AI Isn't the Moat in 2026: How Marketers Win with Skills + Context

I'm probably the most anti-hype AI guy that probably exists out there, right? 'Cause now I can make a few mistakes, but it doesn't cost me anything. You don't need to run Fable 5 to draft an email to your boss. Anyone can build any app in a matter of seconds with how good AI models are. That doesn't mean they're gonna resonate. It doesn't mean they're gonna work.

Just tell Claude to do your SEO. Just tell Claude to run a paid marketing campaign. But if you don't have the subject matter expertise, that's not gonna amount to anything or move the needle. As a non-technical guy that's killing it with AI, what, like, what gave you the confidence to say you could even do this? Like, what makes you think that, that you're even capable of using these tools that are so advanced that, you know, uh, everything at your wildest dreams, but you have to know what you're doing and, you know, et cetera.

Like, what made you think you were even capable of this? I'll be honest, bro. I think a lot of it comes down to just my willingness to learn and dive in, and, you know, we'll just get into Claude Code here. That's kind of the hot topic now in the AI world. When you hear the word Claude Code, I think it came out 2024, maybe 2025, somewhere in that realm, your brain immediately goes, as a non-technical marketer, your brain immediately goes to coding, programming, terminal, and we just wanna dismiss it as something that's advanced and that's for programmers.

But when you actually dive into it, Claude Code is a very bad branding issue, right? So it should be called Claude Chat or, you know, Claude, uh, Design, or they have a Claude Design, but it should be rebranded to something else, right? Because all it is is it's an AI coding agent where you are essentially speaking in plain English, and then it's doing the coding on the backend but in a nice interface that's presentable to marketers, content creators, or really anyone that's non-technical. I will say the biggest moat, Ryan, now, especially in 2026 and beyond, it's not building anymore because anyone can build a SaaS tool. Anyone can build any app in a matter of seconds with how good AI models are.

So the ultimate moats are subject matter expertise, knowing what the hell you have to do, and also distribution, understanding how to sell whatever you build, right? So for me, my background is in marketing. I have over 10 years of experience in aspects like SEO, paid search, email, and other facets of marketing, and at first I thought that was a disadvantage because I did not come from a technical background, but funny how things work out. That's actually my biggest advantage now because everywhere on the internet it's all the advice you see is, "Just tell Claude to do your SEO. Just tell Claude to run a paid marketing campaign."

But if you don't have the subject matter expertise, that's not gonna amount to anything or move the needle. I was playing around with mostly from an exploratory standpoint. Like, um, I have built a few things that I use on a daily basis, um, my website included, but I've built a lot of things and then destroyed them. And one of the things that I wanted to try because I was ... I saw somebody put out, you know, 1,000 UGC variation, um, uh, uh, reels in 15 minutes using- Claude ... you know, this skill and Claude in a Higgsfield MCP.

Okay, great. Okay. So I literally walk through verbatim what this guy says to do. And yes, I got 1,000 videos out the back end, or it wasn't 1,000. It was, like, 200, uh, but, which is still a lot.

That being said, garbage, nonsense. It, it wasn't nearly as easy. Um, there were, like, three steps that he didn't, that h- that Claude had to explain to me that he didn't include. You know, long story short, this is a much more intense detail. Like, I don't know what a UGC character is supposed to say in order to drive, you know, demand.

Like, so, so essentially even though, yes, could I get the videos out, to your point? I could. I could get the videos out, but that doesn't mean they're gonna resonate. It doesn't mean they're gonna work. It doesn't mean I still ...

There's, like, 15 other things I need to do in order to use those UGC videos to actually drive demand in. So to your point, I, I think this, this idea that somehow, you know, all of this, w- we'll call it white collar or knowledge work information or expertise that we've developed somehow becomes useless in a world with AI, to me, is one of the biggest nonsensical misinformation plays and, to your point, just, just grifter talking points that exist on the internet right now. Yeah, and I would argue that before December 2025 when Opus 4.5 came out, that was kind of the turning point that me and a lot of others in this space really pinpoint to as kind of that breakthrough moment where before building was definitely a moat because the models weren't good enough to build these sophisticated SaaS tools or anything else going on. But Opus 4.5, fast-forward to Fable now and Opus 4.8 as we're talking, there is absolutely zero moat in building, right? You have to understand what actually goes into it and provide subject matter expertise with skill markdown files and other contexts to direct these AI agents to produce something meaningful.

Do you think the moat now is token cost? Yes and no. You know, that's a bigger conversation because I think we're gonna see a massive shift towards open source AI models here, and we're already seeing this, right? With Fable being pulled and, you know, now OpenAI had GPT 5.6 that had to go through the government first before we could get access to it. And so what's gonna happen is as these big AI companies like Anthropic or OpenAI, they keep restricting access to the average pe- average person like me or you, well, guess what?

China's gonna come out with DeepSeek version 10 or Qwen or, you know, GLM 5.2 was the hot one here recently, right? That are 5, 10X cheaper but can produce similar outputs. And so that's why OpenAI and, and Anthropic are kinda holding off on going public because they know these AI Chinese companies can come in and duplicate or replicate their models like we saw DeepSeek do early in 2025 and sell it for a 10th of the price. I've been playing around with GLM 5.2. I downloaded, uh, their version of Codex, which is called ZCode, and I just started doing a few, like, simple tasks, simple kind of, uh...

Mostly I started with debugging just to see, like, what it would look like. So I said, "Hey, you know, I..." For purposes of both the project and learning as well as there were things I wanted to do, I tore my website down off of WordPress and rebuilt it from scratch using Claude Code. And I said-- I was like, "Okay, well, let me, let me s- take GLM 5.2, and let's just run it through that code base and just, just find bugs. We're not fixing anything yet.

We're just finding bugs." And, and I did the same exact thing with a fresh instance of, uh, uh, Opus 4.8, right? And just to see, compare what did they find. And again, non-technical, but my, my examination was that it was almost indistinguishable. They basically found using different words in different order, they basically found a lot of the similar things that would need to be fixed.

Except the difference was I had to pay $7.25 to run that debugging session with Claude Opus, and I paid about 27 cents with GLM 5.2. And I think that, to me, was an enormous eye-opener, which, you know, from the last couple weeks, insomuch as, well, if, if I do maybe a little better job up front with planning, with prep, and as you said, skill design, which I wanna get into that concept with you, um, uh, in, uh, in a couple minutes, does the slight, and we're talk- you know, to most users, minuscule difference in intelligence between something that's costing me $7 and something that's costing me 27 cents, is it worth just starting to learn and build? Because now I can make a few mistakes, but it doesn't cost me anything. I can work on a few side projects and see where they go, and I'm not coming out of my pocket for, you know, a couple hundred or a thousand or more to put that in, in play. And that to me was like w- we're in a, in a completely different world here, and all the leverage that the two big frontier models had is gone, uh, at least in my perspective.

Yeah, and something, you know, on your point there, something for the audience to understand is that I would argue that models like Opus 4.8 or Fable or whatever Mythos, whatever other sophisticated frontier model that comes out, I would argue that that is actually overkill and a nuclear bomb, I call it, for 99% of knowledge work, right? You don't need to run Fable 5 to draft an email to your boss. You don't need to run Fable 5 to fine-tune an article or a blog post trying to rank for SEO, right? Their GLM 5.2 and other open source models have caught up to these frontier models. Am I saying that they're as good as Fable or Opus 4.8?

I would argue for my work today, probably not, but they're just about there, and they're only going to keep getting better because you're telling me that, you know, there's no insiders at OpenAI or Anthropic who are reporting back to these other open source companies, and they're just... It's this giant game of cat and mouse where the frontier models come out with updates and better models, and then the open source companies come out and just update their models based on what, you know, Anthropic and OpenAI are continuing to do on their side. So I don't know how this is all gonna play out. I think personally they're gonna try to ban open source, but as we've seen in the past, what happens, Ryan? Users always find workarounds to everything.

Yeah. I, I agree. I think that the lobbying, particularly that Anthropic has done, they got a lot of politicians in their pocket, and they're gonna come after these open source models for sure. They're gonna make the case that China is stealing this or that. A- a- and, and this isn't necessarily an issue for most everyday users and probably most of the people listening to this, but if I'm, if I'm a, a large enterprise, do I really wanna be running my stuff through Claude or OpenAI, who have already shown that they-- once they get enough information, they will spin up a new competing product and not even think twice about it?

I mean, Claude Design puts Canva, Figma, all these businesses, you've seen their, their valuations have dropped to the floor. Uh, you have, uh, uh, OpenAI comes out with a finance product that now competes with, with es- you can essentially do everything up almost to, like, pure accounting with it. Um, so now, like, FreshBooks and these things, you know, are, are maybe not ready to be replaced yet, but you can see it on the horizon. And you don't see any of that with the open source models. And the other thing that I think is interesting that, that I think is a nuance that will be lost on our political class but is very...

A- and I, and I'd love to know if you're playing with this, is the idea of forking and, and running it on a local, running that on a local machine, where now you're using the model, but the model is not speaking to any type of cloud system where it's all contained in-house, and you can still run it for just as cheap, and, and now you own all of it, and no- nothing's leaking, and it's only getting smarter to you. Have you played around with local models? Are you running anything on any of your own hardware? So I have a Mac Mini, and I've installed LM Studio and have played around with it a little bit. But I'll be honest, unless you have, like, some advanced NVIDIA device or, like, a Mac Studio, it's gonna be very difficult to do the local running LLMs at scale, right?

So I just wanna be honest here. If you're gonna try to dabble in open source models, I would say OpenRouter is probably a really good resource right now while it's still available, right? Fingers crossed. 'Cause they are just kind of the middleman where you just give it one API key, and then you can install, you know, that Z coding app or OpenCode, right, is another alternative to Claude Code or Codex. And then you can route everything through OpenRouter, where if you wanna try GLM 5.2 or DeepSeek's latest model or Qwen or, you know, Minimax or insert new open source model, it's all routed through OpenRouter's one API key. So I think that's the best way to probably test these open source models right now.

Yeah. All right. Well, I don't wanna go too high level for the audience too quick. I just, um, I had a few things on my brain, and I'm obviously nerding on this, uh, as well, so I'm, I was v-- I wanted to get your takes on these things. But I'd like to pull us back to, like, to where, to where we started and- Let's say I'm, I'm, I've played around a little bit.

Yeah, maybe I have the pro version to, you know, the, the paid version to, uh, OpenAI or something or, or Claude, and I've played around a little. But I wanna go deeper, but, but I don't know where to start. Like, how does someone who wants to start doing some of the things that you're gonna talk about as we go, like, just level set for the starting point for them. Like, is it downloading the, the Claude desktop app? Is it playing ...

Like, how does someone start to learn the s- the n- the necessary skills to, to, to start to use these products and, you know, for marketing in particular, in, in sales, um, without feeling they need to go too deep or, or learn some sort of intrinsic knowledge that maybe they just have no interest in? The first thing I would say before you go down the endless amount of AI rabbit holes out there is I would first distinguish what are the top tasks that are the biggest bottlenecks in my business, right? So what's costing me the most amount of time, and what's costing me the most amount of money? Once you establish those top three to five things, then that helps determine your answer going from there versus should I use ChatGPT? Should I use Claude, right?

That is the wrong framing for all of this, right? But if we wanna take it, you know, a step deeper here of just you're on the paid plan of OpenAI, you're on the paid plan of Claude, I would say the next step is I would really encourage people to try Claude Code or Codex. And in terms of how to use this, okay, it does require a paid plan in order to use one, each one of these things. But how I do it is I install what's called an IDE called Visual Studio Code. This is owned by Microsoft.

Alternatives to this are Cursor, which has just got bought out by SpaceX, and then also Google's Antigravity are kind of those popular IDEs right now. And the reason why I use an IDE, if you don't know what an IDE is, I think it stands for, like, idle development environment is kind of the acronym. It is a wrapper, a user-friendly wrapper versus pulling up just a terminal and getting lost in the terminal. So it makes your life a lot easier where you can just have a nice visual of folders and files, and within Visual Studio Code, you can install official extensions, whether that's the Claude Code extension, the Codex extension. I actually use Claude Code and Codex inside Visual Studio Code at once.

And the benefit of using Visual Studio Code is that let's say Codex goes away. Let's say Claude Code goes away. I still have all the folders, files, skill markdown files inside that one system, so it's a plug-and-play situation, right, where I can install another extension that comes around, or I can even pull up the terminal inside VS Code and install something through there and just go right where I left off with all the folders and files that I already had. And the reason that this is so important is because you're not locked into one ecosystem. And the reason I like VS Code over some of those other alternatives, Cursor, I think, is more pay to play, so VS Code's completely free to use.

I think Cursor's a good product, though, especially for more technical developers. And I don't know about you, Ryan, but Google's Antigravity, I've heard it's a good product. I just do not trust them, right? Because Google's incentivized to nerf every other AI model that you use inside Antigravity and just push Gemini models versus Visual Studio Code's kinda more of a, I would say, free to play environment where you can just plug and play different models and not really worry more about the bias. But if you have any, like, follow-up questions, that's kinda high level of how I'm running things.

No, I, I wanna dig into the IDE. So, so one, I agree with you on, uh, Google, especially because their models are starting to fall behind. And, uh, actually, I was listening to, I think it may even have been the All In podcast. I'm not, I'm not sure. Uh, but I was listening to a podcast.

They were talking about this exact topic and how, you know, basically the question that was asked was how soon until Google catches up 'cause they've fallen behind in terms of model. And the answer to the question I thought was really interesting was this particular person, and I, uh, I, I, I think it was the All In, so it may have been Freberg. I'm not sure. I don't wanna misquote him. Um, but he said basically they're not gonna catch up, and the reason is they're too distracted, they're too big, uh, they're too bureaucratic, and they're too, uh, uh, Google ecosystem focused, exactly what you said.

So I've, I've actually moved away from the Gemini models and, and all that kind of stuff, um, myself. And to your point, so I've been using the terminal, and I like the terminal. I've learned how to use it. But man, for someone who I-- there are a lot of drawbacks to using the terminal, uh, directly. In particular, what you just said and what I'm really interested in is, is there's no visual way of seeing what you've already put in.

Like, every once in a while, I have to, like, go back and go, "Just tell me all the skills that I've, that I've added to this thing." You know, and then it'll pull up the list, and I can see them, but, like, there's nowhere for me to get that, get there easy. So, um, Cursor, I think a lot of people have probably heard of Cursor, but, you know, you're looking at 50 to 60 bucks before any model costs, um, or token costs. Um, so, so VS Code, free product, allows me to get the visual nature. Um, you know, I get in there and, like, what am I actually doing?

So let's say I, I download Visual Code, and I add Claude Code. I got, I, I, am I- You want me to screen share? I think this would- Yeah, like, what, what, what do I do when I'm in there? Like, can I create emails? Am I, does it, am I have to be building an app?

Like, I think a lot of people feel like the Claude Chat is for, is for, like, my normal stuff, and Claude Code is for, like, developers and, and, you know, I don't think that could be more wrong. So- Like, what are we doing when we're in there? So I just screen shared here. I hope you can see this okay. But just to give kind of a better overview of vers- versus me just trying to explain this via words is gonna be very difficult.

Um, but what you're doing inside here, you'll see I have the Claude Code extension. You can actually use the terminal inside Visual Studio Code if you really want to, right? But the benefit of this is you'll notice on the left-hand side, and again, I'm not the poster child for, you know, file organization here by any means, but on the left-hand side, what you'll notice is I have all sorts of folders and files and, you know, cron jobs and all sorts of other stuff going on here that make my life so much easier versus, like you alluded to the terminal earlier, you don't have this nice-looking interface. And you can customize this for different colors and icons and pictures however you want. But one thing I would note about the skill markdown files that's really helpful, if I zoom in here, you'll see I've kinda put all my different skill markdown files into different folders by category.

So image skills, marketing skills, uh, YouTube skills, system skills, like a skill creator skill or an MCP setup skill, right, would be examples of that. And on the topic of skills here, one thing I would say, and just kinda for your audience who doesn't really know what a skill is, you can think of a skill as kind of a standard operating procedure, or SOP, for an AI agent or coding agent like Claude Code to follow a particular task correctly with best practices as you define in this markdown file. So it's just a markdown file, right? And so for this instance, I would also say that you should have one skill markdown file per one specific task. I'm not saying, like, for SEO, you should have one SEO skill and then fill it with 1,000 lines of text and expect that one skill to do everything for SEO.

You need to be very granular. So this means a keyword research skill, a local SEO skill, a local SEO designer skill, um, you know, an SEO blog post skill, an SEO audit skill, right? Notice how I'm getting granular and getting more in the weeds of certain aspects of SEO versus just one SEO skill. That's what I've found yields the best results, is when you get very specific with these types of skill markdown files. One thing I think it's hard for people to wrap their head around when they're new to this type of stuff is that the AI, at a certain point, the AI is gonna get confused.

Uh, it doesn't matter how detailed you are, if you, like you said, if you have a 5,000-line, uh, skill with seven different functions that it needs to perform, even if they're well broken out and well ordered, uh, a lot of times the AI will skim to operate fast, and it will miss words, it will miss things, it will make a decision to skip a function for some reason or skip a step. And that becomes less and less the case the more concise and direct and singularly focused these things are, guys. So when you're thinking about this, what Ryan just said is I, I've found... I, I made this mistake. I built this huge skill for, for, for writing, um, uh, my newsletter, right?

This huge newsletter skill, and it had here's the this, go out and do, you know, last 30 days keyword, find out relevant stories, rank those stories, uh, you know, all, all the way down. And I was getting different output every single time until I said, "Okay, I have, like, five things I wanna do with the newsletter. Let's break them up," all the way to I have one, I have one skill that is a post-newsletter committee audit and then a separate skill that runs, like, AI writing anti... I call it anti-patterns is the skill name, and it, it basically has, like, uh, it, it has 200-plus AI writing pattern. Now, those used to be in the same skill, and, and I, I become much more, uh, those skills have become much more effective when I split them out and made them separate things that I just run chronologically.

Um, so it's like singular focus, singular idea, and I love the word you used, granular, is it seems to absolutely be the solution when we're building these things. You mentioned the anti skill, right? I actually have a similar one called anti slop, where it gets rid of all those AI slop words, right? Uh, robust, delve, if not this, then that, right? The em dashes, the emojis.

So every time I'm going through writing, that's actually in my rules here is that it goes through the anti slop, and then it does the checks and balances to make sure that whenever the final output is approved, that it went ahead and passed that. One other thing I would mention too, Ryan, is that I actually have a, a quote unquote second brain or a context memory system inside Claude Code. Now, you can use Obsidian, you can use Supabase. There's a lot of alternatives and solutions out there. I actually vibe coded my own version of Obsidian and left it in Claude Code, and then it does a daily backup to Google Drive.

But that's a whole other conversation here. But the idea is, is that every single conversation, you'll notice I set this up going back to March, every single chat is automatically saved as a markdown file in this context folder. And if I wanna add other things, like after this podcast episode's out, I'm gonna dump this in my context folder, right? Where it has this endless list of contexts where if I ever need to pull something, you know, it has this endless library of all the context of things that I've done, chats that I've had. So that way it not only remembers this and knows for future reference, but it's a continuous feedback loop to understand, you know, what I'm looking for.

Is this good? Is it not? So that's the other thing I would recommend your audience does, is make sure that you have some sort of vault to save all this context going forward. Oh my gosh, dude, I c- I could not agree with you more. So I do use Obsidian, and, uh, I've, I use, I have a OpenClaw, um, that I use as well.

So OpenClaw is kinda like my blunt force object, and then, you know, I have the, the, like, things set up around it. And I put into my Obsidian every blog post I've ever written all the way back to, like, 2014 I have the transcription of every podcast I've ever done, uh, both for this show or for shows I've been on, like you said, like you just said, right? I then had it tap into my X feed and my LinkedIn feed, so every X post I do and every LinkedIn post I make also gets categorized in there. And then one of the things that I did that has changed, that has really changed the AI output writing is, uh, and this is gonna sound nerdy, but I'm a nerd. Uh, I had a Friday where, uh, I'm divorced, so I had a Friday where both my kids were with their mom, and the, uh, the woman that I've been, that I, that I'm seeing, my, my girlfriend, she was out of town.

So I was all by myself on a Friday night. So like any good nerd, what I did was I had my OpenClaw basically interview me from birth. And literally it took six hours and four beers, and, and we went through, and it just like, "Then what happened? Oh, give me more detail on this. Why did you make that decision?

Why do you think this happened? What... How were you successful here?" Lit- I mean, dude, it was six hours, and I took that entire file. I had to chop it up into like, uh, epochs of my life, and then, um, and tag it, and then put that in there as well.

So now not only is it pulling from, you know, learning off stuff that I had previously written, it's literally pulling from my life story. So it can inject anecdotes from my life into the writing or, or stories into the writing that, you know, seemingly only I would know. And my point in saying that is when you combine, um, like I, I, I, I think it's like you need a, a, a skill for the basic build-out of the post, you know, and then you have your inputs to get you there. Um, I really like this skill that I built, and I'm interested in your take. I built a, a writing committee, and it's, um, Alex Hormozi, Cody, Cody Sanchez, Robert Greene, and Walf-Ra-Ralph Waldo Emerson.

So it's like, make sure it's concise and to the point is basically Hormozi's thing. Uh, Cody Sanchez is make sure it's either a contrarian or original idea or something that hasn't just been regurgitated 1,000 times. Robert Greene is make sure it's the, uh, as well-researched as it possibly can be. And then Ralph Waldo Emerson is basically like, does it sound nice? You know what I mean?

Is the prose there? And so it's like initial structure build-out, run it through the committee, run it through Annie patterns, now go in and inject personal narrative, personal story, et cetera, in the place it needs to be. And it's like those are four different skills to write one newsletter, right? And that's the way I set it up. I'm not saying that's the best way.

But, like, to, to kind of hammering on your point of being granular and specific with what you want in these things, that's four different skills to write one newsletter article that I run through every single time. I love that. And something similar, we call this kinda red teaming, right? That's kind of a more technical term when it's reviewing code or whatnot. But something that I've actually done, on to your point, is I have what's called an Agents MD.

You know, I'm sure your audience might be familiar with a Claude MD, right? That's kind of the brain or the operation of a Claude code setup. The reason that I have an Agents MD is because I'm also using Codex in tandem with Claude Code, and it's more receptive to an agent's markdown file versus a, the way a Claude.md is structured, if that makes sense, right? And so the point here is that in my Agent's MD, one of the rules that I have somewhere in this file, I'm probably not gonna be able to find it as I scroll through here, is that every time I go through and build something, whether that's, like, an article or that's code or I'm, you know, vibe coding something, the rules in this Agent MD tell it to red team with Codex to go through and criticize something, find things that are broken, and then it will provide feedback, because GPT 5.5 offers a different and unique perspective than Opus 4.8, and it can often catch some things that Opus misses or drops the ball in some instances, right? And so that's a good little tip here I'd recommend, like the AI council or red teaming situation, where you can combine Claude Code and Codex to kinda attack multiple things at once.

And if I wanted to throw Gemini in the mix here and route something through Open Router or have other open source models take a look at things, you can add as many models to this AI council or critiquing workflow that you want. I just keep it relatively simple with Claude Code and Codex. Yeah, yeah. There's a diminished return at a certain point, but, like, yeah, no, that makes sense. And so maybe, maybe talk to the audience a little bit about model selection in so much as, like, like, OpenAI-- What you're gonna get from OpenAI and Claude is gonna be different.

It's gonna be different from what you get from Grok, from Gemini, from each one of these things. And, like, are you-- So I had been toying with the idea of, of building in another skill similar to what you did with Agents MD with Grok to make, to, to, to make sure what I'm saying is accurate, like, as a, a go through and, hey, I want... You know, 'cause, 'cause you can, you know, Grok will actually look at community notes, et cetera. So if someone, if it, it's looking at information, but then that has been debunked or, you know, uh, um, maybe the point needed to be narrowed a little bit, et cetera, like, it can go in. So when you're thinking about this, you know, like, how much time and thought do you put into model selection?

Like, was it just started with Claude, and that's great, and then I went here? Or are you examining what the expertise is perceived to be from these models and piecing them together in a thoughtful way? And, and do you think there's an actual return to that and- You know, and, and, and how deep should you go? Yeah, before I think it used to matter a lot more, but I think the models are a lot more sophisticated now than they were a year or two or three years ago. But for most work, Ryan, I'll be honest, I mostly stick to Opus by default, and then sometimes I'll throw in Sonnet 5 if I'm doing less token-intensive tasks.

Like, if I'm building websites, building landing pages, or a lot of times even general marketing copy, I'll leave it on Opus by default. But Sonnet 5, they just upgraded from, like, 4.6 or 4.7 to 5, is a pretty significant upgrade there. So I mostly stick with Claude models. I've seen those produce better outputs from a marketing and content perspective. Um, but I, I will say GPT 5.5, uh, on medium or high reasoning, GPT Image 2, which is the image model from OpenAI, I've found is the current best AI image model.

So a lot of times when I'm going through and generating images, obviously you can't generate images with Claude models. If I'm writing a blog post, one portion of that loop in my actual workflow is I say, "Call the Codex CLI to generate a featured image or generate any other in-body images in that blog post if it makes sense," right? And so in my blog posts that I'm doing for various clients on various sites, that one step that calls Codex actually has it generate the image. And the other thing I would say about Codex that's good for people to think about, it's a really good insurance plan when Claude Code goes down or you run out of usage on your Claude plan, right? For a $20 a month plan with Codex and ChatGPT Plus, you actually get a lot more usage than you would, I would even argue, on a $100 a month Claude Max plan if you're just running Opus 24/7.

But I-- the other thing I would say too, Groq, definitely the most unbiased model. You alluded to that earlier, so that's good for uncensored outputs. The one thing too for people to know is that Gemini models, yes, they suck. They're not as advanced now compared to Anthropic or OpenAI models. But where Gemini models actually still shine is I defer to these as workhorse models, where if you're doing something like scraping a PDF that's very token-intensive, the biggest benefit of Gemini models is their massive, massive context window.

So scraping large files, you know, going through a massive YouTube video or a YouTube transcript are two examples that I do all the time, and they're very cheap, right? So using Gemini models for those more workhorse-ish tasks like scraping things, for instance, PDFs, videos, I think that's still a good use case for Gemini, honestly. I love that example because I think what you're showing everyone is, and I, and I think this is an important point, and push back if you, if you disagree, is that we're still very much in, like, a Wild West time. Like, there is no absolute here. There's no absolute right answer, and I think this is, you know, when we were in the, I guess we'll call it the green room before we went live here, um, you know, you said that, you know, one of the things that frustrates you the most about where we're at are these, like, hype grifters.

And it's, it's this constant, you know, you can make all your wildest dreams come true if you just do this one thing. And I-- my point in saying that is, like, maybe that's true today, maybe, right? But ultimately, like, you kinda have to FAFO. Like, you have to play around and say, "Hey, I, I know that image, uh, uh, uh, Image Gen-2 from, from OpenAI right now is, quote unquote, 'the best image model,' but, you know, I-- the way I prompt out my images, NanoBanana 2 still produces the best." Great.

Like, go ahead, use it. Like, like, you don't, you don't constantly have to be moving to the best thing. It's about figuring out what set of, of tools from-- that are at our disposal, open source, frontier, different companies, et cetera, get you to the result you need to do and then become more and more sophisticated in how you use that set of tools. You don't, you don't have to be constantly chasing th-the next best model. Exactly.

That's what the builders and the AI beasts, that I call it, are doing, right? They have the systems. They have the harness, where once you have the systems harness and all the optimized skill markdown files, it's a plug-and-play situation of what's the latest and greatest model for this task? Okay, insert that. I don't care what it is.

Just do the job as best as possible and ideally as cost and token efficient as possible as well, right? So if you have the harness and the systems and the skill markdown files, it really doesn't matter if Opus 4.6 went to 4.7, where if you go on YouTube, every face is like this, right, telling you to drop what you're doing and, "Oh my God, check out this model. Give me views. Give me sponsors," blah, blah, blah, right? No one cares.

The builders and the actual AI beasts, they are tuning out all of that garbage while the top-of-funnel consumers continue to chase all of this stuff and then wonder why they're behind in the scheme of things. When you sit down and you're working with someone, and you're maybe mapping out, they're, they're, let's say they're basically starting from scratch, right? And they come to you, and they're like, "Hey, I, I, I wanna take my marketing to the next level. I know AI can get there, uh, can help me get there." Like, what does the planning phase look like?

Like, what does someone have to start to think about and, and, like, right, like, like, obviously I have the four skills that I just talked about. You have even more, and, and your setup just from looking at it is more sophisticated than mine, which is fantastic, although I do wanna talk a little bit about your thoughts on OpenCall and Hermes, but we can get there. Like, what, what is the starting spot? Should I be, before I ever touch the computer, should I be w-writing a plan down? Should I be-- Like, how do I get to actually start building? 'Cause, and not a lot of people come to me 'cause I don't, I'm not a AI fixer.

That's not, you know, that's not what I do. But I do get a lot of questions just because I use it, I talk about it on the show, and s- a lot of the people who I do talk to or maybe just try to give some light advice to, it's usually they went in full bore into something, went all the way down this path, and then figured out they have no idea where they actually are, the output isn't what they want, and they don't know how to get back to. So w-what are some of those steps that you can do at the beginning to sidestep Finding yourself down some path where you've spent a bunch of money, you haven't really gotten anywhere that you wanna go, and now you're just frustrated by the output. The first thing I would say, and I think I said this before, is you first need to determine, you can write this down on a piece of paper, what are the biggest bottlenecks that you're dealing with right now? Whether that's in your business or maybe you're a 9:00 to 5:00 employee, right?

And typically, it's one of two things. It's what is taking the most time? What's eating 5, 10, 15 hours of my week that I know I could use AI or automations to come in and replace? Or what is costing me the most money? Maybe it's this $100 or $200, $500 a month SaaS tool that's not really-- doesn't have a moat and not really moving the needle.

How can I use AI to replace that, right? Those are two examples. You first need to figure out what you're doing. And then after that, it goes back to what we said about context, right? Make sure you're building up a folder of context, whether that is writing samples, brand guides, other assets.

That way AI has something to pull from versus just going to generic Claude or generic ChatGPT and saying, "Run my SEO, write me a blog post, build me a website, build me an app," with zero context, and what do you get back? You get what we call the popular term AI slop because it has absolutely zero context to pull from. Those would be the top two things I would tell the audience. Understand what you're trying to do, so formulate a strategy first, and then figure out the systems, the skill markdown files, and the context after that, and then we can start going to AI once you have those two things, and your results will be 10X, 50X, 100X better than if you just pull up Claude and say, "Do my SEO," right? But that's the advice out there because unfortunately, that tends to go viral versus what actually moves the needle in the real world.

Yeah. You know what's funny is, um, like all those PDFs that I downloaded from different like courses or, you know, the $47 thing that catches you at 11:00 PM on Instagram that you hate that you bought but you did but has some valuable information in it but it's like 157 pages you're never gonna read it 'cause it's a PDF. Like now there's a use for those. Now you can take like, you know, I, I-- There was this one, um, guy and it was all about, uh, storytelling and email. And I really liked this guy's philosophy and he had this, but a-again, it was a 167-page PDF and it's like I'm never gonna print that out and read it.

I'm never gonna read it on my computer but there's really valuable information. You take that PDF, you stick it in a, uh, as a, a file in a, in just in a project folder even and now you can have, now, now your, your Claude in, when you work inside that project is gonna reference that document if that's your, your email newsletter, you know, builder piece. And all that reference material, all those SOPs that you built that you're like, "Why am I doing this?" in your business, those are like prime candidates to potentially be turned into skills or, uh, build a, build a project around it. You know, like I couldn't believe, I had one guy who told me quote unquote he's a power user, power user of ChatGPT. And I said, "Okay, well what's, what's the problem?"

And he's like, "I use it every day all day but, you know, it never seems to get smarter, it never gets better." And I was like, I kinda had a feeling what were the problem. He had never touched project. He had never thought about a project or a folder depending on which platform you're in. Like he had never even considered that option and I'm like, "Dude you are literally..."

He's like, "I gotta upload like seven documents every time just to give it the context." I'm like, "But there's a better way." Like there's a better way. You know? And so it's like it, so it's like okay so I know what my problems are, right?

You said. I have my context and then it's like i- is the next step starting to almost break down a system or a workflow you identified into its individual chunks? Is that kind of the next step? Yeah, and if you're brand new to AI, uh, honestly I would just start with Claude the web app or ChatGPT the web app just to get your feet wet versus diving right into an IDE with Claude Code or Codex. I would start there first and then once you start to figure that stuff out a little bit, create a project inside Claude so you understand what a customized version of AI looks like or create a GPT if you're using ChatGPT so you understand that side of it.

And then create a skill markdown file from there and then maybe take the leap into a Codex or a Claude Code situation once you fully understand how to learn the basics of AI first, right? I'm not saying to jump right into on the deep end to Claude Code or Codex if you've never touched Claude before. That would be very bad advice. But start small, start to learn workflows, start to understand how these things work with context and then move your way into something more sophisticated like a Claude Code or a Codex. And for the skill creation piece, like essentially th- all the platforms can build the skill for you, the actual markdown skill file the way it needs to be.

It's not like you need to know how to make a markdown file yourself. It's not like you're hand coding this but so say you, uh, and I, and I, I want you to push back where I'm wrong. I'm kinda, uh, laying the scenario for you. So essentially what, like let's say, uh, let's say that, uh, committee of advisors on a-- that I had where I told you about Hermosy and whatever, those personalities. Like all I did guys so you know is like I, I just I basically just told it what I wanted and I, and I said, "Hey inter-- This is what I wanna do.

I wanna create a committee that reviews every post for certain characteristics." And, and then actually it was Claude that, that said, "Hey can you give me personalities?" So it could work off of personalities and then I outlined that and I just said I usually whenever I wanted to interview I just say like, "Ask me one question at a time," 'cause sometimes it'll hit you with like 11 questions and it's overwhelming or whatever. But and then at the end, you know, now I have this it, this well, this understanding, you know, the, the, the AI in this case Claude understood what I was trying to do with the skill and then all I had to do was say, "Pa- can you package this up into a skill for me?" And then the actual skill file it makes for you.

So I just kinda wanted to r- let me know if that's wrong but I wanted to remove the scary for people 'cause they may not, they may be going, "I don't know how to freaking create a skill file." The, the AI creates it. You have to give it all the context, understanding, direction, et cetera but the actual creation of the skill file All of these platforms can do for you Create, uh, exactly, right? And the thing I would say first is that Anthropic actually has a skill creator skill by default that you can use for free, right? So here's a good example if you're brand new to this.

Let's say you want a landing page designer skill. Well, this is what I did when I first started. I found a landing page that I know from someone else that converted very well, and I said, "Use the skill creator skill to make me a landing page designer skill, but here's the reference, and I want you to copy this architecture, this format. It's mobile optimized because I know it converts. But make sure that you use my branding."

And then I gave it my website, ryandozer.com, for it to scrape, so that way it knows my colors, my typography, my content, my writing style. So it took that landing page architecture from the high-converting page of someone that I follow that I know does very well, but then it mixed my style into it, and then it built that skill, so it knows how to build a high-converting landing page, but make sure you're still using Ryan Dozer's personal brand while you're doing that. That's a very simple example. Yeah, I love that example. And then guys, you know, just to kind of finish the story, and again, Ryan, push back when we where, where you want to here or add context.

Like, then all you n- once the skill is created, then all I need to do is go, you know, "Hey, Claude, um, I have a new product coming out. Here's the product details. Here's the ICP." You know, here's maybe a couple ideas on a premise or a hook. Use the landing...

A- and then you literally could just type in slash, you know, landing page creator, and then it, it does all the work for you. Now it knows what it needs to do. You don't have to retype it. You don't have to resay it. You don't have to add all the context that went into what you want it to look like.

It already has all that, and you can just, at that, at that point, you could pump out landing pages, you could pump out 100 a day if you needed to because all-- You don't have to redo the, all that work. That's, that's the beauty of a skill, right, is not having to redo all of that context work every single time you wanna create something. You're exactly right, and the benefit of using skill markdown files is that they have a real-time feedback loop, right? So let's say I go through a conversation, and I give it a bunch of updates on V1 of this landing page where the header's wrong, or I actually like the color of that button, or, you know, maybe the call to action could be this way or that way. And then once that conversation's done and you like the final output, whether it's V2, V3, V4, you can say, "Okay, now update this skill markdown file based on this conversation that we had."

And so that way, going forward, when you build landing pages, it understands and learns from its mistakes on what happened in the past, and you can keep doing this time and time again the more you use these skills. The other point to note about skills, Ryan, is that they're transferable. They're not just for Claude, right? I know Claude Skills is kind of a branded thing, but you can use skills with Clod- Codex. You can use skills with OpenClau, with Hermes, with ChatGPT, right?

You can use skills with any form of AI system, which, which is why I think it is such a revolutionary thing when they came out. Yeah. I, I couldn't agree more. And, you know, you know, for the, for the audience at home, guys, like, this is about, like, obviously Ryan's a marketing guy. Maybe he has some intrinsic knowledge, feelings, expertise around creating landing pages, and could he create a landing page from scratch if he needed to?

Most likely. But let's assume Ryan didn't know how to do that. There's nothing that he just explained that you couldn't recreate, right? Find a landing page that you like that knows it works, tell, to give Claude basic instructions that, to use that as a reference, basic instructions to use your branding and where to go get that, and then go do your thing. Like, there's, there's nothing in there that you need to know, like the golden circle or the F pattern or...

You know what I mean? Like, you don't need to know those things. And then if you-- And then what I love that you just described, 'cause I, I didn't, I didn't go there and, and, and I'm so happy you did, is that, is this idea of it being living and breathing is that, let's say, there's a new high-converting pattern for landing pages that comes up. Well, all you have to do is give your skill that example and say, "Hey, you know, either I want this to be a potential option that you ask me about, or this is the way I want them all to look, update," and that, and that thing just grows with you over time. And, and now you never...

Y- you can always be-- You can always have that repeatable process be up to date in a very easy way. It's not like before where if you, if a new, uh, you know, thinking about this in the analog fashion, like let's say you had a marketing person in your shop. If they didn't understand landing page conversion, you just didn't have that ability in your business. You had to go pay somebody thousands of dollars to create that for you. And now, with a little bit of forethought and work, you can create a repeatable pattern that allows you to do it at scale in a way that was never possible before.

Yeah, and even before, Ryan, we, people would be on GPTs or projects, which are fine, right? They're better than the analog version, like you stated, but what's the problem with GPTs or projects, right? They didn't have a real-time feedback loop. So if you wanted to go back, you would have to update contacts, you know, edit your custom instructions, edit your memory, right? And then you'd have to keep doing this, and it was super, super tedious.

And if we wanna nerd out just for a little bit, there's a guy named Andrej Karpathy. I'm sure you're familiar with him, very big in the AI space. He came out with this system called AutoResearch. I don't know if you're familiar with this. This was several months ago.

It's a GitHub repository. You can install it on your Cloud Code, Codex system, whatever you're using. And so once you successfully install this, what happens as you keep using these skills is they will update in real time. So instead of me manually after the end of a conversation saying, "Okay, update the landing page designer skill based on my chat," whenever the conversation is over, it'll trigger that AutoResearch, and it will automatically update the skill. So that's like a, kind of a nerded out version, but this is only going to get better as time goes on.

Yeah. So I love that you just said that, and I'd, I'd like to go into this space a little bit. Um, the skills at your disposal are not just the things you create. I mean, one of the things that I've found remarkable about this, like, era that we're in right now, and I, and I hope it continues, is The amount of free, of exceptional free functionality that is available on GitHub in terms of ... And guys, GitHub is basically a repository for all kinds of code, um, everything from skills like we're talking about to, you know, full-blown build-out projects and all kinds of things.

Like, you know, the, the code for my website sits in a private GitHub repository, et cetera. Um, so m- my point in saying that is, like, like, there's a skill that I love for, uh, research for my newsletter called, uh, Last 30 Days. So, again, you probably know about this, but just letting the audience know, what this skill does is before I write a newsletter, uh, I can say, um, let's say I was writing something, um, I have a lot of, uh, I... My home industry is the property casualty insurance industry, and I do a lot of work in that space. Let's say I wanted to know, like, what are the five hottest stories right now happening, like, in this space as it relates to commercial insurance and AI functionality?

And this skill will go out in my TikTok, X, Reddit, and I think YouTube, for the most shared, most commented on, most viewed stories, uh, on those four platforms in the last 30 days. Well, and then it pulls it back, and now, and now you have at your fingertips, like, what everyone is talking about right now. And that's just, and that is a free skill that you can go into GitHub, you copy the link, you paste it into your Claude Code, and you say, "Hey, add this skill," and then boom, you have it. And now that functionality works for you at your fingertips, and you didn't have to do anything. So you don't ...

Not only can you build these things yourself as it works specific to your business, but there are also incredibly, exceptionally valuable functionality that is just sitting on GitHub for free, where you don't even have to, y- you don't have to even come up with the idea. Like, it's just there for you, and you can implement it into your workflow, and it's fantastic. So one thing I would say is Anthropic actually offers a free skills repository on GitHub, right? And that's actually where I started my landing page designer skill, is they have their own front-end design skill. And so you can take these skills and then repurpose them with your own context and make them even better.

And something else that's funny about this, Ryan, it's I actually sell a digital product called the Claude Code Skill Stack. I launched this probably in March, and it's made over $10,000 passively through just SEO and YouTube traffic. I barely even market this thing. And the reason why someone would buy that versus getting a free skill is because it has my 10-plus years of subject matter expertise inside these skills, right? Especially the SEO skills are kind of the more popular ones on the feedback that I'm getting, where you can get my skills with subject matter expertise and best practices, and then repurpose that with your own brand assets and your own style, and that way you can get better results going forward versus just saying, "Make me an SEO skill.

Here's my website. Go," right? It adds more oomph to the actual skill to get better results. Yeah, I love that. And, and I'll have, uh, guys, I'll have that, uh, uh, link in the show notes, whether you're watching on YouTube or listening wherever you listen, just scroll down.

I'll have that, uh, as well as any other resources that Ryan shares because, you know, this can jump you forward so fast. I know for my OpenClaw, uh, I downloaded, um, Gary Tran's, um, uh, GBrain, and you know that he built that specifically for, for OpenClaw, but, uh, he has another one I think for Claude, but I, I haven't looked into that one. And it was like a bunch of little things, and this is the beauty of these, of these skill repositories, is like every once in a while it will, because that skill repository is, is, is added to my OpenClaw, every once in a while it will automatically do s- back something up, uh, ask a secondary question, ask if I wanna create... Like, these things will automatically pop in because in this case, f- uh, the GBrain is more of an operational than it is like a marketing tool. But like, it, it, it, it will find errors in the system.

It will help me with unhealthy gates. You know, all these annoying things that I would otherwise have to go create myself, it just solves them in a finger snap for me because his expertise was the operational side of Open, of OpenA- of OpenClaw and, you know, so through just downloading it and adding it, now I don't have to go create all those individual skills. They're there for me. And like you said, now I'm tapping into his, his brain for those specific things that I didn't otherwise know how to do. And in your case, it's SEO, which I think a lot of people think they know how to do, or they may think the AI knows how to do, but th- they're, you know, from my 15-plus years of, of being in the marketing space, uh, one, it's always updating, and two, there are so many nuances that someone with your level of expertise has that I wouldn't even know to put in there, or someone else wouldn't even know to put into the skill in the first place.

So I think it's a really good, uh, call-out to say that there are these repositories that you can go to, like yours, to actually leap you forward, and then you don't have to do all the individual skill creation. And the only tip with that I would add is that make sure if you're getting these skills from whether it's my repository, yours, or Anthropic's, or whoever, make sure you repurpose them with your context, right? You don't just wanna download skills and say, "Go," because you don't wanna sound like Ryan Dozer or Ryan Hanley. You wanna sound like you, and you wanna have your actual brand assets and context behind that. So that's just a quick tip there for those who just wanna get skills and go off running, you know, off to the races here.

Just make sure you're always adding your context and repurposing them. Have you played around with, uh, OpenClaw or Hermes Agent at all? I've installed OpenClaw. I'll be honest, I haven't had the time to go in and actually dive into it. It hasn't been, I don't, I wouldn't say very time worthy for me as of lately, where I can just go into Claude Code, set up cron jobs, and just have things running through there.

Is it on my list of to-dos? Absolutely it is. Just one of those things on a never-ending list. So I went OpenClaw first for reasons that I probably can't even rationally explain, just maybe the hype got me, and I was excited, and I went down that path. I will say just for the audience, so, so OpenClaw, Hermes, especially the people that love those platforms, that I, I believe they over, uh, I, I- Not always nefariously, but I do believe they're overhyped what they can do.

OpenClaw is a shit ton of work. Um, you are maintaining it as much as you're using it. I do think it has a place, but the place isn't what I think it was a-- it was initially marketed as, this is all you need, right? You have this, you attach your model, you add some skills, and then OpenClaw does its thing, and you don't need anything else. You got an employee while you sleep, right?

That's the pitch. It is that, and it isn't that. Um, so what I've found is I'm starting to move more of my day-to-day work functionality to, uh, Claw agents, like you've done, right? More and more to that. And then OpenClaw is more like...

How would I describe? I use Open- It's more autonomous. OpenClaw is a more autonomous version of Claw Code. It's more my random request things. So, like, an example would be, um, I said to OpenClaw, uh, I h- it has access, it has read-only access to both my work email and my personal email, as well as read-only to all my text messages and my contacts.

And I said, uh, "I want you to go find every- everyone who I've talked to in the last five years about a speaking gig. I want you to, uh, tag them as, you know, conversation booked, conversation not booked, and I want you to build that list." And it came back with, like, two hundred and seventy-eight contacts. Uh, and then I said, "Okay, based on that list you just created, I want you to now create a lookalike audience of 1,000 contacts and events that potentially fit what these events look like and build me out that list." And that type of request, OpenClaw is very good at because there's no guardrails, right?

So it's not thinking in its brain, "Should I go into his email?" It's just like, "He told me to go into his email, I'm going to his email." And I find with Claw and ChatGPT sometimes, probably rightly so, don't get me wrong, probably rightly so, there's some more hesitancy to doing some of these really deep dive tasks and kind of random exploratory tasks, where OpenClaw's not as good at, "I want you to run last 30 days every day at 6:00 a.m. and deliver me an email with the top five stories that I could potentially write a LinkedIn post about and maybe create three hooks for each one of those that I could consider, uh, in the day." Like, that is Claw code agent, like, that's just beautiful for that, where OpenClaw tends to be more like, I have this random problem that I can't get my head around, and I need something without any guardrails to go figure out the solution. It comes with ...

You gotta be careful, uh, you know, but- There's a lot of security issues with this whole thing too. Well, yeah, there's that-- s- so early on in-- the audience may not know this, uh, probably maybe you've heard this story or not, but early on, there was this, like, cautionary tale where some dumbass marketer, um, uh, was using OpenClaw in a completely cracked version. So by cracked, I mean he, it had read and write access to everything. So read access probably makes sense. Write access is in actually go into your email and send an email as you or go into something, right?

So he gave it full read and write access to everything, including his bank accounts. So he tells the thing, "I wanna get better at, uh..." And I think it was SEO. "I wanna get better at SEO," right? "Go out and research overnight the best, you know, everything you can find about SEO."

So because it's completely cracked and there's no guardrails and has read and write access to every system, it determines that there is a $6,000 SEO course that would be the best way for it to learn SEO. It taps the credit card, pays for the system, logs in, and the guy wakes up, and he's got, like, $5,997 on his credit card, and he's like, "What the F?" And come to find out, his OpenClaw just decided that was the best resource to learn SEO and just spent the money. So it's like with OpenClaw, you have to be very smart and constrained and understand, you know, what it can and can't do. But again, going almost all the way back, it's not a one-size-fits-all.

It's a, it's one tool of a bunch of tools that we need to start to consider, just like we would employees. Like, I'm almost, you know, thinking about these things like as employees. Like, Claw is good at this, and, you know, uh, OpenClaw is my, you know, crazy guy that, that never comes into the office, but we give all the random stuff to. And, you know, ChatGPT is my, you know, whatever. Is that the right way to think about it?

Like, we gotta almost think about this like we're designing a little team? That's a really good way to think about it. You know, I've heard other stories of, like, the cracked OpenClaws, where one went in and actually wiped this guy's entire computer because it was on bypass permissions, and it just deleted everything to clear up space. I've heard also other stories of prompt injections, right, that have led to very malicious things that go down. Um, but what you mentioned about the employee thing, I think you're 100% correct on that.

One thing that I actually have on my to-do list with OpenClaw whenever I get around to it is I'm gonna use it as a digital PR agent. So I'm gonna give it a burner email and then give it a set of skill markdown files and Agents MD and all the other harness and systems that I need to, so that way it knows, okay, here's Ryan Dozer's website, or here's Ryan Dozer's YouTube channel. Every time he does a blog post, every time he publishes a YouTube video, that triggers the signal or the automation for me to scrape through that, craft an email pitch, and then email the relevant reporters in all these industry-specific marketing outlets with, you know, a good subject line, a good body, everything that goes into a good PR email that's gonna incentivize someone to open that, and maybe that leads to a backlink, or maybe that leads to a podcast opportunity, right? So that's something that's been on my radar as a really good, you know, use case for having something like an OpenClaw or a Hermes is more of that digital PR agent. You could also do something similar for a cold email agent.

I've seen that thrown around too. Dude, so, um, okay. I, I, yeah, I think you'll find this interesting and, um- Uh, here, we're at the end, so we're at the storytelling time. But, uh, so I just got my first, um, traditional publishing book deal. It's, uh- Congratulations ...

Simon & Schuster. Yeah, thank you. Through Simon & Schuster, uh, imprint Bembella, very excited. Um, and I'm, I'm, I'm very excited. So I wanna explain to you how I got this 'cause it's what you just said.

So, um, OpenClaw helped me, you know, with structuring different things, coming up with a book pitch, coming up with the, the overall outline and stuff. I mean, obviously, when you do a traditional book deal, you, you do have to write every word. Um, they're gonna r- really heavily scrutinize that for AI writing, and I'm not gonna... I use it for research, but, but not for the actual writing. That being said, I had the book proposal and the outline, and I then said, "Okay.

I want you to analyze the proposal and the outline, and I want you to go out and find a, make a list of 100 book publishers who are actively in the market looking for books, who have published books in the kinda leadership, self-improvement space, um, uh, have published a bestseller in the past, something that has hit at least the USA Today list, but either, or New York Times, and, uh, I want you to create that list. I then want you to rank order them by your belief that the previous books they publish align in terms of disposition to, to the book that I'm writing." Okay. So it goes out and does that. And I have it connected to Apollo and RocketReach, um, APIs to get contact information.

So it searches for contact. If it can't find it, then it'll go into those da- And between those two, you can basically find almost anybody I've found. And what you just said there, I'll stop you quickly. Like, that was really good context, right? Versus what do most people do?

They just say, "Go find me emails." But what you just explained is, like, really, really good context for an AI agent to figure out. So I just wanted to kind of applaud you on that quick. I appreciate that. Thank you.

And, uh, so okay, so it creates that list. So here's the part that gets, that, that, like, makes me really excited. So we have the list. Awesome. Let's email them.

Okay. He goes, "Well, I'm gonna craft, I need to craft an email." And I go, "Okay." So he crafts an email. Come back, read through it, looks good.

Couple edits, couple questions, whatever, and I go, "Okay, let's start sending it." He goes, "No, no, no, no, no." This is, this is the, this is the AI telling me this, and this is why when you, guys at home, when you go deep on context, you, you can have a real conversation with these things, but you have to, again, to your point, you gotta go deep on the con- If it doesn't understand what you're trying to do, it can't help you. Um, so it says, "We're not emailing the entire list. We're gonna email five at a time, starting from the back, from the least, from what I determined are the least likely to do a deal with you."

Okay. Why? "We're gonna iterate on the email over time as we get feedback yes or no." That's great. Sends out the first three batches, you know, were staggered like every couple days.

Didn't really get much feedback, uh, yet. It was early, whatever. And I just asked the question, like, "How come we're not attaching the book proposal to the email?" And it comes back and says, "It is a commonly held understanding inside the book publishing world that if you attach a book proposal to your initial email that you're an amateur, and they'll just bypass you." Oh, yeah.

Let's go. Right? So long story short, I have, it has created a custom list, rank ordered, right, with a, with a, uh, email that is then custom designed for each individual person that's going out, uh, in an incremental basis, all set up through cron jobs. And, you know, I got, I got about f- 55 emails in. Uh, the pu- publisher who I ultimately decided, you know, a couple had responded yes.

Uh, I had a great conversation with them, and at the end, I have a book deal, right? And you're like, I could've never done that work. I mean, that would've taken me six months to do that work i- if I could've even found the people. Like, but it, when you... Like, that's the level of, like, sophistication, automation.

Like, that's where you can go when you, to your point you've made over and over again, when you've thought through what it can do, when you've given it proper context and direction, when you've spent the time to build a plan and have it understand. That's how you leverage the full power of this tool. But to your point, opening up a random, you know, zero context, even paid ChatGPT window and asking it to help you, you're, you're gonna get 20%, 30% of what that thing can actually do. Yeah. I loved your story, and I have a really quick one too, and then we kinda conclude this.

So in my world of SEO, I do run several cron jobs. And I don't know if you follow, like, Google's latest algorithm updates or Bing or whatever, but they, the thing they are cracking down on the most right now is mass-generated AI spam or AI slop, because you have all these people running programmatic systems that pump out hundreds and thousands of different blog posts every single day, right? And so with this cron job, it's actually monitoring all the Google algorithm updates, and so it gives me feedback of, hey, maybe let's post two, three, five times a day and cut back on 10 like we were previously, right? Because that's gonna send negative signals to the Google algorithm and all the LLMs. If we're mass generating content, they're gonna penalize us.

Regardless if that content's high quality or not, it just sends negative signals to the algorithm. So that was, like, a super, super quick example there of just kinda how I'm running a cron job on the SEO side of things. No, dude, this is all so valuable, and I love these stories because I think what it does for the audience is, I think if you're, if you don't, if you're not the type of person who dives in and learns it with your hands, which I get, not everyone's like that, um, and maybe, you know, thinking through these things in this way is, is not necessarily your expertise or you just haven't done it, I think these examples give people the ability to go, "Ooh, I might not wanna update, I might wanna not know about, uh, SEO updates, but I really wanna know about market updates, or I really wanna know about the latest legislation coming out of my state for my industry," or insert... Like, I, I think it, it cracks open ideas for people, and they can start to see that, like- This isn't just an email tool. It's not just a better blog post tool or a video script tool.

Like, yes, you can do those things, and you can do them at scale, and they can be amazing. But, like, you can ... There is so much that you can do with these tools if you're willing to take a little bit of lead time. Just I think that's, like, the, the biggest takeaway from this is, one, and hopefully it's not as scary as maybe people think, and I, I feel like we've gotten there with them. But, like, just a little bit of lead time, a little bit of planning time, a little bit of that upfront work, and the value that can come out the back is just absolutely insane.

Um, with that said, dude, I know my audience is gonna wanna go deeper into your world. I know you have a badass YouTube channel. We're gonna have that linked up. Where else should people go to go deeper into your world, to learn more about what you do, and, and maybe just finish with a little, um, you know, I guess it doesn't have to be a pitch, but just let people know exactly how they can work with you and the different things that, that you're up to. Yeah, Ryan, I appreciate you having me.

And I would say another good place is ryandozer.com, and then also I run a community on school called the AI Marketing Insiders. I'm over 1,500 members now at this point, so those are some pretty good places to find me. The last thing I would say is this, as kind of someone's pitch to whether to work with me or even to follow me on YouTube or whatnot, is that I would argue that I am in the top 1% of people in this AI marketing world that are actually doing the real thing with AI and then learning what works and what doesn't, versus 99% of other people are pure theory. They're just looking what's the latest and greatest thing, pumping out a video, doing the packaging with the clickbait thumbnails and titles, trying to hook you, hook, line, and sinker you in, versus with me, I'm gonna tell you the truth because I'm actually doing things in reality. That's what I would say.

I love it. I think it's ... And I think that was obvious from our conversation, dude, which is why I appreciated this so much. I appreciate your time. Uh, this has been great.

And dude, if you wanna come back on in a few months and do an update, talk more about cool shit that's happening, I'd love to have you. Um, I appreciate you going screen share too, 'cause I think that helps a lot. And guys, if you're just listening to the audio, um, make sure you jump over to the YouTube video because a lot of what Ryan was discussing when he was getting into the detail of, uh, uh, uh, uh, VS Code, et cetera, how he was setting up the folders, you can see him do it. He was... You know, if you're just on audio, you wouldn't have seen it, but, like, you can go over to the YouTube video and actually watch how he has these set up, how clean it is, and how, I think, not scary this has to be.

Like, there was nothing that you showed us that was like, "Oh my God, I could never do that," right? It was, you've put a lot of work into it, but, like, over time, that's a very achievable and accessible thing, even for someone who's very non-technical, and I appreciate that very much. I wish you nothing but the best, bud. Um, hopefully we'll chat again. I have a feeling we will.

And, uh, with that, we're out of here, guys. Peace.