DN #27: Agents as Software, Small Business AI & Building in Public (w/ Julian Fitzpatrick)

Niklas:So let's just continue. What was my last question? Hi and a huge welcome to you, my lovely listener. So glad you're here. Today you're joining me for chat with Julian. Julian is building build with pixel NAI layer of sorts. How does it work? What's it is

Julian Fitzpatrick:about the audience ⁓ having having a grand grand audience. Yeah, I mean like ⁓ the way I think it I think it's great to grow. ⁓ for me in particular it's more of just having people on the same the same page and they're interested in what I'm bringing in terms of value and it gives them value. You know

Niklas:⁓ yeah, yeah, okay, right, yeah, perfect. Yeah.

Julian Fitzpatrick:Okay, so essentially it's ⁓ an AI agent platform and essentially I'm going after small businesses and setting up an intelligence layer within their business. And while I'm doing that, I'm teaching my agent essentially my infrastructure and the tools that I'm building for these companies to give give it to consumers as well. Everyone wants to sell sell their product, you know, and I think with that especially this day age, you you have to provide value to where they wanna use your your your product. Now I don't think you honestly have to get to the level of like Musk or Alex Carp or those people. I think it depends on your your niche and who you're going for. Like and you know, you you have your So essentially the platform, or if we want to go a little bit backwards and where, you know, I come from, like I come from a VFX background, and that's been essentially my career for the last like 17 years. And, you know, two or three years ago was sort of like the start of a or like the onset of AI with ⁓ you know, the chat GPT three and all that. outside like what makes you happy but I think if you like if the thing that I've learned is you know it's very hard when you're starting off to hit the like consumer market so it's very hard to build a product and give it to a mass amount of people. What I've been learning is it's better to go after local and if you think about it And it sort of opened my eyes on the possibilities of what could come in the future. So I started building and learning essentially like how to code, ⁓ because then you didn't really have your own capability to build your own agents. ⁓ so I was learning how to build websites, ⁓ using N-AN, all that good stuff. And I started building a platform that was essentially for image to video generation, like. In terms of scaling, is you really want to start with like almost family and friends. You ask everyone in that group, okay, does this work for your like does the product help you or whatever? Then you expand. And so then it's like your local businesses hitting them up, going and meeting them, like pitching your idea, all that kind of stuff. And then as you grow, then that's when you like when you're going from Creating images, going from image to video, first slip frame, last frame. And I was building all this architecture, and it sort of dawned on me like, ⁓ I need to sort of sell this knowledge, sell this, this, ⁓ what I've built. ⁓ because you know, with with AI, it's it's very easy now to be very creative. let's say a local town to like a bigger city, then you start going out to other cities, that's when you I think you can start going after consumers because then you have an idea that has market fit. And then while you're doing that, if you're growing your audience online, then you can show that. You can show what you're building. That goes with like the building in public is you're you're able to to show the proof of what you're building. On how you tie things together in terms of your infrastructure, your API calls, what tools you give it. And so essentially, Build with Pixel has become this sort of home where I'm working with local small businesses to sort of get them up to speed, what an agent and automation can bring to their business and developing that as well as And then they're more keen to to listen and you build that trust by showing that. Because a lot of a big thing is ⁓ just having that trust between either you and a c ⁓ consumer or you and a brand. my platform where you can speak to my agent and learn how I've built the infrastructure and then tools that I create I'm gonna be giving to the public to where they can use. ⁓ but that's that's essentially what it is. ⁓ yeah.

Niklas:Yeah, so we touch different topics, right? So I would fully agree first of all, that consumer is incredibly hard. So ⁓ the reason is you just have to, consumers turn very fast. It's really hard to build a sticky consumer product. It's a lot easier to build a B2B product. Yeah. We are looking at I think a really interesting moment in time also when we are talking at because we had we just had this, I would say, huge with lower churn and get it profitable. generally believe that it's a lot, lot easier because consumers don't really like to pay for stuff. Usually they want it free. And also there are a few companies in the market like Google Facebook that have a huge distribution mode that will aggressively move into fields that have viral growth. So you're always, or you're relatively fast competing with them if you're looking at a really big upset around closed models, fable tape being taking off the market. We are now seeing really, really good open weight models coming from China. I think the latest type was around GLM five point two, which seems at least on benchmarks to be better than Sonet five, which was released today, and seems to be a little bit below Opus four point eight. And that's also what I'm hearing. So what what I feel and that's been the discussion, I think there's been a really big market and B2B there are just a ton of companies that you can build for and especially now with the cost of building getting so low, you can really target a smaller niche, still build some profitable software, not that the huge like venture scale market and be really, really successful in that. So I think that's easier. Still, you need to build some kind of audience, You're building, you're selling to people. S kind of controversial interview with ⁓ Alex Karp probably in the last days, ⁓ that was around the idea that you have the application layer, you have the LLM layer, you have the compute layer, and kind of the c the LLM layer now gets ⁓ or the AI layer now gets commoditized of sorts, right? So there will obviously be use cases for these frontier models. But it seems we seem to be at a point where the open weight models actually are getting good enough ⁓ for the ma for most use cases. And we are also seeing like on Open Router, which is a platform where you can use different models, that you can ⁓ that the open weight models are actually outperforming the classical closed models. And I think that's a huge shift because if you go back one year, there was a huge gap in between like open weight models. You have to know whom you're selling to, who your customer is, what you're building. think these things still hold true. That hasn't changed too much. Who is your, like, if you look at your ICP, what's your target range? What are you looking at? What are the revenue ranges when you look at companies? Why?

Julian Fitzpatrick:mean honestly I don't I don't really look at it like that. I'm because I'm small and nimble, I just I go I go for it all. So essentially because this is so because I know what I'm building is impactful to any business.

Niklas:And especially China and the US closed models. So I'm really interested in if you look at it with what you're building, what what are the use cases you're looking at? Where does your platform come in and how does it plug into all these different technologies?

Julian Fitzpatrick:So as of right now, like I like agree that gap is getting smaller and smaller, and it's making it to where you know businesses and users can have their own closed system that can perform just like Cloud Code, ChatGPT, all you know, all these top-tier, frontier models. Now, with my system, I of course always use the top, top frontier model because I'm trying to I essentially I want to hit everything. I want because in the beginning you're wanting to get testimonials, you're wanting to get use cases, and the way I've built my system is that it can learn from the stuff that it built, so then I can propagate that very easily to a new system for a different company. So when I'm tackling I and my mindset, like a lot of people know progress my tool base and all that quickly. But whenever I propose to ⁓ like businesses and such is you know, I open up the table to where okay, we can hook this up with like a VPS machine in the cloud, all that type of infrastructure database, but you can also go local. You can keep everything on premise. And I you know, ⁓ I can't remember who said it, but ⁓ all of that lingo when it comes to ⁓ going after businesses and all that that I that's not that's not me and it that's not where I think I need to hold my attention. So and this is a it's an interesting thing. I don't know if you've ever watched ⁓ I think it was Kevin O'Leary and he was talking about Steve Jobs and he was saying about how Steve Jobs like he was amazing at I think it was Shamath. ⁓ I'm probably pronouncing his name improperly. But he was saying like how you know how in the past we went from everything being on premise, like on everything inside the building, ⁓ in terms of hardware and all that, and then we sort of gone to cloud and right now there's a you know, everyone's using these top tier, frontier models, everything's cloud but cloud based. But what we're gonna start to see is Focusing on the signal. And the signal is like the top three things that you need to finish to move forward in in your business or w whatever. And everything else is noise. And to me, that kind of stuff is like noise to me. For me, it's I need to like I I come from a building mindset. So the hardest things for me is selling and There's gonna be a change in those subscription models or pricing tiers, or it could become more expensive. And we're gonna have these businesses transform to have everything local. And these open source ⁓ models that are coming out are making it like capable to where you can produce the same amount of tools, web apps from those local models and have your own infrastructure protected. Going out to companies and actually talking them through it. Now I have experience talking with clients in from VFX, but don't have as much experience with actual like selling, like pure selling. Like that's a whole other beast. Like that that's a that's why there are great salesmen, car salesmen, all all all types of you know salesmen, because that's a a mindset. And so for me and never even touch you know the internet so with in terms of like my platform I'm not local based I'm not at that point yet I don't see a need for for where I'm at it's something that's always on my mind and something that I can build for for businesses but for me it's the the pricing structure for these top tier models is cheap enough to My focus now is okay, how do I take my product, break it down, and sell it to where I can sell without even showcasing it? Like that's that's what I need to get really good at. Because then once I get good at that and really good at that, then I can easily get revenue from any business and then start focusing on verticals and going from there. And then focusing on, you know, Try to outpace everyone else to build the tools and everything that I need, to then where you can set up your own router system. You know, I'm sure you've heard of open router and all that, where you can dictate which model is doing what per agent or workflow. And it's essentially you want to get to that point to where you're setting up. what the revenue is from the businesses I'm tackling, but at this point that's not like my my big thing because if I were to focus on that, if I don't get the sales right, then th none of that matters, if you know what I mean.

Niklas:Yeah, that's actually true. And you also need the access to go back to the other one, right? So whoever you're talking to, who wants to buy, probably you would see as a customer, which I think is also the right look. If we go back to more like product strategy, you're also building like creative tools, largely. If you look at this industry, how it has evolved over the last like 12 months, what do you think?

Julian Fitzpatrick:Agents and ⁓ workflows to utilize certain models that are cheaper, and then using the top-tier models to code and push your platform forward. And then you get to a base where if you document everything properly, then you can go to lower-tier models and they can they can build tools quicker, even being less intelligent because they have all the documentation and context to do it.

Niklas:what has happened, maybe walk us through it and what will happen next.

Julian Fitzpatrick:I mean I honestly I mean short term I honestly think you know agents will take over. If you look if you look at the past in terms of how how these models have progressed to where first like when you're talking to Jet Ch ChatGPT or Claude, you could extract code, but it wasn't able to work within your code base. Then you had like Codex and Claude Code.

Niklas:And and specifically looking at use cases when I now would start ⁓ was built with pixel and aware business, right? So why would I buy it? What is the typical things that that you can help with?

Julian Fitzpatrick:So essentially my use case is it's very generalized right now. I'm trying to attack every industry that I can. And so what I essentially do is whenever I go to a business, I basically tell them, okay, I'm gonna be your your consultant for AI, but what I'm gonna do is I'm gonna set you up with the infrastructure and then I'm going to attack a bottleneck problem. manis these these tools come out to where okay you can you can build within their within their code base their agents and they can build within the ⁓ that world then you got open claw and hermes come out that gives the user base like a capability to build with their own harness and so what you're seeing is you like these within your business. And then I'm gonna tie that with the agent and make it easier and automate that workflow. And then what I do is I go through and I just start hitting all these bottlenecks and problems. So for instance, there's this I I talked to a ⁓ an ⁓ owner of a volleyball academy and their their business is very scattered. So they The the top tier companies, it's shortening in terms of what they're able to give user base and what's coming out open source. And it sort of goes back to what you were talking about before in terms of like local models, the power. And what we're gonna start to see is that businesses are are gonna be able to to ⁓ and users engineer their own harness for their required use case. I talked a an owner of a volleyball academy and their their is very scattered. They post post about having like training clinics, skill practices, and these are on the fly. So ⁓ they'll about it and then they have a month before, let's say there's a tryout. And so between that post and then ⁓ when comes up, they need to output call-to-action banners, posters to social platforms. They post they post about having like training clinics, ⁓ skill practices, and these are on the fly. So they'll post about it and then they have a month before, let's say there's a tryout. And so between that post and then when that comes up, they need to output call-to-action banners, posters to social platforms. So and that and that's like where I see the strategy is is that and you're gonna have a huge amount of jobs where people need to be an a AI operator, orchestrator. You need to know how to to talk to agents in a a ⁓ natural language with very specific And so what I've done for them is basically I've set up this workflow to where they can just put in their intake information and essentially it's at, what it is, the price, where they need to sign up, all that. And then they provide a headshot ⁓ the person they want on the banner, and then a reference CTA. And what it will do is it will create the entire image, to action. And so what I've done for them is basically I've set up this workflow to where they can just put in their intake information and essentially where it's at, what it is, the price, where they need to sign up, all that. And then they provide a headshot of the person they want on the banner, and then a reference CTA. And what it will do is it will create the entire image, call to action. Goals and tasks and understanding how it engineers, and that's gonna be a huge thing coming up. And I think the developing of the harnesses and how they're built for companies will be huge. And so then what happens is software will be removed. You're you're gonna have agents are gonna be what you sell, and that agent for particular. ⁓ the title description and then they're able to post to their socials with the click of a button. ⁓ the title description and then they're able to post to their socials with the click of a button. And so on the back end, you have an agent that is learning from that. So it's able to look at the information, the call to action that's been created, the headshots, and build from that and see what what's being used and categorized and organizing it. Industries. So that that's like where I always tell businesses that's where I come in is that I want to, and that's why I'm hitting a ton of different industries. Because let's say I I work with a business owner within real estate and I go through their whole business and set up an agent to go through every workflow, social, all of that kind of stuff, and I build the framework for that agent for that business. So that's like one I come in, I hit a workflow, but then I grow out that plan and essentially take their infrastructure that is the old school way and add this layer over here and essentially have the agent take over automations, create tools for all these problems, and then they won't be using their old way of ⁓ their old school infrastructure. And then you have a a an agent that knows your business. I can easily then, with that business owner, be like, hey, we developed this agent together. Let's go sell it to other realtors. And then you sell that as a package of like, hey, we can incorporate this agent into your business, tailor it to your needs, and it takes care of basically 90%. So you're gonna have these packages of agents that are gonna be software and they're knows where all the files are at, knows how to build and can keep building on top of that.

Niklas:When you when you look at these companies, do you have like typical use cases? Do you feel something is repetitive, something is creating a lot of value? And which industries do you think are currently easiest to enter for a for an approach like yours?

Julian Fitzpatrick:It's like the next level. They develop their own software for the business and how they operate. And then it'll just if you look if you abstract even further, you'll have an agent. I can have an agent even ⁓ operate those different verticals. So I can have my own agent know how to construct apps per per vertical. So I think it's I think right now, definitely any small business, but I think right now the the ones that I've been most successful with are ⁓ academies that I was just talking about because a lot of those small ⁓ companies that own academies and they have a bunch of different teams and they're trying to get players, they're th they have a lot of constant ⁓ what's the word? Like Like going back to where it's going, I think the that huge gap that's always been between us and open AI and cloud, that's gonna and anthropic, that's gonna r get really small. And if you can tackle small businesses to medium businesses, you could thrive, you know, and then at some point get to an enterprise level. They're constantly updating and posting to socials, and it's something that can be automated quickly ⁓ and and setting up a tool that can redo that process very very fast. Anything with like marketing, because everyone needs socials, like i like everyone knows nowadays, like it's all about retaining ⁓ your

Niklas:If you look at agents, I'm always wondering what's the UI above it, right? So yeah, they work autonomously, but humans want to look at something. So how, how does that work?

Julian Fitzpatrick:So that's a big thing. The human gait, the the approval from the person using it. And that's a an ongoing ongoing thing. So like let's ⁓ if I were to describe how I'm building my agent and my tools, is essentially I when like let right now I'm implementing something into my agent to where ⁓ it can do a CTA. Or gaining the audience and r retaining their attention. And so a lot of these businesses now they don't have a means of posting and ⁓ letting their audience know what is happening. It's all a manual process. So what I've been going for is are these academies, realtors. ⁓ realtors go through the same thing and trying to market their themselves, and it's a whole job. It's a it's a big task. Campaign through my website. So what I do is I give it a goal. Now I've already taught it my infrastructure, how it should check stuff, how it should validate. So I give it a goal, and I essentially say, once you're done with that goal, show me screenshots, show me what you've done, show me the proof of what's built, and then tell me what's next. And that's my like first layer of approval. Then once it's done with that, and I can verify that it ⁓ but I'm also ⁓ working with like an architecture firm, but that that use case is a little bit more complex because they go through a whole different way of like planning ⁓ their floor plans and all that kind of stuff. ⁓ so and right now like AI is very new. So one of the things that I'm built it properly, then I say, okay, now let's ⁓ commit that to like staging so that I can test the tool. So then it goes through, it builds, commits it, pushes it staging, and then that's where I actually go in, I check UI, I you know see if everything functions properly. If it's all good, then I'm like, okay, what are next steps to pushing this to online main production? dealing with is learning which ones are the best to tackle. And so when I prototype a tool for a company, I then have that tool and then I'll reach out to other similar companies and then see if there's a market fit or if that was a one-off. And that's the big thing is like, you know, you can create a tool, but if you don't talk to anyone that could use it, then you don't know if there's truly value within that tool. If there's issues, then I just go through, I screenshot what's wrong, I explain like where something's at that maybe not working or if a function's not working or whatever, and I describe it to the agent, and then it goes through, fixes it, I do a quality check, and then push. And then depending on what you're doing, you should always have that human in the loop. So right now it's me, I'm just going out and I'm creating these tools and then once one sticks, then it's like, okay, I need to propel this one and then I know how to sell it to that vertical ⁓ with different different companies.

Niklas:Think this is a very good approach, right? Because you already know who the customers are. And I remember that fairly recently I've actually ⁓ heard Mark Cuban talk about this topic, quite skeptical about open AI, for example. But what he said is okay, you have millions of businesses who have some benefit.

Julian Fitzpatrick:But once you get to a point where you're doing the same thing over and over, and let's say I did that same thing fifty times, and 98, 99% of the time it got it right, that's when I would be like, okay, that's time for to make it ⁓ automated. I don't need to check it because it's nailing it every time. And of course, you go in and can check whenever you want, but that's when it's

Niklas:if they use these tools like small businesses, exactly what you're you're going after and they don't they need people to help them implementing this kind of software. So I think this the it's very clear that that there's a huge market. What I find really interesting about what you're doing is you decided to build in public, right? And I I this is an a path I would always also choose. ⁓ but I'm very interested why you chose it.

Julian Fitzpatrick:Proof to you, okay, the system's working good, ⁓ good enough to where you can automate it and move on to the next thing. And so that's the that's the goal that you want to get to. And every business they they operate differently. So some things can be, you know, automatic and automated and never looked at again, but there's always gonna be some things that you have to have a human review it, especially if there's like compliance and ⁓ I think it's mainly because, you know, anyone that engineers or builds they they're and you know, there's anomalies, but for the most part we're very to ourselves. So like you build, you build, you build, and then once it's finished, you're like, ⁓ okay, I need to move on and build the next thing. And the whole thing with guidelines because you need to make sure all that's good. But if you look at it overall, if you're able to hit like ninety percent of your workflows and automate all that, that's a huge time saver. And then that business owner or business can look at other areas to where they can forward their growth. me when I started building was I wasn't getting any vision on what I was building. So I had to make the choice, okay, I need to start I need to start putting this out in public and for people to see what I'm building. Whether it's like short little videos of a tool I created or a prototype or whatever it is so that they have vision on my my my knowledge and my skill set. And I think it's a very

Niklas:Yeah, that's really interesting. ⁓ If people now are curious about you and your business, where can they find you? Where can they sign up for Build with Pixel?

Julian Fitzpatrick:⁓ you just go to build with pixel, buildwithpxl.com, sign up. You can talk to my agent there. I'm on ⁓ YouTube, ⁓ it's at Julian underscore builds, Twitter, ⁓ I think it's underscore pixel world. I I I can't remember, but you should be able to find it with a a search Julian Fitzpatrick. And then LinkedIn ⁓ you can find me the same. But if you go to build with pixel, it has my socials at the bottom. ⁓ but it's It's a very strategic way of doing it when you're small. Like, you know, I don't have a ginormous audience. ⁓ of course I want to build one, but I think the only way you can really build that is just to show what you're building and why you're building it. And especially in an age where no one truly knows where AI I mean, we all have our vision of where AI and agents are taking us, but there's gonna be millions of different ways of how these It's really cool because you can talk to my agent and I'm I'm essentially teaching it the products. Well, it's building the products, but also I'm telling it to tell users like they can ask questions about their business and it'll help them guide them through building the infrastructure to build an agent. And essentially, I want this agent to know everything from how I built it. Agents are going to work, and what tools are going to come out of it, and what kind of jobs are going to come out of it. And I think putting that knowledge out there, just showing what I'm building and getting it out there, then it just brings more vision, and you could do things like this where you know I can talk to you about ⁓ what I'm learning from AI and agents, and then the same, vice versa, because to teach other businesses but to also give tools. So it will become a platform where I'll have just various tools on there that I've built that they can use. But then it's something I can show businesses, hey, these tools I've implemented, I can build this for you know, ⁓ for you in whatever you need. I've had some similar talks with people and everyone's tackling it differently. Everyone has a different thought process on how it should be constructed. And I think that's like gold hearing hearing about that and learning from it.

Niklas:Thank you so much, Julian. This was really interesting. And to you, my lovely listener, I'll you next time. Yeah, I think that I mean the title of this very famous paper that comes back up again and again was Attention is all you need, right? And if you if you look at it very rationally, there's a lot of truth in it. So if you have an audience, ⁓ so it's it's easier to build on an audience, however you do it, ⁓ than to build on no audience. So once you And I think it's very visible with a few of the like ⁓ big indie hackers, the people who are building, really building alone, building those one person companies, ⁓ and having ⁓ very decent revenue numbers. Not not like for a large business, but for for something small. And the like for example, Peter Levils or something, for a long time the he sold a lot to his audience. I think it changed over time and now he's pretty famous. But for a long time he actually ⁓ sold sold a lot to the audience. And he said I think a year ago, actually when AI started he thought he had to get famous. And this this remains largely true, I think. Like you have to If you want to to sell somehow, it's a lot easier if you have a public profile, if you have an audience. And it also works really well for people like Alex Carp, for example. It works well for hiring. It works well for people like Elon Musk. They are able through their profile to get the best talent and to also get customers because people just want to work with him. I don't know what your take on this is.

Julian Fitzpatrick:Yeah, no, I I I I f fullheartedly agree in in terms of like the building the audience and it is easier to to be able to sell to a grand audience because you have that that that user base that will look at it. ⁓ I don't necessarily think that you have to get super large. I think, you know, there's you can have a niche and if you have

Niklas:Sorry, one one second in between. I cannot see your video at all anymore. Can you maybe turn your camera on and off once quickly? Like I don't know what this is.

Julian Fitzpatrick:Sure.

Niklas:Might also be a program. Can you still see your image?

Julian Fitzpatrick:⁓ I can see you and I

Niklas:Yeah, then it might might actually be me. Yeah, but you still see yourself, right?

Julian Fitzpatrick:I I see myself and then I see you.

Niklas:Yeah, okay. No, then it will be good in the end. Then it's just probably me, then I'll drop out. But your background is gone.

Julian Fitzpatrick:⁓ let's ⁓

Niklas:Yeah, I don't know, they updated the software, so it might be software glitch.

Julian Fitzpatrick:⁓ it's not letting me Is it because we're recording?

Niklas:⁓ yeah, that can be let me start.