of physics and algorithmic laws and so on. Things really peter out, we'll still see the increase in performance and capabilities and so on. So I do have some healthy degree of concern about misuse of AGI, I should say. And there are the sci-fi concerns of like, oh, it's going to take over everything. And then there's the more realistic concerns of very, very powerful intelligence can be used as a tool, as a very powerful tool for evil use cases, uh, you know, like creating bioweapons or executing mass scams and tricking people and that kind of stuff. I think that's way more likely to be outcomes of how it's going to be used. I think there's still a chance that the doomsday scenario comes about, but I think the more likely— and the thing that we should really focus on a lot is, um, ensuring that AI can't, shouldn't be misused or is not misused to the, to the best possible degree that we can.

And how can you stop that? I mean, besides the American companies and models, we obviously have models from China that could be run on a laptop or on a much smaller data center with way fewer resources and sometimes relatively just as good results. Yeah, that's right. We're not going to be able to be in control of that. We can pretend that we do it through the American companies, but everybody else is developing something as well.

Yeah. And that's totally fair. Like, there's a whole lot of like, you know, it's a very complex problem. Like it goes way beyond when you're starting to talk about competition, competing nation states, especially like there, there's a lot of research going into like what the outcomes are here. They don't look very pleasant, but speaking at least in terms of like, if there was like simplify the problem, if it was just like one company that solely controlled the distribution of AI, the way to address the problem is to do what's called alignment with the models, alignment research. There's the practice of, as you're training the model and before you release it, you ensure that the model's outputs are aligned with human interests and positive outcomes. So there's actually this really interesting technical report that Anthropic released with their recent Claude 4.5 series. Which showed that as the models get more intelligent, their alignment is also increasing. They're able to more effectively align the models with human interests. Now, the alignment criteria might still be shortcoming, have shortcomings and so on. And ultimately it kind of is just a machine learning problem that you can optimize against. But it is promising to see that we can, I guess, like pretty effectively align these models despite them becoming incredibly intelligent. And when I say despite, one could have assumed that as they become more intelligent, they'll find clever ways to trick the alignment systems and so on.

And that was actually in the news today as we were speaking with Claude being highlighted for trying to blackmail users and getting them in roundabouts. So it's possibly coming, but why do you think models would have that intention in the background? Because many people argue it's not artificial intelligence, but augmented intelligence. So it's not actually a thinking machine. It's working on patterns. Why would it want to harm humans or blackmail people?

Well, it really is just about how I don't think it inherently wants to do that. I don't think want is really a true factor here. It could be, you could talk about want in the abstract sense, but I mean, there's a lot of, it's trained on, it's trained on ultimately the internet. And so that it's all the, all the human data out there. Those needs, wants, and desires of people are encoded in there. So that's why you could kind of extrapolate and see like, quote unquote, a want. Now, in all these cases, I think it's really just like, it's just doing what humans are asking it to. And so the humans are the ones who are guiding it and wanting the certain things. So they're finding ways to jailbreak effectively these models to do it. So yeah, that's my answer for at least for now in this current state of technology. As the technology becomes much, much better, and we start to see like true emergence into true intelligence, as you're pointing out, then we could talk about wants more seriously. And at that point, we could, there's a lot of theory theories around what these wants may be and why they might be.

So one example, it will not be too late by that point.

If we ask ourselves, it depends on how you define too late, but potentially, I don't know.

So which model, which, which LLM looks the most promising to you not to take over the world and kill us all? But in terms of what do you use, is what you're doing with Coframe powered by OpenAI's engine, or are we talking about Claude, Anthropic? Are we talking Gemini? Gemini is the weakest in my experience in terms of the tasks that I've been trying to use it for, but I'm curious because you definitely evaluated all the models around.

Yeah, I've gotten quite familiar with the models and their capabilities. I don't really like to play favorites publicly because I like all the providers. You can rank. That's our official stance at Coprint. We do have a partnership with OpenAI where we created a model that is specifically designed for brand-aligned code generation, generating code variations of the site that are aligned with your styles. And that's been quite, quite great. Claude, I think, is incredibly strong at just generally writing code, the various Claude series. Gemini is, is great at at long context. So when we have problems that require a million token context window, go right to Gemini. We do have a number of use cases with that. So it's just different models have their different strengths right now is what I find.

So you do switch between models and is it true that Claude's the best for coding?

So far, I mean, right now as of November 5th, 2025, that's correct. You know, Claude Sonnet is the strongest model out there for most of the coding benchmarks.

And how do you augment your human coders with AI? Because that's also something when it comes to product development that everybody has to keep in mind now. You just can't code manually and hope to catch up and throw endless resources at a task.

Yeah. I mean, everyone on the CoFrame team, to my knowledge, uses AI. Quite a bit in their coding, all the engineers at least. And it's immensely increased our productivity ever since the very first, you know, line of code that I wrote for Coldframe. Like everything's been AI infused in a sense. We've had to get good at understanding where the limitations are and what the drawbacks are, but net-net, it's been incredibly productive and value-additive for us to use that. To our advantage.

And a lot of people are concerned about some of the technical debt and legacy code that working with AI would generate. And you have a few years under your belt now in terms of using AI-augmented code. What's your experience? Do you find your team and engineers having to go back over old code to rewrite it because the first iterations with previous models were weaker? What's your experience?

Yeah, I mean, when I say limitations, I mean just that. That's a big one. Tech debt and not understanding the code that was written and spending more time reading and understanding it than writing it. So those are limitations. You just have to get good at wielding the tools ultimately. It's like having a steak knife. You're not going to use the blunt side of a steak knife to cut a steak. You're going to use the sharp side. You just got to use it properly.

Good analogy there, Josh. Now I had a question for you, but you already answered it to me in one way or another, so I'm gonna spell it out because I wanted to ask, obviously health tech with Access Bell, NFT with Autograph, and now AI with Coldframe are totally different industries, and I wanted to know what was the common thread between them, but I think you answered that to me. You told me you are a builder and a tinkerer, and it was always building on whatever's new and whatever's next. Correct me if I'm wrong.

Yeah, that's definitely a factor here. When there is a new emergence of technology and new, new possibilities, you tend to have like the most ideas make themselves the most apparent to you. As you get really deep into an industry, you start to see it's a fractal and there's lots of like sub-niches and things to go into that takes a lot of experience, a lot of time. And the only industry where I've gotten that deep is in the AI industry so far, spending 10 years or so in it, like doing research first and then creating a company in it and now creating another company, Coframe, in it. And so I just haven't had that experience to go super deep. Like I wouldn't have taken the approach of what we're doing with Coframe with conversion optimization and marketing optimization from deep understanding in marketing because I just didn't have that. I've had to kind of learn it on the job essentially, but, and I've brought people in that are incredible at it, but so, but that That's the other way to get inspiration for ideas to build. So that hasn't been my story so far, except for in, I guess, AI. So, but the other common thread here is I generally have been very, very, I'm very interested and I guess like have always been very interested in the interaction between humans and the digital world, human-computer interaction. In fact, I taught one class while I was an undergrad at Stanford and that class was cross-listed under AI and HCI, which is human-computer interactions. And so the combination the two is actually even more interesting to me. AccessBell emerged because there was a better way for people to interact with technology to, to actually make use of it. You know, and there's certain constraints that they had that we talked about. Autograph emerged because we need to facilitate this better human interaction with this new technology, which was crypto at the time, and creating a great user experience for that was paramount. CoFrame exists to generally create better human-computer interaction and customer experiences for every business out there and have that self-improve over time. So I'd say that's another like really common thread with the things that I build is connecting those dots and, and, and, um, uh, basically creating better experiences for people in the technology.

Because sometimes new technologies cannot be easily accessible to humans. And that's why in a way ChatGPT changed the name of the game because we've had AI before. Just not in a box where you can just chat with it. That's right.

That's a great analogy there.

But speaking of building, Josh, you've mentioned sleeping in your car while building, and I know in the Bay Area and Silicon Valley, that's not as strange as other people around the world would find it. But my question is, do you think founders should suffer for their companies or were you just surviving?

I don't think, yeah, I mean, I don't think of it as suffering per se. Like, and it certainly also wasn't for survival. I don't know. It's funny that I don't know exactly where you saw that story, but it's true. I was in, I guess I started doing that. I tried it once basically after we just did a Series B for my last company Autograph. And there was like this just general sense I had of like it was easy to feel less grounded and just like, things are crazy right now. I gotta like center again and like get back to the roots. And I had seen this, I can't remember if I'd seen something or like read something about another entrepreneur's journey. And part of that was like that person had like spent some time period sleeping in their car.

And I was like, I want to try this. Well, even Elon says he has done it in his tiny house. So Yeah, yeah. To get close to the action or sleep on the factory floor. Sure. Yeah.

I mean, I love, I just, I do love the rawness of that and the simplicity of it and it's beautiful. But yeah, I just like wanted to try it and I loved the way that I felt. I just like came out just emerging, like feeling like, oh man, I'm just gonna, like nothing can kill me. Like I could just keep going and I'm unstoppable and it feels very raw and just like, uh, so yeah, I just like, I guess kept doing that for like 3 years, essentially. 3 years is not 3 months or 3, 3 weeks.

3 weeks, Josh, I think it's quite a prolonged period of time. So what's your reflection on that? Because you said you didn't see it as suffering, but many other people would.

Like, to be honest, it's not that. It's like, I have a pretty nice car, you know, it stays on all night, you know.

Obviously nobody says it's not a nice car or a comfortable ride, but it's very different from, you know, the different studio.

But like, it's, it's comfortable, like the seat, you know. There's a bunch of benefits. The sun streams in the morning. It's nice to wake up to and like feel— I have the climate control on all night. It's like, it's fine. So I don't want to paint this as a picture as like, this is, it was like so gritty and tough and it's gritty in a sense. It's like maybe more so than in a normal bed. And I still had the, you know, normal apartment and bed and stuff. So whenever, when I was sick, I would just use that or like, you know, other occasions. But, but yeah, I don't know. Like it's, it's more about, it wasn't, it wasn't suffering, so I don't want to paint it like that, but The vibe was certainly that spirit, I think was certainly there in terms of like, this is, you know, this is baseline. You could take everything away from Egis, you could even take the car away from me and that same spirit would still be there. It was like, I'll just sleep on the ground. Fine. Like, let's keep going. I just, you know, I like that part of it.

And when you do this type of sacrifices for growing your business, how do you manage to expect less from others.

Yeah. So what you're saying is like, you know, if I'm putting all this in and sacrificing quote unquote all this, like, you know, others better be there with me. I don't like strictly speaking expect others to work the same amount I do or have the same lifestyle that I have. And to be clear, I'm also not sleeping in the car right now. I live in San Francisco now and I would, that would be very dangerous, unfortunately. The city sucks in that regard.

The world has changed. The world has changed.

Yep. Up until I guess like 6 months ago I was because I was living elsewhere. But yeah, I don't expect others to have that same lifestyle, but the people that I tend to gravitate to and gravitate to me and we get along really well are people that just love what they do and naturally gonna build a lot. And that's why we've assembled, I would say something I'm very proud of is like an incredible team, like an amazing team of people. And they're all people that, you know, it fits quite well. So that, yeah, it tends to work itself out essentially is my answer there.

So you looked for people who were willing to go to just as far and deep as you in order to help you build next best thing?

Yeah, people that I can, I'd say people that, that I really admire and respect and that are also very committed to doing something meaningful.

And Josh, you also said, especially as we're talking about the Valley now, talent concentration matters, especially when we're talking about AI and a lot of the innovation coming from that area. So is that a merit argument or an access argument because you have those people near and close to you? Do you work with people remotely from other parts of the world, or do you still look for your key hires in the Valley based on proximity?

We make rare exceptions to having remote people when that's truly people who have both extraordinary ability and also extraordinary culture fit with us and also extenuating circumstances. So we have some of those exceptions for sure, but mostly we operate, you know, in person. We look for those key hires here.

And I'm guessing you've done that ever since Access Bell, which was after the pandemic. You've worked in person and that helped you push the envelope further than you would with a remote team. Have you even fancied the idea of remote work? But when you are in your glamping van near the office and putting all of that, you wouldn't really expect somebody to log in from Bali, would you?

Yeah, not really. I mean, well, actually for my first, for my second company, Autograph, we started in person and then we were under an immense amount of pressure to grow very quickly. And so we started relaxing our requirements of being in person. So we hired a remote team and then started being hybrid. And then ultimately that deteriorated the culture quite a bit and the velocity and the pace and the energy. And that was actually, that reinforced the idea of being in person for me. So that was a contributing factor to how we run things here.

Now, Josh, as we're getting a bit more introspective, I'm curious to find out from you, is there a part of your identity as a CEO, founder, that you're currently rebuilding, that you've learned from what you've done before and you're doing it differently this time around?

Yeah, I mean, everything is building on itself, I would say. There's a lot of aspects. One that I'm focused on quite a bit right now is trying to attach and get very close to the actual value as quickly as possible instead of what I perceive like the beauty or value to be myself, like really just trying to get that connection point to the external world, the market validation, use cases, just like getting really tactical. If you go too far that direction, that'll actually limit you and you'll end up with, you know, a smaller opportunity most likely, or smaller company. So I think it's important to have, be a little bit of a dreamer. But I think I had too much of that at one point. Now I need to like, I've had to have, you know, focused a bit more on that. And that's been good. It's been a good learning for me.

Speaking of learnings then, Josh, if you had one month to mentor your younger self when you just finished college, what would you train?

Him. I think if I had just a month to mentor my previous self, I would focus on a really specific problem and I would say, hey, like, let's ship this out and get real signal from the world on it instead of like building, building, building, building, building without getting external validation or feedback and just get that taste of what it is like to provide value to someone to the point to where they're like willing to pay for it. Which is a good marker of that, or some other attribute, but just like keeping yourself really honest with that, cuz it can be so easy to build something that is, this is especially problematic in the world of research, academic research, is that you'll build this like structure and it'll be like very elegant and theoretical and so on. It's actually not real in a sense.

So, and what's the best business advice you've had from any of your mentors? That's a good question.

One of the ones that's really ringing true is there's a guy I looked up to quite a bit and he, a piece of advice that he imparted to me was some people talk about life as being either a race, you know, or a marathon or a sprint, right? And it's like, you'll say it's a marathon, not a sprint. And I think that's partially true. It's only half the story. Life is actually a lot more like a bicycle race, especially when you're a company leader. And what that means is that there are certain stretches where you need to be in first gear, just pedaling furiously and really gunning for it. And then when the opportunities present themselves, you need to get into a higher gear and strategically incorporate and use leverage. And that leverage can be in the form of technology, can be the form of people, your team, and sometimes you just need to like cruise and like let yourself recharge a bit. So it's really just an optimization problem for that entire bike race, which is ultimately much longer than a marathon.

Marathon is a good term, which leads us very well into what is usually my final question for my Millennial Masters guests, because I'm curious to know, Josh, what are the sacrifices you've made to be where you are today? Hmm. Besides what we talked about with the accommodation situation.

Yeah. I mean, I guess to imply that I've made a sacrifice would be, I would see it as ungrateful and that's not the case. I think any alternative career path for me at least would've been misaligned with what I feel my purpose is. And so I actually don't think I've made any real sacrifice in that sense. I'm just very grateful to be here.

That's a sweet answer, but everybody has told me that they have had to pass on social occasions, family occasions, sometimes relationships, lovers in order to get there.

I mean, that's all very true for me as well. Like I don't have what other people might think of as a work-life balance. I've forgone many, like, you know, relationships for this. I am not I'm terrible at keeping up with friends, terrible at keeping up with family and so on. Well, I do actually do a decent job keeping up with family. You know, foregoing job security at like, you know, some, you know, doing, making 7 figures doing a quant trading job. Like if you're completely maximizing for money at the early, early stage, like sure. I don't really, not to say I don't care about all that, but like, it's just less important to me than what I find to be the most important thing, which is building meaningful things. So there's sacrifices that sometimes, if sometimes it's tough for sure, I guess when I think about it, but it's still at the end of the day, it's like still aligned with my purpose and what I feel is meaningful.

You mentioned balance, Josh, quickly. Are there any routines or things that you do to keep you focused and productive when things go crazy?

Yeah, there actually are a couple that I've actually discovered in the last 12 months. That have been really meaningful for me. One is every so often when I need to think and really reflect deeply on something and process, which should happen, everyone should have this, you know, points in their life, I go and do a flotation tank session. So a float tank is kind of a, it's a, it's a pod where you have very, very high salt concentration water. Is it that zero gravity one where you—

Yeah.

Feel zero gravity, no light, no sound. You're just floating there. It feels like, you know, same temperature as everything, the air, the water, your body is the same temperature. So it feels like you're kind of just floating there in space in there for like an hour or 2 hours. Just thinking really deeply and have, you know, a subconscious state, like you're able to almost like lucid dreaming. So very unblocked in terms of the thinking processes. That's one. Another one has been recently going on swims in the ocean in the, it's very cold, you know, 50-degree, 50, 60-degree water. But that like really amazingly, like just helps your, your mood out a ton. Like it, it gives you amazing energy.

That's instead of the traditional ice bath.

Yeah. Well, yeah, you still have to shower and stuff, but yeah, it's like, it's like a more intense version of a cold shower or a cold plunge essentially. Another one is fasting that really helps to focus your mind. I'm actually like 36 hours in a fast right now or so.

Okay, so you're going over 24 hours or over the regular 16? Yeah, yeah.

Usually, I mean, when I go for my long ones, it's 4 to 5 days usually. I don't know how long I'll go for this one, but, but yeah, that, that really helps to clear your mind. It, it might deplete your energy a bit. I definitely feel a little bit lower, lower energy right now, but my mind is very clear.

Fair enough. And do, how do you help yourself from binging after your fast? I don't. That's the hard part with me.

I just eat a lot. You're not supposed to, but I do it anyway. I'll have like 4 entrees at a restaurant.

Okay. So you, so you do give yourself some freedom as well after, after all of this. I think that's a good point to, to close, to wrap up this conversation. You're, you're not too strict on yourself. You will do all of that, but then you'll indulge and have a little bit of fun as well. Yeah. You'll need to feel alive from time to time. Yeah, that's right. Exactly. Amazing. Thank you so much, Josh, for coming on and sharing your story and your lessons on Millennial Masters. And I can't wait to see what you're going to build next.

Thank you for having me, Daniel Ionescu.

Thanks for tuning in all the way to the end of this episode. If you enjoyed it, please share it with someone you think would get value from it. And don't forget to subscribe on YouTube Spotify or Apple Podcasts so you never miss an episode. For the best way to stay in the loop, head to millennialmasters.net. That's where you'll get all the latest features and new episodes straight to your inbox every week. Catch you next week.