The Exchanges

Q What about security? Like, are those, like, if I install some of those skills, are those, like, do I have peace of mind knowing that that's secure?

A Yeah, no, that's a really good question. Um, so there's a couple of ways around this. One, I think you can always ask the agent to do an audit of your security setup, which a lot of people don't think about. It's kind of almost like the meta prompting it. So Hermes has knowledge of where the keys are stored in your configuration. Um, and you can say like, is this a security setup? Tell me why or why not? And it'll go through and let you know if there are any secret keys exposed on your computer. If they're in plain text, if like a firewall is set up poorly, it'll let you know. Um, the other thing that's very unique to Hermes is that it's built to also out of the box be able to be ran inside of a Docker container in case you want it on your machine, but isolated from the rest of your files. And then you can also run it on modal as like a serverless service as well. Um, so it's really flexible in how you meet. I personally am a little bit risky and I just kind of run it on the bare metal. Uh, and I'm just routinely making sure like every day I'm updating it and I'm also making sure That like I ask it to, you know, secure my own setup.

AI assessment note: “able to be ran inside of a Docker container in case you want it”

Q That's awesome. Is that something that you recommend people, you know, use Hermes agent to help, help you set it up and, and, and productize some, some of the stuff that you're doing?

A Yeah. So I think like the idea of using agents to get things done is like a new paradigm. So the easiest way to like get used to it is to solve like personal problems in your life. So the biggest like personal problem that I first solved with an agent was, um, like figuring out what to cook at home. Because my wife and I, like, we love door dashing, we love eating out, but obviously, like, that's not, you know, that's not the healthiest, and that also costs a lot of money. So the first thing I did was I set up a local speech-to-text model on my actual computer, and I sent a long, like, eight-minute telegram voice message of me going through my fridge and my pantry of every single thing and every single ingredient that's in my pantry, and I said, every day, can you send me a recipe Or three recipes based on what's in my pantry and what, like, my fitness goals are. Um, it seems like something small, but it kind of takes a lot of, like, mental load away from, like, my day-to-day. Um, there's a lot of things, like, I think that if you start doing, like, really basic stuff like that, um, you know, like, we can, uh, you can kind of automate a lot. Also, we gotta, we gotta cut this out, bro. There's a lot of personal stuff in here, so. This knows way too much.

AI assessment note: “easiest way to like get used to it is to solve like personal problems”

Q Just cause I think some people don't know what G stack is. Can you just give a quick primer on what G stack is and you know, how, how, Why you think that people should install G-Stack with Hermes?

A Yeah. Yeah. So G-Stack is basically the way I understand it, um, is it was built for a cloud code and it was made by Gary Tan, the CEO of Y Combinator. Um, essentially the idea is that it takes the Y Combinator style startup process, which is like figuring out what works week over week, asking the right questions about what you should improve about your product and your business, and then helping you go and implement that as code. Um, and make decisions on that. So, um, those types of things were previously only available to people who were in Y Combinator, which was like the best, which is like one of the best startup accelerators in the world. Um, but now a lot of that knowledge has been basically open sourced as a skill that you can bolt onto your agent. Um, and that's something that like was never available before in the past. Um, so if you're obviously working on a startup, like an app or something, um, and you're, you know, not in San Francisco, or maybe like you're not familiar with like the startup methodologies. I highly recommend using something like that.

AI assessment note: “So G-Stack is basically... it takes the Y Combinator style startup process”

Q Yeah. What is the five percent of time I should care about it?

A Proprietary information that like maybe specific to your company or some methodology that is specific to you, that has to be referenced in every single conversation. Cause the annoying part with an agent on MD file is every time you go back and forth with the agent, it's added in the context, right? The cool thing about skills, and I'm going to talk about skills in a second, the way skills are designed, the skills are Used in a way that's called progressive disclosure. Meaning when you have a skill file, the entire thing is an additive context. It's just the title and the description. So the agent has the title and description in the context. And when you, let's say you have a notion report skill, right? And you tell your agent, Hey, I want you to create a notion report. It's then going to check its context and be like, Oh, I have this skill. Let me check out the entire document. So it's not in the context. What's in the context is the name and the description, but that's enough for the agent to be like, oh, this is a skill I need. Let me go use it, which is fantastic. I'm a skills maxi. And I'm going to show later in the episode, like how you craft the perfect skills, but with agent.md and claw.md files, it's context being added at every turn. Right? So let's say you have like a thousand line file claw.md and let's say that's like 7000 tokens. You're spending 7000 tokens on ev…

AI assessment note: “Proprietary information that like maybe specific to your company or some methodology”

Q It's so frustrating too, right? Cause you're like, you give it a task and it seems like so binary, like right or wrong. And then when you tell it, Hey, like, why didn't you look at the trust pilot? Why didn't you see if they've raised money? You're absolutely right. It's like, what?

A And, and the thing is, the reason why this is the case is the models, Um, actually don't think they're predictors of tokens, right? So when you give it English, when I give it English, it maps it on this vector graph, and then it looks for the closest resemblance, and it says, this is the response, right? So when you say, what is the capital of France? It maps it again on this graph, and it says, oh, Paris is pretty close by, then it gives you Paris. It has no, it doesn't think, it doesn't understand. It feels like it understands. It feels like it thinks. Heck, it even feels like it has emotion. That's because it's been trained on so much data. But it actually does not know how to think. And this is where a lot of people will be frustrated, um, with like, why is it not understanding me? You have to walk with it. So I told it, okay, this is how you research. And it's like, okay, it researches. And guess what? This is part of the context. And we're like, okay, now that you're done researching, when it's a good company, these are the qualities you look for. And then when it's really good, send me an email. And then once we had a successful run and we did it again and again, then I converted it to a skill. The reason being is a lot of people create the skills themselves, or I mean, they'll use the AI to create the skill, but it doesn't have the context on what a successful run look…

AI assessment note: “the reason why this is the case is the models, Um, actually don't think”

Q How should people think about security and these different products?

A I like to think of security as in just, like, scoping what they have access to. So, by default, Antigravity, Cloud Code, and Codex, they're very, very secure because they're built by these massive companies that have a lot on the line, um, to protect. And I just, you know, if you, if you're building out these agents to manage different elements of your business, like the other week I built one, um, that does, manages meta ads. And obviously that's quite a risky thing to give an agent control over managing ad budgets. So it just comes down to like what you feel comfortable giving the agent. And also you can control what privileges or you control what, um, tool permissions has access to. So that if it was compromised, for whatever reason, the worst case, like, isn't that bad. And that means, you know, just giving it, like, read-only access to certain important platforms and stuff like that. Does that make sense?

AI assessment note: “I like to think of security as in just, like, scoping what they have access to.”

Q management tool, slideshow tool, and then I'm going to go hire people to go, or do it myself, and I'm going to go do it. But there's this, Shift that's happening right now that instead of basically going to a tool to automate a function, you say to yourself, okay, if this was an AI employee, how can I spin this up? And that's, that's what you did, right, Oliver?

A Yeah, so I just thought of Larry as a AI employee, almost like a virtual assistant, uh, hiring a virtual assistant with one job to do this one thing, and that was his sole purpose, to go research, go find out as much as he could about slideshows in my niche, and figure it out himself, and he started to do a very good job. So, This is the first slide he created, um, and instantly you can tell it looks rubbish. So this was created with the Dali three, um, image model, and then it goes into a slightly better model afterwards, but because the image looked AI, it was a massive turnoff. Users didn't like it, the format was wrong, it had the black bars, and that's why it flopped. And then we started trying more hooks. You can see then he started trying facial reactions because they were trending at the time. The problem with this is humans are extremely good at recognizing what a human is, which makes us extremely good at recognizing what an AI human is. And I still don't think it's fully nailed, uh, Gemini just released Nano Banana II, and I've seen some very good images created on that, but I've not tried it myself to, to go in depth. So we kept, we kept trying, and then we finally hit our first banger of 137,000 views here, and it was the hook.

AI assessment note: “Yeah, so I just thought of Larry as a AI employee, almost like a virtual assistant”

Q to people who, who've seen OpenClaw, who might have installed it, barely use it, um, but have also seen, like, co-work and some of these other cloud hosted, you know, Manus now has a, you know, sort of OpenClaw competitor, like a lot of these, you know, co-work cloud hosted versions. Like, what do you say, like, why, my point is, why is OpenClaw better than whatever cloud alternative exists?

A The key one is you own it. It's in your house, you own the files, you have a lot of control over it. However, I use Manus a lot, and I think Manus is excellent, um, especially with the integrations it has out of the box. If you're sat on the fence about if you think OpenClaw is going to be useful to you, and you're one of these crazy people that wants to buy a Mac Mini just for OpenClaw, I would highly recommend just starting with Manus, seeing what you can achieve with that, um, and You definitely don't need a, a whole Mac mini for open core. The, the minimum requirements are very low. Just get something that hits those minimum requirements and has a lot of storage to, to store all the work that you do on it.

AI assessment note: “The key one is you own it. It's in your house”

Q Absolutely. Uh, You know, for people who want to get started, um, you know, we can leave it at this. Like, what do you suggest and, and just fire people up, you know, about this. Like, what, what can you say that fires them up to motivate them to actually do this?

A I think if you want to get started, Larry Brain is the best place to get started. It is not just a skills marketplace. It is one subscription, and your agent gets the entire context of all the skills. So when you first get OpenClaw, you will ask it questions, and it won't know how to do stuff, and you have to give it skills. If you get Larry Brain, you get the context of the entire marketplace, so you can ask your agent to do something. It will say, oh, there's a Larry Brain skill available for that. Do you want to download it? You download it, and then it can instantly start helping you. The Larry Skill is one of them. The Super X Alternative I mentioned is one of them. They're both free, so you can get Larry Brain, test it, find out how it works, and then we've got over 80 skills now available for you to download, and its main goal is to help you achieve your goals. So you say, I wanna, I wanna do this, I wanna make more money. Oh, here are the skills available to help you do that. I think that will massively help anyone Download an open crawl for the first time. Speed up. And I think a lot of the people who you mentioned who are stuck, they don't know where to go with it. They downloaded it. It's not working for them. I think that's gonna, that's gonna supercharge their, their open crawl account.

AI assessment note: “I think if you want to get started, Larry Brain is the best place to get started.”

Q So I see why it's interesting. But how does this get me better output?

A Exactly. Yeah. So the next thing is Obsidian released this new tool called Obsidian CLI. And what that allows you to do is it allows you to use Claude code and it can go and it can read all of the files in your Obsidian vault, which is a folder of text files. But with the Obsidian CLI, It can give Claude code, not only those files that it can read and access, but it can also give Claude code information about the interrelationships of those files. So you can see, so Claude code can see that, oh, this file is connected to this file and this file and this file. And that gets very interesting in terms of what Claude code can understand about you and what Claude code can understand about Uh, all of the relationships between the things that you're working on. It can start to surface patterns about what you're thinking about that you are not seeing for yourself. Some idea that you might've been writing about for a year in this vault, it could be a latent idea and it can just immediately say like, Hey, did you know that you've been writing about this same pattern in startups or in this particular project you're working on in every single note you're making across these different domains? And, and, and seeing that for the first time, Can be like a huge light bulb effect. It can cause like huge progressions in your learning and your understanding and your point of view on the world, but…

AI assessment note: “It can start to surface patterns about what you're thinking about that you are not seeing”

Q And is that how you started building a lot of your commands? Like it, it sort of, it started suggesting and you're like, just go build it.

A Well, I started off actually like building them myself. Like I'm like, Oh, trying to think about commands myself. But then, yeah, I said, I started asking the agent, like, well, wait a second, what commands do you think would be interesting? And, and just, and this might be useful. Another thing that I'd like to do is I like to move to like, Higher levels of abstraction when I'm using an LLM. What I mean by that is I could say like, oh, make a command that, um, tells me what I should focus on each day. And that's, that's like, um, a command that I thought of. But when I, when I, another thing you can do is you can step back and I can say, based on my obsidian vault and what you know about me, form an understanding of what you think my level of understanding, like, Where you think my skill level is in terms of a person and the projects that I'm working on. And based on that, suggest the kinds of commands I should use that would take me from the level I'm at to a higher level, right? Get like, you know, get it to suggest the commands for me instead of me suggesting the commands and I could pick between them. So look at this. This is, this is, it was, this was the agent's idea based on what it read in my vault, based on the notes I'm taking, right? So let's see what this is. Daily note idea extractor. Ideas, insights, and original thinking accumulate in detail in daily notes, but …

AI assessment note: “Well, I started off actually like building them myself... But then, yeah”

Q it's a really good niche. And then I could build a micro SAS to go in, you know, really amp up monetization and enterprise value. But I, I suspect there are some people who are like not bought in. So for the people that are naysayers about directories and creating a directory right now, specifically in this moment in time in 2026, what do you have to say to them?

A Yeah, that's a good point. And I would actually say if your timeline is To make money in less than six months, I would not build a directory. And I'm just saying that because I've been building them for a few years and I just want to save people the time. Well, that said, I don't know if there's, I'm obviously biased, but I don't know if there's a better playground in structured kind of approach to a project. Where you can learn how Claude code works and learn SEO and learn how to sell and get into the business of lead generation, which will always be around, um, then directories. And, um, the first directory that I built with Claude code was just a playground for me to learn. And because it's so simple, like you're able to. Like use this as like a vehicle to learn that high leverage skill, which I think is AI coding and SEO. So at the end of the day, the way that I think about directories is that it is still to this day, a playground for me to learn how to get people's eyes on my projects. And I've built other businesses and I've spent like a lot of money on Facebook ads and my first business. I would argue the top three ways to get distribution nowadays are ads, organic social, and SEO. Like those are the big three. So choose your distribution. I made the mistake of not focusing on distribution first and directories are a distribution first model. They kind of have an unfair …

AI assessment note: “if your timeline is To make money in less than six months, I would not”

Q Or if you had some sort of lead magnet or CRO, conversion rate optimization skill, you could invoke that, right?

A Exactly, yeah. And to do that, you could say, hey, um, perplexity, go do deep research on lead magnets and conversion rate optimization, uh, or whatnot, uh, compile all the learnings, and then you can say, hey, Claude, create a skill based on this. All right, so we're building it now. And then after we build this lead magnet, we're going to get to sort of like the traffic generation part. So you've got a landing page, you've got a website, you've got unique positioning. It looks good. It feels good. It sounds good. You've got this mechanism to collect email addresses, and then we're going to, we're going to figure out how we nurture those leads and we're going to figure out how to drive traffic into the whole system. And again, complete marketing system. So far we're at I don't know, 54 minutes or something like that, and, ah, you know, typically this would take a lot longer or it would cost a lot of money, ah, to hire an agency to do all this for you.

AI assessment note: “Exactly, yeah. And to do that, you could say, hey, um, perplexity”

Q Do you ever use Uh, Claude on, or Claude code on the phone, iOS?

A I don't. Uh, I tried to do it one time. Uh, so I used Claude, uh, the Claude app, and you can access the same sort of, like, tools from Claude desktop in the app, and, uh, Claude code there could access, like, existing repositories and things like that, but a lot of the work that I do is oftentimes, like, Knowledge work and things like that. Um, so it was nothing that was in, like, an existing repo, and I couldn't just keep working, but I've seen people use, like, Claude, C-L-A-W-E-D, where they're, you know, working with Claude Code and WhatsApp and all kinds of other tools, but I don't know. I feel like I already spend enough time with Claude Code. I don't really need it everywhere I'm chatting.

AI assessment note: “I don't. Uh, I tried to do it one time.”

Q Totally. Yeah. And I think like, I guess one question I, I have is like, is one Is one model better for being more of a beginner, non-technical live coder, or, you know, doesn't really matter?

A Yeah, it's a good question. I mean, I think the fair answer would be probably Codex, because Codex, ah, edged out Opus four, six a little bit on some of those coding benchmarks, and is, is kind of known for writing better production code. Probably codex in that way. Um, at the same time, one of the downsides, and like I said, this is, I could only do this in a totally balanced way because they're so different. Um, you know, at the same time for a vibe coder, knowing when to interject and stop codex and say, oh, wait, you're doing this this way. Uh, can you instead look at doing it this way? They're probably not going to know how to do that, right? And so that's where maybe Opus four, six is better where you could say, okay, spin up four or five agents and let them work with each other. Right?

AI assessment note: “I think the fair answer would be probably Codex, because Codex”

Q Yeah, it makes complete sense. You're also, You're also going to use less tokens in the end, right? Cause you're going to, right?

A Yeah. Cause the thing is the better your plan, the better your input, the better the initial set of documents that you give the model, um, the, the better the outcome. And if the better the outcome, there's no back and forth, right? Most people will have a route loop running. It'll be a basic plan and it'll do what you told it to do, but you weren't specific. So now you're going back and then maybe you're running another loop or you're going back and doing all these changes, but If you get it done right, if you invest the time in the planning stage, I hundred percent believe you'll save a lot more money, and this will help you clear up a lot of ideas. So like, for example, this idea that we just had, this TikTok UGC farm, um, how do we want it set up? Do we want it to be flat with search? Do we want it to be client campaign assets? There's a lot of like these minute details that You're not thinking about, and because you're not thinking about it, you're allowing cloud code to make those assumptions for you, right? Which at the end, after it's burned through a ton of tokens, now you're going back to change, right? We can save so much headache if we do the proper planning from the beginning. And hopefully, um, people see value in this, um, ask user question tool. Make sure you specify it in your prompt. And hopefully, Greg, that, that made sense.

AI assessment note: “Yeah. Cause the thing is the better your plan... there's no back and forth”

Q Because when I, when I read the question, what have you changed your mind about? Um, in the, in the last year, like it's such a big question, right? So I feel like people need like tools to get, you know, get them, actually get to the core of that question. That's the hard part.

A Yeah. And I think the calendar, like I am a big believer in just the simplest tools because look, like every productivity guru on the internet will sell you some expensive productivity solution or some expensive like second brain or whatever the thing gets called. And You know, what I found in my own experience working with some of those things is that, um, you end up moving a whole lot, but not actually making progress on things. You know, you like have this complicated productivity system, but you're not actually, uh, you know, creating progress on the things that you were supposed to work on. And so I pull up the simplest version. Like I, I, I use Google calendar, go look at my calendar. Um, from the beginning of the year, one of my core operating beliefs in January of this year was Achievement is not going to change my life. Like that has sort of been something over the last several years that I've grown to believe that, um, achievements are not going to change my life in any way. So I need to enjoy the process, right? It's like a pretty common thing we hear. And to me, one of my biggest things I've changed my mind on is that actually some achievements are so profound, uh, you've put so much towards them and worked so hard towards them. That they actually do have the potential to change your life and create an inflection point. And that it implies that they are worth leanin…

AI assessment note: “one of my biggest things I've changed my mind on is that actually some achievements”

Q I know LinkedIn suppresses links. Do you post it, and then after a few hours edit the post, and then put the link?

A Sometimes we do that, depending on the post. For this post, we just put it in right away. People have all sorts of ideas about LinkedIn, like you have to post at a certain time, whatever. If it's good content, it will get reach. I'll give you a few other examples just so you can see, because sometimes what we will do is we'll just put the CTA in the comments. We won't even put it in the actual main post. Sometimes we'll go back and edit the main post an hour later, but you can see here, like join the waitlist in the comments. So you can see, uh, this is the post. It's kind of just the post that's meant to go viral, get a lot of attention, and then we plug it in the comments. Here's the other thing, Greg, is that One thing I learned while working with my co-founder, Jake, is that the more subtly you sell your product, the better. So oftentimes for our other companies, he would just create posts, and there would be no CTA, and you would almost have to, like, go to his account and try to figure out what he's selling, and I was doing the complete opposite when I met him. I was, like, really hard selling our stuff, and he was driving, like, all the revenue for our businesses, and I was like, what's going on here? And so that's when I learned it's really about the art of the subtle cell. You have to get trust and then people, the, the less you tell people about your product, the more…

AI assessment note: “Sometimes we do that, depending on the post. For this post, we just put it”

Q the vibe coding mobile app space is like the, you know, next big thing maybe, you know, there's definitely a lot of opportunity there. A lot of people want to be able to do it. You know, Rourke, and there's the VibeCode app are, are two, you know, popular ones, um, in addition to anything. Um, like, have you, have you seen them? Have you played with them? Any thoughts?

A Yeah, I, honestly, I played with them. I still think, at least at the time of recording, um, anything is a bit better, and the reason is, for some reason, the design sense is a little bit better, and its ability to follow instructions is, It's just, it's so good. It's so good, um, compared to the other platforms. But again, you can get away with using any of them. They're all using the same models under the hood. I believe they're all probably using the Claude models under the hood. Um, it just, it's really just also down to preference, I think. So, if you prefer the style choice that Vibecode or Roark has, for sure, you can totally do it. But then if you, at least for me subjectively, I did kind of like the designs that I was getting with that.

AI assessment note: “honestly, I played with them. I still think, at least at the time of recording, um, anything is a bit better”

Q AI feature is really interesting. Impressed that it worked too. I mean, my real question with 4.5 is like, okay, great. Now it's made a landing page, but can it create a SaaS app? You know what I mean? Like, yeah, ultimately I'm interested in pushing 4.5 to the limits and I would, you know, I would be curious if it actually could build an app that I can sell.

A It, it absolutely could. Um, I mean, you know, if you wanted to build this entire product in Claude code with 4.5, you, you could definitely do that. Um, you know, like I, like with, with Claude 4.5 sonnet, like I've taken, um, you know, uh, idea for a SAS from, from zero to production ready. So you, you can certainly do it, um, with either of these models. Uh, now. What I have personally found, and to be clear, I haven't tried to build a full on application with Gemini three yet, but, um, I think that Claude's earlier models, like I would always go back to them. Like when I was, I was trying to use codex and GPT five before 5.1 came out and, uh, you know, Gemini 2.5 pro and stuff like that. And while the logic is good, And the, the thinking and the debugging is good. Um, getting from zero to fully ready was easier with the Claude models. Uh, that, that's just been my experience. So I always kind of try to lean on Claude as the workhorse and then have these other models doing what they're good at as sort of a advisory model, if you, if you will. So here they are like side by side. Now, I mean, if we wanted to, we could, uh, we could see how each of these builds out beyond the landing page and just say, Hey, create a clickable prototype, uh, for, for the concept. And we can kind of get a gauge on how, how it's thinking there.

AI assessment note: “It, it absolutely could. Um, I mean, you know, if you wanted to build”

Q we have this here? Let's go. The only downside is sometimes you get, uh, tripped up and you forget which agents are working on what. Here we go. Uh, looks like it fixed the error. Let's go back in, make sure we're not getting an error on the screen anymore. Is this like kind of percolating, uh, ideas for you, Greg, around Codex and things you can do with it?

A Yeah, totally. I also think, uh, first of all, I agree with the whole MCP complicated workflow to the errors. So I've tried using some of those tools and it's, Oh man, it's like, it just feels like I'm over engineering things. So I'm happy you said that. So that's, that was actually what I was thinking about. Um, the second thing is I was literally, I was thinking about this app and I, you know, one of the beauties about, uh, building an app instead of just like sketching it, uh, you know, on a notepad is it got me thinking that this in a mobile setting would actually probably rip. Um, You know, not that we're going to build an iOS app today, but, you know, maybe if people want in the future, we can.

AI assessment note: “Yeah, totally. I also think, uh, first of all, I agree”

Q So, so listicles, like, being on listicles, the top, you know, 25 restaurants in New York City, that sort of thing, brand mentions, but how do you actually How do you actually do that? Like, which tools do you need? Like, what's your stack to go and actually do these things?

A Totally. I mean, so there's tons of them out there, right? Like, like prompt watch is one, like AI for SEO or AI SEO tracker is another. That's like an indie one. Um, there's, uh, I can't remember the name of Jackie Chow's, uh, company. He just released one of these as well. Um, there is, uh, another one called try profound. That's more enterprise. But all of them, all of them kind of do a similar thing. All you're looking for is the URLs that are being referenced by the AI search for queries that are related to your product. So all of these companies and all of these products, they do almost really similar things. Um, you basically like give your product, it creates all of these like different ways that somebody could query for your product. And then it goes and it finds all of the URLs that are being referenced for those queries. So once you have that URL list, you're like set, you already know what AI is referencing, right? So now your job basically goes and it turns into, I need to reach out to every one of those websites and get our brand placed on each of those individual URLs, right? So just to like, again, uh, to use real numbers, um, I, I gotta keep that company hidden, but this is like what a friend just told me when I talked to him a couple of weeks ago. So imagine you find, like, you're a software company. Um, you basically find all of the queries that are related t…

AI assessment note: “like prompt watch is one, like AI for SEO or AI SEO tracker is another.”

Q Well, thanks for sharing that sauce. What else, what else you got?

A Yeah, the next one, um, I mean, this is a pretty simple one. I'm sure you do this all the time as a writer, but, uh, it's just voice dictated writing and you push it into a cloud, like workflow. Um, So there's no like magic secret other than this prompt. So you, I mean, people can, we can like drop this in the description. Uh, but it's basically just a prompt to train on your writing style. Right. So I grabbed all of my top performing tweets on X ever. Right. Uh, I have like 10 to 20 of them. I dropped this prompt into Claude. Um, and you basically just push all of your best writing and it's going to be trained on your tweets. And then, uh, I'm, this is a pretty popular workflow, but just like, I like going on walks and talking on my phone. I have like my best ideas while I'm walking. So I just like walk and I just have whisper flow open and I just like jam on different things. And then I have a bunch of different notes. And then when I come home, I just like pushed into this chat that's trained on my writing. And then you just get really, really good tweets. Um, so like for this, this example, They're pretty optimized tweets. You still have to go in and edit it a little bit. Like you want to get this. So once I get this, I'm going to like push it into like Twitter, for example, and I'm going to add a couple lines that are more like give more insight, right? The insight part, A…

AI assessment note: “Yeah, the next one, um, I mean, this is a pretty simple one.”

Q a walk and ideas hit you? Um, or do you have, like, I, I know I'm speaking about myself. I have an Apple Notes called Ideas, and in my Apple Notes I have these kind of half-bake ideas that then I'll use Whisperflow to, to actually dictate it in. Um, but, you know, I'm just curious, like, do you, do you start from nothing, or do you start from something?

A I usually start from nothing. The ideas really come to me when I'm walking. I might have like some broader things that I'm just thinking about. And I don't know. I think I work so much that I don't have thoughts until I start walking. Like my mind is almost numb because I'm working so much. So once I start walking and like, I'm not looking at slack, these ideas just come to me and I just, just start jamming, um, from there. And yeah, it's a really good workflow for me to, to, Just have tweets like 90% ready when I come home and start queuing them up. Obviously I'll add like an image or some kind of video because video content does really well on Twitter. Uh, but yeah, this is like a pretty simple one that I think will be useful to a lot of people who are trying to create like LinkedIn content or Twitter content.

AI assessment note: “I usually start from nothing. The ideas really come to me when I'm walking.”

Q Yeah. Um, with, with respect to mobile apps, right?

A With respect to mobile apps. And I mean, it, it all kind of boils down to attention at the end of the day. Like, if, if you're trying to make a product, uh, it's not just someone's money that you're after. Like, even before that, it's their attention. And there's so many fucking distractions. There's so many different things that are pulling you in a million different directions at all time that I find the best way to sort of introduce myself and get someone's attention for the first time is to do something stupid and do something funny, do something simple, not try to fucking explain Socrates to them or something. Um, but just basically tell a small joke or another analogy could be like, give them a small piece of candy. Right. As opposed to like, you know, here's a fucking steak.

AI assessment note: “With respect to mobile apps. And I mean, it, it all kind of boils down”

Q Right. Yeah, because what I'm trying to figure out, what I'm trying to reverse engineer is, how do you create a mobile app that gets millions of downloads? So, stupid simple one. What else?

A Uh, so, uh, in our case, there's always been some kind of viral content creation loop, meaning someone comes in, makes a piece of content, shares that content on other platforms, and that's what drives your user acquisition. Um, and so that whole, uh, that whole component has a lot of extremely important details. Um, so first is the content creation. How are you gonna help someone make a good piece of content? Also, the game has changed since five years ago, uh, with how far generative AI has come, um, and the kind of content that people are interested in making. But still, I think it's applicable today. Um, so here's how I would boil it down. How do you, how do you help someone make a piece of content that either they're going to post on Instagram and post on TikTok for the algorithm, and the algorithm is like, that's a great piece of content, let's give it millions of views, or more for their own, like, personal networks. And the people in the personal network are going to be like, wow, that's really awesome. I want to make that thing too.

AI assessment note: “in our case, there's always been some kind of viral content creation loop”

Q Okay. And, um, like, okay, so it goes number one. What next?

A Yeah, so, just in terms of the trajectory of the company, listen, I had no idea what the fuck I was doing. Like, I'm not, like, one of these, like, Silicon Valley kids who, like, watched every Paul Graham video. I was just, like, smoking weed in my bedroom playing League of Legends, and, like, really, really interested in AI. Um, so, made every mistake we could, uh, but, uh, at that time, now we're sort of at the end of twenty-twenty-one, We had a team of about 20 people all working here in Toronto. Um, got a few apartments that we were using as offices. Um, in twenty-twenty-two, We're like, let's raise money, because that's what you're supposed to do. Uh, ended up putting together a series A, uh, where the lead investor kept us in due diligence for two months, and then pulled out of the deal. And they pulled out of the deal partially because our original Wombo app used music, and there were some sort of unknown liabilities and risks around our use of music. I didn't know if I should bring it up with these Mixi guys, but Fucking the music licensing people are vicious, annoying motherfuckers, but also, you know, it is what it is. Um, and also partially because FTX crashed, and, uh, all this crazy shit was happening in crypto, and Zerp was over. Yeah, there was like a market spook. So we went from having fucking number one app, thinking we're about to get fifteen million dollars,…

AI assessment note: “just in terms of the trajectory of the company, listen, I had no idea”

Q Yeah, this is really smart. I've, I should do this because I do a lot of writing. Um, and I do this manually right now. So the fact that this can just be, you know, somewhat automated is really big. Beyond writing, what are other, you know, common use cases that you can, you know, use rules for?

A Yeah. Yeah. I'm happy to send you this prompt, by the way. Another one is custom commands. So this is a little bit more of a detailed command for things that I like to do when I'm reviewing code. So if you're, if you're newer to coding, some of these things might seem a little overwhelming, but you can kind of dial this up or dial this down. The idea holds true though, which is that you have, The agent and inside of here now I can run slash code review. So I've got this slash command and that slash command is just driven by this file, this markdown file. We can make whatever we want. We can do code review, security review, um, vibe check. Like we could, we could do whatever we want here. And inside of my code review custom slash command here, it's just little gotchas that have kind of bit me in the past. Like, Hey, did you think about what would happen when there's no internet? Did you think about what would happen when the application is loading information? Did you handle that with like a nice loading spinner? Did you, did you write good test? Did you write a like smaller number of tests that are really good versus just spamming a bunch of garbage? Did you change the authentication? If you did, you might want to run and do a security review. Uh, did you have, you know, places where we might want to speed things up with caching, like some of these more advanced, um, software e…

AI assessment note: “Another one is custom commands... things that I like to do when I'm reviewing code.”

Q Before we head out, is there anything we're missing around? If someone is going to use cursor today, you know, how, how can they get the most out of it?

A I think that for a lot of people, they have this tendency to make their conversation, this append only, they just continue in one chat forever. And I don't think they realize, I noticed this with a lot of people getting started. People are just kind of graduating from vibe coding into building a real production app. And they're like, It was great at first, but then it got bad and they're trying to figure out like, why, why did it get bad? Did it get bad? Because there's just a lot more code now. Did they secretly make the models worse? Is there some conspiracy? And almost always when I work with people and try to debug what went wrong, they're just trying to give the model way too much stuff. So like I mentioned with that context window, if it gets up to a hundred percent, cursor can automatically summarize and keep going, but you, you don't really actually want to do that. Because if you, if you get to like 80 or 90%, it's just very likely that there's so much junk in there, so much context that the model can get confused. And there's actually been some early research that has shown, uh, in long context tasks, especially with models that can have a million tokens of context. So just a ton of stuff. You don't really want to be kind of running that as your daily way of coding where it's like, ah, just give it 500,000 tokens of context. It just gets really confused and they, they…

AI assessment note: “for new tasks, for new features, start a new chat”

Q I'm happy that this exists, right? Because a big part of what makes a winning AI ad, or actually a winning ad in general, is how good the hook is, right?

A A hundred percent. I would say like everyone focuses on like the written hook. So this is like probably 50% of the result. And also a super important thing is like the visual hook. What do we see in the first second? And we will talk about that in the, uh, uh, future of this video, uh, because you can also with AI create some crazy stuff to hook people's attention very early on. They decide to scroll on an ad in like a millisecond, right? So you need to hook that, hook them even visually before they Listen to the beginning of the script. But with this toolkit, you can generate the first sentence. And I think it generates very good hooks because it was trained on so many hooks that we had in our database. Um, why did most people give up on learning a new language? I think it's very, very good hook. So you can save multiple ones, uh, save the ones you really like, delete the ones you don't like, and, um, just start making like a collection of hooks, uh, right here. And then you can just copy paste them all and put them into Um, a Google Sheets. Then once you have your hook, and I actually use the hook generator to get this one that I really like. True fact, most people never stick with learning a new language. Once I got that, I just wrote the rest of the script using Claude as well. And I got the script right here that you can then use with an AI actor. So we have this library o…

AI assessment note: “A hundred percent. I would say like everyone focuses on like the written hook.”