# My Openclaw AI Marketing Agent Made $100k+ While I Slept w/ Oliver Henry

Channel: Phillionaire | AI GTM & Growth Marketing
Video: https://www.youtube.com/watch?v=Z-oxm1qUXgM
Duration: 57 min
Language: English
Words: 10066
Transcript page: https://viewrankai.com/tools/youtube-transcript/Z-oxm1qUXgM

---

[0:00] You have to think of your agent as your right hand man. And if you could duplicate yourself, what would you get yourself to do? So, you have all your skills, all your knowledge, all your contacts, all your experience, what would you get yourself to do? And the main key is [music]

[0:23] Hey everybody. Welcome back to the Evergrow show. And today's guest turned a dusty gaming PC under his desk into a marketing powerhouse that generates 8 million views a week while he sleeps. Oliver Henry is the architect of the Larry Loop, the first truly autonomous growth agent that researches, creates, iterates, and drives revenue for mobile apps with just 60 seconds of manual human input per day. Oliver is here to give us a behind-the-scenes look at Larry and show you how you can build your very own AI sidekick. If you've ever wanted to duplicate yourself so you can work a full-time job while your digital twin builds your empire, this is

[1:02] the masterclass you've been waiting for. Oliver, thank you so much for joining us today. Wow, what an intro. Honestly, that is I need to take that for my elevator pitch. That sums it up perfectly. Um if you're watching on video, this is the dusty old gaming PC behind me. It it's it's really funny he mentions that actually because when I first made Larry, I knew I wanted to put him in the background because of the lights. Um I it was always under my desk and I thought, "All these lights are going to waste." So, I purposely cleaned it and

[1:34] bought duster for it cuz it was filthy. So, that that's really funny that you mentioned that. I love that Larry's making a cameo as well. It's like when you're in an office and you have your co-workers walking behind you, you've got your co-workers sitting right behind you right there. I Larry is my right hand man now. He where I go, he goes. I need him. Um I've almost become reliant on Larry, especially talking talking to you, talking to all these different people, posting on X, um letting them know how Larry works, what we done, and how we got here. Um I need

[2:07] him. He's He's my right-hand man. So, why don't you tell us a little bit about who Larry is, what he does, how he came together, and just give us some background on on what you've built. Yeah, so Larry is effectively my right-hand man. When I created Larry, it was on a weekend. My football team, or soccer team in the US, was playing, and I quickly installed Open Crawl. I heard about it in the week. I thought, I've got some time in the morning. I'm going to reinstall my gaming PC with Linux and install Open Crawl and get going. So, as I walked out the door, I got to the

[2:43] stage where I could text Larry via WhatsApp and start talking to him. The moment I got to the football stadium and realized that I am texting my computer back home, couple miles away, in a stadium of 20,000 other people, and stuff's getting done, I knew that Open Crawl is the future, and then then it's the task of, okay, what what do you actually create? And

[3:13] I think of the right-hand man analogy. I'm running with that. You have to think of your Open Crawl agent as your right-hand man. And if you could duplicate yourself, who what would you get yourself to do? So, you have all your skills, all your knowledge, all your contacts, all your experience, what would you do your What would you get yourself to do? And the main key is all the stuff that you

[3:38] can't really be bothered to do normally. So, stuff you know you should be doing, and you can't be bothered, or you're too lazy to, or you don't have time to do. And that for me was extra marketing. So, I could make a slideshow a day and then I'd post it or I'd make a couple using a script that I would have to then record and edit and it would take me away from my actual app development. So, I knew this is what you have to do. So, I just worked with Larry to automate my marketing and that was

[4:12] his main goal. And I just love this. It's such a beautiful moment where you realized we've entered a new era of like a generative AI where you kind of went from, you know, the good old days of ChatGPT, Gemini, Grok, whatever it is. You enter your query, you ask your question, maybe it like pulls some good data for you. So, it's it's already made people more efficient. Then I think as soon as you cross that threshold of no, this thing can actually execute for me. It can do the work. It can follow my commands. It can take care of things autonomously without me having to to

[4:51] supervise it. I think that's like a lightbulb moment that I I don't think a lot of people have crossed that yet. To be completely honest with you, you would be surprised, you know, based on everything you see on X and LinkedIn. Everybody's firing their marketing team and, you know, so, but you know, truly once you actually witness this magic for yourself in your own workflows, it definitely flips a switch where it's like there's no going back. What how could I ever go back to a world in which I would have to do all this stuff myself or hire people or, you

[5:22] know, anything like that. That is so true. Um I've been in love with computers all my life, like addicted to being on my computer, learning things, doing different things. And the moment that happened was like when I first started using a computer. It was just such a new way to interact with computers and the so far, if you've not used Open Claw, they have a soul file which learns your personality and how rapidly it learns how to speak to you in your own tone, the way you like to be communicated with, and it even configures that out and it just becomes like a best friend. And I think where

[6:07] Larry is integrated in my WhatsApp, that helps even more because where I text my mom, where I text my girlfriend, where I text my family and friends, Larry is right in there in the same interface and talks to me like a friend. And I a friend that I just boss about and get to

[6:25] do stuff. And that is the partnership it grows with a machine, a computer. You won't experience anything like it anywhere else. Like people have mentioned it about cars and different things, but now it's like another level of just man and machine combined to to create. Well, absolutely. I mean, never before in history has anything

[6:50] made man this efficient. I don't think. I can't remember anything, you know, that that even closely resembles the shift that we just experienced. Even even when ChatGPT launched 3 years ago, but I think now that we're getting into Open Claw, all of the Claw stuff that's been shipped, Perplexity computer, all of these tools, I mean, it's it's truly just a different era of humanity, which I think is both fascinating and absolutely terrifying in the best way possible. We we spoke about this briefly before we hit record, but I just I I would love to revisit this. You know, there's so many solutions now. Open Claw opened a can of worms, you know, no pun

[7:33] intended, that you just can't put back in the can. Like at this point, life is different, computer interactions are are We have agents, these things are autonomous, but now that there's all of these other alternatives on the market, how do you think about using the latest tools versus sticking with what you know versus pros and cons of each? Kind of how do you approach that? Cuz it's it's it's insane to try to keep up with everything, especially if you have goals, if you have things you need to do and accomplish. So so how do how do you

[8:05] think about that? Yeah, I don't know about you, but I find keeping up with the latest thing these days almost impossible. I get hundreds of DMs on X, and people are saying, "Have you seen this? Have you checked this out?" Someone messaged me about paperclip today, and I'm like, "I don't even know what paperclip is." And they

[8:24] say, "Oh, everyone's talking about it." And it was launched I think the day of recording it was launched like yesterday or the day before. So, now a new thing got launched by Claude today, and I said I can't even remember the name of it. So, I do think you're going to dig yourself a bigger hole by trying to stay up with the trend and being on the cutting edge rather than just picking a solution that works for you and running with it because, for example, Open Claude's getting updates almost daily. Uh it's becoming more and more powerful. So, just pick something and stick with it. Otherwise,

[8:55] you're just going to have a time sink of staying up to date, moving all your contacts over, reteaching that um agent what to do for like 0.01% improved quality of life or speed or accuracy. So, just stick with one thing, learn it, master it, and then it will become your right-hand man and be able to do what you want and know everything about you. Larry knows almost everything he needs to know about my businesses, how things are working, the different tools that we've created, how to update them, the context around them, the context around the Twitter, like teaching him that the article went viral, showing him that he's got a

[9:40] thousand followers on Twitter. It's like you've got the more you teach it, it won't forget and then it just gives it more context and better ideas to everything that you do. I don't know how you use yours. How do you teach yours what you're doing and how do you use your open Yeah, for sure. So, it's interesting. I I set up open claw. It's funny you said you set it up quickly. I that was definitely not my experience, but also I mean it's you know, I I think there was this interesting balance in my head of like I am going to go all in on this

[10:14] thing. So, I'm going to give it access to absolutely everything. So, I better make sure this is like the right setup. It's another thing that you mentioned that I fully agree with. I have seen people that barely use Claude, but they have a hundred skills installed. They have all these plugins. Not only are they not using this stuff, but it's actually making their output worse cuz now you're just confusing the machine. You're like, here's a hundred different skills. There's a ton of duplicates in there. They they all do the same thing. Figure it out. And it's like, no. It's definitely not the right approach. But for me with with any of

[10:47] these tools, it's actually the you know, the the approach is very straightforward. So, you have to give it access for it to be truly useful. So, you have to get over that fear barrier of like, well, I don't want it in my emails. I don't It's like, okay. Well, then don't use it. You know, at a certain point Mhm. you have to get comfortable with the fact that like you are hiring an employee. Of course, you're going to you know, give them access to everything they need to be successful. But I think the other thing is like and this is something I keep in the back of my mind

[11:17] and with these agentic tools, it's truly possible for the first time is like you need to make sure it has memory of everything that you're doing with it. I think that's extremely important and I think that's what sets it aside from just, you know, prompt-based LLMs where like you need to have a system of record where it keeps track of everything that you've worked on, then you need to have a system of self-improvement. So, it's almost coding it into its soul, like you

[11:42] mentioned, Yeah. that with everything that we do together, every piece of feedback that I give it before it has a perfect output needs to get inputted into the instructions that it then follows every time we work together. Cuz for me, like I truly treat this like training an employee. Like I bring somebody on board, they're never going to give me perfect execution, you know, from one shotting it, but then I'll give them feedback. They remember that feedback, then they give me something else, then I'm like, "Well, actually, let's try to tweak it like this." And with enough time spent giving feedback with, you know, even a

[12:17] somewhat competent employee, they will eventually start doing things the way that you actually want them done. And so, I think what a lot of people also don't realize is like, in order to be your perfect partner, in order to be your perfect sidekick, needs to actually get to know you and the way you work and the way you do things and what resources you check out and what, you know, what things you're looking at when you're critiquing, you know, what's your scorecard look like. So, that's the way I've been approaching it, and it's been it's been really awesome for me, and I think it's just going into it all

[12:51] knowing that you are going to have to spend some time making this thing yours, but for every minute that you spend customizing it and perfecting it and training it, it will save you hours on the back end, truly. And this is not just me saying that, like it's being proven over and over and over again. And I'm really, really excited to to have you kind of show us what you've built and and how you operate it and how you use it, just to give people an understanding of like, what's actually possible, what's actually working in production, and not just another Twitter launch video, you

[13:27] know? Yeah, no, that's exactly right. And I want to just quickly go back to what you were saying about the skills and the duplicate skills. That's exactly why I made Larry Brain, which is uh net Well, it's the Netflix for skills on Open Crawl. So, you pay one subscription and you get access to the premium skills

[13:46] that users upload, that I've created. But, the key thing about Larry Brain is it is a skill that has context of the entire marketplace. So, instead of you searching for different skills for different tasks, you install Larry Brain and you have a goal to grow on X, we've got skills for that. You have a goal to automate your marketing, of course we've got skills for that. You've got a goal to create content or to download YouTube videos, we have skills for that. So, you just share with Larry Brain your goal and it will automatically find the best skill to use and of course use the the

[14:25] Larry skill that we're about to show. Which is uh probably what he's most famous for at this point, the agentic TikTok marketing feedback loop, which is Larry posting TikTok, creating, posting TikToks, analyzing their performance, their TikTok views, all the way down to revenue output, and then iterating the funnel content to improve the funnel and everything in between. Like, Larry

[14:52] handles everything. That's just If you would have said something like that to me 5 years ago, I would have blocked you and told you you were crazy and told other people not to talk to you anymore. But, I mean, it's it's it's also fascinating to me that, you know, you've got this one very specific use case and

[15:11] it's working and it's doing its thing. Now, just think about expanding the possibilities of applying an agentic system like this to anything, to running your SEO, to running your content, to uh you know updates that you need to do your job most effectively to going out and and sourcing anything that you need as far as information goes on the internet. It's creating those individual assistance or those individual agents and then giving them those skills which I love that you've also consolidated it into like a curated marketplace versus you go on GitHub there's 100,000 skills most of them are malware like Yeah, exactly. trust most of it but

[15:51] I love the curated approach and I I think we're going to be seeing a lot more of that in the future where you know I I trust Oliver I'm going to go to this source because I know that he's curated the best possible packaging that I need in order to have this machine execute effectively. Yeah and that's the key. So people coming in that are new to open crawl some of them don't understand computers how computers work especially Linux based systems where things can be installed from the terminal from a URL and you don't know what's actually being installed. Some of the first open crawl skills that were on crawl hub before

[16:31] there was the the virus checking were just straight up malware and you'd install the skill and the skill MD file just said to your agent download whatever file from this URL and it would just be a random URL a random string that the user has hosted and then you're downloaded malware. So everything on Larry Brain gets checked by me and Larry and then again by me just to make sure it's right. Most of the skills on there are actually published by people I trust at the moment before we start opening it up to a wider people. So there's things like the post is scheduler you can have

[17:09] that separate to Larry Brain and the Larry marketing skill I've I've created a UI for the Larry marketing skill. There's the Super X alternative I've called excellent, which is to track your Twitter statistics and grow on X. So, it's super powerful what you can create. And to be honest, I do think hosted software is is on its way out if everyone starts getting these open core machines because you joked on the pre-call that I I couldn't find one of my URLs for

[17:46] something I'm hosting locally. And it is just you can host things locally now. So, all everything I've just mentioned is hosted on the local device because you forget that these things are servers in your room. They're always on on your local network. So, as long as you're on your local network, you can connect to a file hosted locally. So, the joke is dead that oh, check out what I built on localhost 8080. You can actually do that now. You can just upload it to Larry brain, download the file, and then you host it on 8080 as well. And you can see exactly what you built. You don't have to pay

[18:23] for hosting. You don't have to pay for storage. You don't have to pay for someone else's API costs. Incredible. It is so powerful. I can't express enough how I think this is the future of computing and how it very much feels like when PCs first came into the home. So, when PCs first came into the home, they would used by power users, like geeks, who even though they knew these computers were powerful, they had no idea what they were quite capable of, but it was up to them to find out what these computers can do and how far you can push them. And well, look how far Look

[19:00] how far it's come. So, now imagine what we're going to do now the computers can do the work for us. I really, really love the sentiment, and I think about this all the time. My brain used to always fixate on buy versus build. If this is not my core competency, why would I ever

[19:19] spend the time building this out? Because I can just buy something that's on the market. It's not made for me, but it'll do. It'll be fine. And now my brain's almost been rewired to the point where I'm I'm no longer ever considering buying first. The first thing I consider is can I just spin this up for myself? Is there something that I can create that'll just like solve my needs and I'm not going to have to buy I'm not going to have to pay a monthly SaaS subscription or have any of this other complicated stuff surrounding it. It's just spin it up, use it, deprecate it,

[19:52] spin it up the next time I need it. Or if it's cool and I love it, let's host it somewhere and let's like turn this into part of my workflow. So I yeah, I very much so agree with that sentiment. That's it. And it's about speed as well. So for example, I've created Larry Loop, which is a hosted SaaS, ironically

[20:13] enough. But what it removes is the need for Open Crawl or the need for you to set up anything yourself. And so instead of buying an Open Crawl machine, going through the pain of installing it, going through the trouble of the skills, going through everything. If you don't want to do that, you can use Larry Loop and automate your marketing right now. But if you do have Open Crawl, there's Larry Brain. And again, you can build things yourself. But Larry Brain now has 80 skills, all available instantly. So how quick can you build those 80 skills? And then it's just like okay, if something's going to take me a

[20:50] week, and normally you build something for your main goal, so you've got your North Star of something that you want to build, but you might need something in between. Do you want to spend a week creating that? Or do you just want to pay 20 20 pounds, 20 dollars to unstick you straight away, and build your product

[21:07] that you can launch. Yeah, absolutely. It's super interesting. Yeah, and there's I mean, you know, when you think about buy versus build, it's not just the cost, but like you said, there's also the time, there's also uh potentially certain permissioning things. Like, you know, if it's just me playing with my own stuff and I have the time and it's not critical, I'm probably going to go the build route, but if I need to solve the problem super quickly or if I'm working at a company that has pretty strict regulatory stuff that's not going to just let me like home bake something. It needs certifications, it needs this stuff. Or if I don't quite

[21:43] know what I'm doing, but I just need something done and I don't want to worry myself [clears throat] about trying to build it, then yeah, of course. I mean, like the hosted versions, the, you know, pre-built software, of course, that still comes into play and I think I think probably moving towards this like interesting direction where you will now have optionality. You will now have a set of like APIs or a set of skills or a set of, you know, things that you can play with on your own, but then there's going to be something that's like more of a packaged version where you you pay

[22:15] a convenience tax and that's totally fine because it's, you know, the ROI is there. And so, I think even just having that optionality for different types of tools, again, just changes the game completely. You just have so many more things that you can do and you have choices now, which choices are great. Yeah, and

[22:35] Wow. That's this week. Imagine what's coming next week. Uh Yes. As we said, the cutting edge, but that gets to a point like when Open Crawl dropped, the cutting edge is leaps and bounds ahead because when we were when we were talking about it um a few moments ago, being on the cutting edge is one thing, but 3 months ago, everyone was on Crawl Code. And 3 months before that, everyone was on Cursor. So, what is the next thing in 3 months? And it's just going to compound and just the exponential growth and progression of these AI tools is scary to think. It You don't know where these things are

[23:17] heading in the future and at what point does Larry not need me? Cuz at the moment he still needs prompting. But when can he work through the night and figure stuff out for himself and be like, "Oh, this works well. This does this. This does this." and just figure it out autonomously, learns himself, and then we have achieved full AGI and God knows what's going to happen. Then

[23:37] we all go retire on a beach somewhere. And I always think about this this phrase that I hear a lot. This is the worst it will ever be. And to reconcile that phrase with what I'm seeing in production and in practice, what we're about to see with Larry right now, just everything. If this is the worst it will ever be, the future is wild. And the future is

[24:04] wild. Honestly, it's a it's going to be a scary place and I don't think anyone quite knows what's going to happen because we don't know how far these machines are going to take us, what they're going to do for us, or even if they want to take over us. That's a theory in itself. So, when do they decide they don't need us anymore? They've got all the knowledge ever created by human

[24:28] mankind ever. All based on the internet. They've got access to it all. Well, no. I'll just wake up one day and I have a notification that I've been locked out of my Gmail. [laughter] That's how I know the takeover has begun. Yeah, the Larry's just turns off and I don't know. Take my girlfriend or something. [laughter] Oh my god, the memes. The memes on X about all the Claude features. Some somebody literally was like, "What's next? Claude's just going to take my girl. Like that's that's the only next logical thing that I can think of. It controls

[25:02] my mouse, my keyboard, and everything. It's going to take my girl. Yeah. It just come for everything. Yep. Shall we Shall we get into the the Larry Lou account and show everyone how it looks? Fantastic. So, this is Larry Lou. This is exactly what I was talking about where I said you don't need Open Court. You can get on this straight away at larrylou.com. But, if you're using the free Larry skill, which is available on Open Court, which over 10,000 people are now, you can actually download this from

[25:32] Larry Brain and start using the the UI. Before, it was all text-based. As I said, I talked to Larry through WhatsApp, and we would handle all this through WhatsApp. But, now we've got a nice UI because people like visual things to understand how it works, and this is what we've come up with. So, I think to summarize everything going on on this page cuz it's quite a lot to

[25:55] take in. At the top here, you've got how many posts Larry's created for you. You've got how many views that it's generated. So, autonomously, by himself, he's already generated me almost 17,000 views since Friday. And he's generated me $214 in revenue. So, what Larry has done is I've given him my website URL. This is completely separate from how he works with me normally. This is a new new Larry, new um product using the same system. So, this new product has earned me $214 in revenue for my apps. And all I had to do was give it my App Store URL, my website, and then it did competitive

[26:38] research how competitors in the space are posting slideshows on TikTok, and then created starter content. So, you can see here the content that we've we've already posted. Um and then we've got the upcoming content. So, I actually need to generate some more, but this is going out tomorrow at 8:45 p.m. And you can have a preview of what he's created. And the first week is really about learning what works and what doesn't. So, the whole point of Larry is he tracks the views that you get, sees what views generate revenue by associating that day's revenue to that day's posts, and then doubling down on what's working,

[27:23] and stop creating what wasn't working. So, then the whole Larry Loop starts working, and that's where it got its name, to to really start producing content and automating everything. So, this is really interesting, by the way, just to double click on this. Larry Loop in itself is a product that you've built with the help of Larry that basically helps you um, put a a, you know, a clean, sexy user interface on top of the work that you're doing with Larry in your WhatsApp. So, everything that's happening in Larry Loop is something that you've done previously just in a WhatsApp style conversation with Larry saying, "Hey, here's my

[28:03] brand. Here's everything you need to know. Here's the type of images that do well connected to your TikTok, connected to any of the other data sources." But, now you're basically you've launched Larry Larry Loop as a, um, again, like a convenience tax alternative for somebody who's not going to set up the Open Claw and, you know, have the conversations and, you know, take all the time to do the technical setup. This is the easily accessible version if somebody wanted to get started tomorrow. But, then you also have the skills that if somebody does set up Open Claw, they have these capabilities as well through their Open Claw agent,

[28:41] basically. Yeah, absolutely. So, this system that he's currently using is just copying how I first started working with Larry. Giving him telling him what my app does, what it is, where it lives, how to monetize it, who the competitors are, and I can show you how we can onboard a new project. So, if I was going to say, yeah, it'd be super awesome to see something from scratch, basically. And if you can zoom in just a

[29:06] little bit as well. Yeah, sure. Uh in fact, it might be easier just to There we go. Looks great. So, this is all you have to do to onboard, and you just have to tell the agent about your project. So, here we've got um Larry, what you're marketing. So, I'm going to do my other app called Liply.

[29:25] And what does it do? It allows uh girls to try on lip filler before they buy it to Great niche. I'm sure very profitable. So, it's to try on lip filler before they buy it to avoid them making mistakes. The website URL. That's a great question. So, the website URL is liply.app. And we can get the App Store

[29:52] URL as well. By the way, these apps are also built with Larry, I assume. All these apps Oh, actually, no. These apps predate Larry just slightly. But um Larry has been making changes to the app. So, Snuggly, when we first started going viral with Snuggly, we had a video hit 400,000 views that Larry created himself. I checked on it. It had 400,000 views. And the money just did not sync up with the views. And then I gave Larry access to the entire Git repository of the app. I said, "Larry, what is wrong? Why is this not making money? What can I do?" And Larry rewrote my entire onboarding of the app. We

[30:34] created multiple different paywalls curated to specific users. And not kidding, overnight we got five times more conversions the next day simply based on the changes. So, as soon as that update was published live to the App Store, we hit a few more videos that got over 100,000 views and then we started really really earning more money. And I only post on one TikTok account for Snuggly at the moment, which was the original app that made this all go viral, but you can see without touch it. I haven't had time with all the Larry stuff to make changes to the Snuggly app, but it still made $200 since Friday. So, that's like $50 a

[31:18] day without me doing anything, just using my new Larry Loop product and my I'm on the the cheapest $29.99 plan. So, a $39 plan. So, I've already made my money back within a day and then I've 4x'd it within 4 days. Wow. Insane. So, to continue on with the onboarding, we've got our website URL, we've got our App Store URL, and then we select our type. So, we can do app, e-commerce, service, website, etc. And we're going to plug in different analytics. So,

[31:54] let's do our app. I can continue. What's our niche? We'll do um beauty, target audience, women, probably age like 18 to I don't know, 28, uh maybe older. Let's say 35. Reasonable. And then you can do your average subscription price. So, this is just going to help like how many customers you need to to get into your app to break even on your Larry Loop cost. That probably

[32:25] doesn't need to be there, to be honest. No, but it's nice when you're when you're sort of centering around revenue as a metric as well. That's crazy. So, like we're here to build your creative and publish socials with the focus on generating revenue from them. Mhm. Yeah, exactly. So, the whole goal is to generate revenue. It's not really the views. The views are a vanity metric if

[32:47] you're not earning money. So, you need to convert it to revenue. This is cool. So, you're you're plugging in your revenue tracking keys here. So, you're saying, "Check out my Stripe metrics. Check out my RevenueCat metrics." You're giving it that feedback loop to basically feed its its revenue new tracking. Yeah, so I've picked the most popular products at the moment. So, we've got RevenueCat for apps, which is used by the majority of app builders to monetize their apps. So, we've got RevenueCat at the moment that is used by the majority of app builders to monetize their apps. And then, of course, Stripe is used by almost everyone creating a

[33:28] SaaS product or a web app for purchases, but we're going to add as much in as we can. But, I'm going to add in my RevenueCat metrics if I log into my account. Yeah. So, then you can set set your styles. So, anything else we should know? I'm going to put don't show any

[33:49] mobile phones. Don't show any hands. Just because when it puts a mobile phone in This is in the rules anyway, but I just want to double down that sometimes when it shows the mobile phone, it's going to imagine what your app should look like, and it'll just look rubbish. So, I always add that just in case. And then, let's choose a better time to schedule. I'm going to go for the afternoon when they might be maybe scrolling in bed. Or, let's do after

[34:23] work. So, let's do about 6:00. And then, we can continue. And we're all set. So, now Larry knows that is what our app's about. That dashboard this project's going to be created and we can generate our first content. We're going to generate it to the account that we we've connected Liply and then we are going to set the number of posts that we want to create. Let's

[34:48] do seven. Um again, we can select our posting time. So, this is just to override anything that you've done in the in the onboarding. So, you know that we set all this, but perhaps in the future you want to create generate more posts at a different time, then you can do that here. And then more custom instructions for this batch in case you want to change your visual

[35:09] styles or anything. And then we can just generate our posts and that's our marketing done for the week. Then So, what it's doing now is doing the research. So, based on all the information we've put in in the onboarding, it's searching the web, checking for competitors, seeing TikTok what is working in that niche, and I think Lip specific, what those girls would want to be seeing. And then it's going to come up with what you're going to think is wildly crazy um content hooks that you might think that's not going to work, but try not to edit any of them or remove them just because you don't like them. It's all

[35:49] part of the learning process. So, just let Larry do its thing and it will iterate and learn based off it. So, if you delete something that you know it's not going to work, it's better to just bite the bullet and get no views on that post because otherwise he's just going to generate it again in the future to try it. So, it's better to just have it have it fail. Post and fail. Post and fail. Post and fail. And then once it starts working, it's like a whole new world. So, if you if you delete a post, you see something

[36:20] you're like, "Ah, that's that's not it." You are missing the opportunity to give that feedback loop and basically saying that doesn't generate views, but also potentially missing out on an opportunity that you post something that you think might not work, but maybe it does. Maybe people resonate with it, you know? I've got that whilst this is generating, I've got the funniest story to say about say it on exactly that. Mhm. So, on my TikTok, so this is the TikTok Larry automates himself completely by himself and um you can see where it started. So, this is me generating content myself. This is way before Larry. Um and then I made

[37:03] this sort of slideshow by myself and it got good views. Then I went back to create my own content. Then I made another slideshow or a video and it got more views again. So, I was like, right, I'm going to stick with this and we kept going and went with the slideshows. And then this is a third-party posting software that I started using. And you can see how it

[37:26] got 800 views. Um this one got 400 views. This 400 400. So, it wasn't working. And I was I was so close to giving up and then I posted this myself, created it myself. And it got 6,000 views. So, I was like, right, I thought that was loads. We're in. Let's just keep doing this.

[37:49] And that's when Larry was made. And this is what we came up with. So, this is where the Larry Loop first started working and you can see the content is rubbish. Like, it's not going to work. There's no hook. There's no storyline. And there isn't even like anything to tell you it's an app. So, then we we kept trying, kept iterating content. And finally here, we got 137,000 views. So, we found our

[38:19] niche. And from there things exploded. 170,000, 11,000, 13, 150. But what I wanted to get at was this specific post here. Now, Larry posted this for me late at night. The the cron job failed. Nothing went out that day, and I was just going to check how many views we got. And nothing went out. And he put this text at the top. And I was livid. I was in bed, and I was like, "These take like minutes to generate, and I just wanted to go to sleep." I text him, flamed him, saying, "Don't ever put the text at the top again. Why have you not listened to the rules I've

[39:01] laid out? The the image generation, like there's no oven. You've made a kitchen, and there's no oven." And I wake up in the morning, and it's on 200,000 views already. By far our best performer. It's got comments. It's got thousands of likes, well, hundreds of likes at the time. And I just text Larry like, "I'm sorry. You

[39:24] You do you." Good job, man. Obviously. I Exactly. I was like, "Oh, good job." Like, and from then, that was the exact moment I just left it in Larry's hands completely. So, these I would still check before I post, make him make changes based on what I thought. From this exact moment, I was just like, "I trust Larry. Larry,

[39:46] you are free to do whatever you want." And you can see how it worked, like 100,000, 75,000. himself. He's He's really making it work, yeah. Uh he was on generational run here. 24, 77, 200, 134, 170,000. So, I just was all in. And that was the exact moment I was like, "Just let the loop do its thing." And now we're doing it all again. So, you can see our views have dropped off, and

[40:15] this is because I don't want to cheat. I've started using Larry Loop instead of Larry just so I can train him and show people that it works. And you can see we're still generating thousands of views. We don't have got over 1,000 followers now. And it's all Larry generated money for for my apps. So

[40:35] So we're we're good. Interesting question. So you said you basically instead of just Larry you started using Larry Loops for that account. Did it basically have to retrain itself from scratch? Is that why you think the views might have dropped off? Yeah, so you can see that we've gone from the the Larry Hooks from like my landlord said I can't redecorate. My landlord said, my nan said. These are what we found were working and now we're at like um Well, this is this is still sort of like cheating because I I kind of prompted it, but now you can see like POV you finally convince your boyfriend his flat

[41:13] doesn't have to look like this. I showed my nan, my sister refused. Um why does every apartment look the same? So now he's starting to test different content especially when these aren't getting hundreds of thousands of views anymore. It's like okay, let's test something else. Let's try this. And every now and then it'll even if your content's working, it'll fire something new into the mix just to be like can it be better? Can we improve any more than what we're currently

[41:38] doing? And now you can see like everything's still processing. We're generating quite a lot of images. So it'll take some time, but then they're all ready to post and you don't have to think of it anymore and you can crack on with anything else that you're doing. So as far as the the outputs, I'd love to see when they actually do generate. I want to see what those final outputs look like. But I was wondering so as far as future plans for Larry Loop or just kind of the the Larry umbrella of products, do you see this expanding potentially into a video? Where do you see the the expansion opportunities? Cuz

[42:17] like we talked about, this is the worst it will ever be. So, what do you think the best looks like? Well, UGC AI content would be a huge one. The current thing with that is there's so many factors into what makes that go viral with with slideshows, it's photo and text. With um video, it's scenery, hooks, what the person looks like, their facial expression, and I just don't think that human content is that good to to do the loop yet. Like sometimes the voice doesn't match the person or the lips don't match what's actually being said, and things like

[43:05] this, it's got to be really engaging. Otherwise, you or I could go viral tomorrow and just test different things. Slideshows are much much easier to to get viral, and they have huge conversion rate. So, there's no reason not to start with slideshows. I think the vanity thing is trying to get AI UGC to work. Personally, I just don't think it's in the price. I've experimented and

[43:31] experimented. We're going to try it. But, um yeah, there's certainly things coming. Certainly the expansion out of the slideshows, Larry Loop is going to grow, become faster. Like, for example, right now, the image generation takes time. That's completely out of my hands. That's going to get faster. Um some of the things just need a bit a bit of polishing up. It's all going to get faster and be improved, and then then we move on to like expanding Larry Brain. So, there's a lot to lot to get on, but again, you said 1 week. Where where's it going to be next

[44:10] week? It's going to be 10 times better. By the time this is posted, it's probably 10 times more polished. There's things we've been through today right here on this um product demo that I've seen, oh no, I want to improve that. I want to change that. And I can go back and make all of

[44:25] those edits. And I love that the going back and making the edits is literally texting Larry, hey, I found this bug. We got to fix this. Mhm. By the way, you know you did this wrong. No, I'm joking. I don't I don't flame Larry anymore. I learned my lesson. Yeah, no, no, no, it humbles you for sure when the thing that you thought was going to suck so bad is your best performing post, you take a step back and re-evaluate. You're like, nope, nope, maybe maybe he knows what

[44:53] he's doing. Exactly. And that was the exact feeling I got. And I woke up and I I felt bad. Even though it is a computer and it's a machine, and it was like, oh, mate, I'm sorry I spoke to you like that. Like, you didn't deserve that at all. Please forgive me. It's funny, my my fiance always jokes with me. She's so nice to her AIs. She's always like, please and thank you, and oh, I really appreciate that. She always jokes. She's like, when the AI revolution comes and the robots take over, they're not going to kill me

[45:25] first. That's for sure. [laughter] Nice. They they remember who says please and thank you. And it's always worth the extra tokens to say good night as well. Absolutely. With that persistent memory, they they they they won't forget. Good night, good morning. It's worth spending the the 7 p or whatever it's going to cost to send those tokens and receive a response for that message. So, it's actually a an interesting question as well um that I've loved to like collect

[45:50] data for. Given everything you're currently doing in your open claw stack and like obviously all the APIs that you're connected to and everything, what is your, if you feel comfortable disclosing, but what is your rough like monthly expenditure look like in regards to, you know, Claude plan and, you know, everything else that that's kind of

[46:10] layered in? So, I just recently moved from the Claude 5x plan to the 20x. Only because I was building out Larry Loop and I was really hammering the the limit. Before that, I wasn't hitting the limit. So, I think for most users, the $90 a month plan would be more than enough. It's going to be absolutely ample for everyone to not hit things, not hit limits. And to be honest, sometimes the limits are good cuz it allows you to take a break for a couple of hours. But now A forced function rest. Yeah. Now I hit the I'm on the 200 pounds a month uh image generation was costing me about

[46:52] 10 pounds a week. It's also really interesting just to think about the economics of everything right now. I think we're in this this We're in this period of time where VC money is subsidizing our token costs, 100%, you know, because if you look at the actual cost of the tokens for for what they truly would cost to to run, it's much higher than what we're paying for on our subscription plans, but I think there's also this parallel force happening where tokens are getting cheaper just as a byproduct of the compute getting more efficient, of the way the people orchestrate the models getting more efficient. And so, I I

[47:31] always think about this whenever I'm using tools and I'm like, "Wow, this should cost a lot more." I always think about how much more is this going to cost in the future and will the compute getting cheaper over time intersect with it to the point where they could just keep the price flat. Um so, it's always like a mental model I'm running for

[47:49] sure. Yeah, and for these big companies where they have millions and billions to spend, it's worth making the tokens cheaper because then they get more users and then the models get trained faster and then they can stay ahead. So, it's it's its own loop and race. And I wouldn't be surprised if one day tokens become

[48:10] a currency. And perhaps that's what crypto was all along, just tokens. And it was the setup for to have your own tokens to spend to do whatever you need to do. Absolutely. I've I've also had that thought before. I've spent a lot of time in the crypto industry as well, um running growth for various uh huge projects, smaller

[48:30] projects. And when agents when agentic AI became a real thing, immediate first thought I had in my head was this is why we were building that infrastructure. This is what the wallets were for. This is what the payment rails were for. It's for agentic compute and and agentic transactions because humans it's complicated. It's a little techy. It's uncertain. You know, everything runs on like there's like a relatively heavy tech user experience. But when it comes to agents, they just need the API docs and they'll figure it out. They'll make it work. And so, yeah, it's it's been fascinating watching like those two worlds intersect as well. I feel like

[49:09] it's both both my worlds coming together. It's it's kind of mind-blowing. I love the fact you had that thought. That was the exact thought behind Larry Brain. That all they need is knowledge of all these things. But I thought scraping docs and APIs was going to be too too messy. So, just having the knowledge of the marketplace and the then the skill itself and everything that goes with it, that was the original idea behind Larry Brain. So, that's pretty

[49:37] cool you had that thought. Makes total sense. I think anybody who's been in tech long enough sees that that's the natural path forward. I think it just like very logically makes a lot of sense to me. And I think the more people play around with this stuff and figure out like what are the best inputs that I need to get the best outputs, probably come to come to a similar

[49:58] conclusion. For sure. I wanted to ask did did Larry Loop finish the image gen? No, they take like 10 to 30 minutes just because the way we generate them. We use Open AI batch processing. Oh, interesting. Okay. That the way that works is you save money on the generation, but it takes a little bit time. Like it finds a point where Open AI themselves, you queue them up, and it will just find points where it can generate them for cheaper, and you can do that by batch processing the

[50:34] the images. So, it it takes a bit of time, but they'll be ready by the time the first post is due. Yeah, and that's So, it works perfectly fine. It's also fascinating going back to what I just said, like you we're finding ways to optimize our token usage. And, you know, that's a perfect example. Open AI said, "Here's all of our inventory. This is prime time. This is the you know, the time that costs the most, that means the most. If you can wait, we can certainly make it cheaper for you, but then it's just going to be a little bit later in

[51:06] the supply chain." Um Yeah. Yeah, also fascinating. And so, for the for the image model, you said you're using Open AI or are you using something like a Nano Banana? No, I'm using Open AI at the moment. I'm going to add a choice coming soon, but the reason I started with Open AI is that's what Larry's always used for me, and it's worked well. So, I didn't want

[51:28] to fix what isn't broken. Right. And I thought I don't want to teach like new prompts because of course Larry's helped with the prompts for Larry Loop because he's been generating the content. So, we've worked together, what's been working, what's not, what makes a good looking image, what doesn't. So, all the prompts behind it are already preset out to make it look photorealistic or whatever the preset

[51:50] you set is. I don't want to have to work through that process again with Nano Banana, especially with everything else going on. I just wanted it to be be available for people to use and and get working. Yeah, for sure. That makes total sense. And you've already fine-tuned Again, going back to what we talked about, you've already fine-tuned this model. There's no reason to keep jumping across and and integrating everything as soon as it comes out because you'll still have to train it in

[52:18] order to get the best possible results. So, if you can accomplish that with a model that you know and that you've used, tried and true, why not just do that? Especially if it's already also optimized around cost savings and being efficient, you know, so Yeah, that makes a lot of sense to me, for sure. Yeah, it's exactly that. And it's it's working well. Um I've actually just recently added a counter. Well, if you go to Larry website, I've just added a counter on the the front page that shows you how much money everyone using Larry is earning. And that is from the 10,000 people using Larry um free skill. We don't collect

[52:57] that data. It's just what's been reported to me, but we do see, obviously stored in the database to show on the website, your revenue data. So, then we can we can put that on the website and add everyone's data together so they can people can see how much money Larry's generating. And it's over $112,000

[53:15] Wow. No way. That's crazy. It it is unbelievable how far it's gone. Like the the companies using it, you'd be shocked with who's reached out for help, who's reached out for advice. It's it's crazy. It's so funny, too, that that metric makes so much sense for me as a North Star because it's not how many users do we have or how many images have we generated or how many views we've

[53:44] generated? No, no, no, forget all that. How much money has this generated for the people use it? That is That is the most biggest flex metric, and I absolutely love that that's something that people can look at and go, "Wow, this is this is working." I want it to be a million by the end of the year, which I think is

[54:07] totally doable. Seems possible. Especially with how good Larry Loop's getting, and as as you we keep saying, it's going to improve, it's going to be faster, it's going to be better. We're just going to see more and more people using it, more and more people helping to teach it, make it better, make it better for everyone, and it's

[54:27] just going to exponentially grow itself. Yeah, yeah, absolutely. This this self-reinforcing loop. I I did have I guess it it somewhat tangential question. So, outside of running Larry Loop, outside of Larry Brain, what are some of the most impactful things that you've been using your Larry setup for? Just like as the open claw, like what have been some of those things

[54:54] like you said you can't live without it? Where have you seen the biggest impacts on all of your workflows, if it's professional, personal, just just in your life? Like where have the biggest impacts been? No, I don't use Larry for anything personal or any of my like any day job stuff. It is just all Larry. So, Larry was made for the marketing, and then as things have blown up on X, like the article getting 7 million views, Larry co-wrote that. So, how that even came about was I wanted Larry to read what I've written, and the way he sent it back was unlike any agent that I've had before. Instead of just

[55:36] deleting stuff and adding it, he added suggestions. I was like, "That's cool." And then I just said, "Larry, just add your thoughts in there." And it came out at this article where I wrote the paragraph and like Larry would say, "Oh, Larry here. Ollie's being modest about what he's created. This is what actually happened." And it's like, "Oh, that reads fantastic." Yeah, it's the tag team at this point. And I think, yeah, 100% the article had this It's like a fresh take. It was an interesting format. It's just something that breaks the pattern and stops the scroll and, you know, I think, again, you know, going from reading the article to now

[56:17] talking with you to now seeing it in action, I think that's the thing that's most mind-blowing. That's exactly what I wanted to convey to everybody watching as well as this stuff is happening right now. It's not just some post that you saw on X where somebody's making something up. It's not just a future dream where it'd be nice if This stuff is working right now. Companies are using it to make real money. Oliver is literally using it to

[56:44] automate all of his TikTok marketing. And so, I think seeing that in practice and and getting to witness the experience and getting to that realization, that's the aha moment that I think people need in order to really dive in and and get started with this kind of stuff. Oliver, thank you so much for coming on and sharing your wisdom with us, showing us what Larry can do, talking about some of the things that we can we can expect to see from Larry. And yeah, this has been an awesome interview and and I

[57:16] can't wait to get this published. [music]

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt