Building in Public as a Solo Founder in the Age of AI

0:00You can't think of the design process as like a one and done. You have to see it as different phases. And the first phase is just inspiration. What are the things that I'm inspired by and how can I build a system around that?

0:13Matt Paige:Welcome to the Talking AI Podcast, where we talk AI with both experts in the field and early adopters. I'm your host, Matt Paige, and we're here to demystify AI for you so you can get some value from it. Let's talk some AI. We type because that's what computers required from us, but it was never natural. It was just the way we always did it. Humans are wired to talk. We speak five times faster than we type, and AIs finally made voice good enough that you don't have to go back and fix everything. And this is driving a massive shift where people are moving from typing to talking as they default with their machines.

0:49Matt Paige:And once you make that switch, typing just feels broken. There's no better way to put it. But today I got Thomas Schlossmacher, who is one of the best individuals to talk about this shift because he's building Resonant, a voice dictation tool for Mac that runs entirely on your device and nothing ever goes to the cloud. And dictation is just the starting point for what he's building. But welcome to the show, Thomas. Great to be here. Thanks for having me. Yeah, this is going to be a great conversation. I've been wanting to talk about this because the podcast name is Talking AI. So what's more fitting than talking about talking AI on talking AI.

1:23Matt Paige:But to be honest, I use voice dictation here and there, but recently I'd say, I don't know, maybe three months ago or so, I just completely made the shift. And if I'm ever in a quiet space, a Starbucks or something like that, I literally, I get angry if I have to type. It's like a visceral reaction to typing now. But why do you think talking to our machines has taken off in the way it has recently? What's driving this change? I think it comes down to people kind of getting frustrated with typing. So I notice whenever I use any kind of AI application, I usually want to be fairly precise about what I'm actually putting into the chat or telling to a specific agent.

2:09And I have to make sure that I include images, I put in all these minor details. and if you think about how quickly you can speak, it's just so much easier. And your train of thought runs pretty effortlessly when you speak. But when you're typing, you're actually utilizing an entirely different part of your brain to kind of synthesize your thoughts and then bring them into chat. So it's really easy to just speak to the computer and have this kind of natural interaction where you can put so much context and so much messaging in just a few short seconds.

2:45Matt Paige:it's so it's interesting you mentioned that because like you can feel the difference like if i'm typing something versus the output of me talking to whatever tool it is i i literally think about what i'm saying differently and the output's different and for me i don't know if this was for you but like i had to make that shift of like being okay with the messiness that came out of me just having like a flow of conversation, but it just, it's more nuanced though, I think is one thing, but the AI models are so good nowadays. Forget about dictation, just like being able to take in your natural language through a chat and have a conversation with it.

3:27Matt Paige:But like, what, what was your, do you like remember a moment where you actually made that shift to talking versus typing? Like, did you have any kind of that weird adoption pain, I guess, as others do? I think I'm similar to you. So, like, I do a lot of dictation on my phone, and I find it, I don't do voice messages, but I'll use the Apple dictation. So I've been very comfortable just expressing myself through my voice. And when I'm in a public place, I have to do the voice thing and, like, click the button and kind of speak to my computer. I feel a bit weird about it, but I've been doing it for quite a long time.

4:08And I don't know where I picked it up. Like I'd seen my doctor and my dad, friends do it. But actually talking to my computer and utilizing different tools like WhisperFlow, Voice from AquaVoice, Willow, like I tried them all out. And it must have been maybe two years ago, but I don't know if there was a specific moment that I recall. Yeah.

4:33Matt Paige:Just a gradual transition. So that's actually a good transition point. Like you mentioned Whisperflow and several other tools, like Whisperflow is taking off like wildfire. So like the obvious question is, okay, why build resonance when it's a voice dictation tool? Like what was your thought process in determining I want to build, you know, voice dictation and how are you thinking about it, I guess, differently than Whisperflow? Because in today's age, anybody can build anything. It's just an interesting topic, I think, when getting into deciding what to build and how do you approach different...

5:12Matt Paige:Quick break in the pod. Our State of AI 2026 report just dropped, and it breaks down what actually is changing in AI, what's hype, and what leaders need to be paying attention to this year. You can grab it right now on our show notes or at hatchworks.com. and whatnot? Yeah, I have a few answers. They have an incredible product. I have so much more respect for their team and what they've built after I tried to do everything locally on device. Originally, my thinking was, I've used a few of these tools and I've built a couple of things. I know and I feel that there's this point coming where voice and agents will kind of come together and you have like customer support agents, you have interview agents.

6:00There's all these different types and I felt like I want something that's private because I know that these companies are also training models based on users who are using their applications. And I've given a lot of support to the Proton organization in Switzerland because I like that they use this zero data retention policy. And I wanted to do the same thing, but for a voice tool, because I use it in my chat apps. I use it in my messages, my emails, everywhere. And I also don't want to be trained on, and I want to give people the same opportunity. So the thinking was, how can I make something that's nice and for me and also make something that is privacy-focused so that people who may have, I don't know, sensitive client information or who have context data on screen that Apple can run OCR on that doesn't get uploaded to Whisperflow or one of these other organ-adjacent servers.

7:02Like there's a lot of sensitivity around it. And I think more and more companies will, or more and more people will try and do more on device as the hardware improves. So that was kind of the, I guess, the thinking. And in terms of like people who can build anything, I've seen that like maybe half a dozen of these, like I just created an alternative to Whisperflow and then people upload an open source project and then never touch it again.

7:32Matt Paige:Yeah. Yeah. That was kind of the early thinking. And now I'm just trying to make a really nice experience with the people who use it. Yeah. And it's interesting. Like I didn't, I was prepping for this podcast and I looked at that on Whisperflow, um, which I guess should have been obvious, but like there, everything's done through the cloud, right? Even though it's a desktop application. So it is this interesting, like nuance around privacy. And I feel like there's a lot of people that either don't care about that now or they're hyperventilating of it, but I think it's going to become more and more important as time goes on and the implications of that start to rear their head.

8:11Matt Paige:But I think the other part of it too is running LLMs locally is just going to get more and more efficient. I think it was at Quinn that just came out recently that's running a model on device and it's just highly performant. So I think that is a future way of working in a sense as well where LLMs are running locally. I'm curious your thoughts there on like the progression toward that. And is there a tipping point or are we already there to where running LLMs locally are good enough? So good question. I've been wrestling with this the last maybe three weeks. So I have a M4 MacBook Air and it's good enough for like some small models that have been quantized.

9:01so like i can run them but the tokens per second are slow and because the models are smaller so they can run on this device they're not as high quality in terms of their answers so i recognize that for the the actual pipeline like if you're doing an application like resonant you have to take in the text which means having a speech-to-text model then you have to format it to make it look nice and kind of be adapted towards you. And then you need to post-process it. So you actually need to run it through a language model to say, hey, this is going to be an email, use this particular tone. And to do all of that in a compressed time period under one second and be highly accurate, unless you have a great setup, it's going to be hard to do.

9:56And that's why doing it kind of off device and doing it on a server is more effective but then there's inference costs which you have to also think about and then your data which you have to also think about so i don't think we're there yet but if you have a good device like an ultra maxed out kind of macbook um or you have a studio or like one of the newer models or even like some gpus on a local wi-fi server you could have an amazing setup and do everything locally with olama or lm studio But to get like the majority of people who have a typical MacBook or who maybe even are on a Windows device, I'm not sure about any of their local language model optimization, but those are completely different scenarios.

10:41So in that case, it might be better to do things on a server. But then you run into the question of companies who use the product and do they want their data leaving? So it's kind of this tricky in-between. but I think the next models of laptops that come out over the next one to two years are going to be great.

11:04Matt Paige:Oh, totally. Yeah, I think that's, it's where it's going. It's just a matter of time, it almost feels like. But on the point of like differentiation and existing products out there, I think so many people, they see something that exists and they're like, oh man, well, that's already out there. I'm just not even going to try. Right? Which just isn't the case anymore. And I think there's a couple elements I'm curious your take. Like, I think the actual act of building that used to be defensible in and of itself, because you needed like deep engineering expertise and that's going away because anybody can build anything.

11:41Matt Paige:Yes. If you're an engineer, you're, you know, more likely to build something that's performant and secure and all that, but the models are getting so good when it comes to building now. what what is the the value point now is there's talk of taste there's talk of brand audience all of these things how are those elevating uh around in this new age that we're like moving into now i think um maybe a couple of years ago you could release something that was like half baked and people would say like oh I see where this is going I'll sign up I'll go in for the ride but now because everybody can create anything and make something look nice using skills or screenshots you really only get like one maybe two chances with an AI product and if you don't win someone over in those like first two interactions they're pretty much gone.

12:44Like people churn so fast and they jump between even models or companies really fast. And I think that makes it difficult for builders. So there has to be like a heavy emphasis on taste, which is one of them that's been pretty common, where you create this, you use this potential that you can now build or reference anything to create these highly unique or individualized creations that make this new experience. So I would say taste is definitely one of the winning factors. Second to that would probably be capability. So recently, Perplexity released this computer product. And if you look at the product, it combined the browser that they have, Comet.

13:32They added voice capabilities so that you can speak back and forth for speech-to-text, text-to-speech. And they created this structure with OpenClaw. So you have this combined modality, which is also something that impresses people. But the number of creations like that are dwindling. And you only have so much before everything is kind of this super mixed modality AI experience, where when that point hits, I don't really know how you stand out. Because everything becomes this computer. So I think it's hard. um you have to really put a lot of effort into refining the product and the visuals and the brand of what you're building and not focus so much on just like hey i built this quick thing i mean that can still work if you're famous right like if you're carpathy or somebody and you just drop a half-baked thing people people see it but it's the that's the innovation that's not just like the product um yeah yeah and that's that's where that audience piece comes in interestingly

14:38Matt Paige:enough, but like, I'm actually curious, like maybe even pull up resonance and even just looking at the homepage of it. Uh, cause you're like, you're a solo founder, you're bootstrapped doing this alone right now, but the, it's just beautifully done. It's just mind blowing that, you know, somebody can create this by themselves in the amount of time that you did. I'd be curious, What was your approach to actually thinking through the brand, the design, the experience, which I think is such an important part with a product like this? Sure, yeah, I'm happy to show you. So this is the site. I actually can't take full credit from this.

15:21So I built it all myself, yes. But artists steal is kind of the famous line. and I really respect the design taste from the cursor team. And if you look at the site, it's very similar. So they have this light beige background. They created these, or I think they took these public works for these references of artistic works, which are very different to these hyper-digital experiences, but they focused on this cleanliness. And I thought, okay, from a website positioning standpoint, How can I create something equally as clean that also allows people to kind of immerse themselves and think about, oh, this is what it's like using the product.

16:07So the instant thing that you show is kind of that first touch. And then this video, I created a logo, like something that sits on screen. I'm not sure if you can hear it, but I'll play it.

16:24Matt Paige:the audio is not coming through but it has this cool like ambient vibe yeah to it the and this is essentially the uh resonance like you know talking you're engaging with it and it kind of has that ambient vibe to it's very very yeah it's cool i guess is the way to describe it it's uh it's like this startup noise that i synced with the animation and a lot of the design let me just see how I can switch. I can take you briefly into what that looks like. Um, so. And while you're doing that, what's you mentioned the artist steel type of thing. So I'm, my new hobby is just building things on the side.

17:01Matt Paige:Like I used to do other things. Now it's just that. And I literally was building something the other day and I was like, I'm going to go take Thomas's thing and say, Hey, this is a good reference for the design I want. And I literally took a full screenshot of your, of resonant and fed that in as context. I love that. It's a great, a great point, actually, when you're thinking about, you know, what elements do you like of something when you're building something else? And that's like great context when you're building. It's become a lot like Legos. And I've put a lot of effort into refining my own skills, both in terms of like skills like MD files, but also thinking of taste as a muscle that you can train.

17:44And this resonant app was originally called Charlie. And I saw like, um, different pieces of components that people had made on Pinterest. And I started to just bundle them into a board so that I could kind of mood board the, um, the whole thing and then pick out the pieces that I really liked. And I mean, these are all done by like professional designers who are refining their craft for maybe a decade. So you can see some of the vibe behind what I wanted and then what I built. I would say that whenever you're building something, if you start with a design system, you can just fly through. But if you've already got something and you're trying to like reorient it back into some new direction, it's not

18:33Matt Paige:going to go well so i would say how do you think about building that design system up front because it's not done in the traditional sense of how we used to build design systems in a lot of ways because you're almost having whatever tool you're using you know cursor placo lovable you're giving it context to help build out the design system how do you like tactically do that in this new like age of building sure um so i have three primary points of inspiration um the first is on twitter or x the second is on pinterest and then the third is based on what i see released into the wild on like hacker news or between friends or whatever i see like on on the internet it.

19:27And I, I learned this from my, my former co-founder. You can't think of the design process as like a one and done. You have to see it as different phases. And the first phase is just inspiration. What are the things that I'm inspired by and how can I build a system around that? And building that system, I use ShadCN primarily for the, um, the actual design components. and then I work on each one until it like feels right and then I'll do a pass with a skill that I made that looks at things like visual harmony, white space, using an eight point grid system, type scaling. So like a lot of the standards that a designer might use and I'll ask it like when I hire someone through an AI chat, I'll ask the AI chat like behave like you work for Apple and Vercel, what are some of the frameworks, methodologies, and like techniques that you would use.

20:30And I want you to reason through them and apply them to this set of components or make a design library or, you know, along those lines. And by pulling the LLM into this thinking of, oh, I have this role. Oh, I have this like world knowledge of techniques and different frameworks I'm going to use, that makes the output so much more effective. And I'm actually just amazed at some of the results that I get. So doing it like that has been really helpful. And then I follow a lot of designers. So I categorize on Twitter into bookmarks. I have a folder of design items that I really like. And I'm just constantly logging them, trying to train this ability to see things and then just recreate them.

21:20and that's kind of how it's been.

21:21Matt Paige:And then you naturally start to figure out your own patterns and your own taste through all of this absorbing of things, right? But it's interesting. So you're leveraging skills in a sense and maybe there's something going in a markdown file somewhere that has context. So as it continues to build, it has this reference of, okay, anything that I build needs to follow this design system or patterns that we've, the human and the AI have aligned on effectively. Yeah, I just, I pointed back. So anytime I have a bit of a hiccup or I feel like something's not working, I'll just say reference the design log or the design library that we have in this code base.

21:59Or I'll pull out the screenshots. Yeah.

22:04Matt Paige:I've never taken so many screenshots in my life as I have in the past two years. But, okay, so you're, you know, solo founder builder. And correct me if I'm wrong, you're not like an engineer by trade. right but having known you you're like decently technical like what has been your journey from you know not being a software developer to now building entire products like how how have you crafted that muscle and a lot of it i think is just system thinking and back to like design thinking from like a technical perspective how have you built that muscle over time so i feel like a lot of people see people building but they they feel like the gap's just too wide to get there and it's just not anymore frankly do you mean like my my journey into kind of the vibe coding agent coding world or just like over time totally yeah okay yeah like vibe coding or you know agentic development if we want to call it whatever term you want to call it but like you're you're building functioning systems and apps and web apps and things like that so there's an element of having to know like how systems work, what an API is, what a function is, things like that.

23:18Matt Paige:But then also the AI and agent covers a lot of that for you too. So what's that progression been learning to work with agents in this new kind of agentic development approach? Um, so I was always, um, interested in computers and startups and, um, websites specifically since I was 11. I used to tell this story to investors when they ask about your background. And I would tell them that my mother called the cops because she couldn't find me when I was 11 because I was in her office on her computer, hacking games. And I was just like a script kitty. I was just using people's bypasses and their creations.

24:01But I'd always been comfortable using tools that were digital and like like messing around on my computer and i built and sold websites on and off as like a side thing from 14 until 22 maybe and um i'd always been comfortable doing like integrations and like all of that just felt very natural to me i had tried to teach myself to code like it must have been like six or seven times i bought courses i went and did these like progression programs and i just felt like the way it was taught was so it just didn't click for me and whenever i just sat down and did something myself and like treated it like some kind of play it tended to work and then um i think five years ago now um And I switched out of like a tech sales job into a finance job where I worked for a family office on their investment team.

25:05And I wanted to do a lot of like digital pushes and like help build them into a platform. And that wasn't their focus. And then I just got frustrated at some point and was like, I'm not going to kind of achieve what I want here. I might as well go do it myself. And I learned from a lot of the clients of that group, a multifamily office, like what that looks like. And then I went to Berlin. And if you're ever a little lost, don't go to Berlin. That's kind of a challenging place to be. And I met people for the first time through this program called Entrepreneur First, who were very kind of gifted technically, or they came from some kind of domain that was very specialized.

25:52So you had this mix of like operators and builders. and you do a little bit of like co-founder dating you try and find your your match in a way and then they fund you with a bit of money to build out a a company based on that idea and around that time they told us don't do anything with ai this is a fad and really what year was this this is 2022 going into 2023 right upon the release of gbt so right about chat gbt yeah and And everybody in this kind of cohort was like, these guys are crazy. Like we see the opportunity immediately. And four of the people who were in my program sold companies, one of them for 90 million.

26:41Another one, his run rate is like 4 million a year. But like those people all did super well. And there was so much knowledge sharing. So for me, it was this place where I could meet people who either became very technical or already had these skills and kind of immerse myself more into that world. And then I just saw the release of different tools. And the reason I started making videos was like, oh, I've learned to instruct a language model. I think it was like 3.5 sonnet, or I can't remember the name, but it wasn't very good. but it was building basic applications and I had like a five minute video it's one of my first few posts like with a piece of paper and a pen this is how you build anything using AI and like I made a few of those and they just really took off and worked and the one thing I really wanted to focus on was how do I teach myself more about like back-end engineering or like more kind of advanced topics as someone who isn't necessarily adept there.

27:51Like I've been very comfortable on front end and like libraries and kind of experimenting. So that was a new world. And I just told as like system prompts, teach me about the process, why? And then more and more, I just kind of picked it up and tried new things. And it's fun because you can just tell an agent to do anything, read all of the thinking how it works you have to grab the keys like it becomes very intuitive once you have this guidance um so yeah that was kind of the um beginning until now and then i just got more into the design side because i felt like i understood what people might want through these like digital interactions and then like the back end stuff i'm not really that great I rely a lot on learning from open source tooling, videos of people who break concepts down.