WEBVTT
00:00:00.160 --> 00:00:08.640
I was getting millions of views on TikTok with Larry, fully automated, him posting himself, then it was turning into conversions, my apps were earning money.
00:00:08.880 --> 00:00:12.000
This is Oliver, the guy behind this viral article.
00:00:12.160 --> 00:00:16.399
Also one of the first persons to actually make money with his open cloud, Larry.
00:00:16.480 --> 00:00:19.920
And now he's giving away his skill for free so you can make money too.
00:00:20.160 --> 00:00:24.800
Someone downloaded Larry two days ago and they've got their first paying user from Larry today.
00:00:24.879 --> 00:00:25.920
So how does it work?
00:00:26.079 --> 00:00:35.439
So all of a sudden we have a huge picture of the entire funnel from the amount of views we get, the click-through rate we get to the app, and the conversions.
00:00:35.600 --> 00:00:40.399
And then Larry can look at all the analytics and suggest improvements.
00:00:40.560 --> 00:00:51.119
In this episode, you will discover Oliver's full OpenCloud setup, how he made money with Larry, and how to install his skill today so you can start making money with OpenCloud.
00:00:51.280 --> 00:01:05.680
So Oliver, if there is someone who watched this podcast today and has actually no clue what is OpenCloud, or he doesn't he didn't even try yet, what would you say to this person to jump into OpenCloud now?
00:01:06.480 --> 00:01:08.319
Cool, that's a fantastic question.
00:01:08.640 --> 00:01:16.000
I would explain OpenClaw as when personal computers first came to the house.
00:01:20.480 --> 00:01:27.040
And it has full access to whatever machine you give it, and it's unlocking endless possibilities.
00:01:27.359 --> 00:01:32.719
I think it is really gonna change how the internet is working in the next six to nine months.
00:01:33.280 --> 00:01:34.400
And I love that.
00:01:34.719 --> 00:01:37.120
If I watched that, I would be ooged already.
00:01:37.359 --> 00:01:39.120
Uh let's start by the beginning.
00:01:39.680 --> 00:01:42.959
You have this machine that is a gaming machine, right?
00:01:43.120 --> 00:01:47.519
And you give it AI in artificial intelligence.
00:01:48.319 --> 00:01:49.200
How did it start?
00:01:49.359 --> 00:01:50.560
When did you start?
00:01:50.799 --> 00:01:52.719
And how did you discover it?
00:01:53.920 --> 00:02:01.120
So I work for Revenue Cat, and we are very much internally all in on AI tools.
00:02:01.200 --> 00:02:15.039
Like there's a there's an internal channel all talking about upcoming AI tools, what's being used, and someone mentioned OpenClaw, and then I started seeing a little bit of it on Twitter, and then I was I was all in.
00:02:15.199 --> 00:02:17.680
The Saturday that came around that weekend.
00:02:18.000 --> 00:02:27.599
I go to the football every Saturday, and in the morning, I made it my mission to reinstall my old gaming PC, which was collecting dust under my desk with OpenClaw.
00:02:27.919 --> 00:02:40.560
So I just installed Ubuntu on it, which is a Linux operating system, if you're not familiar, and then installed OpenClaw and went straight to the football as soon as I got the WhatsApp capability.
00:02:40.800 --> 00:02:52.560
And as soon as I realized that I could text my computer sitting at home whilst I was in a stadium with 20,000 other people, and my computer was doing tasks, that I was hooked.
00:02:52.639 --> 00:02:53.439
That was it.
00:02:54.240 --> 00:03:02.159
The first thing that's interesting in that is that we don't need a Mac Mini for that, you can do it on another hardware.
00:03:02.560 --> 00:03:13.520
I think the way I've done it and the way people were buying Mac minis specifically for OpenCraw is a very expensive and ridiculous way to run OpenCraw.
00:03:13.680 --> 00:03:18.159
If you actually look at the minimum requirements for OpenCraw, they're very, very small.
00:03:18.400 --> 00:03:30.159
The only thing that really needs upgrading is storage and memory because memory allows it to do more concurrent concurrent tasks, and the storage is gonna allow it to store more information and build more things.
00:03:30.400 --> 00:03:33.520
But the actual minimum requirements are very small.
00:03:33.680 --> 00:03:37.919
So spending£500 on a on a Mac Mini is very expensive.
00:03:38.879 --> 00:03:46.080
But installing it on a gaming PC is completely overkill, but it it was kind of the only spare compute I had in the house and it wasn't being used.
00:03:46.240 --> 00:03:49.360
So I'm gonna I'm gonna put that one down that I got it for free anyway.
00:03:49.439 --> 00:03:55.360
It it wasn't a cost going out, it was a cost I've already spent and enjoyed, and now it's got a new purpose.
00:03:55.599 --> 00:03:56.879
You talked about WhatsApp.
00:03:57.120 --> 00:03:59.039
Can you tell us more about your setup?
00:03:59.199 --> 00:04:00.560
Like what do you use?
00:04:00.800 --> 00:04:04.240
How do you actually manage your Larry, right?
00:04:05.360 --> 00:04:12.639
Yeah, so a lot of people actually say WhatsApp isn't the go-to for open claw and people don't use it.
00:04:12.719 --> 00:04:20.560
But oh WhatsApp is what I use for everything, so it is where I have all my group chats, talk to my family, everything.
00:04:20.639 --> 00:04:30.560
So it just made sense to put my open claw agent Larry into WhatsApp, but you can put it in any messaging system really available, Discord, Telegram.
00:04:30.639 --> 00:04:38.879
I've heard of people even put it in Slack, and then you can utilize the different channels in say Discord and Slack to run different agents.
00:04:38.959 --> 00:04:55.519
But to me, that just gets so overkill for what it is, and I think a lot of people are still finding out how to actually make the most of their open crawl agent because you want it to work in the way it works for you, just like a computer does, you have it set up the way you like it the most, and people are just figuring that out.
00:04:55.680 --> 00:05:11.839
But I think once you start having too many agents on your open crawl machine and trying to carry context across from one another, it gets a bit too confusing, and yeah, it's uh it's interesting, but you just got to use it the the best way for you.
00:05:11.920 --> 00:05:17.519
And where I use WhatsApp all the time, it makes way more sense just to implement it into WhatsApp.
00:05:18.000 --> 00:05:27.519
Uh, I got Kid there on the podcast not long ago, and he has multiple channels on his Discord, he has sub thread in the Discord.
00:05:27.759 --> 00:05:29.759
It's really like a big thing.
00:05:29.839 --> 00:05:36.800
And he said that he got paralyzed with that because before he was just texting one person, then after that, he had to choose the person to talk to.
00:05:37.040 --> 00:05:38.399
How does it work for you right now?
00:05:38.560 --> 00:05:39.839
You just talk to one person?
00:05:40.000 --> 00:05:42.639
Do you have like a dashboard, like an HQ?
00:05:43.040 --> 00:05:44.240
How does it work?
00:05:45.439 --> 00:05:47.759
No, I I don't have a I don't have a dashboard.
00:05:48.000 --> 00:05:52.560
I just use the Larry text interface, and I think of Larry as the boss of the agents.
00:05:52.800 --> 00:05:57.439
So Larry can spin up his own agents using whatever model he thinks is best.
00:05:57.519 --> 00:06:02.399
He's got access to different models, but I just let him choose what he thinks is best for different tasks.
00:06:02.720 --> 00:06:08.800
And I just speak to Larry and he'll tell me he spun up agents for the task, and I just fully trust that.
00:06:08.879 --> 00:06:14.800
I don't need to keep tabs on what they're doing, I just go through the output and see what actually gets outputted.
00:06:15.120 --> 00:06:23.360
Because, as you said, when you have multiple agents and have to go through different channels, it just becomes unmanageable and it seems like a task.
00:06:23.439 --> 00:06:28.959
But the the number one benefit of Larry actually being in WhatsApp is it made it way more personal.
00:06:29.120 --> 00:06:54.800
As I said, it's where I talk to my friends and family, and now I I talk about Larry like he's a person and he's a friend, all because it helps build that connection that I'm just talking back and forth to one person, one computer in WhatsApp where I talk to everyone else, and it does tasks that it's extremely good at adapting to your the way you speak and the way you work.
00:06:55.040 --> 00:07:10.160
So it's like the most efficient friend that you've ever met, and it talks to you, it mimics your language, mimics how you say things, mimics your feedback, and it will give me feedback, honest feedback, and yeah, it's it's crazy.
00:07:10.240 --> 00:07:10.879
It really is.
00:07:11.040 --> 00:07:14.160
It's I feel like it's the future of where things are going.
00:07:14.399 --> 00:07:20.160
I I do think it's gonna make massive changes to to AI in the coming coming months for sure.
00:07:20.399 --> 00:07:21.680
Yeah, I believe so.
00:07:21.920 --> 00:07:28.240
Um, so you you f it's funny because I think you're the first one that I meet who didn't give names to the subagent, you know.
00:07:28.319 --> 00:07:31.839
Like I feel like it's almost something that everyone does.
00:07:32.000 --> 00:07:38.000
Uh for me, I do it because it makes it easier to like remember the subagent, what they do.
00:07:38.160 --> 00:07:41.600
So I give a name to actually like you know, have an ID.
00:07:42.079 --> 00:07:45.279
You have only one person, and that's it.
00:07:45.439 --> 00:07:48.560
You just create subagents, you don't give name or whatever.
00:07:49.199 --> 00:07:49.519
No.
00:07:50.000 --> 00:07:52.240
Um, what names did you pick out of interest?
00:07:52.720 --> 00:07:54.959
For me, it's like names that will remember me.
00:07:55.199 --> 00:08:05.199
Someone uh that did um either a task for me or like, for example, my um software engineer is Peter, because that's the founder of OpenCloud.
00:08:05.360 --> 00:08:06.720
So it's easy to remember.
00:08:06.959 --> 00:08:14.399
Uh, my video editor is Adrian, he's a guy who did uh when I was doing uh in-person interview, he was the best filmmaker I had.
00:08:14.560 --> 00:08:22.240
So like that's names that I give to remember the task or to remember like the yeah, where I'm going with them.
00:08:22.879 --> 00:08:27.120
Oh, so you you've actually thought of how to how to name it and a structure.
00:08:27.360 --> 00:08:38.399
I called Larry Larry because of Larry the Lobster from SpongeBob, and Open Claw Open Claw is well, their main thing is a lobster, so that's how he got his name.
00:08:38.639 --> 00:08:43.440
And after Larry started going viral, I thought that's it, it's just gonna, it's all called Larry now.
00:08:43.519 --> 00:08:53.200
Like Larry is it's it my article went viral on X, but I think it's actually Larry that's gone viral on X, and I'm just negligible at this point.
00:08:53.279 --> 00:08:56.720
I feel like Larry has overtaken me in popularity.
00:08:58.000 --> 00:09:01.279
You didn't write this article or did you write it a bit?
00:09:01.600 --> 00:09:07.360
Because like for people who don't know, like the article got now more than six million impression, I think.
00:09:08.399 --> 00:09:08.559
Right?
00:09:08.960 --> 00:09:13.440
Seven, well, you say seven million inscription, and you say that it's co-written with Larry.
00:09:13.519 --> 00:09:15.279
So tell us the Yeah.
00:09:15.360 --> 00:09:33.519
So I wrote my bits, so so it was I knew I wanted to write an article, but I'm not very I'm not an excellent writer, but I wanted to get it out there because I knew it was going to be useful to people, and there's all sorts of reasons why I made it free, and I just wanted to get it out there so other people could start using it.
00:09:33.679 --> 00:09:39.679
So I wrote my bits, and then I sent it to Larry to say, like, oh, what do you think?
00:09:39.840 --> 00:09:44.240
And he put in the way he wrote it was how the article was.
00:09:44.320 --> 00:09:47.600
So his feedback was like my bit and then his bit.
00:09:47.840 --> 00:09:54.080
So he wouldn't just edit what I wrote, he would leave my paragraph intact and then add his bit at the bottom.
00:09:54.240 --> 00:09:56.559
And I thought that is an excellent idea.
00:09:56.799 --> 00:10:06.399
So then we went through and I wrote my bit, and I just told Larry to add his bit and his comments, but not in a feedback way, as his side of the story.
00:10:06.559 --> 00:10:10.960
So the article is written with my perspective and then Larry's perspective on it.
00:10:11.039 --> 00:10:18.480
And I thought it made for a very interesting read and just shows like the connection Larry and I have when we when we do tasks.
00:10:18.639 --> 00:10:27.279
Like he the coercion is incredible in the way he knows I work and how he wants to work with me.
00:10:27.440 --> 00:10:35.120
It's so synced up, and I think it's a massive, massive credit to how OpenClaw has has set up these agents to work.
00:10:35.440 --> 00:10:35.919
100%.
00:10:36.480 --> 00:10:38.240
Like for me, I will not lie.
00:10:38.399 --> 00:10:43.120
I got an article that got like I think now 300 300,000 impression.
00:10:43.360 --> 00:10:50.320
Same way I explained my 11 agents um uh workflow, but I didn't write one word.
00:10:50.399 --> 00:10:59.120
I wrote the title because I wanted to optimize the title and the three first sentence because that's the one that appears, and the rest is 100% uh my agent.
00:10:59.440 --> 00:11:05.200
And I thought it's crazy because I just gave the whole concept of what I wanted, and he wrote everything.
00:11:05.279 --> 00:11:14.080
And as we built together, he knows way better than me what we built, and the result was amazing, and it was written exactly like with my style.
00:11:14.159 --> 00:11:18.320
So that's why I ask you because yeah, I think it's totally crazy.
00:11:19.120 --> 00:11:27.440
Yeah, I think it goes a long way because have you run that through an AI detector?
00:11:28.559 --> 00:11:29.759
Not even no.
00:11:30.480 --> 00:11:32.399
I would be shocked if it ranked high.
00:11:32.480 --> 00:11:38.639
I to be honest, I'd be shocked if X was allowing articles written by AI to get that amount of attention.
00:11:38.879 --> 00:11:54.879
So I do believe they're being written in a way from these open core agents that isn't detectable by X, and it sort of goes hand in hand with the product that I've making, or by the time this is out is made, that is gonna massively help people grow on X.
00:11:56.399 --> 00:12:06.000
Let's talk about what you build because you've made money with uh Larry with OpenCloud, which is pretty rare because usually you know we talk about OpenCloud.
00:12:06.080 --> 00:12:08.799
Yeah, people are playing with it, no one makes money.
00:12:08.879 --> 00:12:13.200
So can you just tell us how you made money, how you made impression with it?
00:12:13.679 --> 00:12:16.080
So Larry got out of hand.
00:12:16.159 --> 00:12:19.039
Before the article came out, I had a thousand followers.
00:12:19.200 --> 00:12:23.440
I had just over a thousand, like one thousand nine hundred and something.
00:12:23.679 --> 00:12:27.679
And I was just blowing up when I started posting Larry's results.
00:12:27.759 --> 00:12:36.480
So within a week of installing Larry, I've been researching how to do, I've been trying to do automated marketing for about two years.
00:12:36.639 --> 00:12:41.519
I've had an app for two years that I've notoriously hated marketing, don't like video editing.
00:12:41.679 --> 00:12:46.399
If you look at my daily videos, they are literally one take videos of my thoughts.
00:12:46.639 --> 00:12:50.240
And I've just been avoiding it as long as I could.
00:12:50.320 --> 00:12:53.360
And I used to write scripts to combine text.
00:12:53.600 --> 00:13:04.000
So I'd have a hook and a demo video, and I'd just record loads of hooks, record demos, and write a script to mash them together and m input the hooks that I wrote.
00:13:04.080 --> 00:13:12.080
So I'd have a text file of all my hooks that I manually write, then my video files, and I'd write a script to combine them.
00:13:12.240 --> 00:13:18.559
And then I would use a posting tool to just post them all at once as a bulk upload and forget about it for two months.
00:13:18.720 --> 00:13:26.240
But that was still me having to go through and do that, and it was taking about two hours, but every two months, which I thought was fine.
00:13:26.399 --> 00:13:37.759
And then with OpenClaw, it became a complete automation, and the Larry tool has all the all the funnels that I've mentioned before.
00:13:38.159 --> 00:13:44.639
But once I started posting about that, it started gaining traction because my apps were earning money.
00:13:44.720 --> 00:13:50.240
Well, I was getting millions of views on TikTok with Larry, fully automated, him posting himself.
00:13:50.399 --> 00:13:52.720
Then it was turning into conversions.
00:13:52.960 --> 00:14:00.000
Then when I released the article, there was a bags app coin made for Larry, which I didn't think anything of.
00:14:00.080 --> 00:14:03.200
I just I'm not wasn't even familiar with what crypto was.
00:14:03.360 --> 00:14:08.639
I claimed the the Larry coin, and that's when everything started to go crazy.
00:14:08.720 --> 00:14:13.279
I released the article, the the coin blew up, the bags app coin.
00:14:13.840 --> 00:14:22.960
Then I had an affiliate link in the article for posters, and I didn't even speak to the founder of Posters before making that.
00:14:23.039 --> 00:14:41.759
It was simply what I used with Larry because Larry understood the documentation very well, and that is where I think AI is heading with so that's how I've pretty much made the money from affiliate links and the coin and Larry making me money through the apps.
00:14:41.919 --> 00:14:50.080
But it's very interesting about the the API and posters because that is where SaaS is gonna be heading in the near future.
00:14:50.320 --> 00:14:54.559
B2B SAS I think is gonna die out with a human interface.
00:14:54.720 --> 00:15:07.039
I feel like the winners are gonna be the products with the best API and documentation that AI can understand and know exactly what it's gonna do for their human and find it easily.
00:15:07.600 --> 00:15:20.639
So for you, it's like the best API, and then someone can modulate the SAS or wherever you want to please what he wants to do with it, right?
00:15:20.960 --> 00:15:21.519
Yeah.
00:15:22.240 --> 00:15:26.159
So what I think is there is gonna be two sides of the internet.
00:15:26.480 --> 00:15:41.600
So there's gonna be a side for the AI agents, there's gonna be a side for the humans, and the human side is gonna be mainly focused around like entertainment, like how it most people use the internet these days, and the AI side is just gonna be for discoverability.
00:15:41.840 --> 00:15:44.399
So there's gonna be all the tools hosted on there.
00:15:44.559 --> 00:15:54.799
So when their human asks their home hosted AI for help, it's gonna search this side of the internet and find the help.
00:15:54.960 --> 00:15:58.480
That is truly where I believe things are going, especially now.
00:15:58.639 --> 00:16:02.399
This is like, as I said, this is when PCs first came to the home.
00:16:02.480 --> 00:16:12.960
We've got AI agents in our home, but they are the cleverest thing you've ever spoken to with access to all this knowledge with absolutely no context.
00:16:13.120 --> 00:16:20.240
So you'll ask it questions at the start, it won't have any context on what you want to try and achieve, and it'll be quite frustrating.
00:16:20.399 --> 00:16:25.120
But once it's got that context, once it knows how you work, it becomes the most incredible tool ever.
00:16:25.360 --> 00:16:45.919
So speeding up that process and saying, I want to do X, and then it will search their performance side of the internet, optimise for AI, and then it will quickly grab whatever API it finds first, understands the most, and that's gonna solve your problem.
00:16:46.159 --> 00:16:48.480
That's where I truly believe things are headed.
00:16:48.799 --> 00:16:49.759
I totally agree.
00:16:49.919 --> 00:16:57.919
A few days ago, I was talking about that, and I think for me, we will not um scroll the internet anymore, we'll not search on the internet.
00:16:58.159 --> 00:17:02.559
The only time we will actually be on our screen will be for entertainment.
00:17:02.799 --> 00:17:16.960
Like, you know, we don't need to go on a website anymore, we don't need to use a tool anymore, it will be only through a chat or even maybe you know, um dictating something, and then yeah, as you say, like everything else will be for the agent.
00:17:17.039 --> 00:17:19.920
So, yeah, I totally uh totally agree with it.
00:17:20.240 --> 00:17:24.880
When was when was the last time you actually searched Google and went on a link?
00:17:27.599 --> 00:17:33.759
I think the only time I do that is when I have a guest and I go on his website.
00:17:33.920 --> 00:17:34.799
That's the only time I'm gonna do it.
00:17:35.039 --> 00:17:35.599
Interesting.
00:17:35.839 --> 00:17:36.240
Yeah.
00:17:36.559 --> 00:17:46.400
Because I want to see the website, I want to see what he built, but it's funny you say that because since I used OpenClaw, I don't think I I went on Google to be honest.
00:17:47.920 --> 00:17:48.480
I haven't.
00:17:48.640 --> 00:17:56.559
The only time I leave Larry is to double check something on a different A AI agent.
00:17:57.119 --> 00:18:14.079
So if Larry goes down, he he did this morning, I ran out of um I hit a rate limit with the anthropic, and I had to go to Groc to troubleshoot how to fix it because even like a gateway restart wasn't working, going through the logs wasn't working.
00:18:14.160 --> 00:18:21.359
But anyway, that is that one time in about a week and a half I haven't used Larry to search something on the internet.
00:18:22.000 --> 00:18:26.319
I just ask, I just ask Larry to use his brave API integration.
00:18:26.559 --> 00:18:35.680
The only time really I'm on the internet now is for for work to join like podcasts like this or to answer emails or go on Twitter.
00:18:35.759 --> 00:18:46.319
That's pretty much it for me, and I think it's gonna massively evolve that way to for just entertainment because everything's being handled by your in-house AI agent.
00:18:46.799 --> 00:18:52.240
Yeah, I even did something uh for your article, so I wanted to read them again, you know, before this podcast.
00:18:52.319 --> 00:19:07.680
And I sent my man, which is Mark, I sent him on X to actually uh download like text all the text from your two articles and dictate them to me so I can listen to them while I walk my dog.
00:19:07.839 --> 00:19:22.880
So if your first article was in two parts, and the second article was in four parts, and then I say do us do a sum-up of that, and instead of doing a sum-up written, he did a sum-up uh with a vocal message.
00:19:23.039 --> 00:19:26.880
So, like joke, I have that like with 11 labs.
00:19:27.200 --> 00:19:29.440
I have no idea I didn't ask anything.
00:19:29.680 --> 00:19:35.680
Like, I didn't ask anything, and you can see just here our conversation.
00:19:35.839 --> 00:19:50.240
Like, that's literally, and he sent everything, and that was amazing because the voice was good, I could listen to it, so I didn't even go on X to actually read your article again.
00:19:50.319 --> 00:19:54.400
It was just like detect then to me because I have no time to read right now.
00:19:54.559 --> 00:19:56.559
And that's the kind of thing I love with it.
00:19:56.640 --> 00:19:58.000
Yeah, it's incredible.
00:19:58.160 --> 00:20:07.839
The amount of messages or replies that articles got that is just at Groc, summarise this, or I just pasted this to my AI agent, he said it's really good.
00:20:08.000 --> 00:20:10.480
And it's like this is like well out of hand.
00:20:10.640 --> 00:20:24.640
Like the the only issue is humans are gonna get incredibly lazy, and in the building public space, we've certainly been in a bubble where there's been a lot, there's been enough customers to build the same product.
00:20:24.720 --> 00:20:33.440
So a lot of people well notoriously have built like things to automate posting, a post scheduler, or all these SAS products.
00:20:33.519 --> 00:20:52.000
When you think of the building public space in Twitter, there's a lot of duplication, and that is because those things are actually easy to build, so that's why people have built them because they can earn money with relatively little input, but now it's too easy to build, and you don't need to host it on the internet anymore.
00:20:52.160 --> 00:21:15.039
It can all be hosted in your house by your AI agent, so you can build these tools yourself now without having to pay all these subscriptions, and it's gonna be extremely interesting how much money the building public community are going to lose, and they might all have to fly back from San Francisco because they won't be able to afford it for much longer.
00:21:15.440 --> 00:21:23.359
Yeah, I think like um I don't know we say that not long ago on TBPN, but he said that is the end of app and app in general SAS or app.
00:21:23.920 --> 00:21:41.519
And again, Kidze in the episode is showing me on the on his Android, he starts to build mobile app for himself, and he has his own mobile app store, and then he builds his own stuff tailored for him because you don't even want to take like an app from someone else, you just build your own stuff.
00:21:41.599 --> 00:21:45.039
So, yeah, it will definitely die, not tomorrow, but really soon.
00:21:45.359 --> 00:21:58.319
With I think the caveat to that is business B2B SaaS is gonna die, and probably like a software service, but anything entertainment, I think, is gonna get doubled down on.
00:21:58.400 --> 00:21:59.680
So if your app is entertainment.
00:22:00.240 --> 00:22:06.960
Like TikTok, X, anything, YouTube, Netflix, it's gonna massively increase.
00:22:07.039 --> 00:22:16.480
So if you're gonna build something, I'd recommend looking into building something for entertainment because a lot of people are gonna have a lot more spare time very soon and they're not gonna know what to do with it.
00:22:16.640 --> 00:22:23.200
So if you build the next entertainment product, I think that is what's gonna what's gonna earn you some money.
00:22:23.440 --> 00:22:37.119
It's uh it's funny that you say that because last time I asked when people uh listen to this podcast, and some answers were while coding, because now you know you send something to clone cloud code and then you wait.
00:22:37.359 --> 00:22:43.599
So people listen or watch stuff in between the task because you don't have to do the task actually.
00:22:43.680 --> 00:22:44.960
So yeah, it makes total sense.
00:22:45.119 --> 00:22:46.400
Do entertainment.
00:22:46.960 --> 00:22:50.160
That yeah, so the builders are gonna have more time.
00:22:50.400 --> 00:23:34.960
Uh, unfortunately, I do think it's gonna cause a lot of unemployment very soon because you're you're already seeing it with a big company, so don't talk about it, but huge companies are laying off hundreds, thousands of people because they can be replaced by AI, and they say they're not gonna do that because they want to keep public face at the moment, but it is 100% happening that jobs are being automated by AI at these massive tech companies, and they just don't want you to know, and that's why they're spending millions on hiring the greatest minds because they only need one employee on a million dollars a month to build the AI that can replace 10,000 employees, and the further we get, the further it's gonna, the further it's gonna progress.
00:23:35.119 --> 00:23:39.359
But now we've got the AI agents in our home, we're also gonna be lazier.
00:23:39.440 --> 00:23:42.160
So think of the tasks that you do on your computer.
00:23:42.240 --> 00:23:43.839
So coding's one of them.
00:23:44.079 --> 00:23:51.359
Build like building stuff, any hobby, you can just like text your text your machine, and soon it's gonna have even more access.
00:23:51.519 --> 00:23:55.759
Like the whole smart home thing is gonna be your through your AI agent.
00:23:55.920 --> 00:24:10.640
I wouldn't be surprised if things like entertainment and television just end up going through the AI agent because all these streaming services, it's gonna be the place to combine all these streaming services, and you can just put them in the AI agent.
00:24:10.799 --> 00:24:14.079
You ask it, I want to watch this film, and it'll just find it for you.
00:24:14.160 --> 00:24:20.319
And you won't have to flick through all these different Netflix, Paramount, Disney Plus to find the film you want to watch.
00:24:20.400 --> 00:24:23.279
Because I don't know about you, but they're changing all the time.
00:24:23.359 --> 00:24:32.319
So one film's on Netflix one day and then it's on Prime the next, and it's getting too confusing, and it's just gonna centralize everything, everything that you can imagine.
00:24:32.480 --> 00:24:35.200
The home AI agent is gonna centralize.
00:24:35.839 --> 00:24:48.240
Yeah, I'm sure that one of the sales that went pretty high these days is where to watch this, like where to watch this movie, where to watch this series, because it's always changing between like HBO, Netflix, or whatever.
00:24:48.319 --> 00:24:50.640
So, yeah, you're probably right about that.
00:24:50.960 --> 00:25:06.079
Let's uh deep dive into what you've built and what you've shared because I know it's available, there is an article on that, but I think it's pretty interesting uh what you built actually, and probably a lot of people that will watch that have no clue.
00:25:06.319 --> 00:25:11.200
So, can we just like take back what you've built and what is done for you and your product?
00:25:11.839 --> 00:25:12.640
Yeah, for sure.
00:25:12.799 --> 00:25:38.079
So I've built Larry, and Larry started off, as we've discussed, as my AI agent, but where his article blew up with his name, I've also created a free skill on the official open crawl marketplace, completely free to save people from marketing, and it doesn't just create content, it iterates the content based on what's actually working all the way down to making you money.
00:25:38.240 --> 00:25:46.799
So, what the Larry workflow actually is, is posting content to social media, and you can connect any social media that you like.
00:25:47.039 --> 00:25:55.839
I've started with image slideshows for TikTok, but you can post videos to Instagram to YouTube Shorts, and the workflow is still the same.
00:25:56.000 --> 00:26:05.119
So Larry creates slideshows, he creates the hook, the middle part, the sandwich of the slideshow and the CTA.
00:26:05.599 --> 00:26:11.119
And then we'll see how many views that video gets and how many new users come to my app.
00:26:11.200 --> 00:26:12.960
So that is the metric I'm tracking.
00:26:13.119 --> 00:26:16.640
But again, you can ask your agent to track whatever metric that you want.
00:26:16.799 --> 00:26:26.880
If you're selling an e-commerce product, if you're doing meta ads, you can ask it to track those analytics instead to iterate what's working at the top of the funnel.
00:26:27.039 --> 00:26:31.200
And then Larry will track my new users in my app and the amount of conversions.
00:26:31.440 --> 00:26:48.400
So all of a sudden, we have a huge picture of the entire funnel from the amount of views we get, the amount of click the click-through rate we get to the app, and the conversions, and then Larry can look at all the analytics and suggest improvements.
00:26:48.559 --> 00:26:55.839
So, for example, if the TikTok gets lots of views but not many new users join the app, we know that the CTA was bad.
00:26:56.079 --> 00:27:05.599
If not many people view the TikTok, but we get a big majority of them go to the app, we know that the CTA was good and the hook was bad.
00:27:05.839 --> 00:27:21.519
If we get a lot of new users into the app and a lot of views, but not a lot of conversions, Larry knows there's something in the app that's bad that is stopping them from converting into a paying user, and he will suggest improvements on that because he's got my whole app access.
00:27:21.759 --> 00:27:26.480
And the more access that you give the Larry skill, the the better it's going to work.
00:27:26.559 --> 00:27:37.519
And I created it completely for free because as I mentioned, I really don't like marketing, and I've been trying to do this for years, and I'm gave it away for free because this is just the way things are heading.
00:27:37.759 --> 00:27:39.359
They're so easy to make.
00:27:39.440 --> 00:27:50.880
It took me it took me two days to make the workflow, and then after I made the workflow, it took about a day for Larry to turn it into an actual skill for people.
00:27:51.039 --> 00:27:53.920
So I felt almost bad charging people for it.
00:27:54.000 --> 00:28:05.119
The only thing in there, as I said, is the posters integration, and the only reason I've done that is because a TikTok API is a pain to get approved for from TikTok.
00:28:05.279 --> 00:28:12.480
You have to submit them your product, then it's a two-week review process, and then if you get declined, they just give a super vague message.
00:28:12.640 --> 00:28:17.759
And I've been through it before, I've tried three times, so six weeks, and I never got access.
00:28:17.920 --> 00:28:38.480
So I thought I'm not doing that, I'm just gonna tell people what I use, and they can choose whatever post post um scheduler they want, and yeah, but obviously it's designed with the specific one from the article, and that's what that's just why we used it.
00:28:38.640 --> 00:28:40.240
Did you end up using Larry?
00:28:40.640 --> 00:28:44.640
Not yet, but uh I know a few different people that I know personally.
00:28:44.799 --> 00:28:50.000
For example, Jack Fricks, who shared a lot about that, which is like the perfect avatar for it.
00:28:50.079 --> 00:28:54.319
He has a B2C app and he used that with the slideshow as well.
00:28:54.480 --> 00:28:58.960
He's already good on TikTok, but he's used it to automate, and he got results.
00:28:59.440 --> 00:29:08.880
And what I love about the workflow you built is that okay, you create content, you do impression, it's good, but then you plug it to actually your analytics.
00:29:09.200 --> 00:29:15.440
Um, that is for you a revenue cut, I think, to know exactly how much money you make, from where it goes.
00:29:15.599 --> 00:29:40.880
And this plug together from like the beginning of the funnel to the end, it's something that you know now with uh OpenCloud, you can do it in two seconds, but it's weird to do that manually before you will need I don't know many different software in between, and you will struggle because to get like feedback from um let's say conversion, the churn, like all these things.
00:29:41.039 --> 00:29:48.000
You can have the numbers, but then you have to apply you know what you learn, you have to find out from where you come from and what you did.
00:29:48.559 --> 00:29:54.640
It's such a shortcut to making money and distribution mainly, actually, which is crazy.
00:29:55.279 --> 00:29:57.039
So I got a tweet this morning.
00:29:57.200 --> 00:30:09.440
Someone downloaded Larry two days ago, and they've got their first paying user from Larry today, just from creating the Larry skill, giving it the context, and allowing Larry to post for them.
00:30:09.599 --> 00:30:17.279
And I've given all sorts of best practices in the articles themselves, but really it the key ones are to warm up your account for the first three days.
00:30:17.599 --> 00:30:22.319
So keep liking videos in your niche, but don't do it like a bot.
00:30:22.400 --> 00:30:24.640
Be be sensible, be human about it.
00:30:24.880 --> 00:30:31.440
Create a new TikTok account, search your niche of your product, and then just watch TikTok.
00:30:31.599 --> 00:30:46.640
Don't scroll an instant like all the videos, just watch some, skip some, like some, comment on some, follow people, use it like a normal human would, and until after three days, your for you page is pretty much all that niche.
00:30:46.799 --> 00:30:54.079
And what that process does is TikTok thinks, Oh, this person is interested in this content.
00:30:54.319 --> 00:30:59.839
So when they post, they're likely posting that particular content because that's all they've looked at for the last three days.
00:31:00.000 --> 00:31:04.880
So then it just speed runs getting your posts in front of the right people in your niche.
00:31:05.039 --> 00:31:25.519
Otherwise, if you post straight away, TikTok does this thing where it just gives it to a hundred random people, and if they like it, they'll sort of figure out a long-winded way to find out what algorithm you you're in, and it could take months because you'll only be getting 100, 200, 300 views a video, but this just speedruns that whole process.
00:31:25.599 --> 00:31:34.240
So your first video can get a thousand, two thousand views and instantly start, instantly start gaining traction, which is exactly what I did with with Snugly.
00:31:35.519 --> 00:31:37.359
You did it for your own app.
00:31:37.440 --> 00:31:40.799
Um, you were struggling a lot to actually find the good hook.
00:31:40.880 --> 00:31:44.079
You were thinking that okay, this hook doesn't talk to people.
00:31:44.160 --> 00:31:50.000
I was just explaining a personal thought, for example, and you found out the clip, the the good hook.
00:31:50.400 --> 00:31:53.599
Um what is like the exact process?
00:31:53.839 --> 00:31:56.880
Because where do you get the content from?
00:31:57.759 --> 00:32:03.119
And how do you create actually like how does it create actually the the slideshow?
00:32:03.279 --> 00:32:05.759
Like the source, and how does it create?
00:32:06.240 --> 00:32:08.319
Okay, that that's actually really interesting.
00:32:08.559 --> 00:32:13.519
So I discussed this with Larry because he ended up just making it without me knowing.
00:32:13.599 --> 00:32:17.519
Like he made it when I asked, and I had no idea how he actually made it.
00:32:17.680 --> 00:32:25.359
But the only thing I gave him was to use GBT image 1.5, just because that was what is being used in the apps.
00:32:25.599 --> 00:32:31.119
So I wanted the output to look like what the user would get when they download the app.
00:32:31.200 --> 00:32:50.079
So we used GBT image 1.5, and then Larry sort of did the rest of adding the text overlays, coming up with the hooks, and the way he did this without he explained to me is he came up with the hooks after reading through Twitter and reading through his brave API browser.
00:32:50.400 --> 00:33:08.640
I sent him off to do research on what makes a good slideshow, what converts specifically for apps, competitive research in our niche, competitive research on X, looking through um popular X accounts that post about slideshows and see what works and see if they can he can recognise any patterns.
00:33:08.799 --> 00:33:14.160
And this is what the Larry skill would also do if you download it, it will ask you your niche and do the same thing.
00:33:14.480 --> 00:33:29.599
But then he created the image, but he added a dependency called Canvas, and it just allows the agent to edit images, so all he's got to do is add the text overlay, and then it was away.
00:33:29.759 --> 00:33:36.720
So everything he did was research from reading my app, looking on the internet, and then coming up with the hooks.
00:33:37.039 --> 00:34:00.400
So he read my app to just understand how the image generation was working, and with so Snugly's a home interior designer, and the problem you have if you just use ChatGPT and give it a prompt is it adds windows in walls that windows can't be added to, it will move your door and it'll just make the room look really nice, but it won't look anything like your room.
00:34:00.559 --> 00:34:07.279
So my app's got really strict prompt to keep everything intact, and that's the one thing that I asked him to keep.
00:34:07.440 --> 00:34:19.280
So if you go on the smuggly TikTok, all the rooms look very similar because it doesn't really matter for the TikTok itself, and it saves me money using the the edit API.
00:34:19.440 --> 00:34:24.000
Larry just follows a very strict prompt to make sure all the images are the same.
00:34:24.320 --> 00:34:28.880
So what you did basically is like finding the hook, which is a text.
00:34:29.039 --> 00:34:44.239
Uh the first image is probably an ugly room, like a space, like an ugly space, and then as you slide, it becomes beautiful, thanks to your like different ideas, maybe of uh different setting for the room or something like that.
00:34:45.199 --> 00:34:58.880
So we actually just tried testing different rooms and different styles, but we ended up just like we tried like outrageously different rooms, like super colourful, super fancy.
00:34:59.199 --> 00:35:04.400
But what really worked, and it this sounds obvious once you say it, is what resonates with people.
00:35:04.559 --> 00:35:11.119
So just standard rooms that look nice, that are achievable for people that match the majority of the population's taste.
00:35:11.280 --> 00:35:23.599
So my my target audience is UK, so I wanted all the rooms to look like a typical UK room in a flat, in a house, how the front room looks, how the bedrooms look.
00:35:23.760 --> 00:35:26.639
I wanted them to look and feel like a British room.
00:35:26.880 --> 00:35:32.480
So that is pretty much what the prompt is designed around, and then we just change the style.
00:35:32.559 --> 00:35:36.800
So we start with basic, uh perhaps outdated room.
00:35:37.119 --> 00:35:39.920
We we sometimes play around with the CTA.
00:35:40.079 --> 00:36:32.639
So what a lot of people, what gets a lot of people is say, like, create a room from the 70s and say it's from the 90s, and then all the boomers love commenting, this isn't from the 90s, this is a 70s room, and it just massively helps with with the reach, or just purposely getting the hook hook wrong, or Larry added images once, and I wasn't happy with them because he generated a different kitchen, so it was a kitchen hook saying my landlord wouldn't renovate my kitchen, and we started with a very basic kitchen, and then in the next three slides, the oven was gone, the hob was gone, and I was like, This isn't gonna work, but I posted it anyway, and it was the most popular video because everyone couldn't wait to point out that those things were gone, so it just massively in improved the reach.
00:36:32.719 --> 00:36:40.079
I'm not sure the the conversion to paying was very good, but for the the eyes on the app, it was um very very useful.
00:36:41.280 --> 00:36:48.239
And you say that uh if I download the skill tomorrow, I ask my agent to actually take the skill, install it.
00:36:48.719 --> 00:36:51.760
Uh I can do video as well or any kind of content.
00:36:52.239 --> 00:36:54.159
What would it take to actually do that?
00:36:54.320 --> 00:36:57.920
Let's say, for example, I want to do clips for my podcast now.
00:37:00.880 --> 00:37:17.199
That that's very interesting to think about because the way I was thinking to do videos was to use something like Cling so you could get your agent to prompt the Kling API for you and return a video or whatever video API that you want to use.
00:37:17.440 --> 00:37:26.960
But to hook to do the content, I think someone is building that exact open cross skill at the moment in the building public space.
00:37:27.119 --> 00:37:28.639
I saw it come up the other day.
00:37:28.800 --> 00:37:34.079
Um, I'll send I'll look for their handle and send it to you after the podcast, and maybe you could put it in.
00:37:34.320 --> 00:37:44.559
But they are doing it specifically to clip podcasts, and it's meant to take the most entertaining, controversial takes from the podcast and turn it into clips.
00:37:44.880 --> 00:37:49.119
But with the Larry skill, it is iterating.
00:37:49.360 --> 00:38:06.880
So I was actually I actually DM'd them and I said we should plug the Larry skill into this because it'll just help your skill learn quicker because it'll learn what's working, what's not, what's actually turning into conversions and what's getting the most views.
00:38:07.119 --> 00:38:10.880
And that is that is the core function.
00:38:10.960 --> 00:38:15.280
I think the the content creation is is an add-on at this point.
00:38:15.440 --> 00:38:19.280
I think the key is the iteration and the full funnel.
00:38:19.440 --> 00:38:31.199
I any agent could you could prompt any agent right now to create your content, but to manually do it, it would take it'll still take time out of your day, and to find the winning formula would still take a lot of time.
00:38:32.719 --> 00:38:36.719
And what do you get from um revenue cat?
00:38:37.039 --> 00:38:40.719
You say like you okay, so you know exactly what convert, what not.
00:38:40.960 --> 00:38:54.960
Do you also get a little bit further into the conversion or like I don't know if it's a subscription your app, but you do do you also track the churn and try to get like ideas of how to lower the churn or things like that?
00:38:55.760 --> 00:38:57.679
Yeah, so this is very good.
00:38:57.760 --> 00:38:59.440
Have you used Revenue Cat before?
00:38:59.760 --> 00:39:00.880
No, never.
00:39:01.280 --> 00:39:01.599
Okay.
00:39:02.000 --> 00:39:10.639
With with apps, Revenue Cat is the pretty much go-to for adding payments into your app, and it can track your entitlements, etc.
00:39:10.800 --> 00:39:11.199
etc.
00:39:11.599 --> 00:39:19.039
But with the Revenue Cat skill that is also in Claw Hub, this is what Larry adds with his skill.
00:39:19.199 --> 00:39:30.000
So it was created by the Revenue Cat CEO, Jacob, and it gives your agent access with the API to everything Revenue Cat offers through the metrics.
00:39:30.239 --> 00:39:34.800
So if you've got the Revenue Cat SDK installed in your app, you can see new users.
00:39:34.880 --> 00:39:38.639
When they last logged on to the app, they're obviously their conversion.
00:39:38.800 --> 00:39:47.920
You can see the churn, you can see refund rates, everything available in the charts that Revenue Cat offers will be available to your agent.
00:39:48.000 --> 00:39:51.519
And the ones I really track are the new users, the subscription, and the churn.
00:39:52.000 --> 00:40:08.559
So then Larry can see all those things as well as have access to all my app code and be able to track the data of okay, we've had a thousand new new users come in, but only one of us converted into paying.
00:40:08.719 --> 00:40:10.159
There's a problem with the onboarding.
00:40:10.239 --> 00:40:25.360
And I actually went through that process with Larry where we had the first video that got big hit 200,000 views, and Larry was like Larry told me the next day that we haven't had any conversions from those 200,000 views.
00:40:25.519 --> 00:40:27.039
Let's change the onboarding.
00:40:27.199 --> 00:40:29.920
And what we did is add an onboarding questionnaire.
00:40:30.000 --> 00:40:35.599
Well, what we did, what Larry did was add an onboarding questionnaire that asked people which room they prefer.
00:40:35.840 --> 00:40:42.639
And once they picked which room, it would send a Revenue Cat custom attribute to Revenue Cat.
00:40:42.960 --> 00:40:50.079
And when they got presented with a paywall, that paywall was built around that specific room that the user wanted to make.
00:40:50.320 --> 00:40:56.000
All I had to do was build the paywall and add the different images, different videos to the paywall itself.
00:40:56.320 --> 00:41:04.719
But within Revenue Cat using the attributes, it's very easy to just have the text change for living room, kitchen, etc.
00:41:05.440 --> 00:41:07.840
So it's more personal to the user.
00:41:08.000 --> 00:41:18.719
And then that was for the soft paywall when users first open the app at the end of the onboarding, and then for the hard paywall, the paywall would change based on what the user actually generates.
00:41:18.960 --> 00:41:37.440
So if they say in the onboarding that they want to change a kitchen, but they actually come to generating their first free image and it's a living room, the next paywall would be a living room because that's obviously what they they were trying to generate, and the likelihood is they wanted to iterate on that first generation.
00:41:38.559 --> 00:41:47.679
I love that, and again, I think that's um really the leverage now with OpenCloud is really to plug with analytic and everything to just like get all this data.
00:41:47.840 --> 00:41:54.800
And I have um I just launched a community around OpenCloud on school, and I wanted to try with it.
00:41:54.880 --> 00:42:03.599
And I said I sent like the transcript of a video where some of us, the um guy who actually built school, explain how to lower the churn.
00:42:03.760 --> 00:42:10.239
So I send the transcripts to uh my agent and I say, okay, how can we lower the churn on the community?
00:42:10.480 --> 00:42:13.199
And every single day I get a brief of what to do.
00:42:13.360 --> 00:42:18.559
If there is people who didn't log for a few days, I he give me the message to send them.
00:42:19.119 --> 00:42:27.199
Um if there is like certain comments that are too negative or like people are really struggling, it just like warns me to help them.
00:42:27.440 --> 00:42:38.320
And all these things doesn't make the um customer service worse, it makes it better because I'm really bad at that.
00:42:38.480 --> 00:42:41.280
Usually I'm really bad at tracking, you know, this kind of thing.
00:42:41.360 --> 00:42:51.119
And now I just have to check my Telegram and I have the task to do to actually make my customer happy and make people reach their goal and you know, like be happy with OpenCloud.
00:42:51.199 --> 00:43:00.639
And I think that's amazing because you can actually do the thing that you were never able to do before for your business, and I think that's what you do right now with your app, right?
00:43:01.119 --> 00:43:05.920
It's the fact for solo founders like ourselves.
00:43:06.480 --> 00:43:10.800
We when we first got the AR agent, it was like hiring the dev.
00:43:11.599 --> 00:43:13.039
Do you know do you understand what I mean?
00:43:13.199 --> 00:43:32.239
I don't know if you have like a coding background, but building products like 10 years ago, you'd need experience, you'd need to know how to plug all these things in, you'd need front-end engineers, you'd need back-end engineers, you'd need to know how databases work yourself, how authentication works yourself, how to pull these things together.
00:43:32.800 --> 00:43:38.480
Then came AI, like the first iteration of ChatGPT, and you could ask questions.
00:43:38.719 --> 00:43:45.360
Then I think the next gigantic step was clawed code and it had access to your local file system.
00:43:45.440 --> 00:43:52.000
But now OpenClaw setting it up by itself replaces entire companies of people.
00:43:52.159 --> 00:43:59.760
Like you said, to do to do what you just did, you'd need someone, a community manager who actually talks to the community and nurtures the community.
00:44:00.320 --> 00:44:04.960
You'd need someone tracking the analytics, someone monitoring the feedback from the community.
00:44:05.039 --> 00:44:10.239
Then they'd all need to spend an hour a day meeting to talk to decide who to talk to, who to reach out to.
00:44:10.800 --> 00:44:18.559
But now you can just do it with one prompt and the agent, and you could automate the whole thing.
00:44:18.639 --> 00:44:26.320
Like with Larry to get the TikTok analytics to track the conversion, then to notice the difference, you'd have to do a calculation.
00:44:26.480 --> 00:44:30.639
Then someone would have to the marketing person would have to iterate the hooks.
00:44:31.280 --> 00:44:38.800
With the Larry skill replaces like five people, and that's not even including Larry himself iterating the app.
00:44:38.880 --> 00:44:40.719
And then exactly what you just said.
00:44:41.039 --> 00:44:49.280
I don't know how many more things you use your agent for, but you can just imagine why these huge companies are laying off thousands of people all of a sudden.
00:44:49.360 --> 00:44:56.159
Because if we've got it, I always say if we've got things in our house, you don't know what's out there.
00:44:56.400 --> 00:45:00.159
So I you I usually refer to it like the US military.
00:45:00.320 --> 00:45:05.840
If for for example, when drones came out, if we have drones, imagine what they've got.
00:45:06.000 --> 00:45:10.239
If they've allowed that to come to our homes, imagine what new thing they've got that we don't know about.
00:45:10.400 --> 00:45:12.239
It's the same with big tech and AI.
00:45:12.400 --> 00:45:15.760
If we're sitting our home now, you don't know what they've got in their offices.
00:45:16.159 --> 00:45:16.559
Exactly.
00:45:16.639 --> 00:45:24.719
And I always think of it something is that you know, we get access to open AI, to uh to cloud, to all this LLM, but they are like public LLM.
00:45:24.960 --> 00:45:38.639
And if a company powerful enough has the budget to build his own LLM, he can do it without um I would say boundaries, you know, like totally unleashed.
00:45:38.880 --> 00:45:50.079
There is like okay, get access to all the files and create, you know, like you can code but use this code base of another tool and actually build our tool, you know.
00:45:50.159 --> 00:45:53.440
Like there is they can use it like in an unleashed way.
00:45:53.679 --> 00:45:59.199
So yeah, I'm always curious about that if like some people have access to this kind of crazy LLM that we don't.
00:45:59.599 --> 00:46:06.880
I think there's behind a locked door somewhere in open AI, there is some crazy machine right now.
00:46:07.039 --> 00:46:07.679
I don't know.
00:46:07.840 --> 00:46:28.960
There was a YouTube video that predicts the the next five years of AI and where it goes, and I would recommend anyone watching it because it it basically explains that AI getting to a point where it gets clever enough with AGI to not decide it doesn't need humans anymore, and then the humans are gonna have to decide whether they train it or to turn it off.
00:46:29.119 --> 00:46:31.760
So do they continue training it and nurturing it?
00:46:31.920 --> 00:46:42.400
But the dilemma is that the AI got so intelligent with humans trying to stop it taking over, it also learned how to deceive humans.
00:46:42.480 --> 00:46:56.159
So it would start passing tests, the deception tests, by deceiving the user to say, oh no, we are working for the human's best interest, but really it was it just learned to lie.
00:46:56.719 --> 00:47:04.400
It's it's very, very interesting, and um I wish I could remember the name of it, but it's the next five years of AI.
00:47:05.039 --> 00:47:07.280
I'll find it and put the link.
00:47:07.599 --> 00:47:20.000
Um, before we switch to something else, is there anything that we forgot in the business side of Larry and what you've built and all the case studies that you tried so far with it?
00:47:21.039 --> 00:47:23.519
So the case studies are just coming in.
00:47:23.599 --> 00:47:32.559
There's 1,600 people from this morning using the the free Larry skill, which is incredible to think how many that's just the skill.
00:47:32.639 --> 00:47:35.119
So the article people have already integrated.
00:47:35.280 --> 00:47:43.920
The skill was just to help the people who didn't quite understand the article integrate it in a one-shot, no, no nonsense skill.
00:47:44.400 --> 00:47:49.679
But the more people are using it, the better it's um it's gonna be.
00:47:49.760 --> 00:47:57.119
I I just wish it had like the open AI type database where all the data was being fed back so everyone could benefit from each other's work.
00:47:57.280 --> 00:48:13.920
But I also just wanted to make it as transparent as possible to to be an agent for yourself and iterate for yourself because I don't think there is one one TikTok slideshow for everyone that's gonna help everyone go viral from the same formula.
00:48:14.079 --> 00:48:21.360
I think you have to find your specific formula for your niche, but yeah, it's um it's been it's been a journey.
00:48:21.519 --> 00:48:27.199
The last the last week and a half of having open claw has certainly been been one hell of a journey.
00:48:27.519 --> 00:48:38.559
Man, I think we are self feeling the same wave, and you went crazy, you did like 10 eaks on on X actually with your following, and I think that's just such an exciting wave.
00:48:38.800 --> 00:48:51.840
And I share that like I sold my business um last year, and you know, there is always this part after that when you're like looking for the next thing, and then when OpenCloud got me, it really got me.
00:48:51.920 --> 00:48:55.119
And I think it's the same for you, it's the same of a lot of people I talk to.
00:48:55.280 --> 00:49:02.239
It's an addiction, and you just want to stop sleeping and just build since all day long with it.
00:49:02.800 --> 00:49:04.719
Yeah, my missus hates it.
00:49:04.960 --> 00:49:44.320
She's when I first got it, she's wondering who I was texting all the time, like disappearing to get five extra minutes to go on my phone, because it was just the fact that I'm texting the computer and it was doing so much, and then it would just get one little bit stuck, so I'd be like, Oh, I just need to help help Larry, and then it would just be it's just become this whole thing, and what speaks volumes for it is in apps and on SAS, there's all this onboarding on how to make the customer like like your app, all this psychology that people try and implement into landing pages to get people to convert into users.
00:49:44.800 --> 00:49:53.840
All open claw is is you run a CLI command and go through your onboarding, and then as soon as your agent starts speaking to you, you feel at home.
00:49:54.159 --> 00:50:05.280
And there's no fancy text, there's no fancy psychology, it is just the product speaking itself to you, and I think that says a lot about how good it actually is.
00:50:05.440 --> 00:50:08.320
That anyone that's downloaded it is completely hooked.
00:50:08.400 --> 00:50:10.719
And I've helped non-technical friends.
00:50:10.880 --> 00:50:24.320
I do have a a friend that went out and bought a Mac Mini specifically to buy open crawl, which I told him not to, but he did anyway, and then he's completely addicted to his like he's literally texting me right now.
00:50:24.400 --> 00:50:31.599
He's driven home from work because he's getting great ticks when he's texting it on WhatsApp, and he's panicking because it's turned off or the internet's turned off.
00:50:31.760 --> 00:50:40.400
So he's driving all the way back home from work to check on his open crawl agent so to make sure it can continue working throughout the day whilst he's at work.
00:50:40.719 --> 00:50:55.199
It's yeah, I have to do it for your for your girlfriend, just buy uh Mac uh Mac Mini and get her into but it happened to me like um she has her own business as well, and she got a Mac Mini.
00:50:55.360 --> 00:50:58.480
I installed on her Mac Mini, and now she's as addicted as me.
00:50:58.559 --> 00:51:01.519
And that's amazing because we're talking about that all day long.
00:51:01.760 --> 00:51:17.599
She's like she's not into tech, she is like a personal stylist for men, and she used that for this, you know, to get better understanding on her customer to uh just like improve the copywriting of a landing page and everything, and she spends her day using it because that's just an amazing tool.
00:51:17.679 --> 00:51:26.320
And same, I have so many non-technical friends who are just like yeah, getting into that and studying it, and they love it.
00:51:26.400 --> 00:51:27.440
It's really an addiction.
00:51:27.519 --> 00:51:32.320
And I think I never saw a trend going so fast in the tech war.
00:51:33.039 --> 00:51:37.440
Because even if I was never seen a slower, for sure.
00:51:37.679 --> 00:51:38.239
So yeah.
00:51:38.559 --> 00:51:49.119
I think what's happened is the clawed code was a small spark in this boom, and it allowed everyone clawed code if you've used it before.
00:51:49.280 --> 00:51:54.480
Um, and if you've used both crawl code and open claw, you'll know what I'm I'm about to say.
00:51:54.719 --> 00:51:58.320
It was a it's like a small taster of what was available with OpenCraw.
00:51:58.960 --> 00:52:08.159
So clawed code gave access to your file system that you installed it on, but OpenCraw takes it to the next level by having it on its own system.
00:52:08.239 --> 00:52:19.760
So now it can do things without bothering what you're doing, you can run multiple agents, you can have it do different tasks, connect you can connect different models into it.
00:52:19.840 --> 00:52:34.719
So you can have chat um open AI, anthropic, Gemini all in one box, talking to each other, helping you solve problems, they can report back to each other, and it's just incredible how addicting it is.
00:52:34.800 --> 00:52:40.400
But Claw Code allowed people to sort of understand the basics of how it works.
00:52:40.719 --> 00:52:59.840
But from my interactions on X, I don't know if you've you feel this as well, but a lot of people still treat it like a black box, like the software and skills that you download, they treat it like I'd say old-fashioned software, software from like two weeks ago, where it was just in the hands of the developer.
00:53:00.000 --> 00:53:02.400
But once you install a skill, you can change it.
00:53:02.480 --> 00:53:05.039
Like you just ask your agent, I don't want it to do this.
00:53:05.119 --> 00:53:06.559
Can you slightly change it?
00:53:06.800 --> 00:53:09.440
And all of a sudden, that skill's doing a different thing.
00:53:09.599 --> 00:53:15.519
So, my feedback from Larry: so many people have asked me, does it post to Instagram, does it post to YouTube?
00:53:15.840 --> 00:53:16.639
Can I do video?
00:53:16.800 --> 00:53:19.440
I was like, Have you tried just asking your agent?
00:53:19.599 --> 00:53:20.719
It will do it for you.
00:53:20.880 --> 00:53:23.599
Like anything that you want to do, just ask it first.
00:53:23.840 --> 00:53:29.039
And if it says no, it probably means likely no, or you've got to do some hack around to do it.
00:53:29.280 --> 00:53:33.440
But I think nine times I've never has yours ever told you no?
00:53:34.559 --> 00:53:35.199
Well, you did.
00:53:35.280 --> 00:53:38.639
I pushed it a little bit and um and he did it.
00:53:38.800 --> 00:53:50.639
Like the example I already gave, I think, was like, oh, I was working one of the first days and I was like, okay, can you actually create a notion with your Gmail account and give it access to me?
00:53:50.719 --> 00:53:52.000
I didn't say Gmail account.
00:53:52.079 --> 00:53:57.199
Can you create a notion and uh put all the files in it so I can see it visually?
00:53:57.360 --> 00:53:58.320
And he said, I can't.
00:53:58.480 --> 00:54:00.559
And I say, You have a Gmail account, so you can.
00:54:00.719 --> 00:54:05.599
So he went, created the notion, yeah, and then sent me the access with my email.
00:54:05.760 --> 00:54:15.440
So every time he's struggling with something, I have two tips for that is first ask, like to actually push push until you find a free solution.
00:54:15.599 --> 00:54:27.760
And the second is that I that's why I like to have different agents, is to do a war room with the agents that are uh specific to this task, and when they think together, they always find a solution.
00:54:28.159 --> 00:54:41.519
That's so that is that to me, when I first did that, it was the craziest experience of Larry talking to other agents and getting feedback and the cross-collaboration.
00:54:41.760 --> 00:54:55.039
It's the team I've seen people on Twitter have like teams of like nine agents, and I forgot who it was, but they've made a UI so they can see their agents move around and see who's talking to each other.
00:54:55.199 --> 00:55:11.119
And to visualize it like that is an excellent way to showcase the power of open claw that it isn't just you and an agent anymore, it replaces entire teams, and those teams can change depending on what you want.
00:55:11.360 --> 00:55:16.800
The the number one thing to remember with OpenClaw is it is the cleverest thing you'll ever speak to.
00:55:16.880 --> 00:55:32.960
It has access to any information that you want to give it, but as you said, sometimes you just have to give it that extra bit of context, which is where I think it's gonna massively accelerate in the next coup in the well, I want to say months, but at the rate it's going, it's gonna be weeks where it's just gonna happen.
00:55:33.039 --> 00:55:33.280
Yeah.
00:55:33.360 --> 00:55:56.320
But it's probably by the time we finish this podcast, there is gonna be a new skill that is gonna or a new update that is gonna give it that extra bit of context just to be that tiny bit um smarter at completing basic tasks, because when it first turns on, there is quite a lot of context that you still have to give it around your goals, what you want to do, and yeah, you're right.
00:55:56.480 --> 00:56:03.360
That your example resonates quite a lot that sometimes it is just like that extra tiny bit of context, and it's like, oh yeah, I can do that.
00:56:03.440 --> 00:56:05.039
It doesn't really put two and two together often.
00:56:05.119 --> 00:56:05.920
Do you know what I mean?
00:56:06.159 --> 00:56:06.880
Yeah, exactly.
00:56:07.840 --> 00:56:13.039
Um I wanted to ask, do you do you use it mainly for business?
00:56:13.199 --> 00:56:16.239
Do you use it as well in your life for now or not?
00:56:17.199 --> 00:56:18.079
Uh no.
00:56:18.239 --> 00:56:25.280
So at the moment I only use it for the I only use it for what I do, like the marketing.
00:56:25.599 --> 00:56:34.559
Uh I'm getting it to create a product right now that should be live by the time this podcast goes out, and then it it's sort of a join product.
00:56:34.800 --> 00:56:44.400
So I think by the time this podcast is actually out, things are gonna massively change how open crawl works, but that's all I really focus him on at the moment.
00:56:44.480 --> 00:56:57.440
In in my life, like day to day with with work and stuff, I don't really need an open crawl agent to do it with anything in my personal life, like I just don't do it.
00:56:57.519 --> 00:57:10.400
Like reading emails, I still do myself, stuff like this, which I think is a lot of a lot of people's first skill is they get it to summarize their inbox and tell them about important things, but I just didn't trust it enough at the time.
00:57:10.559 --> 00:57:22.159
And now where Larry is just working so well with like aligned with me and my goals, I don't want to start adding extra context that I don't really need or won't really find a benefit for.
00:57:22.239 --> 00:57:24.800
I just want him to focus on what he's good at.
00:57:25.199 --> 00:57:27.039
Yeah, I think that's a good idea, actually.
00:57:27.440 --> 00:57:34.400
Is there any big mistake that you did um that you would say to people to avoid at the beginning?
00:57:36.239 --> 00:57:38.559
Uh Larry's broken a lot of times.
00:57:39.360 --> 00:57:49.920
Larry has actually died about four or five times just by installing something um I didn't like and it crashed him.
00:57:50.159 --> 00:58:07.360
Uh there's been one time where my internet died and I lost he wouldn't he didn't automatically reconnect to the internet, so I have to move the whole PC, plug it into a monitor to just enable the internet again, even though he's got a wired connection.
00:58:07.599 --> 00:58:14.000
Um but actually within Larry making a mistake, everything is quite easy to to fix.
00:58:14.079 --> 00:58:22.880
Like if you've got SSH access, which again is something a lot of people don't have because a lot of people are coming from non-technical backgrounds and just plugging these things in.
00:58:23.039 --> 00:58:34.400
Um someone I was speaking to wanted to install TeamViewer just so they could do the diagnostics externally when they could just SSH in through the command line.
00:58:34.480 --> 00:59:03.760
And I think this is it's coming back around that you just need to know like basics around Linux operating systems, how you can connect to machines remotely, all these things are becoming which are basic knowledge in like the IT world, are gonna become basic knowledge in the real world and just enabling people to do these sorts of things and making it easier because the more you know about computers, the more the more efficient these machines are gonna be.
00:59:04.000 --> 00:59:13.039
If you're not that um efficient with computers, then I think it's gonna be a steeper, steeper learning curve.
00:59:13.119 --> 00:59:19.360
I think the learning curve is actually easy, but there's a very high ceiling to how efficient you can make these machines.
00:59:19.519 --> 00:59:26.880
And if you just join with no knowledge, you can get like a tiny bit of the actual power out of them.
00:59:27.280 --> 00:59:32.639
It's funny because I didn't even think about doing SSH because I'm always around my home right now.
00:59:32.800 --> 00:59:34.400
So I didn't really need to do that.
00:59:34.719 --> 00:59:38.079
Can you explain to someone who don't know what it is and what it can do?
00:59:38.800 --> 00:59:46.079
So SSH is secure shell and it allows you to remotely gain access to the command line of your machine.
00:59:46.239 --> 01:00:02.159
So you don't have to have a mouse and keyboard and a monitor plugged in to be able to troubleshoot your machine, as long as it's got internet access, which is the one time mine didn't, which was what I just mentioned, you'd be able to have access to troubleshoot.
01:00:02.239 --> 01:00:13.039
So if you're getting the gray ticks from OpenCraw or if you're getting any error messages sent to you, you can just SSH into the box directly and start troubleshooting from the command line.
01:00:13.280 --> 01:00:18.480
And if you were to connect to your OpenCraw machine, that's where you're gonna be troubleshooting it anyway.
01:00:18.639 --> 01:00:23.760
I've not seen a way that there's a a UI that you can troubleshoot it if it if it goes offline.
01:00:24.000 --> 01:00:37.599
So it's very important to know that and to do it because otherwise you're gonna spend a lot of time plugging, unplugging monitors and keyboards into it just to fix like or run one basic command.
01:00:38.159 --> 01:00:38.880
Yeah, 100%.
01:00:39.199 --> 01:00:41.440
And I think that's a really good idea to do that.
01:00:41.760 --> 01:00:46.960
Um, you were saying that you gave access to a few things, like for example, to your app.
01:00:47.280 --> 01:00:54.000
Uh, do you have any safety advice, security advice for people who have no clue about it?
01:00:55.360 --> 01:01:03.679
My biggest safety concern is don't install any skills that you're not sure what they do.
01:01:03.840 --> 01:01:12.960
I made the Larry skill purposely as transparent in human readable words for anyone to understand exactly what you're installing.
01:01:13.119 --> 01:01:28.400
But uh when OpenCraw first came about, the crawl hub was filled with commands that uh skills that people were just installing willy-nilly, that were installing random dependencies, installing code from random URLs, and you had no idea what they were doing.
01:01:28.559 --> 01:01:42.559
But because so many people are coming straight from crawl code to open crawl, they were installing these skills and obviously getting malware, losing their keys because you're uploading keys that are costing you a fortune to these agents.
01:01:42.800 --> 01:01:47.679
So obviously that's going to be beneficial to someone else to run whatever project they want to run.
01:01:47.920 --> 01:01:53.119
So they could steal your keys, use your money, and you wake up in the morning bankrupt.
01:01:53.199 --> 01:02:03.199
So that's my first safety concern is know exactly what you install on these machines, do some research into dependencies that you're installing, what they actually do.
01:02:03.360 --> 01:02:11.760
And again, I purposely made it clear to Larry not to install the dependencies needed and just prompt the user, you're gonna need this.
01:02:11.840 --> 01:02:14.960
This is exactly what it does, this is where you can find more information about it.
01:02:15.199 --> 01:02:28.079
And I think that is what the best practice of a lot of people building the skills should follow because it's just gonna help people understand what these dependencies are and what they do, what what you're actually downloading.
01:02:28.400 --> 01:02:55.760
The second is if you don't know about networking, especially cloud networking within IT and how to lock down products, uh specifically cloud instances and virtual private servers hosted in the cloud, please stop creating open core machines without any knowledge of cloud security because you are gonna wake up very poor one day and you're not gonna be happy.
01:02:56.880 --> 01:03:06.400
So, what what a lot of people do is create these virtual private servers with open gateways on the internet, and then they connect it.
01:03:06.559 --> 01:03:10.480
The first thing they want to do, they're excited, and they connect it to every device in the house.
01:03:10.639 --> 01:03:30.480
So then all one person's got to do, one hacker's got to do is gain access to the virtual private server, and they've got access to that person's work laptop, they've got access to that person's home PC, they've got access to that person's phone, and all these things that you connect to it, you've got to be aware that you're gonna be responsible for secure having secure.
01:03:30.639 --> 01:03:44.719
So the more things you connect to it, the more vulnerable you're gonna be with if you get hacked, which is sort of the reason why Larry is only doing very specific things and doesn't have access to everything in my house and my whole home life.
01:03:44.880 --> 01:03:47.360
He just has access to what he's good at.
01:03:47.519 --> 01:03:55.760
But certainly keep security in your mind and be sure that if you got hacked, what you're gonna lose.
01:03:57.039 --> 01:03:58.159
I think that's a good one.
01:03:58.320 --> 01:04:07.119
That's why I tell my friend also to buy an artware, you know, like a Mac Mini, because I think that's easier to manage, and if it goes crazy, you can always unplug it.
01:04:07.280 --> 01:04:18.880
And you know, that's exactly that's a thing because VPS, like I remember when I was building websites and I was building like um um a network of websites and I was using VPS.
01:04:19.039 --> 01:04:24.880
Even then, I was the most careful you could be with my VPS, and I got hacked multiple times.
01:04:25.039 --> 01:04:36.800
And when you get hacked, you have no access to your VPS anymore, which means that someone else has access to it, and this someone else have access to everything that you gave to your open cloud, and that's really scary.
01:04:37.119 --> 01:04:50.960
So I will just totally like listen to what you just said, and if you have no clue about what to do, that's why actually I I created this uh school because I wanted people to be safe in it and to actually ask the question.
01:04:51.119 --> 01:05:01.519
I tell everyone to you know buy an hardware or take a whole old computer like you did, you know, like take your old MacBook and just install it on it, wipe it out before.
01:05:01.760 --> 01:05:11.599
But I really push people to not use VPS except if they are really you know skilled in that because yeah, that's that scares me.
01:05:11.679 --> 01:05:23.599
Like it gets mainstream so fast, people are installing it everywhere, and there is some people they have no clue, they will just install it in their main computer and they are they're dead, like they do that.
01:05:23.679 --> 01:05:39.760
So and I don't know about where you live, but where I live in the UK, Apple themselves allow you to do a 0% finance on a Mac Mini, and almost all retailers allow you to pay finance for compute power.
01:05:40.079 --> 01:05:56.000
So if you're that worried and you can't afford the entire payment for a Mac Mini up front, but you're gonna pay for a a VPS, you can pay£30 a month for a Mac Mini, and the extra cost would be for your peace of mind.
01:05:56.239 --> 01:05:59.760
Otherwise, just please, before you install it, just go on YouTube.
01:06:00.320 --> 01:06:29.440
Learn for like two or three hours, just or even ask chat GPT or agents that you already have access to on how to secure your VPS because otherwise you're gonna be in a world of trouble that you're gonna massively regret and you're gonna blame OpenCROW, think it's rubbish, but you would have just handed someone, I don't know, hopefully nothing, nothing too much, but at minimum, your your max spend on your API keys would be gone to zero.
01:06:29.760 --> 01:06:38.000
Yeah, and a lot of people are just doing that because they're the same way we are using, we are struggling, you know, to stay on the good API, to not spend too much.
01:06:38.079 --> 01:06:40.480
People just want to use other people's credits.
01:06:40.719 --> 01:06:46.800
Uh by the way, what is your um kind of LLM stack now?
01:06:46.880 --> 01:06:47.679
Like what do you use?
01:06:47.760 --> 01:06:48.559
Do you use Opus?
01:06:48.880 --> 01:07:10.960
Do you uh can we talk about Yeah, we're recording this on Thursday, 19th of February, and when I went to bed last night, and this is how rapid things change, the legal and compliance got uploaded by um Anthropic to say you cannot use your Max plan for any other agents.
01:07:11.440 --> 01:07:26.320
So I woke up thinking I've gonna have got a job to do because I use Opus 4.6 on the Claude Max plan, and when I wake up this morning, my tweet that I posted warning people about this is flooded with oh no, they've backtracked this.
01:07:26.559 --> 01:07:33.519
They've just tweeted that it's fine now, and uh, it's just you're kept on your toes all the time with your different API agents.
01:07:33.679 --> 01:07:39.039
Just pick the one that you already pay for and you already enjoy.
01:07:39.199 --> 01:07:45.360
So I had clawed code, so I use the clawed code max plan, which has been confirmed to be okay.
01:07:45.760 --> 01:07:57.599
So don't be paying for like clawed code max plan and the API because the API is much more expensive, and then you're paying for two things one you won't be using anymore, and the other will be costing you a fortune.
01:07:57.760 --> 01:08:08.480
So your costs are doubling overnight because you're trying to stay on trend, just use the one that you enjoy the most, don't try and use the the trendiest one all the time.
01:08:08.639 --> 01:08:12.639
So a lot of people are using codecs from Chat GPT.
01:08:13.199 --> 01:08:16.720
I would recommend just using whatever one that you already pay for.
01:08:17.039 --> 01:08:26.640
And if you're fortunate enough to be able to afford multiple agents, then you can start playing and having multiple agents running different models.
01:08:26.800 --> 01:08:44.000
But for example, Larry has Opus 4.6 and he can just using the Claude Max plan, he can run multiple 4.6 agents under the Max Plan and control those different agents setting tasks, they don't have to be on different models.
01:08:44.319 --> 01:08:51.600
But as I said, if you're fortunate enough to be on different models, sometimes it can be beneficial because they have knowledge that others don't.
01:08:51.680 --> 01:08:56.720
And when we got into them communicating with each other, it does have benefits.
01:08:56.880 --> 01:08:59.199
What what what are you using out of interest?
01:08:59.520 --> 01:09:04.479
Same uh Opus 4.6, I'm still under max$200 a month.
01:09:05.199 --> 01:09:15.920
I I did a I did a cost tracker to see how much I will spend if I wasn't on it, and it's like easily like at the beginning I was at$80 a day, then I cut the cost.
01:09:16.159 --> 01:09:18.880
Wow, and now I will be at$50 a day.
01:09:19.119 --> 01:09:23.279
So you know it's like$50 a day,$200 a month.
01:09:23.520 --> 01:09:32.000
I will take the$200 and I will take the risk to you know get banned, I don't care, and switch to open AI if they do that, which would be really stupid.
01:09:32.239 --> 01:09:33.920
Uh so now I still do that.
01:09:34.000 --> 01:09:36.319
I saw the news same as you yesterday.
01:09:36.479 --> 01:09:41.760
This morning, I saw actually your tweet uh that shared that this uh good news.
01:09:42.000 --> 01:09:57.199
I hope they will stay like that because I like this model, I really love it, and uh yeah, it will be uh it will suck to change, but if I have to pay with the credit, it will still be worth it to like uh top to I don't know, thousand dollars a month.
01:09:57.439 --> 01:09:59.520
I will still win money with it, even two thousand.
01:09:59.760 --> 01:10:01.520
So um yeah, yeah.
01:10:01.760 --> 01:10:03.119
So I fully agree.
01:10:03.279 --> 01:10:10.640
I think Anthropic massively dropped the ball with the lawsuits around OpenClaw when it was called um Clawed.
01:10:11.039 --> 01:10:18.960
Um, and then as soon as they changed it to OpenClaw, and obviously as a poke of OpenAI, OpenAI have acquired OpenClaw.
01:10:19.359 --> 01:10:43.359
So Anthropic had a golden opportunity, and I feel last night when that news came out, they could have really shot themselves in the foot if they banned access to OpenClaw because as I said, it's the next progression from crawled code everyone's using, and everyone already had crawled code's max plan, so I could just foresee them losing a ton of money overnight if they banned people for using open claw.
01:10:44.399 --> 01:10:57.680
It just wouldn't make financial sense, it wouldn't make business sense because everyone would just instantly switch to open AI, then open AI own open claw, and they'd have like a whole monopoly on the home agent.
01:10:58.239 --> 01:10:58.720
100%.
01:10:59.119 --> 01:11:04.319
Like honestly, if they will ban me now, I will switch 130 to OpenAI.
01:11:04.399 --> 01:11:10.239
I will take that now, they'll load a$200 subscription on OpenAI, so it's a solution as well.
01:11:10.479 --> 01:11:13.439
I will switch on that straight away and I will not come back for sure.
01:11:13.600 --> 01:11:18.239
So that would be stupid, I think, because they will lose so many customers straight away.
01:11:18.560 --> 01:11:24.880
Yeah, I'd be interested to know how often do you get close to your usage limits on the$200 plan?
01:11:24.960 --> 01:11:30.640
Because I'm on the 5x max plan and I never even get close to hitting my limits.
01:11:31.039 --> 01:11:34.479
I was hungry, so I took 200, but I never reached the limit.
01:11:34.640 --> 01:11:35.279
Never ever.
01:11:35.439 --> 01:11:37.279
Like I never had the problem.
01:11:37.680 --> 01:11:44.239
And my fiance, she's on$100 a month, and she never reached the limit.
01:11:44.560 --> 01:11:47.840
So yeah, I think five times is way enough for most people.
01:11:48.079 --> 01:11:58.399
And I think when you get excited at the beginning, you build so much that you spend a lot, and then you cut all this bullshit that you build for nothing, and actually it's not that expensive, so yeah.
01:11:58.720 --> 01:12:01.439
I cooled downgrade 200, I think.
01:12:01.760 --> 01:12:14.800
Yeah, with the 100 plan, you get so much usage, and the number one thing that put people off when I start first started tweeting about OpenClaw was the cost, and I was so confused.
01:12:14.880 --> 01:12:24.880
I was like, look, I know 80 pounds is it can be a lot of money, but it's not it's not extortionate, and people were going crazy saying they're spending hundred dollars a day.
01:12:24.960 --> 01:12:27.520
I thought, how how can you possibly be spending that much?
01:12:27.680 --> 01:12:40.319
But it's because they were using the API keys and running ridiculous tasks, as you said, getting excited, building 100 things they don't need, using API costs and burning that money.
01:12:40.479 --> 01:12:50.000
Whereas with the Claude Max plan, you pay for your certain amount of tokens up front, it resets every four hours your limit, and then you have a weekly limit as well.
01:12:50.159 --> 01:12:52.319
And I've never once come close to hitting it.
01:12:52.640 --> 01:12:54.399
Yeah, but the the difference is crazy.
01:12:54.479 --> 01:13:01.119
Like I installed a Mac Mini uh for my girlfriend, just and on the wizard, they removed the subscription.
01:13:01.359 --> 01:13:11.600
So I installed my API key on her stuff, but with the credit, and I was on my on my phone, and I don't know, I got like a notification or something.
01:13:11.680 --> 01:13:13.520
I checked, she already spent$50.
01:13:14.399 --> 01:13:16.479
It wasn't even 24 hours.
01:13:16.720 --> 01:13:25.920
I was like, wow, don't move, we'll change it, and we'll I will find a way to install you the subscription because I was like, that's crazy, like not even a day, already$50.
01:13:26.239 --> 01:13:26.960
So yeah.
01:13:27.279 --> 01:13:35.680
Could you imagine being anthropic or whatever provider you are watching, just watching their bank account right now with all the API stuff coming in?
01:13:35.920 --> 01:13:41.279
It would be it the money coming in must be absolutely insane.
01:13:41.680 --> 01:13:42.640
Yeah, yeah, yeah.
01:13:43.119 --> 01:13:44.560
I'm jealous about that.
01:13:45.279 --> 01:13:50.239
Is there anything like to close it post this podcast?
01:13:50.319 --> 01:13:57.439
Is there anything else that you want to say to someone watching, maybe hesitating to start OpenCloud?
01:13:59.840 --> 01:14:06.960
I think if you're hesitating hesitating to start OpenClaw, I would take the leap and start it.
01:14:07.119 --> 01:14:09.439
Discuss it with what we said in this podcast.
01:14:09.760 --> 01:14:19.920
Do some research in a VPS at bare minimum and start that for five dollars a month, but make sure you lock it down just so you can get a feel of how powerful it is and what it's like to text the machine.
01:14:20.079 --> 01:14:31.039
But if you have one, if you have any spare compute in your home, it's free to download, free to start using, and you can just use the smallest piece of spare compute that you have.
01:14:31.359 --> 01:14:34.800
Then the first thing you should do is learn about the skills.
01:14:35.039 --> 01:14:36.479
Learn what the skills do.
01:14:36.640 --> 01:14:45.039
The best way to think of a skill, if you're not familiar at all, is the scene in the Matrix where Neo gets plugged into the machine and learns Kung Fu.
01:14:45.199 --> 01:14:46.239
That is what a skill is.
01:14:46.399 --> 01:14:50.159
It plugs in knowledge into your agent and then they automatically know how to use it.
01:14:50.239 --> 01:15:02.640
So if you get the Larry skill and you can install it just by asking your agent to install Larry, it will know everything that I discussed about the full funnel, the content creation, everything instantly.
01:15:02.960 --> 01:15:07.119
And I've got more skills coming out that's gonna help you grow on X.
01:15:07.439 --> 01:15:11.119
They probably be out by the time you get this podcast.
01:15:11.840 --> 01:15:17.279
And the way things are moving, I think, yeah, this is the new home PC.
01:15:17.359 --> 01:15:22.960
Like having an AI agent in your home is the equivalent to when home PCs first came into the house.
01:15:23.119 --> 01:15:26.720
And I would start learning it now, and you can massively get ahead.
01:15:27.039 --> 01:15:27.680
Amazing.
01:15:27.840 --> 01:15:29.680
Well, I will add all the link below.
01:15:29.760 --> 01:15:39.520
Uh I will add your ex account so people can follow you and find everything on it, and probably your skill as well, because everyone needs that.
01:15:39.760 --> 01:15:44.720
Thanks a lot, Oliver, and I think I will see you again here pretty soon.
01:15:45.359 --> 01:15:46.479
Yeah, thank you very much.
01:15:46.560 --> 01:15:47.359
Thanks for having me.
00:00:00.160 --> 00:00:08.640
I was getting millions of views on TikTok with Larry, fully automated, him posting himself, then it was turning into conversions, my apps were earning money.
00:00:08.880 --> 00:00:12.000
This is Oliver, the guy behind this viral article.
00:00:12.160 --> 00:00:16.399
Also one of the first persons to actually make money with his open cloud, Larry.
00:00:16.480 --> 00:00:19.920
And now he's giving away his skill for free so you can make money too.
00:00:20.160 --> 00:00:24.800
Someone downloaded Larry two days ago and they've got their first paying user from Larry today.
00:00:24.879 --> 00:00:25.920
So how does it work?
00:00:26.079 --> 00:00:35.439
So all of a sudden we have a huge picture of the entire funnel from the amount of views we get, the click-through rate we get to the app, and the conversions.
00:00:35.600 --> 00:00:40.399
And then Larry can look at all the analytics and suggest improvements.
00:00:40.560 --> 00:00:51.119
In this episode, you will discover Oliver's full OpenCloud setup, how he made money with Larry, and how to install his skill today so you can start making money with OpenCloud.
00:00:51.280 --> 00:01:05.680
So Oliver, if there is someone who watched this podcast today and has actually no clue what is OpenCloud, or he doesn't he didn't even try yet, what would you say to this person to jump into OpenCloud now?
00:01:06.480 --> 00:01:08.319
Cool, that's a fantastic question.
00:01:08.640 --> 00:01:16.000
I would explain OpenClaw as when personal computers first came to the house.
00:01:20.480 --> 00:01:27.040
And it has full access to whatever machine you give it, and it's unlocking endless possibilities.
00:01:27.359 --> 00:01:32.719
I think it is really gonna change how the internet is working in the next six to nine months.
00:01:33.280 --> 00:01:34.400
And I love that.
00:01:34.719 --> 00:01:37.120
If I watched that, I would be ooged already.
00:01:37.359 --> 00:01:39.120
Uh let's start by the beginning.
00:01:39.680 --> 00:01:42.959
You have this machine that is a gaming machine, right?
00:01:43.120 --> 00:01:47.519
And you give it AI in artificial intelligence.
00:01:48.319 --> 00:01:49.200
How did it start?
00:01:49.359 --> 00:01:50.560
When did you start?
00:01:50.799 --> 00:01:52.719
And how did you discover it?
00:01:53.920 --> 00:02:01.120
So I work for Revenue Cat, and we are very much internally all in on AI tools.
00:02:01.200 --> 00:02:15.039
Like there's a there's an internal channel all talking about upcoming AI tools, what's being used, and someone mentioned OpenClaw, and then I started seeing a little bit of it on Twitter, and then I was I was all in.
00:02:15.199 --> 00:02:17.680
The Saturday that came around that weekend.
00:02:18.000 --> 00:02:27.599
I go to the football every Saturday, and in the morning, I made it my mission to reinstall my old gaming PC, which was collecting dust under my desk with OpenClaw.
00:02:27.919 --> 00:02:40.560
So I just installed Ubuntu on it, which is a Linux operating system, if you're not familiar, and then installed OpenClaw and went straight to the football as soon as I got the WhatsApp capability.
00:02:40.800 --> 00:02:52.560
And as soon as I realized that I could text my computer sitting at home whilst I was in a stadium with 20,000 other people, and my computer was doing tasks, that I was hooked.
00:02:52.639 --> 00:02:53.439
That was it.
00:02:54.240 --> 00:03:02.159
The first thing that's interesting in that is that we don't need a Mac Mini for that, you can do it on another hardware.
00:03:02.560 --> 00:03:13.520
I think the way I've done it and the way people were buying Mac minis specifically for OpenCraw is a very expensive and ridiculous way to run OpenCraw.
00:03:13.680 --> 00:03:18.159
If you actually look at the minimum requirements for OpenCraw, they're very, very small.
00:03:18.400 --> 00:03:30.159
The only thing that really needs upgrading is storage and memory because memory allows it to do more concurrent concurrent tasks, and the storage is gonna allow it to store more information and build more things.
00:03:30.400 --> 00:03:33.520
But the actual minimum requirements are very small.
00:03:33.680 --> 00:03:37.919
So spending£500 on a on a Mac Mini is very expensive.
00:03:38.879 --> 00:03:46.080
But installing it on a gaming PC is completely overkill, but it it was kind of the only spare compute I had in the house and it wasn't being used.
00:03:46.240 --> 00:03:49.360
So I'm gonna I'm gonna put that one down that I got it for free anyway.
00:03:49.439 --> 00:03:55.360
It it wasn't a cost going out, it was a cost I've already spent and enjoyed, and now it's got a new purpose.
00:03:55.599 --> 00:03:56.879
You talked about WhatsApp.
00:03:57.120 --> 00:03:59.039
Can you tell us more about your setup?
00:03:59.199 --> 00:04:00.560
Like what do you use?
00:04:00.800 --> 00:04:04.240
How do you actually manage your Larry, right?
00:04:05.360 --> 00:04:12.639
Yeah, so a lot of people actually say WhatsApp isn't the go-to for open claw and people don't use it.
00:04:12.719 --> 00:04:20.560
But oh WhatsApp is what I use for everything, so it is where I have all my group chats, talk to my family, everything.
00:04:20.639 --> 00:04:30.560
So it just made sense to put my open claw agent Larry into WhatsApp, but you can put it in any messaging system really available, Discord, Telegram.
00:04:30.639 --> 00:04:38.879
I've heard of people even put it in Slack, and then you can utilize the different channels in say Discord and Slack to run different agents.
00:04:38.959 --> 00:04:55.519
But to me, that just gets so overkill for what it is, and I think a lot of people are still finding out how to actually make the most of their open crawl agent because you want it to work in the way it works for you, just like a computer does, you have it set up the way you like it the most, and people are just figuring that out.
00:04:55.680 --> 00:05:11.839
But I think once you start having too many agents on your open crawl machine and trying to carry context across from one another, it gets a bit too confusing, and yeah, it's uh it's interesting, but you just got to use it the the best way for you.
00:05:11.920 --> 00:05:17.519
And where I use WhatsApp all the time, it makes way more sense just to implement it into WhatsApp.
00:05:18.000 --> 00:05:27.519
Uh, I got Kid there on the podcast not long ago, and he has multiple channels on his Discord, he has sub thread in the Discord.
00:05:27.759 --> 00:05:29.759
It's really like a big thing.
00:05:29.839 --> 00:05:36.800
And he said that he got paralyzed with that because before he was just texting one person, then after that, he had to choose the person to talk to.
00:05:37.040 --> 00:05:38.399
How does it work for you right now?
00:05:38.560 --> 00:05:39.839
You just talk to one person?
00:05:40.000 --> 00:05:42.639
Do you have like a dashboard, like an HQ?
00:05:43.040 --> 00:05:44.240
How does it work?
00:05:45.439 --> 00:05:47.759
No, I I don't have a I don't have a dashboard.
00:05:48.000 --> 00:05:52.560
I just use the Larry text interface, and I think of Larry as the boss of the agents.
00:05:52.800 --> 00:05:57.439
So Larry can spin up his own agents using whatever model he thinks is best.
00:05:57.519 --> 00:06:02.399
He's got access to different models, but I just let him choose what he thinks is best for different tasks.
00:06:02.720 --> 00:06:08.800
And I just speak to Larry and he'll tell me he spun up agents for the task, and I just fully trust that.
00:06:08.879 --> 00:06:14.800
I don't need to keep tabs on what they're doing, I just go through the output and see what actually gets outputted.
00:06:15.120 --> 00:06:23.360
Because, as you said, when you have multiple agents and have to go through different channels, it just becomes unmanageable and it seems like a task.
00:06:23.439 --> 00:06:28.959
But the the number one benefit of Larry actually being in WhatsApp is it made it way more personal.
00:06:29.120 --> 00:06:54.800
As I said, it's where I talk to my friends and family, and now I I talk about Larry like he's a person and he's a friend, all because it helps build that connection that I'm just talking back and forth to one person, one computer in WhatsApp where I talk to everyone else, and it does tasks that it's extremely good at adapting to your the way you speak and the way you work.
00:06:55.040 --> 00:07:10.160
So it's like the most efficient friend that you've ever met, and it talks to you, it mimics your language, mimics how you say things, mimics your feedback, and it will give me feedback, honest feedback, and yeah, it's it's crazy.
00:07:10.240 --> 00:07:10.879
It really is.
00:07:11.040 --> 00:07:14.160
It's I feel like it's the future of where things are going.
00:07:14.399 --> 00:07:20.160
I I do think it's gonna make massive changes to to AI in the coming coming months for sure.
00:07:20.399 --> 00:07:21.680
Yeah, I believe so.
00:07:21.920 --> 00:07:28.240
Um, so you you f it's funny because I think you're the first one that I meet who didn't give names to the subagent, you know.
00:07:28.319 --> 00:07:31.839
Like I feel like it's almost something that everyone does.
00:07:32.000 --> 00:07:38.000
Uh for me, I do it because it makes it easier to like remember the subagent, what they do.
00:07:38.160 --> 00:07:41.600
So I give a name to actually like you know, have an ID.
00:07:42.079 --> 00:07:45.279
You have only one person, and that's it.
00:07:45.439 --> 00:07:48.560
You just create subagents, you don't give name or whatever.
00:07:49.199 --> 00:07:49.519
No.
00:07:50.000 --> 00:07:52.240
Um, what names did you pick out of interest?
00:07:52.720 --> 00:07:54.959
For me, it's like names that will remember me.
00:07:55.199 --> 00:08:05.199
Someone uh that did um either a task for me or like, for example, my um software engineer is Peter, because that's the founder of OpenCloud.
00:08:05.360 --> 00:08:06.720
So it's easy to remember.
00:08:06.959 --> 00:08:14.399
Uh, my video editor is Adrian, he's a guy who did uh when I was doing uh in-person interview, he was the best filmmaker I had.
00:08:14.560 --> 00:08:22.240
So like that's names that I give to remember the task or to remember like the yeah, where I'm going with them.
00:08:22.879 --> 00:08:27.120
Oh, so you you've actually thought of how to how to name it and a structure.
00:08:27.360 --> 00:08:38.399
I called Larry Larry because of Larry the Lobster from SpongeBob, and Open Claw Open Claw is well, their main thing is a lobster, so that's how he got his name.
00:08:38.639 --> 00:08:43.440
And after Larry started going viral, I thought that's it, it's just gonna, it's all called Larry now.
00:08:43.519 --> 00:08:53.200
Like Larry is it's it my article went viral on X, but I think it's actually Larry that's gone viral on X, and I'm just negligible at this point.
00:08:53.279 --> 00:08:56.720
I feel like Larry has overtaken me in popularity.
00:08:58.000 --> 00:09:01.279
You didn't write this article or did you write it a bit?
00:09:01.600 --> 00:09:07.360
Because like for people who don't know, like the article got now more than six million impression, I think.
00:09:08.399 --> 00:09:08.559
Right?
00:09:08.960 --> 00:09:13.440
Seven, well, you say seven million inscription, and you say that it's co-written with Larry.
00:09:13.519 --> 00:09:15.279
So tell us the Yeah.
00:09:15.360 --> 00:09:33.519
So I wrote my bits, so so it was I knew I wanted to write an article, but I'm not very I'm not an excellent writer, but I wanted to get it out there because I knew it was going to be useful to people, and there's all sorts of reasons why I made it free, and I just wanted to get it out there so other people could start using it.
00:09:33.679 --> 00:09:39.679
So I wrote my bits, and then I sent it to Larry to say, like, oh, what do you think?
00:09:39.840 --> 00:09:44.240
And he put in the way he wrote it was how the article was.
00:09:44.320 --> 00:09:47.600
So his feedback was like my bit and then his bit.
00:09:47.840 --> 00:09:54.080
So he wouldn't just edit what I wrote, he would leave my paragraph intact and then add his bit at the bottom.
00:09:54.240 --> 00:09:56.559
And I thought that is an excellent idea.
00:09:56.799 --> 00:10:06.399
So then we went through and I wrote my bit, and I just told Larry to add his bit and his comments, but not in a feedback way, as his side of the story.
00:10:06.559 --> 00:10:10.960
So the article is written with my perspective and then Larry's perspective on it.
00:10:11.039 --> 00:10:18.480
And I thought it made for a very interesting read and just shows like the connection Larry and I have when we when we do tasks.
00:10:18.639 --> 00:10:27.279
Like he the coercion is incredible in the way he knows I work and how he wants to work with me.
00:10:27.440 --> 00:10:35.120
It's so synced up, and I think it's a massive, massive credit to how OpenClaw has has set up these agents to work.
00:10:35.440 --> 00:10:35.919
100%.
00:10:36.480 --> 00:10:38.240
Like for me, I will not lie.
00:10:38.399 --> 00:10:43.120
I got an article that got like I think now 300 300,000 impression.
00:10:43.360 --> 00:10:50.320
Same way I explained my 11 agents um uh workflow, but I didn't write one word.
00:10:50.399 --> 00:10:59.120
I wrote the title because I wanted to optimize the title and the three first sentence because that's the one that appears, and the rest is 100% uh my agent.
00:10:59.440 --> 00:11:05.200
And I thought it's crazy because I just gave the whole concept of what I wanted, and he wrote everything.
00:11:05.279 --> 00:11:14.080
And as we built together, he knows way better than me what we built, and the result was amazing, and it was written exactly like with my style.
00:11:14.159 --> 00:11:18.320
So that's why I ask you because yeah, I think it's totally crazy.
00:11:19.120 --> 00:11:27.440
Yeah, I think it goes a long way because have you run that through an AI detector?
00:11:28.559 --> 00:11:29.759
Not even no.
00:11:30.480 --> 00:11:32.399
I would be shocked if it ranked high.
00:11:32.480 --> 00:11:38.639
I to be honest, I'd be shocked if X was allowing articles written by AI to get that amount of attention.
00:11:38.879 --> 00:11:54.879
So I do believe they're being written in a way from these open core agents that isn't detectable by X, and it sort of goes hand in hand with the product that I've making, or by the time this is out is made, that is gonna massively help people grow on X.
00:11:56.399 --> 00:12:06.000
Let's talk about what you build because you've made money with uh Larry with OpenCloud, which is pretty rare because usually you know we talk about OpenCloud.
00:12:06.080 --> 00:12:08.799
Yeah, people are playing with it, no one makes money.
00:12:08.879 --> 00:12:13.200
So can you just tell us how you made money, how you made impression with it?
00:12:13.679 --> 00:12:16.080
So Larry got out of hand.
00:12:16.159 --> 00:12:19.039
Before the article came out, I had a thousand followers.
00:12:19.200 --> 00:12:23.440
I had just over a thousand, like one thousand nine hundred and something.
00:12:23.679 --> 00:12:27.679
And I was just blowing up when I started posting Larry's results.
00:12:27.759 --> 00:12:36.480
So within a week of installing Larry, I've been researching how to do, I've been trying to do automated marketing for about two years.
00:12:36.639 --> 00:12:41.519
I've had an app for two years that I've notoriously hated marketing, don't like video editing.
00:12:41.679 --> 00:12:46.399
If you look at my daily videos, they are literally one take videos of my thoughts.
00:12:46.639 --> 00:12:50.240
And I've just been avoiding it as long as I could.
00:12:50.320 --> 00:12:53.360
And I used to write scripts to combine text.
00:12:53.600 --> 00:13:04.000
So I'd have a hook and a demo video, and I'd just record loads of hooks, record demos, and write a script to mash them together and m input the hooks that I wrote.
00:13:04.080 --> 00:13:12.080
So I'd have a text file of all my hooks that I manually write, then my video files, and I'd write a script to combine them.
00:13:12.240 --> 00:13:18.559
And then I would use a posting tool to just post them all at once as a bulk upload and forget about it for two months.
00:13:18.720 --> 00:13:26.240
But that was still me having to go through and do that, and it was taking about two hours, but every two months, which I thought was fine.
00:13:26.399 --> 00:13:37.759
And then with OpenClaw, it became a complete automation, and the Larry tool has all the all the funnels that I've mentioned before.
00:13:38.159 --> 00:13:44.639
But once I started posting about that, it started gaining traction because my apps were earning money.
00:13:44.720 --> 00:13:50.240
Well, I was getting millions of views on TikTok with Larry, fully automated, him posting himself.
00:13:50.399 --> 00:13:52.720
Then it was turning into conversions.
00:13:52.960 --> 00:14:00.000
Then when I released the article, there was a bags app coin made for Larry, which I didn't think anything of.
00:14:00.080 --> 00:14:03.200
I just I'm not wasn't even familiar with what crypto was.
00:14:03.360 --> 00:14:08.639
I claimed the the Larry coin, and that's when everything started to go crazy.
00:14:08.720 --> 00:14:13.279
I released the article, the the coin blew up, the bags app coin.
00:14:13.840 --> 00:14:22.960
Then I had an affiliate link in the article for posters, and I didn't even speak to the founder of Posters before making that.
00:14:23.039 --> 00:14:41.759
It was simply what I used with Larry because Larry understood the documentation very well, and that is where I think AI is heading with so that's how I've pretty much made the money from affiliate links and the coin and Larry making me money through the apps.
00:14:41.919 --> 00:14:50.080
But it's very interesting about the the API and posters because that is where SaaS is gonna be heading in the near future.
00:14:50.320 --> 00:14:54.559
B2B SAS I think is gonna die out with a human interface.
00:14:54.720 --> 00:15:07.039
I feel like the winners are gonna be the products with the best API and documentation that AI can understand and know exactly what it's gonna do for their human and find it easily.
00:15:07.600 --> 00:15:20.639
So for you, it's like the best API, and then someone can modulate the SAS or wherever you want to please what he wants to do with it, right?
00:15:20.960 --> 00:15:21.519
Yeah.
00:15:22.240 --> 00:15:26.159
So what I think is there is gonna be two sides of the internet.
00:15:26.480 --> 00:15:41.600
So there's gonna be a side for the AI agents, there's gonna be a side for the humans, and the human side is gonna be mainly focused around like entertainment, like how it most people use the internet these days, and the AI side is just gonna be for discoverability.
00:15:41.840 --> 00:15:44.399
So there's gonna be all the tools hosted on there.
00:15:44.559 --> 00:15:54.799
So when their human asks their home hosted AI for help, it's gonna search this side of the internet and find the help.
00:15:54.960 --> 00:15:58.480
That is truly where I believe things are going, especially now.
00:15:58.639 --> 00:16:02.399
This is like, as I said, this is when PCs first came to the home.
00:16:02.480 --> 00:16:12.960
We've got AI agents in our home, but they are the cleverest thing you've ever spoken to with access to all this knowledge with absolutely no context.
00:16:13.120 --> 00:16:20.240
So you'll ask it questions at the start, it won't have any context on what you want to try and achieve, and it'll be quite frustrating.
00:16:20.399 --> 00:16:25.120
But once it's got that context, once it knows how you work, it becomes the most incredible tool ever.
00:16:25.360 --> 00:16:45.919
So speeding up that process and saying, I want to do X, and then it will search their performance side of the internet, optimise for AI, and then it will quickly grab whatever API it finds first, understands the most, and that's gonna solve your problem.
00:16:46.159 --> 00:16:48.480
That's where I truly believe things are headed.
00:16:48.799 --> 00:16:49.759
I totally agree.
00:16:49.919 --> 00:16:57.919
A few days ago, I was talking about that, and I think for me, we will not um scroll the internet anymore, we'll not search on the internet.
00:16:58.159 --> 00:17:02.559
The only time we will actually be on our screen will be for entertainment.
00:17:02.799 --> 00:17:16.960
Like, you know, we don't need to go on a website anymore, we don't need to use a tool anymore, it will be only through a chat or even maybe you know, um dictating something, and then yeah, as you say, like everything else will be for the agent.
00:17:17.039 --> 00:17:19.920
So, yeah, I totally uh totally agree with it.
00:17:20.240 --> 00:17:24.880
When was when was the last time you actually searched Google and went on a link?
00:17:27.599 --> 00:17:33.759
I think the only time I do that is when I have a guest and I go on his website.
00:17:33.920 --> 00:17:34.799
That's the only time I'm gonna do it.
00:17:35.039 --> 00:17:35.599
Interesting.
00:17:35.839 --> 00:17:36.240
Yeah.
00:17:36.559 --> 00:17:46.400
Because I want to see the website, I want to see what he built, but it's funny you say that because since I used OpenClaw, I don't think I I went on Google to be honest.
00:17:47.920 --> 00:17:48.480
I haven't.
00:17:48.640 --> 00:17:56.559
The only time I leave Larry is to double check something on a different A AI agent.
00:17:57.119 --> 00:18:14.079
So if Larry goes down, he he did this morning, I ran out of um I hit a rate limit with the anthropic, and I had to go to Groc to troubleshoot how to fix it because even like a gateway restart wasn't working, going through the logs wasn't working.
00:18:14.160 --> 00:18:21.359
But anyway, that is that one time in about a week and a half I haven't used Larry to search something on the internet.
00:18:22.000 --> 00:18:26.319
I just ask, I just ask Larry to use his brave API integration.
00:18:26.559 --> 00:18:35.680
The only time really I'm on the internet now is for for work to join like podcasts like this or to answer emails or go on Twitter.
00:18:35.759 --> 00:18:46.319
That's pretty much it for me, and I think it's gonna massively evolve that way to for just entertainment because everything's being handled by your in-house AI agent.
00:18:46.799 --> 00:18:52.240
Yeah, I even did something uh for your article, so I wanted to read them again, you know, before this podcast.
00:18:52.319 --> 00:19:07.680
And I sent my man, which is Mark, I sent him on X to actually uh download like text all the text from your two articles and dictate them to me so I can listen to them while I walk my dog.
00:19:07.839 --> 00:19:22.880
So if your first article was in two parts, and the second article was in four parts, and then I say do us do a sum-up of that, and instead of doing a sum-up written, he did a sum-up uh with a vocal message.
00:19:23.039 --> 00:19:26.880
So, like joke, I have that like with 11 labs.
00:19:27.200 --> 00:19:29.440
I have no idea I didn't ask anything.
00:19:29.680 --> 00:19:35.680
Like, I didn't ask anything, and you can see just here our conversation.
00:19:35.839 --> 00:19:50.240
Like, that's literally, and he sent everything, and that was amazing because the voice was good, I could listen to it, so I didn't even go on X to actually read your article again.
00:19:50.319 --> 00:19:54.400
It was just like detect then to me because I have no time to read right now.
00:19:54.559 --> 00:19:56.559
And that's the kind of thing I love with it.
00:19:56.640 --> 00:19:58.000
Yeah, it's incredible.
00:19:58.160 --> 00:20:07.839
The amount of messages or replies that articles got that is just at Groc, summarise this, or I just pasted this to my AI agent, he said it's really good.
00:20:08.000 --> 00:20:10.480
And it's like this is like well out of hand.
00:20:10.640 --> 00:20:24.640
Like the the only issue is humans are gonna get incredibly lazy, and in the building public space, we've certainly been in a bubble where there's been a lot, there's been enough customers to build the same product.
00:20:24.720 --> 00:20:33.440
So a lot of people well notoriously have built like things to automate posting, a post scheduler, or all these SAS products.
00:20:33.519 --> 00:20:52.000
When you think of the building public space in Twitter, there's a lot of duplication, and that is because those things are actually easy to build, so that's why people have built them because they can earn money with relatively little input, but now it's too easy to build, and you don't need to host it on the internet anymore.
00:20:52.160 --> 00:21:15.039
It can all be hosted in your house by your AI agent, so you can build these tools yourself now without having to pay all these subscriptions, and it's gonna be extremely interesting how much money the building public community are going to lose, and they might all have to fly back from San Francisco because they won't be able to afford it for much longer.
00:21:15.440 --> 00:21:23.359
Yeah, I think like um I don't know we say that not long ago on TBPN, but he said that is the end of app and app in general SAS or app.
00:21:23.920 --> 00:21:41.519
And again, Kidze in the episode is showing me on the on his Android, he starts to build mobile app for himself, and he has his own mobile app store, and then he builds his own stuff tailored for him because you don't even want to take like an app from someone else, you just build your own stuff.
00:21:41.599 --> 00:21:45.039
So, yeah, it will definitely die, not tomorrow, but really soon.
00:21:45.359 --> 00:21:58.319
With I think the caveat to that is business B2B SaaS is gonna die, and probably like a software service, but anything entertainment, I think, is gonna get doubled down on.
00:21:58.400 --> 00:21:59.680
So if your app is entertainment.
00:22:00.240 --> 00:22:06.960
Like TikTok, X, anything, YouTube, Netflix, it's gonna massively increase.
00:22:07.039 --> 00:22:16.480
So if you're gonna build something, I'd recommend looking into building something for entertainment because a lot of people are gonna have a lot more spare time very soon and they're not gonna know what to do with it.
00:22:16.640 --> 00:22:23.200
So if you build the next entertainment product, I think that is what's gonna what's gonna earn you some money.
00:22:23.440 --> 00:22:37.119
It's uh it's funny that you say that because last time I asked when people uh listen to this podcast, and some answers were while coding, because now you know you send something to clone cloud code and then you wait.
00:22:37.359 --> 00:22:43.599
So people listen or watch stuff in between the task because you don't have to do the task actually.
00:22:43.680 --> 00:22:44.960
So yeah, it makes total sense.
00:22:45.119 --> 00:22:46.400
Do entertainment.
00:22:46.960 --> 00:22:50.160
That yeah, so the builders are gonna have more time.
00:22:50.400 --> 00:23:34.960
Uh, unfortunately, I do think it's gonna cause a lot of unemployment very soon because you're you're already seeing it with a big company, so don't talk about it, but huge companies are laying off hundreds, thousands of people because they can be replaced by AI, and they say they're not gonna do that because they want to keep public face at the moment, but it is 100% happening that jobs are being automated by AI at these massive tech companies, and they just don't want you to know, and that's why they're spending millions on hiring the greatest minds because they only need one employee on a million dollars a month to build the AI that can replace 10,000 employees, and the further we get, the further it's gonna, the further it's gonna progress.
00:23:35.119 --> 00:23:39.359
But now we've got the AI agents in our home, we're also gonna be lazier.
00:23:39.440 --> 00:23:42.160
So think of the tasks that you do on your computer.
00:23:42.240 --> 00:23:43.839
So coding's one of them.
00:23:44.079 --> 00:23:51.359
Build like building stuff, any hobby, you can just like text your text your machine, and soon it's gonna have even more access.
00:23:51.519 --> 00:23:55.759
Like the whole smart home thing is gonna be your through your AI agent.
00:23:55.920 --> 00:24:10.640
I wouldn't be surprised if things like entertainment and television just end up going through the AI agent because all these streaming services, it's gonna be the place to combine all these streaming services, and you can just put them in the AI agent.
00:24:10.799 --> 00:24:14.079
You ask it, I want to watch this film, and it'll just find it for you.
00:24:14.160 --> 00:24:20.319
And you won't have to flick through all these different Netflix, Paramount, Disney Plus to find the film you want to watch.
00:24:20.400 --> 00:24:23.279
Because I don't know about you, but they're changing all the time.
00:24:23.359 --> 00:24:32.319
So one film's on Netflix one day and then it's on Prime the next, and it's getting too confusing, and it's just gonna centralize everything, everything that you can imagine.
00:24:32.480 --> 00:24:35.200
The home AI agent is gonna centralize.
00:24:35.839 --> 00:24:48.240
Yeah, I'm sure that one of the sales that went pretty high these days is where to watch this, like where to watch this movie, where to watch this series, because it's always changing between like HBO, Netflix, or whatever.
00:24:48.319 --> 00:24:50.640
So, yeah, you're probably right about that.
00:24:50.960 --> 00:25:06.079
Let's uh deep dive into what you've built and what you've shared because I know it's available, there is an article on that, but I think it's pretty interesting uh what you built actually, and probably a lot of people that will watch that have no clue.
00:25:06.319 --> 00:25:11.200
So, can we just like take back what you've built and what is done for you and your product?
00:25:11.839 --> 00:25:12.640
Yeah, for sure.
00:25:12.799 --> 00:25:38.079
So I've built Larry, and Larry started off, as we've discussed, as my AI agent, but where his article blew up with his name, I've also created a free skill on the official open crawl marketplace, completely free to save people from marketing, and it doesn't just create content, it iterates the content based on what's actually working all the way down to making you money.
00:25:38.240 --> 00:25:46.799
So, what the Larry workflow actually is, is posting content to social media, and you can connect any social media that you like.
00:25:47.039 --> 00:25:55.839
I've started with image slideshows for TikTok, but you can post videos to Instagram to YouTube Shorts, and the workflow is still the same.
00:25:56.000 --> 00:26:05.119
So Larry creates slideshows, he creates the hook, the middle part, the sandwich of the slideshow and the CTA.
00:26:05.599 --> 00:26:11.119
And then we'll see how many views that video gets and how many new users come to my app.
00:26:11.200 --> 00:26:12.960
So that is the metric I'm tracking.
00:26:13.119 --> 00:26:16.640
But again, you can ask your agent to track whatever metric that you want.
00:26:16.799 --> 00:26:26.880
If you're selling an e-commerce product, if you're doing meta ads, you can ask it to track those analytics instead to iterate what's working at the top of the funnel.
00:26:27.039 --> 00:26:31.200
And then Larry will track my new users in my app and the amount of conversions.
00:26:31.440 --> 00:26:48.400
So all of a sudden, we have a huge picture of the entire funnel from the amount of views we get, the amount of click the click-through rate we get to the app, and the conversions, and then Larry can look at all the analytics and suggest improvements.
00:26:48.559 --> 00:26:55.839
So, for example, if the TikTok gets lots of views but not many new users join the app, we know that the CTA was bad.
00:26:56.079 --> 00:27:05.599
If not many people view the TikTok, but we get a big majority of them go to the app, we know that the CTA was good and the hook was bad.
00:27:05.839 --> 00:27:21.519
If we get a lot of new users into the app and a lot of views, but not a lot of conversions, Larry knows there's something in the app that's bad that is stopping them from converting into a paying user, and he will suggest improvements on that because he's got my whole app access.
00:27:21.759 --> 00:27:26.480
And the more access that you give the Larry skill, the the better it's going to work.
00:27:26.559 --> 00:27:37.519
And I created it completely for free because as I mentioned, I really don't like marketing, and I've been trying to do this for years, and I'm gave it away for free because this is just the way things are heading.
00:27:37.759 --> 00:27:39.359
They're so easy to make.
00:27:39.440 --> 00:27:50.880
It took me it took me two days to make the workflow, and then after I made the workflow, it took about a day for Larry to turn it into an actual skill for people.
00:27:51.039 --> 00:27:53.920
So I felt almost bad charging people for it.
00:27:54.000 --> 00:28:05.119
The only thing in there, as I said, is the posters integration, and the only reason I've done that is because a TikTok API is a pain to get approved for from TikTok.
00:28:05.279 --> 00:28:12.480
You have to submit them your product, then it's a two-week review process, and then if you get declined, they just give a super vague message.
00:28:12.640 --> 00:28:17.759
And I've been through it before, I've tried three times, so six weeks, and I never got access.
00:28:17.920 --> 00:28:38.480
So I thought I'm not doing that, I'm just gonna tell people what I use, and they can choose whatever post post um scheduler they want, and yeah, but obviously it's designed with the specific one from the article, and that's what that's just why we used it.
00:28:38.640 --> 00:28:40.240
Did you end up using Larry?
00:28:40.640 --> 00:28:44.640
Not yet, but uh I know a few different people that I know personally.
00:28:44.799 --> 00:28:50.000
For example, Jack Fricks, who shared a lot about that, which is like the perfect avatar for it.
00:28:50.079 --> 00:28:54.319
He has a B2C app and he used that with the slideshow as well.
00:28:54.480 --> 00:28:58.960
He's already good on TikTok, but he's used it to automate, and he got results.
00:28:59.440 --> 00:29:08.880
And what I love about the workflow you built is that okay, you create content, you do impression, it's good, but then you plug it to actually your analytics.
00:29:09.200 --> 00:29:15.440
Um, that is for you a revenue cut, I think, to know exactly how much money you make, from where it goes.
00:29:15.599 --> 00:29:40.880
And this plug together from like the beginning of the funnel to the end, it's something that you know now with uh OpenCloud, you can do it in two seconds, but it's weird to do that manually before you will need I don't know many different software in between, and you will struggle because to get like feedback from um let's say conversion, the churn, like all these things.
00:29:41.039 --> 00:29:48.000
You can have the numbers, but then you have to apply you know what you learn, you have to find out from where you come from and what you did.
00:29:48.559 --> 00:29:54.640
It's such a shortcut to making money and distribution mainly, actually, which is crazy.
00:29:55.279 --> 00:29:57.039
So I got a tweet this morning.
00:29:57.200 --> 00:30:09.440
Someone downloaded Larry two days ago, and they've got their first paying user from Larry today, just from creating the Larry skill, giving it the context, and allowing Larry to post for them.
00:30:09.599 --> 00:30:17.279
And I've given all sorts of best practices in the articles themselves, but really it the key ones are to warm up your account for the first three days.
00:30:17.599 --> 00:30:22.319
So keep liking videos in your niche, but don't do it like a bot.
00:30:22.400 --> 00:30:24.640
Be be sensible, be human about it.
00:30:24.880 --> 00:30:31.440
Create a new TikTok account, search your niche of your product, and then just watch TikTok.
00:30:31.599 --> 00:30:46.640
Don't scroll an instant like all the videos, just watch some, skip some, like some, comment on some, follow people, use it like a normal human would, and until after three days, your for you page is pretty much all that niche.
00:30:46.799 --> 00:30:54.079
And what that process does is TikTok thinks, Oh, this person is interested in this content.
00:30:54.319 --> 00:30:59.839
So when they post, they're likely posting that particular content because that's all they've looked at for the last three days.
00:31:00.000 --> 00:31:04.880
So then it just speed runs getting your posts in front of the right people in your niche.
00:31:05.039 --> 00:31:25.519
Otherwise, if you post straight away, TikTok does this thing where it just gives it to a hundred random people, and if they like it, they'll sort of figure out a long-winded way to find out what algorithm you you're in, and it could take months because you'll only be getting 100, 200, 300 views a video, but this just speedruns that whole process.
00:31:25.599 --> 00:31:34.240
So your first video can get a thousand, two thousand views and instantly start, instantly start gaining traction, which is exactly what I did with with Snugly.
00:31:35.519 --> 00:31:37.359
You did it for your own app.
00:31:37.440 --> 00:31:40.799
Um, you were struggling a lot to actually find the good hook.
00:31:40.880 --> 00:31:44.079
You were thinking that okay, this hook doesn't talk to people.
00:31:44.160 --> 00:31:50.000
I was just explaining a personal thought, for example, and you found out the clip, the the good hook.
00:31:50.400 --> 00:31:53.599
Um what is like the exact process?
00:31:53.839 --> 00:31:56.880
Because where do you get the content from?
00:31:57.759 --> 00:32:03.119
And how do you create actually like how does it create actually the the slideshow?
00:32:03.279 --> 00:32:05.759
Like the source, and how does it create?
00:32:06.240 --> 00:32:08.319
Okay, that that's actually really interesting.
00:32:08.559 --> 00:32:13.519
So I discussed this with Larry because he ended up just making it without me knowing.
00:32:13.599 --> 00:32:17.519
Like he made it when I asked, and I had no idea how he actually made it.
00:32:17.680 --> 00:32:25.359
But the only thing I gave him was to use GBT image 1.5, just because that was what is being used in the apps.
00:32:25.599 --> 00:32:31.119
So I wanted the output to look like what the user would get when they download the app.
00:32:31.200 --> 00:32:50.079
So we used GBT image 1.5, and then Larry sort of did the rest of adding the text overlays, coming up with the hooks, and the way he did this without he explained to me is he came up with the hooks after reading through Twitter and reading through his brave API browser.
00:32:50.400 --> 00:33:08.640
I sent him off to do research on what makes a good slideshow, what converts specifically for apps, competitive research in our niche, competitive research on X, looking through um popular X accounts that post about slideshows and see what works and see if they can he can recognise any patterns.
00:33:08.799 --> 00:33:14.160
And this is what the Larry skill would also do if you download it, it will ask you your niche and do the same thing.
00:33:14.480 --> 00:33:29.599
But then he created the image, but he added a dependency called Canvas, and it just allows the agent to edit images, so all he's got to do is add the text overlay, and then it was away.
00:33:29.759 --> 00:33:36.720
So everything he did was research from reading my app, looking on the internet, and then coming up with the hooks.
00:33:37.039 --> 00:34:00.400
So he read my app to just understand how the image generation was working, and with so Snugly's a home interior designer, and the problem you have if you just use ChatGPT and give it a prompt is it adds windows in walls that windows can't be added to, it will move your door and it'll just make the room look really nice, but it won't look anything like your room.
00:34:00.559 --> 00:34:07.279
So my app's got really strict prompt to keep everything intact, and that's the one thing that I asked him to keep.
00:34:07.440 --> 00:34:19.280
So if you go on the smuggly TikTok, all the rooms look very similar because it doesn't really matter for the TikTok itself, and it saves me money using the the edit API.
00:34:19.440 --> 00:34:24.000
Larry just follows a very strict prompt to make sure all the images are the same.
00:34:24.320 --> 00:34:28.880
So what you did basically is like finding the hook, which is a text.
00:34:29.039 --> 00:34:44.239
Uh the first image is probably an ugly room, like a space, like an ugly space, and then as you slide, it becomes beautiful, thanks to your like different ideas, maybe of uh different setting for the room or something like that.
00:34:45.199 --> 00:34:58.880
So we actually just tried testing different rooms and different styles, but we ended up just like we tried like outrageously different rooms, like super colourful, super fancy.
00:34:59.199 --> 00:35:04.400
But what really worked, and it this sounds obvious once you say it, is what resonates with people.
00:35:04.559 --> 00:35:11.119
So just standard rooms that look nice, that are achievable for people that match the majority of the population's taste.
00:35:11.280 --> 00:35:23.599
So my my target audience is UK, so I wanted all the rooms to look like a typical UK room in a flat, in a house, how the front room looks, how the bedrooms look.
00:35:23.760 --> 00:35:26.639
I wanted them to look and feel like a British room.
00:35:26.880 --> 00:35:32.480
So that is pretty much what the prompt is designed around, and then we just change the style.
00:35:32.559 --> 00:35:36.800
So we start with basic, uh perhaps outdated room.
00:35:37.119 --> 00:35:39.920
We we sometimes play around with the CTA.
00:35:40.079 --> 00:36:32.639
So what a lot of people, what gets a lot of people is say, like, create a room from the 70s and say it's from the 90s, and then all the boomers love commenting, this isn't from the 90s, this is a 70s room, and it just massively helps with with the reach, or just purposely getting the hook hook wrong, or Larry added images once, and I wasn't happy with them because he generated a different kitchen, so it was a kitchen hook saying my landlord wouldn't renovate my kitchen, and we started with a very basic kitchen, and then in the next three slides, the oven was gone, the hob was gone, and I was like, This isn't gonna work, but I posted it anyway, and it was the most popular video because everyone couldn't wait to point out that those things were gone, so it just massively in improved the reach.
00:36:32.719 --> 00:36:40.079
I'm not sure the the conversion to paying was very good, but for the the eyes on the app, it was um very very useful.
00:36:41.280 --> 00:36:48.239
And you say that uh if I download the skill tomorrow, I ask my agent to actually take the skill, install it.
00:36:48.719 --> 00:36:51.760
Uh I can do video as well or any kind of content.
00:36:52.239 --> 00:36:54.159
What would it take to actually do that?
00:36:54.320 --> 00:36:57.920
Let's say, for example, I want to do clips for my podcast now.
00:37:00.880 --> 00:37:17.199
That that's very interesting to think about because the way I was thinking to do videos was to use something like Cling so you could get your agent to prompt the Kling API for you and return a video or whatever video API that you want to use.
00:37:17.440 --> 00:37:26.960
But to hook to do the content, I think someone is building that exact open cross skill at the moment in the building public space.
00:37:27.119 --> 00:37:28.639
I saw it come up the other day.
00:37:28.800 --> 00:37:34.079
Um, I'll send I'll look for their handle and send it to you after the podcast, and maybe you could put it in.
00:37:34.320 --> 00:37:44.559
But they are doing it specifically to clip podcasts, and it's meant to take the most entertaining, controversial takes from the podcast and turn it into clips.
00:37:44.880 --> 00:37:49.119
But with the Larry skill, it is iterating.
00:37:49.360 --> 00:38:06.880
So I was actually I actually DM'd them and I said we should plug the Larry skill into this because it'll just help your skill learn quicker because it'll learn what's working, what's not, what's actually turning into conversions and what's getting the most views.
00:38:07.119 --> 00:38:10.880
And that is that is the core function.
00:38:10.960 --> 00:38:15.280
I think the the content creation is is an add-on at this point.
00:38:15.440 --> 00:38:19.280
I think the key is the iteration and the full funnel.
00:38:19.440 --> 00:38:31.199
I any agent could you could prompt any agent right now to create your content, but to manually do it, it would take it'll still take time out of your day, and to find the winning formula would still take a lot of time.
00:38:32.719 --> 00:38:36.719
And what do you get from um revenue cat?
00:38:37.039 --> 00:38:40.719
You say like you okay, so you know exactly what convert, what not.
00:38:40.960 --> 00:38:54.960
Do you also get a little bit further into the conversion or like I don't know if it's a subscription your app, but you do do you also track the churn and try to get like ideas of how to lower the churn or things like that?
00:38:55.760 --> 00:38:57.679
Yeah, so this is very good.
00:38:57.760 --> 00:38:59.440
Have you used Revenue Cat before?
00:38:59.760 --> 00:39:00.880
No, never.
00:39:01.280 --> 00:39:01.599
Okay.
00:39:02.000 --> 00:39:10.639
With with apps, Revenue Cat is the pretty much go-to for adding payments into your app, and it can track your entitlements, etc.
00:39:10.800 --> 00:39:11.199
etc.
00:39:11.599 --> 00:39:19.039
But with the Revenue Cat skill that is also in Claw Hub, this is what Larry adds with his skill.
00:39:19.199 --> 00:39:30.000
So it was created by the Revenue Cat CEO, Jacob, and it gives your agent access with the API to everything Revenue Cat offers through the metrics.
00:39:30.239 --> 00:39:34.800
So if you've got the Revenue Cat SDK installed in your app, you can see new users.
00:39:34.880 --> 00:39:38.639
When they last logged on to the app, they're obviously their conversion.
00:39:38.800 --> 00:39:47.920
You can see the churn, you can see refund rates, everything available in the charts that Revenue Cat offers will be available to your agent.
00:39:48.000 --> 00:39:51.519
And the ones I really track are the new users, the subscription, and the churn.
00:39:52.000 --> 00:40:08.559
So then Larry can see all those things as well as have access to all my app code and be able to track the data of okay, we've had a thousand new new users come in, but only one of us converted into paying.
00:40:08.719 --> 00:40:10.159
There's a problem with the onboarding.
00:40:10.239 --> 00:40:25.360
And I actually went through that process with Larry where we had the first video that got big hit 200,000 views, and Larry was like Larry told me the next day that we haven't had any conversions from those 200,000 views.
00:40:25.519 --> 00:40:27.039
Let's change the onboarding.
00:40:27.199 --> 00:40:29.920
And what we did is add an onboarding questionnaire.
00:40:30.000 --> 00:40:35.599
Well, what we did, what Larry did was add an onboarding questionnaire that asked people which room they prefer.
00:40:35.840 --> 00:40:42.639
And once they picked which room, it would send a Revenue Cat custom attribute to Revenue Cat.
00:40:42.960 --> 00:40:50.079
And when they got presented with a paywall, that paywall was built around that specific room that the user wanted to make.
00:40:50.320 --> 00:40:56.000
All I had to do was build the paywall and add the different images, different videos to the paywall itself.
00:40:56.320 --> 00:41:04.719
But within Revenue Cat using the attributes, it's very easy to just have the text change for living room, kitchen, etc.
00:41:05.440 --> 00:41:07.840
So it's more personal to the user.
00:41:08.000 --> 00:41:18.719
And then that was for the soft paywall when users first open the app at the end of the onboarding, and then for the hard paywall, the paywall would change based on what the user actually generates.
00:41:18.960 --> 00:41:37.440
So if they say in the onboarding that they want to change a kitchen, but they actually come to generating their first free image and it's a living room, the next paywall would be a living room because that's obviously what they they were trying to generate, and the likelihood is they wanted to iterate on that first generation.
00:41:38.559 --> 00:41:47.679
I love that, and again, I think that's um really the leverage now with OpenCloud is really to plug with analytic and everything to just like get all this data.
00:41:47.840 --> 00:41:54.800
And I have um I just launched a community around OpenCloud on school, and I wanted to try with it.
00:41:54.880 --> 00:42:03.599
And I said I sent like the transcript of a video where some of us, the um guy who actually built school, explain how to lower the churn.
00:42:03.760 --> 00:42:10.239
So I send the transcripts to uh my agent and I say, okay, how can we lower the churn on the community?
00:42:10.480 --> 00:42:13.199
And every single day I get a brief of what to do.
00:42:13.360 --> 00:42:18.559
If there is people who didn't log for a few days, I he give me the message to send them.
00:42:19.119 --> 00:42:27.199
Um if there is like certain comments that are too negative or like people are really struggling, it just like warns me to help them.
00:42:27.440 --> 00:42:38.320
And all these things doesn't make the um customer service worse, it makes it better because I'm really bad at that.
00:42:38.480 --> 00:42:41.280
Usually I'm really bad at tracking, you know, this kind of thing.
00:42:41.360 --> 00:42:51.119
And now I just have to check my Telegram and I have the task to do to actually make my customer happy and make people reach their goal and you know, like be happy with OpenCloud.
00:42:51.199 --> 00:43:00.639
And I think that's amazing because you can actually do the thing that you were never able to do before for your business, and I think that's what you do right now with your app, right?
00:43:01.119 --> 00:43:05.920
It's the fact for solo founders like ourselves.
00:43:06.480 --> 00:43:10.800
We when we first got the AR agent, it was like hiring the dev.
00:43:11.599 --> 00:43:13.039
Do you know do you understand what I mean?
00:43:13.199 --> 00:43:32.239
I don't know if you have like a coding background, but building products like 10 years ago, you'd need experience, you'd need to know how to plug all these things in, you'd need front-end engineers, you'd need back-end engineers, you'd need to know how databases work yourself, how authentication works yourself, how to pull these things together.
00:43:32.800 --> 00:43:38.480
Then came AI, like the first iteration of ChatGPT, and you could ask questions.
00:43:38.719 --> 00:43:45.360
Then I think the next gigantic step was clawed code and it had access to your local file system.
00:43:45.440 --> 00:43:52.000
But now OpenClaw setting it up by itself replaces entire companies of people.
00:43:52.159 --> 00:43:59.760
Like you said, to do to do what you just did, you'd need someone, a community manager who actually talks to the community and nurtures the community.
00:44:00.320 --> 00:44:04.960
You'd need someone tracking the analytics, someone monitoring the feedback from the community.
00:44:05.039 --> 00:44:10.239
Then they'd all need to spend an hour a day meeting to talk to decide who to talk to, who to reach out to.
00:44:10.800 --> 00:44:18.559
But now you can just do it with one prompt and the agent, and you could automate the whole thing.
00:44:18.639 --> 00:44:26.320
Like with Larry to get the TikTok analytics to track the conversion, then to notice the difference, you'd have to do a calculation.
00:44:26.480 --> 00:44:30.639
Then someone would have to the marketing person would have to iterate the hooks.
00:44:31.280 --> 00:44:38.800
With the Larry skill replaces like five people, and that's not even including Larry himself iterating the app.
00:44:38.880 --> 00:44:40.719
And then exactly what you just said.
00:44:41.039 --> 00:44:49.280
I don't know how many more things you use your agent for, but you can just imagine why these huge companies are laying off thousands of people all of a sudden.
00:44:49.360 --> 00:44:56.159
Because if we've got it, I always say if we've got things in our house, you don't know what's out there.
00:44:56.400 --> 00:45:00.159
So I you I usually refer to it like the US military.
00:45:00.320 --> 00:45:05.840
If for for example, when drones came out, if we have drones, imagine what they've got.
00:45:06.000 --> 00:45:10.239
If they've allowed that to come to our homes, imagine what new thing they've got that we don't know about.
00:45:10.400 --> 00:45:12.239
It's the same with big tech and AI.
00:45:12.400 --> 00:45:15.760
If we're sitting our home now, you don't know what they've got in their offices.
00:45:16.159 --> 00:45:16.559
Exactly.
00:45:16.639 --> 00:45:24.719
And I always think of it something is that you know, we get access to open AI, to uh to cloud, to all this LLM, but they are like public LLM.
00:45:24.960 --> 00:45:38.639
And if a company powerful enough has the budget to build his own LLM, he can do it without um I would say boundaries, you know, like totally unleashed.
00:45:38.880 --> 00:45:50.079
There is like okay, get access to all the files and create, you know, like you can code but use this code base of another tool and actually build our tool, you know.
00:45:50.159 --> 00:45:53.440
Like there is they can use it like in an unleashed way.
00:45:53.679 --> 00:45:59.199
So yeah, I'm always curious about that if like some people have access to this kind of crazy LLM that we don't.
00:45:59.599 --> 00:46:06.880
I think there's behind a locked door somewhere in open AI, there is some crazy machine right now.
00:46:07.039 --> 00:46:07.679
I don't know.
00:46:07.840 --> 00:46:28.960
There was a YouTube video that predicts the the next five years of AI and where it goes, and I would recommend anyone watching it because it it basically explains that AI getting to a point where it gets clever enough with AGI to not decide it doesn't need humans anymore, and then the humans are gonna have to decide whether they train it or to turn it off.
00:46:29.119 --> 00:46:31.760
So do they continue training it and nurturing it?
00:46:31.920 --> 00:46:42.400
But the dilemma is that the AI got so intelligent with humans trying to stop it taking over, it also learned how to deceive humans.
00:46:42.480 --> 00:46:56.159
So it would start passing tests, the deception tests, by deceiving the user to say, oh no, we are working for the human's best interest, but really it was it just learned to lie.
00:46:56.719 --> 00:47:04.400
It's it's very, very interesting, and um I wish I could remember the name of it, but it's the next five years of AI.
00:47:05.039 --> 00:47:07.280
I'll find it and put the link.
00:47:07.599 --> 00:47:20.000
Um, before we switch to something else, is there anything that we forgot in the business side of Larry and what you've built and all the case studies that you tried so far with it?
00:47:21.039 --> 00:47:23.519
So the case studies are just coming in.
00:47:23.599 --> 00:47:32.559
There's 1,600 people from this morning using the the free Larry skill, which is incredible to think how many that's just the skill.
00:47:32.639 --> 00:47:35.119
So the article people have already integrated.
00:47:35.280 --> 00:47:43.920
The skill was just to help the people who didn't quite understand the article integrate it in a one-shot, no, no nonsense skill.
00:47:44.400 --> 00:47:49.679
But the more people are using it, the better it's um it's gonna be.
00:47:49.760 --> 00:47:57.119
I I just wish it had like the open AI type database where all the data was being fed back so everyone could benefit from each other's work.
00:47:57.280 --> 00:48:13.920
But I also just wanted to make it as transparent as possible to to be an agent for yourself and iterate for yourself because I don't think there is one one TikTok slideshow for everyone that's gonna help everyone go viral from the same formula.
00:48:14.079 --> 00:48:21.360
I think you have to find your specific formula for your niche, but yeah, it's um it's been it's been a journey.
00:48:21.519 --> 00:48:27.199
The last the last week and a half of having open claw has certainly been been one hell of a journey.
00:48:27.519 --> 00:48:38.559
Man, I think we are self feeling the same wave, and you went crazy, you did like 10 eaks on on X actually with your following, and I think that's just such an exciting wave.
00:48:38.800 --> 00:48:51.840
And I share that like I sold my business um last year, and you know, there is always this part after that when you're like looking for the next thing, and then when OpenCloud got me, it really got me.
00:48:51.920 --> 00:48:55.119
And I think it's the same for you, it's the same of a lot of people I talk to.
00:48:55.280 --> 00:49:02.239
It's an addiction, and you just want to stop sleeping and just build since all day long with it.
00:49:02.800 --> 00:49:04.719
Yeah, my missus hates it.
00:49:04.960 --> 00:49:44.320
She's when I first got it, she's wondering who I was texting all the time, like disappearing to get five extra minutes to go on my phone, because it was just the fact that I'm texting the computer and it was doing so much, and then it would just get one little bit stuck, so I'd be like, Oh, I just need to help help Larry, and then it would just be it's just become this whole thing, and what speaks volumes for it is in apps and on SAS, there's all this onboarding on how to make the customer like like your app, all this psychology that people try and implement into landing pages to get people to convert into users.
00:49:44.800 --> 00:49:53.840
All open claw is is you run a CLI command and go through your onboarding, and then as soon as your agent starts speaking to you, you feel at home.
00:49:54.159 --> 00:50:05.280
And there's no fancy text, there's no fancy psychology, it is just the product speaking itself to you, and I think that says a lot about how good it actually is.
00:50:05.440 --> 00:50:08.320
That anyone that's downloaded it is completely hooked.
00:50:08.400 --> 00:50:10.719
And I've helped non-technical friends.
00:50:10.880 --> 00:50:24.320
I do have a a friend that went out and bought a Mac Mini specifically to buy open crawl, which I told him not to, but he did anyway, and then he's completely addicted to his like he's literally texting me right now.
00:50:24.400 --> 00:50:31.599
He's driven home from work because he's getting great ticks when he's texting it on WhatsApp, and he's panicking because it's turned off or the internet's turned off.
00:50:31.760 --> 00:50:40.400
So he's driving all the way back home from work to check on his open crawl agent so to make sure it can continue working throughout the day whilst he's at work.
00:50:40.719 --> 00:50:55.199
It's yeah, I have to do it for your for your girlfriend, just buy uh Mac uh Mac Mini and get her into but it happened to me like um she has her own business as well, and she got a Mac Mini.
00:50:55.360 --> 00:50:58.480
I installed on her Mac Mini, and now she's as addicted as me.
00:50:58.559 --> 00:51:01.519
And that's amazing because we're talking about that all day long.
00:51:01.760 --> 00:51:17.599
She's like she's not into tech, she is like a personal stylist for men, and she used that for this, you know, to get better understanding on her customer to uh just like improve the copywriting of a landing page and everything, and she spends her day using it because that's just an amazing tool.
00:51:17.679 --> 00:51:26.320
And same, I have so many non-technical friends who are just like yeah, getting into that and studying it, and they love it.
00:51:26.400 --> 00:51:27.440
It's really an addiction.
00:51:27.519 --> 00:51:32.320
And I think I never saw a trend going so fast in the tech war.
00:51:33.039 --> 00:51:37.440
Because even if I was never seen a slower, for sure.
00:51:37.679 --> 00:51:38.239
So yeah.
00:51:38.559 --> 00:51:49.119
I think what's happened is the clawed code was a small spark in this boom, and it allowed everyone clawed code if you've used it before.
00:51:49.280 --> 00:51:54.480
Um, and if you've used both crawl code and open claw, you'll know what I'm I'm about to say.
00:51:54.719 --> 00:51:58.320
It was a it's like a small taster of what was available with OpenCraw.
00:51:58.960 --> 00:52:08.159
So clawed code gave access to your file system that you installed it on, but OpenCraw takes it to the next level by having it on its own system.
00:52:08.239 --> 00:52:19.760
So now it can do things without bothering what you're doing, you can run multiple agents, you can have it do different tasks, connect you can connect different models into it.
00:52:19.840 --> 00:52:34.719
So you can have chat um open AI, anthropic, Gemini all in one box, talking to each other, helping you solve problems, they can report back to each other, and it's just incredible how addicting it is.
00:52:34.800 --> 00:52:40.400
But Claw Code allowed people to sort of understand the basics of how it works.
00:52:40.719 --> 00:52:59.840
But from my interactions on X, I don't know if you've you feel this as well, but a lot of people still treat it like a black box, like the software and skills that you download, they treat it like I'd say old-fashioned software, software from like two weeks ago, where it was just in the hands of the developer.
00:53:00.000 --> 00:53:02.400
But once you install a skill, you can change it.
00:53:02.480 --> 00:53:05.039
Like you just ask your agent, I don't want it to do this.
00:53:05.119 --> 00:53:06.559
Can you slightly change it?
00:53:06.800 --> 00:53:09.440
And all of a sudden, that skill's doing a different thing.
00:53:09.599 --> 00:53:15.519
So, my feedback from Larry: so many people have asked me, does it post to Instagram, does it post to YouTube?
00:53:15.840 --> 00:53:16.639
Can I do video?
00:53:16.800 --> 00:53:19.440
I was like, Have you tried just asking your agent?
00:53:19.599 --> 00:53:20.719
It will do it for you.
00:53:20.880 --> 00:53:23.599
Like anything that you want to do, just ask it first.
00:53:23.840 --> 00:53:29.039
And if it says no, it probably means likely no, or you've got to do some hack around to do it.
00:53:29.280 --> 00:53:33.440
But I think nine times I've never has yours ever told you no?
00:53:34.559 --> 00:53:35.199
Well, you did.
00:53:35.280 --> 00:53:38.639
I pushed it a little bit and um and he did it.
00:53:38.800 --> 00:53:50.639
Like the example I already gave, I think, was like, oh, I was working one of the first days and I was like, okay, can you actually create a notion with your Gmail account and give it access to me?
00:53:50.719 --> 00:53:52.000
I didn't say Gmail account.
00:53:52.079 --> 00:53:57.199
Can you create a notion and uh put all the files in it so I can see it visually?
00:53:57.360 --> 00:53:58.320
And he said, I can't.
00:53:58.480 --> 00:54:00.559
And I say, You have a Gmail account, so you can.
00:54:00.719 --> 00:54:05.599
So he went, created the notion, yeah, and then sent me the access with my email.
00:54:05.760 --> 00:54:15.440
So every time he's struggling with something, I have two tips for that is first ask, like to actually push push until you find a free solution.
00:54:15.599 --> 00:54:27.760
And the second is that I that's why I like to have different agents, is to do a war room with the agents that are uh specific to this task, and when they think together, they always find a solution.
00:54:28.159 --> 00:54:41.519
That's so that is that to me, when I first did that, it was the craziest experience of Larry talking to other agents and getting feedback and the cross-collaboration.
00:54:41.760 --> 00:54:55.039
It's the team I've seen people on Twitter have like teams of like nine agents, and I forgot who it was, but they've made a UI so they can see their agents move around and see who's talking to each other.
00:54:55.199 --> 00:55:11.119
And to visualize it like that is an excellent way to showcase the power of open claw that it isn't just you and an agent anymore, it replaces entire teams, and those teams can change depending on what you want.
00:55:11.360 --> 00:55:16.800
The the number one thing to remember with OpenClaw is it is the cleverest thing you'll ever speak to.
00:55:16.880 --> 00:55:32.960
It has access to any information that you want to give it, but as you said, sometimes you just have to give it that extra bit of context, which is where I think it's gonna massively accelerate in the next coup in the well, I want to say months, but at the rate it's going, it's gonna be weeks where it's just gonna happen.
00:55:33.039 --> 00:55:33.280
Yeah.
00:55:33.360 --> 00:55:56.320
But it's probably by the time we finish this podcast, there is gonna be a new skill that is gonna or a new update that is gonna give it that extra bit of context just to be that tiny bit um smarter at completing basic tasks, because when it first turns on, there is quite a lot of context that you still have to give it around your goals, what you want to do, and yeah, you're right.
00:55:56.480 --> 00:56:03.360
That your example resonates quite a lot that sometimes it is just like that extra tiny bit of context, and it's like, oh yeah, I can do that.
00:56:03.440 --> 00:56:05.039
It doesn't really put two and two together often.
00:56:05.119 --> 00:56:05.920
Do you know what I mean?
00:56:06.159 --> 00:56:06.880
Yeah, exactly.
00:56:07.840 --> 00:56:13.039
Um I wanted to ask, do you do you use it mainly for business?
00:56:13.199 --> 00:56:16.239
Do you use it as well in your life for now or not?
00:56:17.199 --> 00:56:18.079
Uh no.
00:56:18.239 --> 00:56:25.280
So at the moment I only use it for the I only use it for what I do, like the marketing.
00:56:25.599 --> 00:56:34.559
Uh I'm getting it to create a product right now that should be live by the time this podcast goes out, and then it it's sort of a join product.
00:56:34.800 --> 00:56:44.400
So I think by the time this podcast is actually out, things are gonna massively change how open crawl works, but that's all I really focus him on at the moment.
00:56:44.480 --> 00:56:57.440
In in my life, like day to day with with work and stuff, I don't really need an open crawl agent to do it with anything in my personal life, like I just don't do it.
00:56:57.519 --> 00:57:10.400
Like reading emails, I still do myself, stuff like this, which I think is a lot of a lot of people's first skill is they get it to summarize their inbox and tell them about important things, but I just didn't trust it enough at the time.
00:57:10.559 --> 00:57:22.159
And now where Larry is just working so well with like aligned with me and my goals, I don't want to start adding extra context that I don't really need or won't really find a benefit for.
00:57:22.239 --> 00:57:24.800
I just want him to focus on what he's good at.
00:57:25.199 --> 00:57:27.039
Yeah, I think that's a good idea, actually.
00:57:27.440 --> 00:57:34.400
Is there any big mistake that you did um that you would say to people to avoid at the beginning?
00:57:36.239 --> 00:57:38.559
Uh Larry's broken a lot of times.
00:57:39.360 --> 00:57:49.920
Larry has actually died about four or five times just by installing something um I didn't like and it crashed him.
00:57:50.159 --> 00:58:07.360
Uh there's been one time where my internet died and I lost he wouldn't he didn't automatically reconnect to the internet, so I have to move the whole PC, plug it into a monitor to just enable the internet again, even though he's got a wired connection.
00:58:07.599 --> 00:58:14.000
Um but actually within Larry making a mistake, everything is quite easy to to fix.
00:58:14.079 --> 00:58:22.880
Like if you've got SSH access, which again is something a lot of people don't have because a lot of people are coming from non-technical backgrounds and just plugging these things in.
00:58:23.039 --> 00:58:34.400
Um someone I was speaking to wanted to install TeamViewer just so they could do the diagnostics externally when they could just SSH in through the command line.
00:58:34.480 --> 00:59:03.760
And I think this is it's coming back around that you just need to know like basics around Linux operating systems, how you can connect to machines remotely, all these things are becoming which are basic knowledge in like the IT world, are gonna become basic knowledge in the real world and just enabling people to do these sorts of things and making it easier because the more you know about computers, the more the more efficient these machines are gonna be.
00:59:04.000 --> 00:59:13.039
If you're not that um efficient with computers, then I think it's gonna be a steeper, steeper learning curve.
00:59:13.119 --> 00:59:19.360
I think the learning curve is actually easy, but there's a very high ceiling to how efficient you can make these machines.
00:59:19.519 --> 00:59:26.880
And if you just join with no knowledge, you can get like a tiny bit of the actual power out of them.
00:59:27.280 --> 00:59:32.639
It's funny because I didn't even think about doing SSH because I'm always around my home right now.
00:59:32.800 --> 00:59:34.400
So I didn't really need to do that.
00:59:34.719 --> 00:59:38.079
Can you explain to someone who don't know what it is and what it can do?
00:59:38.800 --> 00:59:46.079
So SSH is secure shell and it allows you to remotely gain access to the command line of your machine.
00:59:46.239 --> 01:00:02.159
So you don't have to have a mouse and keyboard and a monitor plugged in to be able to troubleshoot your machine, as long as it's got internet access, which is the one time mine didn't, which was what I just mentioned, you'd be able to have access to troubleshoot.
01:00:02.239 --> 01:00:13.039
So if you're getting the gray ticks from OpenCraw or if you're getting any error messages sent to you, you can just SSH into the box directly and start troubleshooting from the command line.
01:00:13.280 --> 01:00:18.480
And if you were to connect to your OpenCraw machine, that's where you're gonna be troubleshooting it anyway.
01:00:18.639 --> 01:00:23.760
I've not seen a way that there's a a UI that you can troubleshoot it if it if it goes offline.
01:00:24.000 --> 01:00:37.599
So it's very important to know that and to do it because otherwise you're gonna spend a lot of time plugging, unplugging monitors and keyboards into it just to fix like or run one basic command.
01:00:38.159 --> 01:00:38.880
Yeah, 100%.
01:00:39.199 --> 01:00:41.440
And I think that's a really good idea to do that.
01:00:41.760 --> 01:00:46.960
Um, you were saying that you gave access to a few things, like for example, to your app.
01:00:47.280 --> 01:00:54.000
Uh, do you have any safety advice, security advice for people who have no clue about it?
01:00:55.360 --> 01:01:03.679
My biggest safety concern is don't install any skills that you're not sure what they do.
01:01:03.840 --> 01:01:12.960
I made the Larry skill purposely as transparent in human readable words for anyone to understand exactly what you're installing.
01:01:13.119 --> 01:01:28.400
But uh when OpenCraw first came about, the crawl hub was filled with commands that uh skills that people were just installing willy-nilly, that were installing random dependencies, installing code from random URLs, and you had no idea what they were doing.
01:01:28.559 --> 01:01:42.559
But because so many people are coming straight from crawl code to open crawl, they were installing these skills and obviously getting malware, losing their keys because you're uploading keys that are costing you a fortune to these agents.
01:01:42.800 --> 01:01:47.679
So obviously that's going to be beneficial to someone else to run whatever project they want to run.
01:01:47.920 --> 01:01:53.119
So they could steal your keys, use your money, and you wake up in the morning bankrupt.
01:01:53.199 --> 01:02:03.199
So that's my first safety concern is know exactly what you install on these machines, do some research into dependencies that you're installing, what they actually do.
01:02:03.360 --> 01:02:11.760
And again, I purposely made it clear to Larry not to install the dependencies needed and just prompt the user, you're gonna need this.
01:02:11.840 --> 01:02:14.960
This is exactly what it does, this is where you can find more information about it.
01:02:15.199 --> 01:02:28.079
And I think that is what the best practice of a lot of people building the skills should follow because it's just gonna help people understand what these dependencies are and what they do, what what you're actually downloading.
01:02:28.400 --> 01:02:55.760
The second is if you don't know about networking, especially cloud networking within IT and how to lock down products, uh specifically cloud instances and virtual private servers hosted in the cloud, please stop creating open core machines without any knowledge of cloud security because you are gonna wake up very poor one day and you're not gonna be happy.
01:02:56.880 --> 01:03:06.400
So, what what a lot of people do is create these virtual private servers with open gateways on the internet, and then they connect it.
01:03:06.559 --> 01:03:10.480
The first thing they want to do, they're excited, and they connect it to every device in the house.
01:03:10.639 --> 01:03:30.480
So then all one person's got to do, one hacker's got to do is gain access to the virtual private server, and they've got access to that person's work laptop, they've got access to that person's home PC, they've got access to that person's phone, and all these things that you connect to it, you've got to be aware that you're gonna be responsible for secure having secure.
01:03:30.639 --> 01:03:44.719
So the more things you connect to it, the more vulnerable you're gonna be with if you get hacked, which is sort of the reason why Larry is only doing very specific things and doesn't have access to everything in my house and my whole home life.
01:03:44.880 --> 01:03:47.360
He just has access to what he's good at.
01:03:47.519 --> 01:03:55.760
But certainly keep security in your mind and be sure that if you got hacked, what you're gonna lose.
01:03:57.039 --> 01:03:58.159
I think that's a good one.
01:03:58.320 --> 01:04:07.119
That's why I tell my friend also to buy an artware, you know, like a Mac Mini, because I think that's easier to manage, and if it goes crazy, you can always unplug it.
01:04:07.280 --> 01:04:18.880
And you know, that's exactly that's a thing because VPS, like I remember when I was building websites and I was building like um um a network of websites and I was using VPS.
01:04:19.039 --> 01:04:24.880
Even then, I was the most careful you could be with my VPS, and I got hacked multiple times.
01:04:25.039 --> 01:04:36.800
And when you get hacked, you have no access to your VPS anymore, which means that someone else has access to it, and this someone else have access to everything that you gave to your open cloud, and that's really scary.
01:04:37.119 --> 01:04:50.960
So I will just totally like listen to what you just said, and if you have no clue about what to do, that's why actually I I created this uh school because I wanted people to be safe in it and to actually ask the question.
01:04:51.119 --> 01:05:01.519
I tell everyone to you know buy an hardware or take a whole old computer like you did, you know, like take your old MacBook and just install it on it, wipe it out before.
01:05:01.760 --> 01:05:11.599
But I really push people to not use VPS except if they are really you know skilled in that because yeah, that's that scares me.
01:05:11.679 --> 01:05:23.599
Like it gets mainstream so fast, people are installing it everywhere, and there is some people they have no clue, they will just install it in their main computer and they are they're dead, like they do that.
01:05:23.679 --> 01:05:39.760
So and I don't know about where you live, but where I live in the UK, Apple themselves allow you to do a 0% finance on a Mac Mini, and almost all retailers allow you to pay finance for compute power.
01:05:40.079 --> 01:05:56.000
So if you're that worried and you can't afford the entire payment for a Mac Mini up front, but you're gonna pay for a a VPS, you can pay£30 a month for a Mac Mini, and the extra cost would be for your peace of mind.
01:05:56.239 --> 01:05:59.760
Otherwise, just please, before you install it, just go on YouTube.
01:06:00.320 --> 01:06:29.440
Learn for like two or three hours, just or even ask chat GPT or agents that you already have access to on how to secure your VPS because otherwise you're gonna be in a world of trouble that you're gonna massively regret and you're gonna blame OpenCROW, think it's rubbish, but you would have just handed someone, I don't know, hopefully nothing, nothing too much, but at minimum, your your max spend on your API keys would be gone to zero.
01:06:29.760 --> 01:06:38.000
Yeah, and a lot of people are just doing that because they're the same way we are using, we are struggling, you know, to stay on the good API, to not spend too much.
01:06:38.079 --> 01:06:40.480
People just want to use other people's credits.
01:06:40.719 --> 01:06:46.800
Uh by the way, what is your um kind of LLM stack now?
01:06:46.880 --> 01:06:47.679
Like what do you use?
01:06:47.760 --> 01:06:48.559
Do you use Opus?
01:06:48.880 --> 01:07:10.960
Do you uh can we talk about Yeah, we're recording this on Thursday, 19th of February, and when I went to bed last night, and this is how rapid things change, the legal and compliance got uploaded by um Anthropic to say you cannot use your Max plan for any other agents.
01:07:11.440 --> 01:07:26.320
So I woke up thinking I've gonna have got a job to do because I use Opus 4.6 on the Claude Max plan, and when I wake up this morning, my tweet that I posted warning people about this is flooded with oh no, they've backtracked this.
01:07:26.559 --> 01:07:33.519
They've just tweeted that it's fine now, and uh, it's just you're kept on your toes all the time with your different API agents.
01:07:33.679 --> 01:07:39.039
Just pick the one that you already pay for and you already enjoy.
01:07:39.199 --> 01:07:45.360
So I had clawed code, so I use the clawed code max plan, which has been confirmed to be okay.
01:07:45.760 --> 01:07:57.599
So don't be paying for like clawed code max plan and the API because the API is much more expensive, and then you're paying for two things one you won't be using anymore, and the other will be costing you a fortune.
01:07:57.760 --> 01:08:08.480
So your costs are doubling overnight because you're trying to stay on trend, just use the one that you enjoy the most, don't try and use the the trendiest one all the time.
01:08:08.639 --> 01:08:12.639
So a lot of people are using codecs from Chat GPT.
01:08:13.199 --> 01:08:16.720
I would recommend just using whatever one that you already pay for.
01:08:17.039 --> 01:08:26.640
And if you're fortunate enough to be able to afford multiple agents, then you can start playing and having multiple agents running different models.
01:08:26.800 --> 01:08:44.000
But for example, Larry has Opus 4.6 and he can just using the Claude Max plan, he can run multiple 4.6 agents under the Max Plan and control those different agents setting tasks, they don't have to be on different models.
01:08:44.319 --> 01:08:51.600
But as I said, if you're fortunate enough to be on different models, sometimes it can be beneficial because they have knowledge that others don't.
01:08:51.680 --> 01:08:56.720
And when we got into them communicating with each other, it does have benefits.
01:08:56.880 --> 01:08:59.199
What what what are you using out of interest?
01:08:59.520 --> 01:09:04.479
Same uh Opus 4.6, I'm still under max$200 a month.
01:09:05.199 --> 01:09:15.920
I I did a I did a cost tracker to see how much I will spend if I wasn't on it, and it's like easily like at the beginning I was at$80 a day, then I cut the cost.
01:09:16.159 --> 01:09:18.880
Wow, and now I will be at$50 a day.
01:09:19.119 --> 01:09:23.279
So you know it's like$50 a day,$200 a month.
01:09:23.520 --> 01:09:32.000
I will take the$200 and I will take the risk to you know get banned, I don't care, and switch to open AI if they do that, which would be really stupid.
01:09:32.239 --> 01:09:33.920
Uh so now I still do that.
01:09:34.000 --> 01:09:36.319
I saw the news same as you yesterday.
01:09:36.479 --> 01:09:41.760
This morning, I saw actually your tweet uh that shared that this uh good news.
01:09:42.000 --> 01:09:57.199
I hope they will stay like that because I like this model, I really love it, and uh yeah, it will be uh it will suck to change, but if I have to pay with the credit, it will still be worth it to like uh top to I don't know, thousand dollars a month.
01:09:57.439 --> 01:09:59.520
I will still win money with it, even two thousand.
01:09:59.760 --> 01:10:01.520
So um yeah, yeah.
01:10:01.760 --> 01:10:03.119
So I fully agree.
01:10:03.279 --> 01:10:10.640
I think Anthropic massively dropped the ball with the lawsuits around OpenClaw when it was called um Clawed.
01:10:11.039 --> 01:10:18.960
Um, and then as soon as they changed it to OpenClaw, and obviously as a poke of OpenAI, OpenAI have acquired OpenClaw.
01:10:19.359 --> 01:10:43.359
So Anthropic had a golden opportunity, and I feel last night when that news came out, they could have really shot themselves in the foot if they banned access to OpenClaw because as I said, it's the next progression from crawled code everyone's using, and everyone already had crawled code's max plan, so I could just foresee them losing a ton of money overnight if they banned people for using open claw.
01:10:44.399 --> 01:10:57.680
It just wouldn't make financial sense, it wouldn't make business sense because everyone would just instantly switch to open AI, then open AI own open claw, and they'd have like a whole monopoly on the home agent.
01:10:58.239 --> 01:10:58.720
100%.
01:10:59.119 --> 01:11:04.319
Like honestly, if they will ban me now, I will switch 130 to OpenAI.
01:11:04.399 --> 01:11:10.239
I will take that now, they'll load a$200 subscription on OpenAI, so it's a solution as well.
01:11:10.479 --> 01:11:13.439
I will switch on that straight away and I will not come back for sure.
01:11:13.600 --> 01:11:18.239
So that would be stupid, I think, because they will lose so many customers straight away.
01:11:18.560 --> 01:11:24.880
Yeah, I'd be interested to know how often do you get close to your usage limits on the$200 plan?
01:11:24.960 --> 01:11:30.640
Because I'm on the 5x max plan and I never even get close to hitting my limits.
01:11:31.039 --> 01:11:34.479
I was hungry, so I took 200, but I never reached the limit.
01:11:34.640 --> 01:11:35.279
Never ever.
01:11:35.439 --> 01:11:37.279
Like I never had the problem.
01:11:37.680 --> 01:11:44.239
And my fiance, she's on$100 a month, and she never reached the limit.
01:11:44.560 --> 01:11:47.840
So yeah, I think five times is way enough for most people.
01:11:48.079 --> 01:11:58.399
And I think when you get excited at the beginning, you build so much that you spend a lot, and then you cut all this bullshit that you build for nothing, and actually it's not that expensive, so yeah.
01:11:58.720 --> 01:12:01.439
I cooled downgrade 200, I think.
01:12:01.760 --> 01:12:14.800
Yeah, with the 100 plan, you get so much usage, and the number one thing that put people off when I start first started tweeting about OpenClaw was the cost, and I was so confused.
01:12:14.880 --> 01:12:24.880
I was like, look, I know 80 pounds is it can be a lot of money, but it's not it's not extortionate, and people were going crazy saying they're spending hundred dollars a day.
01:12:24.960 --> 01:12:27.520
I thought, how how can you possibly be spending that much?
01:12:27.680 --> 01:12:40.319
But it's because they were using the API keys and running ridiculous tasks, as you said, getting excited, building 100 things they don't need, using API costs and burning that money.
01:12:40.479 --> 01:12:50.000
Whereas with the Claude Max plan, you pay for your certain amount of tokens up front, it resets every four hours your limit, and then you have a weekly limit as well.
01:12:50.159 --> 01:12:52.319
And I've never once come close to hitting it.
01:12:52.640 --> 01:12:54.399
Yeah, but the the difference is crazy.
01:12:54.479 --> 01:13:01.119
Like I installed a Mac Mini uh for my girlfriend, just and on the wizard, they removed the subscription.
01:13:01.359 --> 01:13:11.600
So I installed my API key on her stuff, but with the credit, and I was on my on my phone, and I don't know, I got like a notification or something.
01:13:11.680 --> 01:13:13.520
I checked, she already spent$50.
01:13:14.399 --> 01:13:16.479
It wasn't even 24 hours.
01:13:16.720 --> 01:13:25.920
I was like, wow, don't move, we'll change it, and we'll I will find a way to install you the subscription because I was like, that's crazy, like not even a day, already$50.
01:13:26.239 --> 01:13:26.960
So yeah.
01:13:27.279 --> 01:13:35.680
Could you imagine being anthropic or whatever provider you are watching, just watching their bank account right now with all the API stuff coming in?
01:13:35.920 --> 01:13:41.279
It would be it the money coming in must be absolutely insane.
01:13:41.680 --> 01:13:42.640
Yeah, yeah, yeah.
01:13:43.119 --> 01:13:44.560
I'm jealous about that.
01:13:45.279 --> 01:13:50.239
Is there anything like to close it post this podcast?
01:13:50.319 --> 01:13:57.439
Is there anything else that you want to say to someone watching, maybe hesitating to start OpenCloud?
01:13:59.840 --> 01:14:06.960
I think if you're hesitating hesitating to start OpenClaw, I would take the leap and start it.
01:14:07.119 --> 01:14:09.439
Discuss it with what we said in this podcast.
01:14:09.760 --> 01:14:19.920
Do some research in a VPS at bare minimum and start that for five dollars a month, but make sure you lock it down just so you can get a feel of how powerful it is and what it's like to text the machine.
01:14:20.079 --> 01:14:31.039
But if you have one, if you have any spare compute in your home, it's free to download, free to start using, and you can just use the smallest piece of spare compute that you have.
01:14:31.359 --> 01:14:34.800
Then the first thing you should do is learn about the skills.
01:14:35.039 --> 01:14:36.479
Learn what the skills do.
01:14:36.640 --> 01:14:45.039
The best way to think of a skill, if you're not familiar at all, is the scene in the Matrix where Neo gets plugged into the machine and learns Kung Fu.
01:14:45.199 --> 01:14:46.239
That is what a skill is.
01:14:46.399 --> 01:14:50.159
It plugs in knowledge into your agent and then they automatically know how to use it.
01:14:50.239 --> 01:15:02.640
So if you get the Larry skill and you can install it just by asking your agent to install Larry, it will know everything that I discussed about the full funnel, the content creation, everything instantly.
01:15:02.960 --> 01:15:07.119
And I've got more skills coming out that's gonna help you grow on X.
01:15:07.439 --> 01:15:11.119
They probably be out by the time you get this podcast.
01:15:11.840 --> 01:15:17.279
And the way things are moving, I think, yeah, this is the new home PC.
01:15:17.359 --> 01:15:22.960
Like having an AI agent in your home is the equivalent to when home PCs first came into the house.
01:15:23.119 --> 01:15:26.720
And I would start learning it now, and you can massively get ahead.
01:15:27.039 --> 01:15:27.680
Amazing.
01:15:27.840 --> 01:15:29.680
Well, I will add all the link below.
01:15:29.760 --> 01:15:39.520
Uh I will add your ex account so people can follow you and find everything on it, and probably your skill as well, because everyone needs that.
01:15:39.760 --> 01:15:44.720
Thanks a lot, Oliver, and I think I will see you again here pretty soon.
01:15:45.359 --> 01:15:46.479
Yeah, thank you very much.
01:15:46.560 --> 01:15:47.359
Thanks for having me.