He Sold a Solo-Built A Privacy App with AI Here’s What Actually Works | Alex Calish

What's the best way to get this thing to build in a way that's sustainable? I guess, because really, like, if you, if you just keep prompting and hitting enter, prompting, hit and enter, you'll, you'll end up with an app that works, but it's not going to be able to be maintained, it's not going to be secure, most likely, and you're just going to have this potentially like crazy code base where you have no idea where all the stuff is.

Hey, I got Alex Kalish here. He's a very close friend of mine. He has a ton of experience in D2C. He's helped brands like Osea, Barnana, Element doing things like supply chain optimization. And he actually just sold his, his own personal data privacy platform that he solo built and he put it on Acquire and that got sold. But I'm really stoked to talk with Alex because he and I grab a coffee maybe three, four times a week and talk about business, talk about AI, talk about how we're using it. And so we're wanted to bring them on to chat. What's up?

Stoked to connect my first Vibe coded project and actually Erase Me was really built before Vibe coding was even coined. It was sort of in the earliest days of like, how can I paste this piece of code into ChatGPT and hopefully get a good answer? And so, yeah, it was, it was the early days of Vibe coding that, that was built and just sold last week.

You're, you're actually nailing it. We got everybody who's doing like the content influencing stuff, and now you're, you're zooming in from Kona and you're just building stuff. You're doing what everybody wants to be doing.

Trying, doing my best.

So I don't know, maybe. Do you want to talk about how you built Erase Me and like how you used ChatGPT and cursor and all these tools to build Erase Me and then we can talk about some of the Shopify stuff that you were stoked on.

Yeah, I mean, I, I taught myself how to code with Chat GPT, kind of like most people are doing these days. But it was really back in like 2022, 2023, before, you know, we had the integrations with the IDs when the models were a lot less impressive than they are today. And yeah, I mean, I think the idea is like, how can we use AI to be, to make us superhuman in a way? Like, I went from not knowing how to code to being able to put together a project pretty quickly. I do think one of the keys there was really focusing in on a specific language as well as a specific framework for that language. So really figuring out like, what language do I code with the best. I chose Python. I love Python. It's really easy to use, tons of libraries, huge open source community, and obviously Python's powering the majority of AI infrastructure today as well. And then choosing a framework that goes along with the language that you like. Python's not great for web development, but I figured out how to finagle that through the few frameworks that it does have. And that's just kind of the path that I took. And I think I really focused in on like being really good at this one framework and going really deep there versus kind of maybe spreading yourself out thin throughout all the. There's just a lot of different options. So that was sort of my. In hindsight I realized like that was a direction I went and it was really helpful when it came to building larger apps and sort of continuing to just expand on the, the amount of knowledge that I had in the development ecosystem. Ecosystem. Because I did not know how to develop, you know, three or four years ago. That's a brand new tool for me.

And so, yeah, that's pretty cool. Was there because a lot of AI is either built on GO or Python and there's kind of the debate of like, will AI agents be more Go or be more Python in the future. Was there anything you did or saw or like doing research where like, nope, I'm not going to build on Go. Like Python is better? Was there anything like that? Or did you just think that Python was the most popular? So you do learn that?

Yeah, I think it just had to do with my threshold for being able to learn how to code as well. You know, a lot of developers can find themselves knowing 10 different languages and really understanding them well. I've pretty much known Python, CSS and HTML and that's about it. I've been, I've been fairly poor at JavaScript, which is unfortunate because that's one of the best languages from a web development perspective. So I just really focused in on the language that made the most sense to me and that I could really understand what was going on. I think I would have learned Go had I needed to. But they're, you know, the trade off from learning a brand new language versus just going deeper into the one that I was, was connecting with just made more sense.

Yeah, I mean, I feel like you're always ahead of stuff. I remember like a year ago, you Were maybe even a year and a half you were showing us new websites, like almost daily that you were building. And it seemed like you were ahead of the trend because now you have the lovables and you have the things that can build a website really quickly, you know, so what are you doing now that's gonna be cool in six months? Like, what is the cool stuff that you found that you're working on or using it for?

It's funny because I feel like I've regressed a bit, you know, in today you can log on to Twitter or whichever, choose your social network every, every morning basically, and there will be a new launch of a new tool and a new platform to do a new thing on. And I think this is where maybe going, you know, a mile deep versus a mile wide isn't a good thing. You know, right now going a mile wide is great if you want to learn a new platform every single day, discover what's the best. I think the social platforms also make lead you to believe there's a new best thing every single day. Whereas I decided to go a mile deep with Python and the web development framework Django and really just get really, really good at building apps. And like you mentioned, I was spinning up a new website every day, seeing, you know, where I could take a new idea, a new concept, and really just, you know, it's, it's kind of a, it's like a, a builder and their, their tool belt. Right? Like you're trying to figure out what tool belt is going to be best for me so that I quickly scale, quickly, understand my platform easily, so, and really have like a deep understanding of how everything's functioning. And so my tool belt, you know, I've sort of been collecting different pieces of that tool belt over time and eventually you kind of end up with this, this sort of, you know, framework that you've put together of like, okay, I know what my database is, I know what my hosting platform is, I know what my code base is going to look like. And just being able to push a new idea quickly through that versus, you know, I see new new tools that are coming out on, you know, regular basis and I, I get a little, you know, I want to try them, but there's a part of me that doesn't want to get too distracted going down a route that might lead me astray and might not at the same time. So you kind of have to be experimenting and checking these new tools out while at the same time also letting them sort of age a little bit. Sometimes the tools get a little bit better as, as they've been out for a while. I've. I avoided some of the Google AI tools until recently and now it's like nano bananas, my favorite thing.

So, yeah, I mean, you said something that was really funny. This week I saw an article that like, the bad AI softwares are winning because they've gotten really good at marketing and they've figured out how to promote themselves on social media. And so everyone's finding out about some of these tools that are maybe not the cutting edge of AI, but they're really good at marketing and they figured they'd go to market out. And I'm like, yeah, that, that works. Marketing's real.

Yeah.

So with that, you mentioned you built something pretty cool for Shopify themes. And you have like this GitHub repo that you can edit in cursor and then it updates the Shopify themes. So you can basically like Vibe code Shopify really easily, which sounds like a dream for anybody using Shopify.

Yeah, that's. So that's exactly the setup. And I'm very recently getting back into CPG after sort of having a hiatus in purely coding and just building so taking the experience that I have from building these Python Django applications. By the way, Django as a frame web framework is very similar to Shopify Liquid. They're almost, they're like almost identical in a lot of ways. So I had a lot of previous experience digging in through Shopify themes, reading lots of liquid syntax, and that made learning Django really easy. And now I'm kind of going back again into the Shopify themes. Shopify also has a much more advanced development framework. I think it's called Helium. That's for like React native web applications and development. That's just something I don't know how to do. And so I think for like the average user who today it's really cool because your Shopify theme can be connected to a GitHub repository and with that connection you basically have direct access to your entire Shopify themes code base through GitHub, which is the standard version control platform for most developers. And with that GitHub connection you then have a connection directly into your IDE or your development software development platforms. Basically you, you just, you, you immediately have access to cursor Claude code. Whichever your favorite AI coding platform is, that has a GitHub connection, which they all should. You need version control that gives you a hundred percent AI powered editing to your Shopify theme code base, which I just think is something that's very new and very powerful to have as a sort of juice or maybe solopreneur Shopify developer.

It, it sounds so smart and so obvious, but I hear no one talking about this, you know, like of course you can connect it and then have these tools reading your code and then you can update it in small like of course, yeah, but no one's doing it. Why do you think anyone's doing this or no one's talking about this?

Yeah, really it is interesting. I haven't really seen anyone talking about it. I think there's probably a lot of reasons. Shopify's app ecosystem is just so powerful. You might not need to do all these things. You know, asking Claude or you know, Cursor to code a specific, you know, piece of an E commerce store. Most likely there's an app that's already created for that. AI is also very bad at front end development. The design side of, of coding of websites. The AI just hasn't figured out how to make really good looking websites yet. A lot of times when you're trying to update front end know CSS HTML to look really nice, you just get not great results I find. So I haven't quite figured that piece out yet. I'm sure there's a way to get Figma involved there so you can take your Figma designs and just have Claude implement them. But I think there's just a lot of small updates and little interesting tricks that people can do on their own. Now with, with Cursor integrated into their Shopify platform, couple things we're doing is right now on our E commerce store that we're launching, one is like a, a live stream YouTube. So we have a database of YouTube videos that we want to live stream to our front end. So I asked Cursor to code up a player that automatically collects the time of day, sort of syncs it with our playlist of YouTube videos. And that way that every time somebody logs into the website there's like a live streams with a, with a sort of like a TV feel to it so that, you know, whatever, whatever piece of media is playing at that point in time, it'll just jump halfway through the media like oh, you're, you're 30 minutes into this live broadcast. So just little, little fun things like that that add a personal touch to the website, make it stand out as something that's not just a pure Shopify sort of duplicate that you get a lot of times and without having to pay a development agency, you know, 50, $200,000 to do that.

I mean, I remember when all these brands were launching a new site and if they had a big name by all the DC brands, they had a big name behind them, they would pay these agencies 150, $200,000 to build their Shopify site for them. And now it's like you can do almost all that work for one smart person and the $20 a month subscription to pick your LLM model. It's pretty crazy. Or go ahead.

That's exactly it. Like I'm using a free theme and so you don't really have to pay for the theme anymore, which isn't a huge fee to pay. But with the free themes you get a pretty substantial default template. And from there I think using Claude or Cursor, you can really get pretty far with customizing it and making it look really good.

So I got something for you to try. I don't know if it will work with the language that you're best in, but there's a tool called Anima and Anima will take Figma Designs and it will create components of those designs and then you can drop those into Claude really easily. So it's that bridge between Figma and Quad code or Cursor. So try it out because I agree like every single front end for an AI generated website, they all, they have a look and feel. They're almost, they're now to me, they're really easily identifiable and so the quicker you can get away from that, the better, I think.

Yeah, that's, that's amazing. Yeah, being able to take someone, that's someone's graphic design and then get that into your app through cloud or Cursor would be super powerful.

So what else are you using Cursor for within Shopify? Like the video sounds awesome. Like that's a great piece. What else are you trying to get your site to do or what are like the little wins?

Yeah, I think, I mean there's a, there's a bunch of little wins right now. We're, we're still in the graphic design phase on this site. So I've just been playing around with my sort of ideas that I would want it to look like. We, we have a bit of an editorial layout. So I was having Cursor help me create different layouts for like our HERO section layouts for. Yeah, like how, how can I organize these different parts of, of the, of the page and just basic CSS that somebody who, you know, knows how to do these things would do really well, but it's just not my core competency. So outsource that to the AI and it does a pretty good job at it. But honestly, I think the, I mean the most exciting thing that we're going to be doing with AI is just the content side of the business. Right. Like the, the content has just, I think all of the AI tools are just on a vertical path right now. Exponential and Nano Banana has just been incredible with the just photorealism that it can do and the way that it's able to put your product into the images really have fine tuned editing control. You know, you can ask it to remove like a very specific piece of an image. So there's just like a ton of new capability on the content side that we haven't even touched or scratched the surface yet because we're still a little bit early. But as I'm beginning to play around with it, it's, you can see just how powerful. You know, I think the idea with all these things today is like you can have a three or four person team doing the work of like 20 now.

Yeah.

So we get one, we get one good graphic designer that's able to guide us or brand director that can guide us in the right direction. And with the right sort of prompting on these tools, you'll, you'll be able to put out content that looks really good and is consistent. I think that's another piece that a lot of the previous, you know, image generation models, they were kind of all over the place. And you're getting to this point now where you can be very specific and consistent across multiple generated images where all of a sudden you have like a portfolio of images that all look similar rather than kind of just being a little bit all over the place with how they all were being, were, were coming out from each, each side.

How do you keep it? Like that's one thing I struggle with. How do you keep it consistent? Do you have like a brand Bible? And then you turn that into a markdown and you say Nana Banana, follow this brand bible anytime you're creating an image or like what? How do you do it?

Yeah, I mean I'm still figuring this part out as well, but part of it is just putting in an image from earlier and you say, okay, take the same image, the same characters from this image. And now, you know they're going to do something different. So for example, like we're you, we're working on a sports brand and so I'll have an image of say, you know, a finish line of a race and then I'm like, okay, now I need these same people posing on the podium. And nanobanana is really good at taking the same people from that first image and recreating them into a second image where they look fairly close to one on one. I still think there's a bit of variance, but not compared to some of the previous image generation models where you would sort of prompt it and the image that would come out is just completely different and there's no context. So I think like Nano Banana is holding onto a lot of context and translating it between scenes, which is just like opens your whole world up on the content creation side. And it can do, it can do, it can do logos, it can do brand. You know, I haven't even gotten a chance to put our packaging into the, into the models yet, but I think once we do that you're going to be able to see like, wow, I've got my packaging, my products in these images and they're fully AI generated.

So crazy. I mean like when I was doing shoes, you'd have to get the shoe, you'd have to go call a model, organize the model to show up, you'd have to go get a photographer, you'd have to get a space, you'd have to coordinate all these people, drive to downtown LA to go shoot in the space, make sure the model showed up on time, make sure that your photographer is there, have all the scene ready and before you know it you spent five grand to get some shoe shots. That's just one scene and maybe you don't even like that scene. And now it's like you can create whatever you want with just a really good prompt and then just tune those prompts and then like I, I think what you said of uploading the previous one works really well. I'd imagine there's probably some like drift over time but like I'll take that all day, Save me five grand to shoot makes getting product to market so much more efficient.

Yeah, and I think we're, we're lucky to be building right now because it's going to allow us to launch a fully AI native brand. Obviously there, there'll probably be some pushback from, from some customers about having full AI generated content, but I think leading with the AI generated content right out of the gate will just allow us to continue doing that, you know, for the lifetime of the brand.

I'm so excited to see it. I am so excited. It's gonna be amazing. But what about the flip side? Like what have you seen not working with AI? Where is it Just all smoke and mirrors. There's a.

There's a lot I think of. Of hype. You know, I read. I'm on Twitter all day, every day. I think. I think Grok itself actually is an interesting example. Like, when I'm on Twitter, I'm seeing Elon, I'm seeing all these people posting about, oh, this model's so great, this model's so great. I do think there is a ton of promotion going on behind the scenes that's not. That's hard to discern. And I've seen myself looking and then, you know, using OpenAI and using Grok and people are talking about Claude. Everyone's kind of touting all these different models, and I think you kind of just need to use whatever you know, works the best for you. I've pretty much been on OpenAI from the beginning, and every time I try another one, there's just a different feeling. There's a different, you know, temp, the way that it speaks. Like they're. They're just little things. And the models themselves, even within a company, change a lot over time. But I. I do think there's a ton of hype online that's pushing you into these different areas to try to get you to use different tools and different platforms. And I think. I think Vibe coding is like, massively overhyped. You see people on there all the time. They're like, I spent three hours and I have a fully functioning app that does this, this and this and that. And, you know, as someone that's gone pretty deep into, you know, my own vertical, I've seen just how complex coding is. You know, coding is extremely, extremely complex from how you structure your database, how you protect, you know, the secrets for all of your deployments, you know, how you. How you actually deploy the product, what systems it being deployed on. And from what I've seen with Cursor when I've been using Cursor is that if you set it into agent mode and just let it go and don't pay attention to what it's doing, you end up with a massive code base that is, one, probably unrecognizable to you, like, you have no idea what the hell's going on. And two, you know, it can be written in any sort of syntax like that Claude or Cursor decided to come up with at that time. And by syntax, what I mean is like, code is like language, right? Like, if you read my diary and I read your diary, like, the way that we write is just going to be Completely different, right? Neither of us is going to. Or an email. Like, if I read all your emails that you write it and you read all my emails, you'd be like, alex, why do you use so many exclamation points and everything? You're always saying like, hi, you know, write it. And so just something as simple as that. You know, I think a lot of people skip over the fact that, like, when you write your own code, you actually write it in a way that makes sense to you, because there's about a thousand different ways you can write similar code, all from how the comments look across the code base to how it's structured within the file system of the code base. And so there's all these little pieces that make a, that are extremely important if you're going to be deploying a, like, enterprise level or, you know, an app that you expect to make a lot of money and have a lot of users. Because debugging, you know, becomes a challenge when you have no idea what's in the code base. And the AI knows what's in the code base because the AI has, you know, a million token context now, and it can read your entire code base in a half of a second. But for you as the owner and the manager of that code base, like, it can be really hard. You know, I, I built a custom web application for my, my previous employer, and I spent a ton of time just making sure I knew where everything was. You know, okay, where does this function get called? This function gets called over here. And why did I build this over there? And as I've been working with Shopify and Cursor, I'm honestly, like insanely impressed at how well it does what it does, but I do think significant guardrails on it. So when I, when I'm working in Cursor, I make sure it creates a new file for any new thing that I'm doing so that it's not going and it's not going and editing the Shopify theme files that already exist, you know, because it'll just do that. It'll start going into all these files and it goes crazy. It starts writing tests and, and markdowns to tell me what it's doing. And you really have to put some guardrails on and say, no, just focus on this file. You know, I want to make sure that this implementation that I'm working on right now is housed within sort of has, has guardrails on it, because otherwise the code, the AI will be in all the different pieces of code and It'll start changing things that you don't want to necessarily be changed. And so it's, it's, it's a funny experience working with these tools right now because you have this hyper, super intelligent thing that wants to go, and it has a mind of its own, right? It wants to go do a lot of things and be really impressive and it'll. And all of a sudden I find myself saying, why did you do that? You know, I didn't ask you to do those things, but you decided through your thought process that it would be a good idea to do those things. So really, like putting strict guardrails on. I almost went to just the ask mode rather than the agent mode because I was so sick of, I was just sick of it implementing things that were way above and beyond what I had intended it to be working on. So I'm still sort of figuring out, like, what's the best way to get this thing to build in a way that's sustainable, I guess, because really, like, if you, if you just keep prompting and hitting enter, prompting, hit and enter, you'll. You'll end up with an app that works, but it's not going to be able to be maintained. It's not going to be secure, most likely, and you're just going to have this potentially like crazy code base where you have no idea where all the stuff is and the AI will just be looking at you like, of course I know where everything is. I wrote, you know, I can, I can read this whole thing in a, in a millisecond.

I mean, it's like you almost have to have a cheat sheet of like, here are all the prompts that I. My list of prompts as I've built the project so I can go back for the next time and change up the prompts. And then maybe here's another cheat sheet of all the decisions that I made along the way and why I made them so that I know how to tell the AI to go fix it when I need it fixed. I mean, I think what you're saying is really well thought out. It's. It's like those are some of the new challenges of how do we build. When you have this like, monolithic AI slop code base, you know that you need to go fix this one problem because now you actually have paying customers and you got to support them and you got to go fix bugs and you got to go fix little things and edge cases need fixed. But you don't want your entire app fixed. You just want that one little edge case and then you want it changed. I mean, it's an interesting thing to go go solve for. And I've heard big companies like Apple, they're using AI just for the last mile. They're using real humans for all the real engineering that needs to be done and then maybe the AI for like, you know, debugging and like testing and routing and doing all that kind of stuff. So it's interesting. I really appreciate you doing this. Anything else? Any other secret AI sauce that you want to share?

I don't think so. I think that's it.

Amazing. Good stuff. I really appreciate. I can't wait to get out to Kona and come hang with you. Thank you. And this brand is going to be epic when you launch it.

Yes. Keep an eye out for unspecified brand because it's going to be super fun.

I'll be one of the first orders.

Nice. Love it.

Thanks, Alex.

All right, see you, James.

See ya.