# 162 Cracking the X algorithm with Claude Code

# 162 Cracking the X algorithm with Claude Code Speaker 2: If you want to use AI to create viral content, then listen up. I brought on Tom Crawshaw, who's been building automations for nearly a decade. He's attributed over $25 million in revenue for his e-commerce clients, and now he runs a content system that pulls in millions of views. Tens of thousands of followers a month across X and LinkedIn. In this episode, you're going to learn how he built a Claude Code skill that scrapes his X data every single week, updates his voice profile automatically, and writes posts that actually sound like him. If you stick around to the end, he reveals a hidden Claude Code command called Insights that audits every session you've ever run, tells you exactly which workflows to turn into skills next. You're going to love this one, so let's dive in. Tom, what are we learning today? Speaker 1: All right, we're going to get into some juicy stuff around skills, building skills in Claude Code. We can get into my content creation flow that generates me millions of views across platforms, tens of thousands of followers, and some other cool stuff that I think we can get into as well with, you know, Claude's desktop and projects and things like this. Speaker 2: Awesome. Yeah. And I love Claude. I spend all my time in Claude. That's my LLM of choice. And I spend most of my time building skills in Claude as well. So I think this is going to be a super useful deep dive into that ecosystem. And so, Tom, before we jump into your screen and you kind of show us step by step, why don't you just tell us why the audience should listen to you? Like, what's your background? Speaker 1: Yeah, so unlike a lot of people, I've actually been building automations for eight, nine years. And not just in tools like NA10, but I'm going back to Zapier, which has been around for a long time. I was working in the e-commerce space with eight, nine figure brands. And we were building automations for email, SMS, in Klaviyo and these kinds of platforms. So yeah, automation doesn't just mean like agents and all this like, APIs, HTTP requests, all this stuff. Automations, we've been able to do those for quite a while. And they've been incredibly effective. I mean, we attributed over $25 million in revenue for our clients over that period. So they do work. And the next evolution of that was getting into NA10. And that's kind of what I've gone deep on the last couple of years. And then more recently with Claude Code and Lovable All of the OpenClaw, which is kind of a cool new tool that's been dropped. So yeah, that's a bit about my background. And yeah, in terms of like the stuff I enjoy to do, you know, guitar, skydiving, base jumping, freediving. I love kind of the adventure side, because I think you've got to balance it, right? If you spend enough time in front of the computer, you've got to get out, move your body. I like to dance and sing and other things. So you've got to keep the balance, at least for me. Speaker 2: Yeah, absolutely. No, I think that's just good general advice, too. So I'm excited, man. Let's dive in. Let's see what you've got going on under the hood. And again, we're going to look at some Claude Code and some skills specifically. So let's check it out. Speaker 1: Right, I'm just choking on my water here. Speaker 2: No worries. Speaker 1: I literally just inhaled it. I wasn't expecting a bit of comedy today, but just have it roll. So, I'm going to share my screen and show you one of the skills I've built. Let's just find this button. I don't know if this is going to do the... The screen of death here where it zooms on everything. Let's just do the entire screen. Speaker 2: Yeah, that works. Speaker 1: Yeah. Unknown Speaker: Cool. Speaker 1: So, sorry about that. That happens. Let me just go. So, I mean, most people know this is Claude Code. It's incredibly powerful. You can actually access this now in the desktop app. So you can just load up a folder or a branch from GitHub and you can just work in here. I mean, I personally prefer this interface because It's just a little bit nicer than working in the term and I just find it's a bit easier to read and When we talk about skills, we're talking about These It's essentially a series of context files, right? So in the in the in the beginning, right when before skills we had we had projects and And so, you know, these are like core projects. I've got a bunch of them for different use cases, right? We have this, let's say, Cinta NA10, and this is connected to the Cinta MCP, and I can just tell this project what I want to build in NA10. It will literally build it in NA10 without me having to I don't put any nodes in or anything. It's just, it's connected to my account, right? And in here we have instructions and on the backend we've got an MCP integrated with this. So projects, you know, you can simply, I've got this copywriting one. I use this a lot and we've got, you know, copywriting guides and this is all your context, right? But the issue with the project is every time you put a message in that chat, it's loading all of that context and it's taking up a lot of your, Context window, so when it comes to building skills, the great thing is it's like a book. The LLM, Claude in this case, will read the contents. So what chapters have we got? And it will show you all the files that you have access to in that skill. And instead of loading all of them, it's going to pick the one that it needs to do that specific job. In Claude Code, you can just type forward slash, and I've got this one called Content Create. And if I just hit it, I'm not gonna give it any instructions or anything like this. It's now gonna ask me, it's gonna read that file, and it's gonna figure out, okay, he's just starting out this project. Whether I'm writing content for X, whether I'm writing an article, whether I'm writing something for LinkedIn, It's got a bunch of context in here. So I'll go into how I created this and how you can replicate this for yourself because the issue is with content is I would never automate it. Like with a workflow or something, it's just such a terrible idea because you're not going to get the level of quality that you want. And my background is in marketing and copywriting and writing emails and things like this. And so taking that lens, I can read the output and be like, no, I want it this way. And so having that experience and perspective can really help. And if you don't have that, that's totally cool, but you can learn that yourself. I think the issue of relying on LLMs for absolutely everything means that you end up not using your brain. So if you don't understand copywriting and what is a good hook, and you don't do some research to pull in some examples of what works, then you're not going to be able to develop that critical eye. And that's what makes the difference between you creating AI slop and you creating something that's going to have a higher chance of going viral and people sharing it. Does that make sense? Speaker 2: Absolutely. Yeah. I mean, I think the best point that you made is it needs references, right? Like you can't just ask it to write you an article or write you a tweet or write you a post. If it doesn't have really good reference material that it can look at inside of that skill to determine what does good look like, then it's just going to give you slop. So that's a great point. Yeah, continue. Speaker 1: Yeah. So before you even try and create this skill, that's what you need to do is, and I've put out a few videos on my channel about how I go through this process. I've not done anything on creating skills, so this is kind of new and fresh, because this is something I only recently created. I was still, you know, like a month or so ago, I was just using these Claude projects and getting good results. Like I had my process, But I wanted to convert it over to Claude Code with this skill that I could then develop and add to. And so that's what we've got here, right? So it's loaded my voice profile, okay, which is, I have a XAPI MCP that basically can connect to the XAPI. It's not an MCP, sorry. The XAPI, I've written a script for it. It'll pull all my posts from the last seven days. And it'll find the ones with the highest engagement, the highest reach. And it's going to run that job on a weekly basis. Because you can only get the last seven days of posts right. It's going to run on a weekly basis. It's going to update my voice profile on a weekly basis based on the top performing posts, right? I have a Salton Inc. principal. Is it Alan's Salton Inc. or Salton-ish? I don't know how you would say his name, but I think he's Eastern European or something. He's an absolutely, he's a fantastic copywriter. I've followed his work for quite a while and I just love the stuff that he shares. So I, you know, and you can do this as well. You can, you can grab like emails or articles or posts from copywriters and start building up a resource and you put that into, into a, a chat called chat and say, Hey, can you distill these lessons and can you give me as a markdown file? And then boom, you can feed that into your, into your skill as like principles to, to use in your posts. Speaker 2: Right. Again, just basically telling. You're telling Claude Code, like, hey, every time you use this skill, I want you to make sure you're writing in my voice, which is what the voice profile's for. And I want you to make sure you're following these principles, which, again, you took from a master copywriter. So it's like every time it's running that skill, it's doing it in your voice and it's following best practices for copywriting. And again, I think the purpose of the skill, right, is so that you don't have to give it that context every single time you want It to use it. It can just call the skill and boom, it's all right there. And then it's good to go. Is that kind of the point of this being a skill versus like a super long prompt? Speaker 1: Yeah, exactly. You don't have to be grabbing all those documents every single time. Speaker 2: Right. Speaker 1: And previously, I had those loaded into my Claude projects. So this is just a different way of doing it. Both have their pros and cons. This is going to take up less, you know, tokens and things because it's more efficient. And I can do everything in one chat here, whereas I was kind of moving from a couple of different projects in Claude. Speaker 2: Right. Speaker 1: I have a voice evolution report. So this, like I mentioned, it's similar to what I mentioned earlier about taking the most recent articles and then updating. My voice profile, so this report gets generated. We've got a Nano Banana system for creating images. I'm having mixed reviews, I'm having mixed thoughts on this. I could probably improve this process, but getting a good image for your social media is, it's hard, even with the tools that we've got. In my experience, to get it, How you want, I mean, if I just, oh, if I just like bring up my, my X profile and just kind of give you some, some examples, like this was pretty straightforward, but stuff like this. Speaker 2: And that's a good image. Like, I like that. That looks good. But I bet that took what four or five, seven iterations probably to get it to that point. Speaker 1: Yeah, I like stuff like this, getting those logos in. I mean, I had to add some of these in post in Canva. That was pretty straightforward. This was a little challenging, but you can see I have this kind of theme with these icons and arrows, and that's something that's done pretty well for me. So I'm like, okay, I'll repeat this. Speaker 2: Right. Yeah, those look really good. Speaker 1: Yeah, so things like this, they can be tricky to get right. And this one got over a million views, right? And yeah. Speaker 2: Oh, I literally saw that on my timeline. Yeah, the six. So yeah, within the last couple of weeks, I'm like, that looks familiar. Speaker 1: Yeah, so, so that's the issue with image generation right now, I generally use a cloud project, where I've, I'll just show it real, real quick. I generally use a project with. Speaker 2: Claude's chat's down right now. It has been for the last hour, so that's probably the issue that you're going to run into. I was trying to use it an hour ago and it's down for, I think, for everybody. Speaker 1: Really? Speaker 2: Yeah. Or maybe it's back up, but yeah, it wasn't working this morning. Speaker 1: Oh, wow. I mean, it was fine for me like an hour ago. That's funny, isn't it? Like imagine how many people are like absolutely just stuck right now. Speaker 2: I'm like, they can't do anything. I mean, that was literally me this morning. It's so funny because I was, yeah, I'm like, what's wrong with Claude? And I'm like, oh my God, am I going to have to go back to ChatGPT? Like I was, I just had like, it was such like a let down moment. Speaker 1: Back to the dark side. Speaker 2: Yeah, exactly. So. Speaker 1: Okay. Speaker 2: But Nano Banana 2 just came out yesterday too. I don't know if you saw that. Speaker 1: I don't think I've tried it. I mean, I've just been using in Gemini, right? I don't know if it gives you access to Nano Banana or it's just using Pro, but I mean, I'll just show you like a previous chat, like what's this? Oh, how do I move that? Sorry. Okay, you can see this is one of my images that you can see the progression. Yeah. One of the articles I was working on, I was like, okay, I've got this reference, I like this. And it's like, I'll generate five, 10 prompts, ideally with a reference, sometimes not. And you can see it's just, I'm like, oh, that's terrible. That's terrible. That's terrible. And I'm like, okay, back to the drawing board again. And I was like, okay, let's try and replicate this, but change the text and have these people fighting. I'm like, okay, boom, that's what I need. Speaker 2: Yeah, that looks good. Yeah. And that's, I think to your point, like that's the struggle with Image generation, even with an incredibly powerful model like Nano Banana, it's rarely is it a one shot to get like a really good image. Usually it's, you know, you got to come up with a prompt and you got to iterate two, three, five, seven times. And then, then maybe you've got something that's workable, but there's other times, like you said, where it's like, well, actually I've got to go back to the drawing board and start from scratch. Speaker 1: Yep, exactly. And you know, it's taste at the end of the day and it's difficult to, It's a replica that so great thing with these tools is you can you can get a lot of variations done and these generations of super quick. So, I can I can do like five to 10 variations, pretty quick, but Yeah, and he gave me that and then I landed on this. And I'm like, okay, you can tweak this in Canva, you can remove the watermark, you can size it correctly, all that stuff you can do after the fact. Another thing you can do is you can put it into Canva and you can select, like you could select all this text and then you can move it around. Speaker 2: I didn't know that. Speaker 1: You can delete stuff. Yeah. Speaker 2: Interesting. So that's actually something I wanted to kind of ask you about. So are you on a Canva paid plan and do you use that to kind of, so let's say like Nano Banana gives you an image and that gets you like 80 to 90% there, but then you use Canva to kind of get that last 10 to 20%. Is that kind of your strategy? I feel like that's what a lot of people do, right? Speaker 1: Yeah, it's just, if you're trying to just force Gemini to resize it and get it to where you want it to get, you end up just pulling teeth at that point. Speaker 2: Right. Speaker 1: Because it won't listen to you to get it to the specific size that you want. So I'm just like, all right, I'll just throw it into Canva, create the Canva size that I want, remove the watermark. Like with this one here, it wasn't generating the gamma logo so you could see. Speaker 2: Yeah, I've had to do that too where it's like it'll just hallucinate the logo or sometimes you give it the reference logo and it still doesn't put it in the right place. Speaker 1: Yeah, exactly. I mean, I gave it all three reference logos. It just didn't pull that one in for some reason. So yeah, little tweaks like this. Just give it, I don't know, it's got to be done at this, where we're at right now with this tool. Speaker 2: Yeah. Speaker 1: And so, yeah, what I was saying is if you go to edit and you can go, right, okay, you can do magic grab. So it's gonna analyze the image and you can be like, okay, I want to grab this and like grab and It should allow you to just just move this logo. So you're like boom. Speaker 2: Oh, wow That's now are you on the is that require a paid plan? Like are you on a Canva paid plan? Speaker 1: I am on the plan. I can't remember what planet plan I'm on, actually, but I am on a premium plan. I've had it for the longest time, I think. I don't know if I should share this on your podcast or on your channel or not. It didn't come from me, but you might be able to find a way to get an account for like really cheap or fiber or something and just like, I'm not, this is like, someone shared me this plan with me and I can use it and I'm not sure exactly how it works, but that's an option you can go for. Speaker 2: So you're on like a legacy plan of some sort it sounds like? Speaker 1: Yeah, honestly, I'm not entirely sure. Look, I'm on this, it's like a team and this team has given me access so you can... If you want like a top plan, like that's how I did it. But I've had this for a while. So it's not the most, I mean, it's legit. So you could do that. You'd probably find it on Fiverr or something. But the other cool thing you do here is magic arrays, right? And you might have heard about that. You can just select stuff to arrays and it will delete it for you. It's pretty, pretty cool. Speaker 2: So Tom, is it safe to say that, again, just to kind of clarify my earlier point, and this is more so for my reference, because it's funny, right? Before we jumped on here, I was looking, I was doing some searches to try to find a really good AI forward carousel creator for like LinkedIn carousels. And so is it safe to say that Canva is almost like a requirement if you want to create high quality images slash carousels? And again, like, you know, the image generation models can get us 80, 90% of the way there, but to really get them to 100%, something like Canva is required. Like, is that fair to say? Speaker 1: I would say yes, yeah. I don't know what plan you would need to be on to have these features, but yeah, if you want to avoid playing tug of war with Gemini and kind of force it to do the things that you want, Probably do yourself a favor and grab a Canva account. Speaker 2: Yeah, that makes sense. And are you using it to create like, I don't know how active you are on LinkedIn, but have you ever used it for like carousel creation or carousel design for LinkedIn or for any platform really? Speaker 1: I've not really been doing carousels, so I just repurpose my X posts over onto LinkedIn, so I'll use the same image, basically. I mean, these are article images for X, so I'll just, maybe I'll crop that for LinkedIn, but I'm basically using the same images across those platforms. Speaker 2: Nice. Well, yeah, you've sold me on a Canva account. So I think as soon as we get off here, I'm going to go upgrade because, yeah, like you said, it's like playing tug of war with Gemini. Oh my God, there's nothing more frustrating. Speaker 1: Yeah, totally. And when it comes to content, you know, your copy is important, but your image is also, like that's the scroll stopper, right? People will read the hook, but the image will often grab attention if people are just flicking through the feed. Speaker 2: Right, absolutely. Speaker 1: They both work together really well. You might be asking, well, how do I know what a good image is? Again, you have to build up a folder of references. Speaker 2: Right. Speaker 1: And eventually, if something goes viral, then I've got a folder of all my viral images. Speaker 2: Yep. Speaker 1: And so I can use those as references for new images. Like I did with the one, the agent teams one that I, this one here, like I've been using this style because yeah, why not? Why not just try keep doing something that works? Speaker 2: Right. And is that kind of your workflow of like, hey, all right, I want to create a new image for this X article that I'm writing. Do you almost always start with a reference file of an image that was successful for you in the past and you kind of just iterate off that reference? Or are you ever really starting from scratch? Speaker 1: I'll try and find a reference where possible. But I might just use the same reference, right? I often do that, like I'll use the same reference or I've got a handful of references that I'll try. Yeah, so there was a time when I was doing, I don't know if I can find some of it. Yeah, I was doing more of these, like with something on either side. You might see, look, I had the same reference for this one. Speaker 2: Yeah, and those look good too. Speaker 1: Same for that one. Yeah, same for that. It's just experimentation, right? That's all it is. But yeah, when it comes to, I mean, just jumping back to the skill for a second, you know, we can actually start it. We can actually get it up and running in the terminal. Hopefully this is going to work because this is connected to my Claude account, but yeah, it might just be the desktop app that's not working. Speaker 2: Yeah, I think it's just chat that's down right now. I think code is up and I think co-work is up, but chat is down for some reason. Speaker 1: Right. All right, we'll see what pops up. Yeah, okay, so it's gonna ask me for, you know, the product or topic, audience, the goal, content type, and anything else relevant. So I literally just, I mean, if you can see on my screen, I have this Whisperflow. Speaker 2: Oh, yeah. Speaker 1: It's, if anybody's watching this and you're not using Whisperflow in 2026, You need to look yourself in the mirror and ask yourself this simple question. Do you talk faster than you type? It's a simple question. Chances are you do. Everybody does. Especially if you're like me where when I'm typing and I make a mistake, I have to backspace to correct it. It's just one of those things, right? So imagine doing that and like the other day I was in a coffee shop and Whisperflow wasn't like, it wasn't picking up my audio that well. Because there was like music in the background and I think I was a bit further away from my laptop and I was trying to speak quietly. Speaker 2: Yeah. Speaker 1: It wasn't, it was just struggling to turn what I was saying into a transcript. And so I had to type everything out. I was like. Speaker 2: And you feel like you're in the stone ages. Unknown Speaker: This is slow. This is well slow. Speaker 1: So yeah. Next time I go to the coffee shop, I'm going to take this microphone with me and I'm going to plug it in and you can just literally hold it up to your mouth and just literally just whisper in. And that's why it's called Whisperflow. You can literally just whisper into a microphone and it'll pick it up perfectly. It doesn't just transcribe, it actually like Format your text or your speech, which you're turning into text. It will delete any ums and ahs and it will correct. But if you repeat yourself, it'll correct and it won't have the same thing twice. And what I've found with using Whisperflow is it's actually improved my thinking and my speech. Because when you're typing, you have a bit more time to think about what you're typing and how to structure it. Whereas when you're speaking, what I found in the beginning is that when I would, let's say, dictate and copy a response to an email, I would end up having to go into the transcript that I got and just edit it and delete stuff. I'm like, that doesn't quite make sense, or that doesn't flow as well as I would like it to be. And that was because my thoughts were a bit jumbled. So that's one of the really cool benefits I found is I'm actually getting better at formulating my ideas in real time and pausing instead of Like stumbling over words or just rambling tracking, you know, yeah. Speaker 2: Oh, yeah. I mean, dude, you're preaching to the choir. I am the biggest Whisperflow fanboy. I feel like I probably converted 30 people to pro plans, you know, not even as an affiliate just like from from talking about it and I mean, shoot, if Whisperflow is out there listening to this and wants to sponsor the pod, we'd love to have them because it's my most used tool by far. It's the one tool that I'm using 24-7 all day, every day on my phone and on my computer. And to your point, it has forced me to Be a lot more concise with my speech because when I first started using it, like you said, it's like I'm just kind of yapping into the microphone. And then I've got to go in and delete sentences that didn't make sense. And it's like, oh, I repeated myself here. But after a couple of months of using it, you're just you're shortened to the point. And then and then you're done. It's really a fantastic tool. Speaker 1: What I'm referencing is different context, right? If I'm responding to an email, I want to be short and concise and to the point, but if I'm just brain dumping about an article that I'm about to write, I'll just put it on hold and just go for it, right? Speaker 2: Right. Speaker 1: For six minutes and sometimes longer. That's the maximum you can go in one recording. So yeah, it's fantastic. I think we both agree on that. And anybody who's watching or listening to this, if you're not using it, just try it out and get used to it. Because the way that you interact with these AI tools, you can either type or you can have a talk. Talking is just quicker. So you're going to get more stuff done and you're going to breeze through things. It's just quicker. That's all it is. And it's just going to give you that advantage, especially when you're When you're trying to ask an LLM to do something for you, you might, it's the end of the day, you're a bit tired and let's say you're not using Whisperflow, you're just using your keyboard, you might just half-ass the prompt. You might just type in something really brief and you're like, oh, whatever, I can't be bothered typing it all out. Whereas if you speak it, It's a lot easier to add additional context in the same amount of time it will take you to type it out. So you're going to get better outputs and that's how it works. And cool thing is, right, when I'm dictating, I could be, you know, say if I'm writing an email, but I want to reference, let's say, a transcript, right? I go over to Fireflies while I'm, you know, see if you're typing the email out, you'd have to stop typing the email, go over to Fireflies, check what's in there, go back over to the email and continue typing. Whereas doing it with voice, you can actually switch between apps and tabs while you're recording. Speaker 2: Right. And then eliminates a lot of the context switching. Speaker 1: Yeah. Once you've finished recording, you can place that text anywhere. So you can actually check. You can check other windows and tabs for information that's going to help you write that email without just staying in the Gmail tab, for example. Yeah, I could go on about this, but it's very good. It's very good. Speaker 2: Yeah, no, it's a fantastic tool. And so let's jump into, can we do like an example with your content create skill here in Claude Code and kind of see what that output looks like? Speaker 1: Yeah, definitely. Let's give it a whirl, shall we? And there is one other thing I want to show you here in Claude Code. I don't know how many people know about this, but it's a gangster. But yeah, why don't we write a post for Whisperflow? And full disclosure, I have already written a post for Whisperflow that's going out today on my profile, but let's just imagine I haven't done that. I'm just going to voice note in. So yeah, I'm looking to create a tweet about Whisperflow, which is a I'm a voice AI tool that allows you to dictate and talk instead of typing. So I want you to check out the Whisperflow website, grab all the information that you might need. for the post like stats and things like this key benefits and That's yeah, let's get that rolling I'm not not giving it. I'm not giving it much there right and so what it's what it's been told to do is Is to start with the hooks right and Best thing you can do with content is start with a hook. Otherwise, where are you going to start? It's probably the thing that I spend the most time on. Because if that's not good, then no one's going to read the rest of your post. So I spend most of my time on the hook and the image, to be honest. Speaker 2: Yeah, because if those suck, then it doesn't matter how good the actual post itself is. Nobody's going to read it or watch it if it's a video. Speaker 1: Yeah, exactly. And so you can see here, it's actually going out to the website, and it's scraping the website and checking for some information. So we've got a response here. Key anchors for the post, reporting a free trial, et cetera. OK, before I generate the hooks, is it organic or sponsored? Audience, angle. So it's asking me for more clarifying information. Very smart, and it's going to help with the creation of the post. It's going to be organic. The audience is, I think, all of the above. Anybody that's still typing is basically the audience. People who run an online business, for example. I'm a solopreneurs. The angle based on my voice profile, yes. Let's see. Discovery journey. Unknown Speaker: Let me just pull this over a little bit. Speaker 1: I've been typing into F. So these are like the angles, right, that it's generating me. So let's have a quick read through Discovery Journey, Timeline Precision. I just tested this third-person case study. Let's go with Discovery Journey. Speaker 2: Yeah, that's what I was thinking because I feel like people can relate to that because I mean that was my experience It's like I was typing manually for 30 years and I discovered with the flow and it completely changed how I work You would think that we were sponsored, but we're I mean, I'm not I don't know about you, but I Have I have received sponsorship funds for one post But Yeah, I couldn't think of a better company really to work with honest. Speaker 1: And fun fact did you know the keyboard that you're typing on right now? Speaker 2: Was designed 200 years ago, and it was designed to slow down your typing Interesting so but to 200 years ago before the computer was a thing like what's the backstory on that? Like while Claude code is here going to work typewriters. Speaker 1: That was the original reason for the keyboard layout. Typewriters, if you typed too fast on a typewriter, it would get jammed and stuck. Speaker 2: Oh, yeah, makes sense. Speaker 1: So in order to slow people's typing speed, they spaced out all of the common... So if you look at your keyboard, you look at A, O, U, E. They're all spaced out. Which makes it slower. So that's what we've inherited, which is pretty insane. I don't think it'll ever change now. Speaker 2: At this point, I feel like we're too locked in. And I mean, I feel like if we changed it, it would go back to being slower than it is now. Speaker 1: Yeah. Well, you know, you'd have people with different keyboards. And then if you learn one keyboard layout and you go to your friend's place and he hands you a keyboard, you're like... Like you just have to slow right down. Speaker 2: Yeah, exactly. Yeah. I'm excited to see what this puts out because I mean, I know we're having it, we're having a generated tweet, right? And like your, your tweets have been crushing, like your articles and your tweets, like you're just getting a ton of engagement in general on X. And, you know, I don't know if you're willing to admit this. I certainly will. But I mean, these days, Of course, I curate it. I don't just auto post what AI creates, but like 100% of my content is AI generated. And ever since I started using AI to generate my content with a really strong underlying brand voice profile and reference files and all those things, my engagement is through the roof. I'm talking up Thousands of percentage points in the last few months when I implemented this system. I mean, are you kind of in the same boat where you're always having like you're never really starting from scratch? You're always starting with an AI draft? Speaker 1: Yeah, basically. I'm always starting with a voice note, essentially. Speaker 2: Yeah. Speaker 1: Here's what I'm thinking. I want to write a post about this, and then maybe I'll pull in some references. Maybe it's a Reddit post that I want to take some ideas from and expand on. Or there's a new release, right? Anthropic are dropping loads of new stuff. That's an opportunity to put your own spin on it and to break it down a little bit in a more easy to understand way. For example, so it's given us a bunch here. So let me tell you what it's doing here. I've got it to score all of these hooks. You can see phase two hook generation. So it's got some stats and stuff here from the website. 16 hooks ranked by score. And I've got Curiosity Loops, Specificity, Assault on a Tweak, Sensory, Now Problem, Credibility, Voice. So it's rating all of these hooks based on this. And if you're not doing this with your copy, Like asking to rate it based on like copyright and principles and stuff. It's, this just gives you a better idea of what the AI thinks is good. And then you, obviously sometimes it's not right. Like I don't go with the, I don't just default to the number one. Like this here, I build client automations for a living. Last month I was typing every Slack message. There was a gap that I hadn't noticed. That's actually pretty good. It's like that contrast. Right in number in second silver players like it's got the silver medal and So he's added this 2026 tweak. I'm not a big fan of that I've used that quite a bit recently Yes, so some of these are pretty pretty decent I would say it's kind of Elaborate on different on the same kind of angle I've been using voice notes and air tools for a year thinking I was ahead and I realized I was still money and typing every app So yeah, it's given us a lot and I would always suggest to generate a ton because Sometimes it's just just not that good right nice. I splice between different ones. I'm like, okay I like this first line, but the rest of it's not great. Speaker 2: I do the same thing, especially for my X articles. So I have an X article writer skill and part of the skill generates seven different headline variations similar to yours, like based on a reference file that I've given it and based on copywriting best practices. And you know, sometimes the suggestions it gives me, I'm like, oh, like this one's really good. Like I'm just going with this one. And then other times they all suck and we've got to iterate three, four, five times. And then other times it's like, oh, I really like the first part of suggestion two. And I really liked the last part of suggestion five. So we're going to kind of fuse those together or I'm going to have it or I'm going to tell it, hey, Take suggestion two and suggestion five and give me some additional variations just based on those two. Right. And then again, you, like you said, it's, it's taste. It's not just taking what it gives you. It's, you've still got to use your intuition to determine like what's good, what isn't, and where do we need to need to iterate? Yeah. Speaker 1: A hundred percent. A hundred percent. And, you know, people think, oh, you use AI to create content and you just put it out a slop, blah, blah, blah. And you can see it's a little bit more involved than just. Speaker 2: It's not a copy paste like people when people hear. Yeah. When people hear like AI content, they just assume like, oh, you're literally just saying, hey, write me a tweet and then you're just copy pasting it. Like it's so much more than that, because, I mean, you know, as well as I do the. The heavy lifting for this process, like 80% of the heavy lifting is on the front end and creating the skill the first time. Like that's really the biggest workload. Once you have that, most of the foundation is in place. But even then, when you're calling the skill, you still got to iterate on the output, right? It's still not going to be perfect every time. So I think that's something that just most people don't realize because they haven't gone through that process yet. But once they do, Yeah, it's still a skill. It's the same writing skill that has always, to be a good writer, it's always been a skill, but now it's just the skill part has just changed. It's just a different skill, but it's still a skill in terms of an ability. Yeah. Speaker 1: And like I mentioned earlier, the other aspect is that I have a good eye for what I think is We're going to work. And that takes time and data to figure that out. So we can see here, it's written me the full post, which I think is pretty decent. Claude, get a response, copy it over. The things that I don't like here is this, open the app, full stop, paste it, full stop, send. That's AI kind of shit for me. And so, this is not the end. This is not the end. Let's just check if there's any other... Still typing. Unknown Speaker: Okay. Speaker 1: I mean, the rest of it, I mean, it's on the whole pretty good, right? Speaker 2: It's a good start, is what you're saying, right? Speaker 1: So, gate2 does postfill right. I'm just going to say, apply the slippery slope and humanizer. So I have two things that I run in my poster. One is called the Joe Sugarman Slippery Slope Effect. It's a technique that copywriters use. So the goal of your copy is to get the next line read. Literally it. So the slippery slope effect just, for me, it makes it more readable and makes the lines just, it just feels a bit more effortless just to read the next line. And then the humanizer is going to make sure there's no M dashes. It's going to make sure there's no cliche AI slope like, What's the classic one? Speaker 2: It's not X, it's Y. Yeah, I hate that. Speaker 1: You know, no walking, no running, no jumping, like the three no's, the triple negatives. Speaker 2: It'll say like, no fluff, just XYZ. I'm like, oh my God. I hate when it says no fluff. Like I had to add that to the skill. I'm like, never say no fluff. Speaker 1: Yeah. And what I will say is that this is not perfect. Sometimes you do have to remind Claude Code to actually, hey, This is not in the skill. Why are you ignoring stuff that I've told you to do in the skill? I don't know if you've found that, but I've found that sometimes if the chat goes long enough, it'll start to go a little bit off script and I'll have to rein it in. You're putting em dashes in here. I thought we just went through the whole process of humanizing it. It's like, oh, sorry. And then it all reverts. Speaker 2: Yeah. So I have my OpenClaude agent running a lot of my skills versus Claude Code running the skills. And one thing that I had, so she has a self-improving agent skill where she's just able to log certain learnings to long-term memory. And one of the learnings I had her log, I'm like, hey, never use an em dash again ever, ever, ever. And of course she logs it and then I'd say 95% of the time she remembers, but then there's sometimes she'll slip one in there and I'm like, nope, remember what we said? And she'll be like, oh, I'm sorry, like I forgot. So it's not perfect, but it's mostly there. Speaker 1: It's not, no. So we now have... It's actually giving me the structure here, which is kind of cool. And we can see the revised post. Speaker 2: Let's see. Speaker 1: Open the art, paste in, send. Okay. See, I would accept that. Speaker 2: That's better, yeah. Speaker 1: Or maybe in X, I would just put a comma and not have it just a capital and a full stop. But that's describing like, open the app, paste it in, send. To me, when I read that, that sounds okay to me. Speaker 2: It's more natural, yeah. Speaker 1: That's what maybe I would say. I'm always in the back of my mind going, would this sound natural in a conversation? If I read this out loud, does this sound normal? And it's a good check to do. Yeah, I'd be pretty happy with this. I'm not going to read through it all right now. And yeah, there's a call to action at the bottom. But yeah, you could put in whatever you want. In this case, I'd be, you know, copy this into my scheduler and you're good to go. Speaker 2: Yeah, that's awesome. Speaker 1: The other thing that I wanted to show you real quick, because I know I'm just looking at the time. I don't know if you've ever tried this, but forward slash insight. Speaker 2: Insights. Is that a pre-installed skill? Is that something you had to go out and download or does that come default with Claude Code? Speaker 1: That's default native. Yeah. Speaker 2: Interesting. Yeah, I'm excited to see what this does. I've never seen this in action. So what do you use it for? What does it do? Speaker 1: It'd be better just to show you what it gives you. It's going to analyze, I don't know how far back it goes, but it's going to analyze all of your Claude sessions, your Claude Code sessions. And it's going to give you some, no surprises here, it's going to give you some insights on your usage. What kind of patterns it's seeing and then what it did for me. I mean, this is going to be unique for everybody. Speaker 2: Right. Speaker 1: But it's going to give you some prompts that you can put into Claude Code to build skills or to build scripts or to improve how you use Claude Code based on your usage and your patterns. So it's going to see things that you probably don't see yourself. Speaker 2: That's insanely valuable. I had no idea that was a thing, but I could already see the value there. It might say like, hey, you're spending a lot of time prompting X, Y, Z. Why don't you just turn that into a skill? Of course, it's essentially infinite intelligence and it's going to tell you like, hey, here's all the things that you're doing that we could be doing better. So if you just run that insights once a week or however often you're using Claude Code, I'd be shocked if you don't get some insanely actionable insights that allow you to just amplify your workflow. So is that what it gave you right there? Speaker 1: Yeah, so let's run through real quick. So it's giving you at a glance of what's working. Okay, you've built something genuinely impressive. I've just shared with you guys. What's hindering you? The biggest issue is it keeps violating the very strict rules you've painstakingly documented. I'm like, yes. Speaker 2: It says producing em dashes, right? And it literally doesn't give you an em dash right before it says reintroducing em dashes. Speaker 1: Yes. Quick wins. Try creating a dedicated command. Custom skills that bundle your voice profile. I mean, I think I've already done that, so. Ambitious workflows. As modules get better at self-correction, be able to run a content pipeline where Claude drafts self-audits against your voice profile and revises autonomously. We'll see about that. And then it's got all these sections. So what you work on, so it's kind of telling me the stuff that I've been working on. How you use Claude Code. It's giving you an overview of how I use it. Everybody uses it in a different way. Like, I don't write a ton of code. I write automations, sure. But I don't, I'm not really, I don't create any software products or such right now. Let's see. Impressive. Speaker 2: This is so cool. And it's so in-depth. I was not expecting a full report. I was expecting it to be like, hey, here's a couple of high-level things. But this is literally a full report of your use in very granular detail. And that's just by running a slash command insights inside of Claude Code. Speaker 1: Yeah. And the best thing yet to come, I believe, where things go wrong. So how can we fix these? Speaker 2: And could you realistically feed this report back into Claude Code and say, hey, take the where things go wrong section and build a plan for how we can fix that. Like that's feasible, right? Speaker 1: Look at what we got here, right? Speaker 2: Oh, so that's crazy. Speaker 1: Suggestions to add to ClaudeMD. Just copy all of these checks. It's a writing style. Oh my god content constraints workflow discipline Custom skills, right? It's giving you you've already been this is insane formalize your most common multi-step workflows Boom, I thought I thought that was actually a skill book anyway hooks, it's copy and paste that in MCP servers You can copy and paste that in. New ways to use Claude Code, right? Just copy this into Claude Code, and it'll walk you through. So yeah, a bunch of this stuff on the horizon. OK, your data reveals a sophisticated content system. The next leap is autonomous agents that self-correct. And it's giving you prompts to try and build this. Based on like your biggest friction source. Misunderstood requests. Parallel hook generation. End-to-end article factory with checkpoints. See ya. Ah. User got so fed up with Claude's copywriting that they rage quit and wrote a better Twitter thread themselves in a different tool. Speaker 2: Oh, my God. Dude, it's so... Unknown Speaker: I actually did that. Speaker 1: That's the thing. I actually did that and I told Claude Code, I said, hey, Your version sucks. I went over to my Claude project and I created this and it's better. And I showed Claude project, my new version. And I said, you know, what's better about this? And can you, and one of the things you should do at the end of every session is what did you learn about? What did you learn? Did you learn anything new about that you could tweak the skill with? Something like that. Speaker 2: Right. It's like how can we, based on this session, yeah, based on the session, how can we optimize the skill is what you're saying. Speaker 1: Yeah. And so I'll ask you to tell me those things. And then if I feel like, because you don't want to just automatically add them to your skill, but you do want to see what it comes up with. And if there is something that's useful to add to it, that's where you come into this iterative process of improving it. Speaker 2: Dude, it's, I mean, so first of all, I've never heard anybody mention the insights slash command before. Like you're the first person I've ever seen talk about it or show it. I'm literally going to take the transcript of this podcast we're doing right now, give it to my brand voice profile and have it write a tweet on the insights slash command. And I'm going to post it and I'm sure it's going to do well because again, if I hadn't heard of it, I'm sure most people haven't either. And it's insane. Like that's insane. What you just showed me is crazy. So if people are using Claude Code and they're not running that insights report at least once a week, like you're just leaving so much on the table. So that was a huge, huge tip right there. Speaker 1: Yeah. Exactly. Exactly that, yeah. So yeah, tag me if you want. I mean, I think I wrote an article about it. Yeah, I did write an article about it as well. So yeah, it's just a hidden feature that nobody really has heard of. But yeah, you can see he's kind of pulled these out now. As I was watching this, it was like, I'm making those changes now. I'm like, no, you don't. Nope. Speaker 2: Yeah. Speaker 1: So hopefully it's not made any changes, but I will, I'll, I'll check these and, and, uh, and just check if this is okay. Slippery slope is only mandated for articles, right? So not posts. So I'm like, okay, we should, we should mandate that for posts as well. Speaker 2: Right. Speaker 1: Yeah. That's, that's the process. I know I said that I do have a mentorship where I teach people AI automations and how to be AI operators. But I know we've not really spoken about automation, but, you know, building skills is just one tool in, you know, I call an AI operator toolbox. And I think being able to use Claude Code, being able to use N8n, knowing when to use skills and projects and when to Building Lovable versus Claude Code. These are all decisions you make based on the problem you're trying to solve. I don't think there is a one size fits all. And that's where the nuance is. Like, N8n is not dead. It's just, it's useful in certain scenarios, but so is Claude Code and knowing when to use those. In different situations is, I think, you know, a smart way to approach this. Speaker 2: I say the same thing. I mean, I tell people that. They're always like, what tools should I use or what tools should I learn? I'm like, well, what problem are you trying to solve? Because that's going to directly influence what tool you should use. So I think that's really good advice. And Tom, I mean, thank you so much for taking the time to jump on today. I mean, like I said, I learned a ton. I know you mentioned your mentorship. Where can people connect with you or if they want to work with you directly, where do you want to send them? Speaker 1: Yeah, so you can find me on Twitter at Tom Crawshaw, C-R-A-W-S-H-A-W-0-1. On X Twitter, if you want to call it. I do have a YouTube channel, the AI Growth Lab with Tom. And yeah, you can DM me on X. I have a website, learnna10automation.com. And yeah, there's information about the workshops that I have and the mentorship that I run on there. Speaker 2: And of course, as always, we'll link all of those in the description on YouTube and in the show notes on podcast platforms or if you're listening to this episode. So Tom, thank you so much for the time. For everybody else, we'll be back next week. Unknown Speaker: Awesome.