How I Run A 0-Employee Business With Ai Agents

Ever wished you could clone yourself to run a company? John Rush pretty much just did that. In this episode, we talk with the founder behind what might be the most automated organization on Earth. 80% of his business is run by AI agents he built himself. And he plans to eventually automate the entire thing. From prototyping with lovable to launching SEO agents that work while he sleeps and actually create content that ranks. John walks us through his entire process. So if you're a builder, founder, or just curious about the future of solo entrepreneurship, this is a mind blowing look into what's next.

Hey John, thanks for coming on.

Thanks for having me.

So I saw your Twitter profile. It really caught my attention. You know, your profile says I run the most automated org on earth using the AI agents I built, which is a huge claim. But I checked out your website and some of the projects and it looks like you, you're legit. So I'd love to like learn more about how you're doing that.

Yeah, thanks. So I've been running VC backstory tips with a lot of people, like big teams and a lot of overhead for like 10 years. And eventually I was actually tired of working with people and then the AI happened and then I saw this whole opportunity for me to actually start working solo. So I went solo. I delegated my startup that was VC back that I hired cto, I hired CEO, so I left the startup and then I started learning AI and I started building this little workflows, tools and then later little agents that would just do the most repetitive work that I do myself. So first I had to do everything myself completely like the accounting, paperwork, design, coding, marketing, sales, everything. Right. So I had a lot of pain that I wanted to solve. Then I start using AI to solve everything one by one and my goal is to basically be the first who solves the whole thing. And basically my whole organization will run just me and AI. I'm not there yet, so I'm about 80% on the way, so it's 20% left. Now.

Do you have some people helping you right now or is it, is it just you?

I have people helping with, with coding, so everything else is just me. And this year I even done the whole tax work, like, like tax papers and accounting. Everything was done by me and I. So I'm, I'm proud. I hope, I hope it will work. But I had the professionals checking the whole thing out after and there was no, no mistakes. So I was pretty surprised. I was sure there will be some errors, but then for coding I use co Makers. So I find people who are really good with AI and then it's just me and one person and we build a project together. So no more people. Just I do everything that's on coding and the co maker does the coding by using the AI coding tool and the real coding as well.

I see. So for most of these, these 20 projects, you have a kind of like a co founder who does all the coding and all that kind of takes care of all that side.

Yeah, exactly. I usually start all by myself while I validate the idea and I build the mvp and often it just dies out because nobody actually wants that. But then when I manage to actually get attention and I sell at least 10 licenses for that, then it's a sign that it works. Then I go and find somebody who will build with me further.

I mean, maybe if we could just like show people like any, any way you're using AI agents because like people are listening to this and they're like, oh, that sounds great. How do I actually do any of this? Right, yeah, sure.

So one of my favorite AI agent is called Lovable. So everybody knows it. And the way I use it is that I use it for prototyping. So when I have an idea, I use Lovable instead of Figma. So I pitch my idea to Lovable and then I spend the whole day talking to Lovable about my idea. And so I use it as a prototyping tool. But also it's an AI that's smart, right? So you can also talk to it and you can ask questions and you can ask it to come up with ideas which, which works really, really well. So in this case, I've been building this inbox bot. It's an agent that will manage my whole inbox across all channels, like emails, LinkedIn and. And I built this prototype where basically it unites all my messages into one place and then it shows them here. And then I have AI assistant where I can just ask questions about this chat or I can ask it to reply and then it can draft the replies for me. And I built like around 10 different prototypes using lovable within one day with different kind of ideas in place, the different designs, different ux and then I bring that to my friends. So, so I find people who are willing to spend an hour with me just testing that out. So then I send them 10 links, we jump on a call and then they share their screen and then I watch them use all 10 prototypes. I don't ask any questions, I just tell them, try that, don't talk to me, just try them all. And it works so well. So in one hour I see which prototype works best, where they can actually understand what's happening. And then after that I have one winner. So then I take that winner and then I now continue into building that into a real, real product. So that's. That's probably my favorite use of AI because, you know, as an idea guy, as a founder, I have so many ideas and then, yeah, I can test them so quickly, like literally in hours I can test them. And in the past, I would have to hire a designer and then, yeah, I would have to hire a developer who would code that interior designer.

And you work with them in Figma and they. You pay them a lot and you might get a design. It's totally not what you want. You got to go back and forth and then hire an engineer and then maybe nobody wants.

Exactly. I mean, like, imagine like now my iteration cycle is like three minutes. So I sent message to here, like I have messages to lovable. And then, you know, it spends three minutes and the whole job is done. With a designer, it takes you like three days for the same cycle. And now three minutes. Like, it's crazy. You can't even compare that. So I think we will see like crazy growth in new ideas and new takes on the old ideas.

Did you say 2020 in a week? Is that your current rate? Like, roughly.

Yeah, yeah, but. But most of them don't work out.

Yeah, of course.

Basically, you know, so when I have an idea, the first thing I do when I have an idea, I often lovable and I pitch the idea and I see what happens. Like. Like that's my habit now.

Yeah. When you say pitch lovable, do you mean do you actually see what level thinks about your idea or do you just like, see. Yeah, what's the feel of the landing page? Whatever. And does it have a good feel and possibly something. Yeah, or both.

Yeah. You know, I treat AI as humans and I think that's one of the reasons why it works so well for me, because a lot of people use AI as technology and they try to treat it as software where it expects you to write in a certain way, but I treat it as a human employee. And I think that lovable should understand the idea not just on a technical level, but also why it's important for a world. Like, who are the users and why do I want to build it? And if you do that, you get surprised by the ideas. Lovable sends you back. Not the ideas, but by the UX it built like it's not that boring. Like it actually builds cool stuff.

I wonder if we could come up with an idea right now and just like do something in lovable and just show people. Yeah. Kind of play through like what your process might be.

We can do that. Let me start a new product, new project. Go to dashboard.

Okay. AI agent or AI experts directory.

I mean why not? So I would say like AI expert.

Directory and maybe have like different categories. Whether it's like influencers or like actual, you know, scientists or you know like maybe like two or three categories that are.

Make it look.

Make it look good, nothing bad.

Build the most popular directory. So it sounds silly, but whenever you add these emotions into your text when you talk to AI it tends to do a better job. I don't know why. Even when you say please, there's even research where they measured the difference between please and without please. And it was 7% better, more accurate answers when you had please. People built cool stuff with lovable. Only today I saw somebody raising VC money with an app built on lovable. Then I saw a guy who was making 30k or 4k mrr with lovable app. So I think this is like, like a magic thing for, for the modern entrepreneurship. In the past you would have to raise money first on a pre seed and then you would find a co founder or outsource that and the whole thing takes months or years or maybe never happens. And now basically in one day you can do the whole thing yourself using lovable. And that's crazy.

Yeah, it's so much faster to prove out concepts. Yeah. Like you said. I mean I lived in San Francisco for 13 years and had a similar thing where I've done two VC backed startups with mixed results and a lot of pain of raising a ton of money and going, you know, having the crazy ups and downs and, and things taking longer than I would have liked. I would have loved back in the day to be, to have something like this.

I'm jealous. You know how easy it is for, for the founders today. Yeah. Here we are.

Oh wow.

So it actually works. We can filter the things.

Also the cool part that you see it, it even has pictures. These are the stock pictures. But. But it looks good from the start. Yeah, with the fake data. Look, AI is completely changing the game for every business right now. And I'm seeing the biggest wins happening in marketing. Most people are using AI all wrong or barely scratching the surface of what's possible. That's why I'm sharing this AI marketing automation playbook. It shows you how you can eliminate 80% of your manual marketing work while.

Actually improving your metrics and conversion rates.

And it breaks down everything from content to customer journey automation. If you're in marketing, you'll definitely find a few gems in this. Get it right now by scanning the code or clicking the link in the description. Now let's get back to the show.

How good is Lovable on code? Because one thing I was thinking about is, you know, recently I've been playing with things like Factory and other things and it feels like they're a little bit more hardcore on the code. I wonder if there's like some kind of like ideal workflow where depending on the product, maybe some products like a directory, may not need anything else. Right. But if you're doing a more complicated product, maybe there's some workflow where you kind of like prototype the idea out with something like level or V0, which are kind of typically better at UI UX kind of stuff, and then hand it off to like a more serious coding AI agent like Factory or something else. Right.

I mean, you probably seen my AI coding threads. Yeah, right.

Where.

Where I have these 40 tools I compare. So I actually tried all of them because not in one day, just whenever I see one, I try that and that's why I ended up trying all of them. And you have tools that are great for prototypes that, that basically replace Figma and the front end development. So v0lovable are probably the best for prototypes. And then you have tools that are the best for full stack apps. And replit is probably the most powerful. I built stuff with Replit that I run in production and it works pretty well. And then you have tools that are good for mobile apps. Then you have tools that are great for business logic or for heavy logic where it's not about the UI but rather the logic. For example Cursor or Windsurf, they are great for building logic, but they're not as great for vibe coders. So you have to know how to code where with Lovable you don't have to know how to code. And Bolt is probably in the middle of everything. So it's kind of, it's good for the front end, it's good for real apps, it's good for vibe coders also it's good for coders. So I think it's the best kind of middle, like the average tool that does everything okay. But again, my favorite is Lovable just because I see that in most cases the UX is the most time consuming thing in the product because you iterate on UX the most. And also UX usually decides whether your product going to win or not in most cases. And that's why I use Lovable. But also I have this other tools like this Unicorn platform.

What's the Unicorn platform? What is that?

For anyone listening, It's a website builder and directory builder. The difference here is that Lovable can build everything, whatever you ask. And the Unicorn platform can build only specified products, for example, like wait list or Airbnb, like website. So whatever you have in the gallery it can build. Whatever you don't have in the gallery, it can't build. And that's why if your idea exists in the gallery, then you're going to get much better results compared to Lovable. So for example, here's the real estate directory and basically you can build yourself something similar but with your data for your use case. And it will look pretty much like Airbnb almost.

And you're pretty confident that it will work because this template has been made by the experts and then it works well on security and everything. So I tend to use specialized agents for the tasks because I think specialized agents, they work better because people have built them and they have fixed all the corner cases because AI is great for a lot of the tasks, but whenever you leave it unattended and when, whenever it does the work on itself and there's no human storing it, then it does some stupid mistakes pretty often. Like I use tools like Manus or Genspark and they are really good for some tasks. But if you compare them to the special agents, for example, there is a tool called Autogrid that helps you to build this data files like for directories, you can create a bunch of data and it was really, really great for that. Somebody bought that tool. So now it's, it's not available anymore. But Ganzburg and Manus their average for these tasks. So that's why I think the best way to use AI agents is to basically find specialized agents for every task. And then you start by using general agent, you get some data in and then you test that, use that as a prototype or whatever. But whenever you go for real work and you need real full data, you go into specialized agents. And that's what I do. So basically I move to, to the next agents.

Have you tried anything at any of the platforms? Like, I know there's like Lindy and there's Mind Studio and a few different platforms and Innate in. Yeah, and we actually just had a flow from Lindy on and he was showing me some demos, I was like, oh, I didn't know you could have like different specialized agents for all these different use cases. And then you can kind of tinker with it underneath the hood and pick like, oh, do this in this order. And then here's your context for doing that and which model and you can pick which model you use and like everything that's kind mind blown.

Yeah. Linde is probably the best for non coders and marketers to build these workflows and agents where things work based on triggers and they react on messages and then, you know, they, they search something on the Internet, they send an email, they send connection requests and all those things. So I saw people building huge sales and marketing workflows on Lindy. But it is like my main issue with things like Linda is that it is very limited. So you don't really code that much there. So I'm a big fan of writing code for agents because then you can make them kind of smarter. But I think what LYNDA does is that it lets great marketing people or great salespeople or great operational people automate the junior people work or automate their own repetitive work. So it's really good tools. But I use usually Zapier and N8 for the same thing because they were in the game for a long time. I'm used to them. But I think Linda is even better for that. So I'm planning to switch to that. So the other tool I always use and I actually built myself is SEO Bott. So basically this is a true agent. So you give it URL like I'm using your own URL now. And I don't know what's going to happen when I click that.

Probably. It's probably a mess. So give me some advice. I've been meaning to do SEO lore because like it's so prime for SEO and I just haven't done anything significant with it.

Yeah, you have to start somewhere. Yeah. So basically this agent is probably my favorite one. Like I built it myself, but still I really like it because I wish other agents were built the way I built this one. Because when I was building this agent, my goal was to remove all the work from the user completely. So basically now the only thing I will do from here on is to watch it work. So there is nothing more for me to do.

Almost like swatch the terminal and it's just going at it.

Yeah, yeah. And it takes forever. Like it will keep working like every day from now on until I turn it off. And that's so cool because I think a lot of the agents, they end up building not the agents, but actually tools where AI helps you. And I like that. But that doesn't change the game for me. For example, like I'm so busy I don't really have time to use the tools. So I would rather turn them on and just let them run. And I think most agents still think that their main users are the teams and teams want to have tools to become more productive. But my main users are the busy founders like me who don't want to become more productive. I just want to do nothing. And the whole thing happens because I have other things to do, right? Like I have more important things to do. So let's see what happened here. So it does some research. So it found you.

What's it doing now? Is it like analyzing any problems and maybe telling me things I could content I could create or.

Yeah, yeah, it has done that already. So it suggests you these 10 headlines. So it does research for the headlines. It found low hanging fruits for you and it looked up on Google for the search traffic. So you can land 1k clicks for this article and 2k clicks for this one. Basically it has done the whole work of laying out the strategy for SEO for your tool and it has the headlines. And if I click proceed then it will just activate the work and then it will start building the headlines. It will send the headlines to your website and then it will keep doing that every day until you stop it.

That is amazing. How do I get this? I need this.

I will send it over to you so you're in it. So I will just invite you after.

That's awesome. If someone's listening, how do they find this? Because I think a lot of people are going to see this and be like, oh my God, I want that. I mean it's like any business owner, I don't know why you wouldn't want to try this.

Yeah, it's seobotai.com.

Okay.

Seobotai.com and they can try. And I really urge agent developers to try too because I had the idea with while building SEO Bot and all my other agents is that I think the, the founders and the software world is really slow on adapting for the new paradigm shift where we have to move from SaaS and you utility tools into AI agents that run on autopilot where you as a user is acting more like a moderator or like a boss. Right. And the boss usually gives feedback and the boss doesn't do the work. Right. And I'm Disappointed by seeing most of the agents not moving that way. Like they're still looking like the old software. And I think they should be kind of bolder and take more risks and actually make agents that do the whole thing for the user because that will simply accelerate the founders. Like now we have, you know, hundred people building the same company and imagine if 100 people could build hundred companies, maybe that will not never happen, but at least you can increase the number of founders building the products. And that's kind of my dream, that my tools will enable solo founders to build startups without raising VC money, without hiring people, at least at the start and they get somewhere.

You in theory could go through the entire process of just like you know, some kind of new kind of AI agent powered studio where you're just cranking out ideas and the ones that work you like double down into and put more resources into.

Yeah, I mean that's exactly where we're going, where agents will be able to do everything humans can do. And of course humans can do things better than the agents the best humans can do things better than the average agents. But in most cases you don't need that. Like when you're validating idea or at early stage it's not that important the quality of work, it's rather the cost is important. And you can always bring human on board later or you can be the human who, who you know, improves the outcome. And that's what I do. So almost everything that my agents do I end up revisiting and improving myself at some point. But for example, look at SEO content. So it generates me 100 articles a month, for example, for one product. And then after the last day of the month I go to the analytics and I see which are the three articles that did the best. And then I spent an hour to improve those three articles. Yeah, so basically I'm still involved. But you know, I do like the AI helps me. I'm almost like a VC fund that funds 100 articles and then it's, it's the seed funding and now I see who is doing the best. And then I kind of.

You're gonna pile into the series A and the series B to go in on those articles.

Yeah, exactly. And you can do the same with everything else because a lot of people are opposed to AI and they say that AI is doing a bad job with content and the quality is low. And, and my take on that is that yes, that's true, AI is not as good as humans today. But very often in the first run you Care more about quantity than the quality. And then later you can bring the quality yourself and the quality is going.

To get better and better automatically. Like most of these tools, like, oh, a new model came out like, okay, now it's just like 20% better and.

Cheaper and you know, like in, in the unicorn platform. So I have this thing, which is pretty cool. So I won't let people hire agents. So I have this.

Oh, man, you're, you're, you're, you're, you're competing with my friend Matt Wolf, his Future Tools site.

But here, these are not the links. These are actually agents that run within the platform. So if I hire the, the agent that can monetize my website, it will actually do the whole work. So when I enable this, it will outreach all the potential sponsors and it will pitch them my website and then eventually land me a deal with a sponsor for sponsoring my website. And same thing with other things. So I think we will see agents and humans competing for the work. We're seeing that already and I think that's kind of the future.

Oh, wow. I should actually be using this to get sponsors for my newsletter. I should be mainly reaching out to people. I should literally just have like an agent doing all that.

Yeah, yeah, exactly. Because that work is easy. Right? It's just time consuming. Like, it's not like you're doing something crazy. It's not like you're renting a personalized message to every sponsor. You're actually just, just doing boring job over and over again.

Yeah, yeah. Typically giving them like a PDF with the information or sending them a link to like. Yeah, it's pretty simple.

Exactly.

One question I had is, you know, I saw your episode with like, Greg Eisenberg where you're talking about like using AI for directories, and I think maybe that's what you were kind of known for before. What do you think about directories? Like, are they still useful in the future? I guess that's one question I have is because I could see the argument for them being useful or them not being useful because people will just use AI and AI will tell them everything.

Yeah, I think that directories will be useful because AI is good at answering questions, but it all comes down to the data it uses as a rag as their context. And right now the biggest kind of progress among AI chatbots is their context and where they get the data. Because AI is great at answering questions, but it doesn't always know the right answer because it wasn't present in their training data.

And now Grok is doing a good job because it uses social media and its own tweets for that. So I think.

The future of directories is that the directories will represent.

The highest quality of the content within the niche. For example, there are people building the Future Tools or futurepedia or. Or other directories for AI tools. Like those people spent hours and days and months and years on curating the list of AI tools, on finding their pros and cons and comparing them to each other and finding the reviews about those tools. So this data is so important. And if you feed that data into LLM and then ask a question and like, what should I use for this problem that it has really good data from the directory that it can use to answer the question. So then the value of good data is just growing. So I think in the future, the LLMs will pretty much do all the same thing. We're seeing that now already. Like, they're almost the same now. Like we had sometimes when one was better than the other, but today I think all four players are pretty much the same. But. And now is to race for the data. Like, who has the best data. And I think if you were the one who can provide LLMs with the best data, you can charge for it, you can monetize it. And that's why I think directories have big future.

Yeah, that's kind of my thesis with directories is that people are going to care more and more about who's curating it. Right. Like, for example, Future Tools is Matt Wolf.

Right.

My co host of this podcast. Right. And that's a big reason that people go to Future Tools is because, like, yeah, Matt has some agents helping him behind the scenes. I'm not going to tell of his workflows. Maybe he shared some of them on the podcast, but there's still a human element of Matt and team checking things before they go live. Right. And actually I'm working on something like that for lore.com right now. Like, I'm going to have, like, maybe by the time this is out, it'll already be out. But like, you know, leaderboards for AI curated by experts.

Yeah, I saw it.

It's cool.

It's cool idea. I like it.

Yeah. Like, what does Matt Wolf think? What does Nathan think? What is, you know, what are my other friends, Rowan at the rundown, like, all these kind of experts, what do they think? And I think over the time that'll probably be a big trend is like, directories and you really care who's running the directory.

Yeah, exactly. Like as we enter the world where the amount of data will grow exponentially, there will be so much data that the problem is what do you pick from this large variety? Right, right. And they already use directories a lot. So for example, just today I was searching on Grok for, for indie makers and then it gave me suggestions and then there was a link and I clicked the link and it links to indie maker list. And that was my directory. Right. So I was so happy, actually, I forgot I had that directory. So you can see that LLMs, when they have the task to pick the right place to pull the data from, they tend to choose directories. Because that makes sense. That makes sense that they will go for directories versus blog posts because it's obvious that directories are higher effort content rather than the blog posts. It takes months to build directory or some directories took years for people. And blog post is just like one evening work. Somebody puts 15 tools together.

Now five minutes with your, your little agent that you're running.

Exactly like no, no time at all. Right? And I think LLMs understand that because they are smart and they will go for directories. And then if I'm searching, if I'm asking LLM who is the best indie maker, and then it, it links me to indie maker list, I would probably click it and I would probably spend my time on the directory to do more search. So people are concerned that LLM will just hide away all the websites, but not really, because my intent now is to do something around edit makers. And probably now it makes sense to just go to the website and just browse that directly without having LLM as a proxy. And that's what I did in this case and that's what people do. So I think directories will be still discovered through LLMs.

I do wonder if it's going to make it harder though. Like, this is one reason I, you know, I have a very short domain name, you know, because I know it's easy to type in. And in the future I know having a long one can be good in some ways for like, for Google and whatnot, like really precise. But I also do wonder if over time Google's just going to answer it for you more and more. And so yes, the LLMs will suggest you. And it's really good to have a name that people can remember that they can go type in by themselves right into the browser.

That's true too.

What do you think about like, so you know where SEO is? I know, I know before you used to have a lot of content about SEO and obviously some of the stuff do you create. You got your SEO bot. What do you think about SEO? And I forgot what they're calling it now. I think I called it aio, but I don't know. There's different terms for this where basically how do you optimize for the LLMs? Yeah, yeah, I think a 16Z came out with one, somebody else came out with the term. But how, you know, how are you thinking about that? And in terms of like optimizing for.

LLMs, I think that it is very similar. So it's rarely when I see LLM answering with a web link that's not on top of Google search. So I think that if you're good on SEO, if you're being shown on top of Google and on top of Bing in most cases you will be also within the.

Within the context and rag of LLM. So I think that the game is very similar. So it's not a new game. A lot of people sell that as a new game and there are new rules and you have to do new things. But I think it all comes down to the same thing. You have to have good content. Your link should answer the query, people should share it. There should be backlinks to that. And I don't think that will change because that ultimately corresponds to high quality data and you can probably find a way to hack that. But that will not work for too long. I think in the long term still the best data will win and the best data is the, is the one that people share.

I think, you know, you're talking about like the playbooks not being different for SEO and aio. I mostly agree. I think that the difference may be though that domain authority and things like that and reputation probably will be more important just because of the flood of content that'll come out. Right? So it's like that they'll have to lean more and more into like who is putting out this content, right? It's some random AI agent from some random person over wherever versus this person who has like a big reputation on. They'll probably look at social signals too, like how many followers do they have on X and maybe do they have actual engagement? So it's like real followers, you know, they'll probably eventually look into that as well as well as like what's the actual domain authority? And I think they'll also probably look at more and more too like what associations do they have publicly with other, other people that are trusted, right? Whether it's other experts, individuals or other Big brands. That's. That's kind of my theory. And what I'm working on with lora.com is trying to build more alliances between big companies who interact with lora.com and. And help them. And hopefully over time, that tells ChatGPT that, like, oh, Lora is like, the big. The source that you should trust for, like, AI information, you know, and whatnot.

Yeah, I mean, that. That makes sense. I had another test where I had new website and I had a tweet, where I had a link to my new website and tweet went viral, and the website had zero backlinks except my tweet. And the next day I had a lot of traffic from Google, which meant that in Google's algorithm now they look into backlinks from social media, and not just from social media, but also social media and popular profiles. Because I had.

Well, right now they're looking into Reddit a lot, is my understanding. And people are manipulating the hell out of that, is my understanding. Like. Like, people are going into Reddit and, like, tons of people are doing fake comments on Reddit. That's like, the big thing now is to go on Reddit and just create tons of fake accounts and spam all your links. And apparently Google looks at that a lot. So I think at some point they're gonna have to, like, that's why I'm saying the whole authority thing, because, like, they went for, like, you know, looking at your backlinks and all that, and now they're more and more going to, like, okay, well, what are people saying on Reddit and whatnot? And that's getting super spammed. So they will have to go to authority very soon because, like, it's. Reddit is just. It's horrible right now. I don't know if you've seen it, but, like, there's so many fake comments.

Yeah, I saw. I. I saw it. And it's happening also on LinkedIn and Twitter and. And nobody knows what to do with it. So.

Yep, true. Like, what do you do? It's like, it's actually an idea. Back in the early days, I was involved in crypto early on, and we had this one thing called Quick Coin back in the day, which. Which was this thing where you could shoot bitcoin to a friend and you could. You could actually see their face. It was like kind of the. Kind of the opposite of everyone else was doing about being anonymous. We were being the opposite of anonymous, but making it super simple. It went crazy viral for, like, a month or two. It was one of the biggest things in crypto for like a month or two and then died off. And then after that, we're like, okay, what do we do now? And then one of the experiments we ran or we were thinking about was like, well, in the future, you know, AI and stuff's gonna get way better. And I was like, okay, yeah. It was my talking about co founder. I was like, okay, yeah, in the future, like, what, in like 20 years? You know, I didn't realize how fast this was gonna happen. Right. But what we were talking about is like, oh, people are gonna be using AI to like, spam online. There's gonna be tons of spam. And so maybe what you do is you. You have to have some kind of cost to post. But the challenge there is just like, if you have cost to post, I mean, there's just so many people who are not going to post and, you know, just, you know, challenging.

In most cases, though, the people who do run those bot farms, they're willing to pay for it because they are probably promoting their tool. They're going to make money on that. So they will just see it as a marketing cost. And what you will see is that the real people will not be willing to pay for that and the growth hackers will be willing to pay for that. So the whole Internet will be fake.

Fake growth hackers, yeah. Fake Internet theory, right. Like on ST steroids. Like, it's all just like Lindy's and Claude's and whatever. Just like chat with each other.

Well, as long as they chat with each other, then it's fine. So we're not the ones who have to do that. But, but I think big changes will happen. A lot of changes will happen. And, and, and my best take on that is that platforms like X, they, they have the best position on that because since X uses its own feed and its own content to pull the data in, then it's easy for them to actually backtrack the whole thing. Like, who is writing this reply? Who is this? How many followers they have? Who is following them? Are they verified? Have they ever had other bungers on X before? So it's so easy for them because it's their own data, right?

Yeah, I think, I think, I think other signals too, that they probably haven't announced. I'm convinced that they also check DMs. I've actually tested this before.

And so not sure if I should say this on the podcast or not, but I've definitely. So probably over the last year, two. The two of the things, one of the things that really boosted my account on X was having Elon Musk interact with my post. He's interacted with probably about seven or so of my post two or three times shared my post. And the first time that ever happened, I told friends right before it happened, I'm like, he's going to interact with my post today. And an hour later it happened. And the way I did it was, you know, I was connect, you know, or at least my theory about why it happened. Maybe I'm totally, maybe it was just, you know, luck or whatever. But my theory was I bet they look at who you actually talk to in DMs and see if there's actual interactions happening, not just like you sent them a message and they never responded. I assume that's a strong social signal to, to, to them. And I assume if you're connected to someone and they're also connected to that person and it's, and it's, they have data that you are a close friend or, or like a, you know, colleague of that person, it's probably way more likely that it shows up in their algorithm than when, when they open their ex. And so I, I, I have like two or three friends who know Elon pretty well and one of them was more just like somebody that Elon shares his stuff all the time and I, I told him what I was doing, but I messaged them and on purpose had, you know, conversations, curious if that would influence the algorithm. And, and lo and behold, one hour later, Elon Musk responded to my tweet. My friends were like, what? How did you do that? I'm like, I'm pretty sure that's right. I assume that that was my experiment and it worked and I've tried it a few times after and every time I did worked almost every time.

So you also tried that.

Sorry for anyone listening to this. Now if you get spammed by random people who are like, hey, how's it going today? You know, this kind of stuff. I'm so sorry. So before we get off here, I always like to ask a few kind of like rapid fire kind of questions, you know, what's your most controversial belief about AI?

We will not get to AGI using LLMs. I think we have pretty much achieved most of the progress already. And more progress will come from software itself, not from LLMs.

Really? Wow. What makes you think that? I think the opposite. So where's that coming from?

Because when you.

Look at the progress in AI, you don't really know Exactly. Whether that new leap forward happened because of the LLM or because of the software on top of LLM. And in my case, since I'm using a lot of AI directly, and I use LLMs directly, and in that case you see clearly the difference between the models. So the difference between the models isn't that huge. So basically in 2023 we had GPT4 released. And since then, as a user of their APIs in all of my tools, I haven't seen a crazy change. Like, it hasn't improved as much as I would expect. I would say that the improvement between GPT 3, 5 and 4 was greater than between 4 and everything else that happened after. But then if you use the tools, the applications and the agents, it feels different.

Right.

Like the change there is huge. So it suggests that it happens because of the software.

Yeah, the model is getting way better at using different tools and knowing which tools to use. And it does feel like that's how we get to AGI, is through that, through the actual smart, like knowing which tools to use and using properly.

Yeah. And it is similar to the computer. Like the computer itself has been getting better and better and smarter and smarter before AI, but it would never lead to AGI. It was just a software that makes the hardware kind of look smart. And that's what we will see a lot more progress moving forward because there will be a lot more software built on top of AI and tools and they will be connected to each other, etc. But to get to AGI, you need the foundation to be stronger. The LLM itself and LLMs are not moving up. There's. There's a graph. Actually, I will, I will send you after the call where they look at the growth. And it has basically went flat, like almost linear, not exponential, since 2023 for LLMs.

Yeah, I don't know. I'm convinced. I mean, I've talked to a lot of people who work at the companies and heard about things that are in development that haven't been released yet. And my feeling is that a lot of the limitations that we're seeing are more around the infrastructure, whether it's like, lack of chips, energy. I think energy and chips are like the big thing that's actually slowing down the progress. I think the actual Labs, I think OpenAI has, you know, they have better models that have not been released and a lot of it's just because, you know, the website would probably die and they also just would not have enough chips to actually run the thing. Right. Yeah. And enough energy Another thing I like to ask. I think, I think you have kids, right? I think, I think I saw that on a podcast.

Yeah.

But I'm always thinking, you know, like, you know, my son's 11 and, you know, in the age of AI, what should I be teaching him? Should I be teaching him to code? Should I be teaching him? Are there other skills that are like, super critical that I should be teaching them?

I think for kids, it's not that important what they learn. It's rather that they have to train their brain muscle. So whatever helps to train their brain muscle.

Video games.

Great. And I mean, well, that's. It makes kids smarter, like, way smarter. But there's downside that they may just find it so, so fun that everything else will look boring. But I think the coding is single best way to, to train your brain. Like, it's almost like gym. It's almost like lifting weights. Right. And also it's addictive. So it's one of the best ways of training your brain where you actually like doing that, similar to the computer games, but without a downside. Like you will never see someone, you know, going mad because they write code too much. So at some point you get tired of it. But with video games, you never get tired. It's like, it's too, too cool to do that.

Yeah.

You know, I would still learn to code.

Yeah. There's a lot of people who've been highly successful that started video games, though. And that was almost like a gateway drug into coding and just getting interested into more technology. Last question. Imagine you had a time machine, John, and you know, you travel to the year 2050 and you get out. What, what do you see? What's. What's different, what's not different, you know, what's the world like?

All right, that's interesting. I think we're going to see nothing, almost.

Wait, what do you mean?

I think there will be.

Buildings totally locked, and there will be people in the buildings. Buildings are going to have no windows, nothing. And there will be people inside there connected to the computers using neuralink. Direct brain to the computer, basically. Yeah, exactly. So basically it's similar to the Wally.

That's not the future I want, but something tells me that that's inevitable, that we're going to end up having the Wally like the future, because we're just heading there now. And AI will make everything. AI and humanoid robots will make everything super cheap and eventually free. And then humans will have to invent games so that they, they can do something because there's nothing Actually, they have to do, like, everything is optional. Okay. Then we make games where we actually compete and there is some game status, game, whatever, game. And then when they invent the games, the games will be first offline and online. But eventually online games will win versus the offline because they'll be more interesting because things like Bell Tree will help.

I. I think what you're saying, like, I. I get part of it, and I agree with part of it. I. I think, you know, I love the book Ready Player One. I read the book way before, you know, the movie came out and all that.

Yeah.

And. And you know, at the end, you know, you know, people kind of get off of the technology in the games a little bit. Right. It's like. And there may be like a transitionary period where, like, yeah, people definitely are using it way too much. But, you know, I've seen this in San Francisco where people, yeah, they use technology, a lot of people. But also people love to go out for hikes and stuff. Now, like, it's. It's super common that you go out for meetings, you know, hiking and whatnot. Right. And get off of technology as much as possible. I think you'll see that a lot of people will actually spend less time with technology. And then the technology does a lot of the boring work they don't want to do, and they can actually get out more. And then they'll be like, some people will be like, crazy about the games and they'll do it, like, all the time. And for other people, be like, yeah, I'm going to go do that with my wife tonight for like two hours. We're going to, like, kind of just do some game or some Watch some movie or some new kind of, you know, interactive experience.

Exactly. So, like, one of the thing I do is, like, I. I promote kids for founders. So whenever I meet founders, I feel like, you should have kids. And then they tell me I have no time, and I tell them, just use more of AI and then you have more time.

If you have a company, that's actually like one of the best times to have a kid. Like, you can actually. It's like amazing education they would never get anywhere else if you actually somewhat include them. Like, not every day. They're not like going into the office every day or something, but if you include them, it's like one of the best experiences they possibly could have.

Yeah, I mean, I work from home, and when it's day and not night, my kids come over and they just, what are you doing, Daddy? What? I Do. Yeah. Now it's cool with AI. So for example, I'm doing these prototypes with Lovable and then they ask me what I do and I tell them what I do and then I let them prompt it and they prompt it and then it brings some fresh ideas into what I do. So some of the things I have, I actually ask my kids to give me some random ideas or random prompts and then it makes things more creative for me.

You're going to have to increase their allowance. Right. Like if one of their ideas takes off.

I'll put them as co founders. In that case.

Yeah, yeah, yeah, yeah. Put their names at the bottom, like a little heart or something. You know, kind of like maybe tell the people listening where they can find you. Like where they can find you online. If there's a certain website you want to push people to, whether it's the, you know, I think there was the SEObot, AI.com which I'm going to be checking out, but anything else as well.

Yeah, it's. I'm on X and LinkedIn and Substack. It's John Rush X the handle. And I share everything I do every day, every month. I share my plans for the month, every last day of the month, I share what I have done during the month. And then people tell me, oh my God, it's a lot you've done. And I say it's because of the agents I built. That's basically my life. So you can find me there and follow me there.

Yeah. And for anyone listening, whether you like own a business or you're like an individual, I mean, I think there's like so much you can learn from this episode. Like if I was, if I, if I ran a business, I mean, you know, even at the Cubs box, be doing this. No, you probably should have like a division in your company, right, where you are using AI agents to crank out ideas that are beneficial to your main company and then you could spin those off and have AI try those and the ones that work then actually double down and put resources and people on those projects. I think, I think so many companies should be doing that. I don't know why they're not.

Yeah, I think this is really good idea. I think all corporations should have this little AI factories within where people just try to innovate within rather than waiting, you know, for innovation from outside. That's so smart.

So easier, so much easier to do right now too. Right? Like one person or less or zero people to get started.

Start with one person. Like find the most creative person in your organization and let them do that full time.

Crank out the 20 ideas a week. Maybe you find one good one a month, and then actually, like, turn that into, like, a new business unit. Yeah, I think everyone should be doing that, but it's been awesome, John.

Yeah, thanks.