Claude Code Converted Me (I'm Not a Developer)

Lud code has converted me. Yes, it finally happened and I am using it basically all day every day to do most of my work now after how many months have you been telling me to use this for Gail?

>> Like more than six months. Yeah, >> more than six months. Well, for context, Gail is like the early adopter of the early adopters. Like he's the first person to use most things. And despite him uh yeah telling me to use this for so long, I don't always pay attention to everything he says. But as a non- early adopter myself, I think you know we've hit that second wave now and and you're seeing kind of normal business people start to use this more.

Why exactly is that? Because the models got so good that the gain in productivity is undeniable at this point and that if you're using it all day most day you decided this is essentially better than what you were doing before and you're still doing it now so you you you either would have found value or you would have gone back to your old ways right. I think it's Opus 4.5, it's really Opus 4.5 that was kind of like that jump in capabilities uh and it made CL code more than just a coding.

Yeah, and I think you've mentioned this before, but because it's called CL code, every non-technical non-developer person thinks, "Oh, that's not for me. I shouldn't use it." >> And you know, in the beginning, those fears are almost feel justified in a way because it looks like a tool for developers. It sometimes feels like a for developer. Yeah. Yeah. But it's like it's it's it's difficult. There's that kind of like mental hurdle. Like I switched from PC to Mac a couple years ago and I felt like a complete idiot, like a beginner trying to do just basic stuff again and like it's it's not a nice feeling and it's like there's something similar in using a tool like VS Code with CL code and we'll show you how to use this uh later on.

>> How do you feel now though? Like do you feel like you would go back? Do you think you're going to stick to it? Do you think like >> Yeah, I mean like it's this is how I work now. That's >> Could everyone do this? Like could everyone do this? Every single person doing any kind of knowledge work, any kind of marketing, online marketing, running an online business, uh you know, if you use a computer for work, you should be using this.

>> Um and I think this is the year when everyone who is using it will start to get ahead by leaps and bounds when everyone is not. It's like when computers came along or the internet came along and people who started using that early on just destroyed everyone that was doing you know calculations by hand.

>> Yep. I agree. I think by the end of this year if you haven't started using this tool it's like I see I see that it's like a lot of people now we talk to for accelerator for example, they've everyone has used ChatGPT. It's not you have no edge using a chatbot. Like everyone is using a chatbot. If you're on an edge, you need to jump out of the chatbot into these tools. And if you it's this is the point of this adaptability that we always talk about. It's like, you know, you have to adjust with the market etc. That's the adjustment. Like if you're not making the adjustment, you're [ __ ]. It's like I think uh >> I I will say it's difficult to know which adjustments to make because there have been plenty tools come along.

>> It's solidifying though. It's >> but this is like the market really is yeah solidifying around CL code specifically but you know >> next I think >> there are other ways of yeah there are so let's just give some context here. CL code is not the same as CL. CL being the the desktop or the app >> but CL code is inside the the app actually so it's like if you really want to go that way it's like you actually have you have it inside the app now but it's not as powerful. It's getting there. It's getting better. If I share my app now, for example, you'll see that this is a chat and I'll tell you why I sent hi to code in a minute uh all the time. Then you have this co-working which is kind of an in between between code and the chat and you have code here that allows you to walk like for example, you see I have this thing and uh and you can do some stuff there like now they brought stone but you can see it's actually buggy like it doesn't even work anymore. So the point is like you could do some of it but this version is the shittiest version of CL code and that's not how you should use it. Essentially what it is it I mean correct me if I'm wrong here but it's like a very high thinking reasoning um model.

>> The same model. >> So so okay, it's the same model as a chat. The only tools it can use is a terminal and a terminal is the gateway to do anything on your computer. Like actually every time you click on an icon on your desktop, what happens really is there's a terminal command in the background that executes and does and opens the app you wanted to open. Like a view representation.

>> All CL code is is uh Opus or whichever model you select >> with more access with like root access to your computer >> and I'm smart enough to use it properly and to know what to do. Like it knows all the commands by hand etc. So it knows what to do and that's the >> that's it. But but like for example, when I'm using CL code and we'll we'll show you some things we've built with it and what you can do just a second. But like when I'm using it, it often kind of thinks to itself back and forth and goes through like, you know, 5, 10, 15, some 20 minutes um before it finishes something.

>> But when I'm using like the CL desktop app with the same model, >> it's not like it might take like 30 seconds, a minute or something, but it's faster. So >> still the same model. The difference is when you run CL code, it has access to a file system. So now when you run it, you have a lot of essentially knowledge and context scattered across files inside your system. So you have, you know, a brief summary of what the company does. You have like a description of like how what makes a good podcast episode. You have all these kind of documentation that it has access to. It has access to tools as well. So it can call, you know, Notion. It can call Google Drive. It can call all that. And so it's utilizing the context it has. When you're in the chat, it only has the past discussion that is automatically sent and it has the >> That's not strictly true. If you're in a project, it has all the context.

>> Sure, but it's kind of loaded as you send your message. >> Yeah, but you still have like MCPs. You >> The point is like when you send a message in a chatbot, it gets basically all the context at once. When you load CL code, it has no context and it has to pick where it's going to go get context. So it uses search and it's like, "Oh, is there a document about the podcast? He's talking about the podcast." It's going to run a search. There's going to be a line that says, "I'm searching for documents about the podcast." It's like, "Oh, I found three documents. Let me read them."

>> But to come back to the point where I guess I I misspoke when I say it uses the same model. But like it's fundamentally there is something like there's some extra like coding. >> That's what you call agentic. Like, you know, when people say agentic etc. Like it basically that's what it means. It means using tools to build its own context. Instead of you building the context in the chatbot, it just figures out figures it out. It figures out what it has access to. It's like, "Oh, I need more information to do this task. Let me spend some time gathering information." That's why you'll see it kind of go for a while and be like, "Oh, let me search that. Oh, I read this in this document. Maybe I should also search for that. He mentioned that, etc." And so it has none of that in your chat. Um, what it has is if you have a project has a document, but it kind of like loads them in the first prompt. Just destroys your context window and just loads all of it in the prompt whereas it doesn't do that in CL code. It will go and read it if it thinks it needs it. Uh and that means it takes extra time.

>> Fine. Fine. So let's talk about some practical examples of this. So last week uh on Friday morning I built a dashboard and you know, whoopdedoo. Um >> Can we see it? >> Yeah. >> What does this dashboard do? So last week I built this dashboard using CL code. Uh and it connects to our Stripe API and pulls in all of the sales. It actually connects to two Stripe APIs because we had two accounts uh over the last six months and it pulls in all our sales data. It breaks it down by plan. Um it works out all our fees, our processing costs, our taxes, our VAT costs. I'll get into that in a second.

>> You have a map as well. A map of where a member >> Who who drew that map? Did it. >> Okay. >> We've got kind of like uh breakdowns of like where our revenue comes from, member counts, subscribers, churn, um things like that.

>> Cool. >> And you know, very simple like you you make a you ask it to make a dashboard um showing you basic information and it can do that. Okay, fine. It's not how long it's not crazy complicated, but so I I spent three or four hours um >> Okay >> doing this, but >> 90% of that like I had a working dashboard in 10 minutes I think after I'd ironed out a couple bugs.

>> I spent most of the time um working on two things. First is I wanted to get a clear picture of all our processing fees to understand how much we're being charged and second I wanted to get a clear picture of our VAT because we collect VAT which is like a sales tax and for some European customers and it's very complex who who has to get charged and how much they get charged. Um, but it figured out all of that and it figured out how much we needed to be setting aside to pay our next VAT bills. So, there was no like surprise. Oh, you have to pay $10,000 tomorrow.

>> How did you do these before? Like before you built this? >> I didn't. I just guessed it because it was like it's really difficult to calculate. So, I would have like Yeah, I would have like floating funds of money just in case we had a big bill come in.

>> So, were you keeping more money aside and being less flexible with money? You know, less less of that was being invested or spent on the business or whatever. >> Whereas now, I know this is our position. Like I know on the 20th of April, we owe this amount and on the 7th of May we owe this amount. And when that's paid, I can simply uh check that it's paid and it'll reduce.

>> Does it remember when you restart your server? >> Yeah, it does.

>> Okay. >> So, uh this is fantastic. Like really, really helpful. The other thing it did um and I know you're mocking me for the map. So we do like a weekly group coaching call with our plus and max members and you know the time zone for that you can see right here. Most of our customers are in Europe or North America. Um so we optimize for you know a time zone that will work for both those folks. Occasionally we have people in Asia or sometimes even Australia ask us about it. And I think twice in the last few weeks we've had someone say, you know, oh, do you have anyone else in Australia or anyone else in Singapore or Asia um on there as well. And now I can just pull this up and say, well, yeah, we do. We have three in Australia or we have three folks in Singapore, one in Thailand, one in Hong Kong. And you know, they can see they can show you in the map. And that that was like a five minute job to do that. Very niche thing. It's a very niche thing that's like it it's hardly ever going to be come up and you would never like hire a developer, pay a developer to do that. But the fact that it's so abundant and cheap to code anything these days means you can do all these little annoying niggly things that you wish you could do but couldn't justify it previously because of the cost, right?

>> Do you understand the tech behind it? Like do you know what it's built on for example? >> I don't have a clue. Yeah, it's built on XJS for example, like the little N at the bottom and the local host 3000 gives it away. But essentially, CL code figured it out for you, right?

>> Do you need to know what is? >> No, no, I'm just asking. It's interesting because it's like some people want to know like it's like how much do I need to know to create something like this and uh and it's like apparently not a lot because you don't even know.

>> When I started building the dashboard, it mapped out this, oh, you know, I'll look for some hosting and I'll do this and like no, no, I don't need to do any of that. Just fire up on my local machine when I need to access it.

>> And then I was like, "All right, okay, cool. I'll do the quick and dirty approach here. Here you go."

>> Uh, and it was like pulling I saw there was some Tailwind CSS at some point in there, but again, I don't particularly know exactly what that does. Um, but it used it. And >> So you put it in plan mode. You put it in plan mode. You told it what you wanted and then you said go and then you got 90% of this.

>> Yep. >> Yeah. So that's powerful. And it's like you can connect to your so for example I don't have the dashboard yet but like I have a skill that connects to analytics etc. Like if a lot of the apps we use they have data and I think the main problem with small businesses is we don't use that data um for knowledge for example. So one thing that I've realized for example doing something similar with CL code is I pulled like our email ISP uh data into CL code and then we used to email the podcast and I was like, "Oh, look at this. Look at the broadcast and just tell me what's interesting. Literally just like go go figure it out." And we figured out that basically almost nobody clicked on the links for the podcast and so we stopped emailing it because of that. And so that's the point. We have all that data and we were making mistakes but because we had no time or even sometimes not the skill to essentially put it together and or it wouldn't be worth the time like it wasn't very important. But now you can connect your APIs to these things through a dashboard or through a skill and we'll talk about skills in a bit I'm sure and you can have code be your data analyst and translate that data into actions. For example, the action was stop emailing the podcast people don't click on it. But when you email like short videos Loom style, you get like 20 times more clicks. So like do that. Great. Like, and I love that it translates all this complicated data in all this complicated software into just [ __ ] do this, which is essentially my job. Like, my job when I create processes is basically that. It's like just [ __ ] do this. Uh, people love that. And CL is awesome at this. And so now I'm like obsessed with collecting as much data as we can everywhere for everything and then plug it into LLMs and have it figure out what's the best course of action based on the data basically. And you know, you can even uh have it like so for example like with with this dashboard stuff right, we're pulling information from one source which is Stripe, very well-documented API. It's like it's very easy for it to get everything it needs. But you know, in future when we expand this to look at like our sales funnel, we might want to look at, you know, how our YouTube videos are doing or how um, you know, our marketing performance is and we might have to go to like different places to get the data and there's not necessarily an API for it. But you can go indirectly through something like Ampify to get social media metric data if you want to pull that.

>> If you do links and if you do GTM on your links like GTM parameters which I've done. I've plugged analytics into it as well. It's like awesome. Like I >> I think you know for when uh Google Analytics 4 came out, everybody wants to use this. Sucks. I don't know how to use it. I don't know what it means. So but you don't need to anymore. Just use AI to interpret it for you.

>> That's actually a skill that's coming out into accelerator really soon. It's one of the ones on my list. Um, it's like literally just plug your analytics API, put your DTM on your links, like set up your conversion tracking. By the way, also helped me fix conversion tracking. Uh, because I had the code of the website and I had um, I had analytics. It was like, "Oh, you have this newsletter thing, but conversions not tracking. Do you want me to do it?" And it just went and edited the website, set up conversion tracking. I push the code, boom, done. Conversion tracking was done. And now I can analyze what's happening. Can we talk just very briefly about skills? Because I'm sure there's at least one person who heard that and doesn't understand what you mean by a skill and think it's more like a traditional you know >> skill issue >> a skill issue or knowledge.

>> It's not a skill issue. >> But um >> it's a it's like let me try and explain this cuz I'm very uh newbie uh at this. >> A skill is a file and sometimes uh set of files uh in markdown format. Sounds like a text file which explains how to do something. And one of the benefits of having it in this way is you can say like, "Okay, here's step one, do this. Step two, do this. Step three, do this." As a business process, it will remember how to do that. So whenever you ask AI to do that thing, it will do it for you. But within that file, you can also say, you know, like access use this MCP to pull the data from Notion or use this this script to um edit the image or whatever it might be. Um, so it's really just like an SOP, but for AI.

>> There's a bit more to it. Um, I think the the thing with skills is like, okay, these are basically pre-written prompts that you break down into chunks and then AI kind of opens the chunk it needs. Like you make one main file that's like kill the workflow and then you kind of have different like sub SOPs basically that explain the subprocess and sometimes it's a conditional thing like if this happens then do this, if then go read that document, if not, don't. Essentially, it's like essentially what you're saying is like the skill file, the skill.md file is like, here are the five steps and just like, you know, a few lines about each one.

>> Yeah, and it needs to be short because it's optimized for minimal context. So it's not like overflowing your context with the entire >> Progressive disclosure basically. >> When it doesn't need to to access it at that point. When it does, it'll go and access the more detailed file about how to do step one or step.

>> There's still more to skills there. That's the one part, but the other part is it can contain uh scripts, Python scripts. And so it's very useful for like like MCPs are good, but they're very token hungry. Like you need to kind of load all the tool definitions and the model needs to operate it. So let's say you want to extract 10,000 keywords and then you can only extract a 100 keywords per API call. For example, like you need to do 100 API calls. That would destroy your context window and you run out of context on code. You can make a Python script that will do these API calls for you and save everything into a file and it just runs outside. It's just like a piece of software that runs and then the model just has to press the button to run it basically. Maybe just put a variable which is like the main keyword or whatever you want and it will just run. And the point is that you can essentially get more done outside of the limit of the token windows of these LLMs through that. So for example, an example of a skill that we use that has a script is the thumbnail one. So it's like if you see the thumbnail for this podcast, it's probably created by that. It basically creates the background with nano banana and then it overlays one of our photos on top and edits it onto it. And to edit it onto it is a Python script. It's not the LLM doing some magic or whatever. It's just has a script that it uses. It can just grab a photo and say I want it at this position on top of this image and then essentially it just plays with the X and Y of where the images and moves it around. Um, and so that's a script. And so that's the power. Like you can combine these MCPs, these SOPs basically and traditional programming for deterministic outcomes. And that's >> I'm not >> I want to add something here because you know when you start talking about traditional programming for deterministic outputs like I think there's probably a few non-technical people out there who are thinking like that just went over my head. What the hell is he talking about?

>> Um I want to like explain a few much more simple business use cases um of why >> this is relevant. Um there's two things. One is you can share these processes, these skills with other people. So Gail, you created the newsletter uh skill file, showcase newsletter uh skill. You shared it with me. I then used it to create a newsletter.

>> Mhm. >> Great. But in doing so, I also wanted to edit the output slightly so it was, you know, a little bit better, a little bit more to what I was trying to achieve. So afterwards I asked it to update the skill file to reflect the different direction I I took it in. And so it went and it updated itself. It updated it its own skill file. It rewrote the process in here to add more information about what it could see as the difference between it outputted and was in the skill file. And that is like really a key improvement versus using you know a project in CL uh or uh ChatGPT to do it because I don't know about you but I would spend a long time making like really good system prompt and really good set of context files. I'd then run it a bunch of times and after a few weeks what I was outing would be quite a bit different and I would and I was like, "Oh, I've got to go and like spend all that time to update the system prompt again." And it's it's a hassle. But now you just say update your system prompt with or update the skill file based on that the output >> an employee you essentially fix it with them and then they learn from that experience. It's very similar. That's the problem with projects. It's like they kind of stuck and then you have to redownload the files, go edit some text file like [ __ ] that time for that and quite often you just settle right. It's like it's kind of like 90% of you're like, "Oh, I just have to tweak it at the end etc." and you just it's not worth the effort to redownload to slightly tweak it and it still won't be perfect. But if you just say look at everything that I gave you as feedback and what propose some edits to the system prompt for example, then that's low effort after running a skill basically and that's kind of the point. It's like I tell people the first time you run it, it will be 70% as good, just the same as like a project. But the point is like you will run it and it won't be good and then you tell it, "No, change this, change that, edit this, edit that." And you tell it, "Okay, now fix your instructions." And next time it will be 80% and you do that again, there will be 85%. There's kind of like a diminishing return right as you go. It's not like it ever goes to 100%. Um, but you get to a point where every output requires less time, which means that frees up more time for more things and you end up essentially leveraging your time more and more because the other takes more burden away, you know.

>> Which is exactly how it should work with an employee, right? You know, you have to invest all this time when they start to train them up and share your process and like how you like to do things and then probably the first time they do it, it's not great and you know, needs quite a bit of feedback. But then the next time it is slightly less feedback and then slightly less and slightly less and then eventually they can more or less do it themselves without too much handholding from you. And that's that's exactly what AI has replicated.

>> Yeah. And that's the thing. It's like you're basically creating digital employees or rather kind of like specialized freelancers that know how to do one task square. When you create a skill, it keeps improving. I would say it's like it's not too far. It's not as proactive as an employee would be. Like it's not going to spontaneously tell you how to make it better or look at your competitors if you don't tell it or whatever. Which I think eventually will come. Um, but in terms of like improvement rate, it's not so far off from working with junior people on these kind of marketing tasks at least.

>> It's like it's not even junior people. Like it's it's because in a way some of the things capabilities do like it's very very senior. >> But it's more like uh I mean That's exactly what it is. It's like working with a robot that has no common sense. Like it's very smart. It has access to all the data on the internet. It's relentless. It works 24/7. Um, but like it really does lack some basic common sense sometimes. Um, so you have here steer it a little bit. Yeah.

>> Yeah. It's like I don't think it's going to replace you, but like, you know, I was talking to Tim from Hrefs. Uh, and he's like, you know, Dimitri, who's the CEO of Hrefs, like forces us all of us to use CL code and do this. It's like, it's mandatory at this point and create skills etc. It's like, everyone, it's like, you need to do this this year. Like, if you don't do this this year, you're like the boomer who didn't know how to use a computer that arrived in an office in the 2000s, basically. That's about the same.

>> Yeah, they will like as painful as it probably is for all the non-technical folks uh in in the company at the moment, they will be thanking them thanking their CEO so much in a year's time for that because yeah, this is this really is the future. Yeah, you have it's it's non non-negotiable. And it's like, let me ask you, I mean, you didn't really write newsletters before, but like how long do you think it would take you to write a newsletter now? Like, how long does it take you versus how long would it have taken you if you did it without even with a chatbot? If I was if I was even with the chatbot, like an hour or so to do it back and forth, like to really really get it down. Like I'd probably get a few examples of ones that we've done before. Maybe like a few examples of some newsletters from folks that I know do good newsletters. Um, brain dump a bunch of my thoughts in there. You know, get a shitty version and then do like three rounds of feedback to it. And then eventually I get that, take it and copy paste it into into Bento and send it. Whereas now um >> You didn't explain that part actually. You didn't what what does it do exactly? >> Well, yeah, whereas now uh with a skill file, it's like, here is the transcript of the video which is based on. Go and write the newsletter. And if there's any kind of direction I want to give, I can give it. But it's so good, it usually doesn't need it. So it will sometimes ask some clarifying questions. I may uh get the opportunity to steer it if I want to. And then it will create it. I can edit it within CL code in the um text file there. I save it, tell it it's ready to go and then it actually goes and connects into Bento and uploads it as a draft. It doesn't send it. I would never let it send it because that's like could go terribly.

>> The MCP doesn't let you anyway. Like it they built it on purpose so that it can't go crazy.

>> Bento, by the way, um, not sponsored at all. It's an email tool, very similar to sort of Active Campaign, except it's kind of like an indie tool. It's uh, it's very like AI native. So like they have one of the most robust MCPs of out of all of these uh email tools out there.

>> And it's just it just works very well with this stuff. So if you want CL code to go upload your draft email into it, like it it knows exactly what to do. It gets all the fields correct. Um, it's ready to go basically. So, all I would do at that point is just to send myself a preview, check it, check the links, because, you know, I've made that mistake before, not putting links in there, and then send it and it's uh, press send in inside Bento and it's it's good.

>> There's actually one thing in this scale that you don't I don't know if you even understood it does that because I built it. But the point is in through the Bento SP, you can't upload images. So, when we have like a thumbnail or something like that, um, actually the image needs to be hosted somewhere else. So within the skill, I've built a subprocess that is uploading images and it's connecting to our Cloudflare. It uploads images to Cloudflare, gets a URL back when it's uploaded it and then when it generates the HTML for the email, it puts the essentially the image our directly inside the HTML which seamlessly adds the image to the email. And that's the kind of like sub workflows that you can build within these skills. You will understand if there's an image or not. It will just figure it out basically.

>> And that reminds me because there's a I have a complaint about this uh so >> Okay, go ahead. >> Um >> I don't know if it's Bento or if it's like the deliverability thing but it suggests that images should be or the whole email should be below 100 kilobytes and sometimes when you upload images which are bigger than that it's not so good. So it's almost like you would want a image compression process in there. You don't even need like a third-party service to do that. You just tell CL code to compress the image and it will like for you to do it. But >> it is that it's simple to no it's not hold wait. It's simple to compress the image in CL code. I think what you're saying is to do that as a skill is not so simple. >> No, that's not what I'm saying. What I'm saying is the reason Bento tells you you need images below 100 kilobytes is because for mobile specifically, people who are on like 3G um, so it's not it's the whole email below >> Let me finish. Yeah, but it doesn't matter. Text takes no space. Um, the point is that um Cloudflare detects the size of the viewport. So if it's easy on your phone, it will actually serve a smaller image automatically through the service that we use. But Bento when it tests, it tests through desktop resolution. Um, which is allowed to have bigger images than than that. Um, so the point is Bento is actually not very smart in the way it detects these things. But in reality, Cloudflare fixes a lot of that stuff for you by compressing the image automatically and also serving the right size for the right viewport. Um, so it's like, yeah, it's not.

>> So, so you you you chose to ignore it basically because it's not necessary. >> I I'm not saying we pass it all the time. It's like if the image is really big, sure. But >> It does come off like, you know, if you have a 4 megabyte image, it's like seems a bit over the top.

>> It's a bit much, but like I mean, you can just ask code. You can be like, "Hey, before you update, upload the image, uh, just compress the image." And it's like, honestly, you just tell it that it will just install the Python library, update the scale, and then next time you run it, it will just do it. Just test it before so the image doesn't look like [ __ ]. But that's it. Yeah.

>> Um, that's the beauty of it. It's like you had that the time it took you to explain this to me, you probably could have edited the scale, right? It's like, and that's the point as well. It's I think that's quite important. That's an important distinction because it means you don't need to go back to me when you need things like that done anymore. You can literally just tell the chatbot and it does it. And I think that's an unlock that a lot of people don't have yet. Like I even see it with questions in like accelerator, etc. Like I would say it's less and less but 20 to 30% of questions I'm like, "Just ask code." Like it's like, "Just just tell the bot because it will actually solve it for you." Even if you don't understand the tech behind it, the same way you did not understand the tech behind the dashboard, it's fine. It will translate your non-technical language into technical stuff and specs and then do it for you. And we're at that level of abstraction now for at least simple things. Like you're not going to build a full SAS in one prompt, but simple marketing workflows like that, that easy. Honestly.

>> Reminds me of uh I saw this video years ago. It was talking about electricity generation and how they saw this future of like abundant electricity and if you were just able to imagine a world where generation was so cheap. Um what uh benefits that would bring, what problems that would new problems that would face. And then the problem stopped being like, "How do we make energy or electricity?" and and the problem became, "How do we transfer it around or transmit it, move it?" And so it just creates this whole different perspective of the world. I think we have a similar situation like that with with coding and almost like work to an extent.

>> Whereas it's so cheap and so fast and so easy for anyone to build anything these days that >> utilizing existing code or existing processes >> doesn't matter >> it doesn't matter, right? You built a a V1 of this dashboard. It was much more simple, but it worked. And I was like, "Oh, do we need to like, you know >> You pass it to me. I'm like, "No." >> Put this on GitHub and share it and da da da." And you were like, "Dude, just [ __ ] rebuild it." And I was like, "All right, fine."

>> I had I had a working version in 10 minutes. Like that was did 90% what that did. So >> but I think also it's quite interesting because it means that you because you're not a developer historically, you've solved most of your problems without code and you're kind of trained at looking for a solution or SAS or something like that, something that you can use off the shelf that will do 80% of what you want and then you kind of have a shitty system to make up the last %.

>> Or some manual work. >> You need to change your head around that. You can code. Everyone can code now. At least personal tools that you run locally. There's really little risk if you don't upload it on the internet and you can solve your problems with code now. Uh even if you understand none of it, provided you just learn the tool. Which how long did it take you to kind of like feel comfortable using CL code from the day you opened it to like it's okay. Like I don't know. It was probably about three or four weeks, but that's not of usage. That was that's like total time. I was probably in those three or four weeks used it like four days or something. Um >> Mhm. >> And it was really one when I spent a whole morning like three hours messing around with this dashboard and then starting to like build out context files and things like that for other work that really like, "All right, this is my new thing now." As soon as I had that mental shift from like, "All right, I'm going to stop using all these projects I have in the cloud desktop app um or in Gemini Gems and just rebuild everything." Basically, anytime I came up with a task that I needed to do,

>> Um I I would migrate it over to CL Code. And then, you know, that was Friday, it's Tuesday, and like most of the stuff I use is on here now. And yeah.

>> Okay. Okay. I can if you want, I can show you one of the skills I just built recently because I think people need to see more use cases. Um, so let's do that. Um, so actually what I'm going to do is I'm first going to share a Chrome tab for Marketing Pros. So Marketing Pros, recruiting company we work with in South Africa. They recruit marketing professionals. It's cheaper than the US and people are pretty good. That's like the high-level thing. And the thing is like one thing that they need to do right now is they need to run some ads. Like people who go on their website, if they saw ads, they would get more leads, basically. That simple. Everyone should do that. We've never lost money doing that.

>> About retargeting ads like >> Yeah, I mean, it's not really called retargeting anymore because it's not pure retargeting. Like it broadens a bit, but the idea is like, yeah, people who interact with their brand on social as well, that kind of stuff. So this is my VS code. Looks very similar to yours. Chatbot on the right, whatever content in the middle, files on the left. Right? Like it's pretty simple. And so I have this skill called Meta Ads, Meta Campaign, sorry. And you can see literally I started the ski the the thing by saying /meta campaign and I just gave Marketing Pros. And the thing is like it's quite hard to come up with like good angles for ad campaigns, right? So like how do I catch people? How do I stop the scroll etc. And how do I make it like interesting and fun while still selling my product? And so I built that skill that basically Well, it creates a folder which you see here, Marketing Pros. That's the run. And it kind of like scraped the site to understand what they do. It also, you know, we talked about scripts. It has this script that basically scraped their CSS. So it gets their colors, for example. So, you see it got all the colors. It also, if I actually go in research, you will see that it actually downloaded their logo for example. Um, it just that's what you do a Python script for, for example, on a scale. And again, I don't know how it works. I just talk code, make a script that does that, right? Um, so any reviews it by the way, the logo is way too big on their site. I discovered that running that anyway. It made a brief and it was like, "Oh, they're basically a staffing company in South Africa." Does it look good? My input is, looks good. Then after that, it makes itself a little to-do list and it interviews me on the uh on this. So actually, you can't really see the questions, but it's like, you know, who is your primary buyer? Have you run ads before? Do you have stuff that worked? What is, you know, what's your best customer success, best testimonial like? Ask me a bunch of questions basically. And based on that, it basically writes a whole actually it has some notes here, you can see it just it writes this file basically uh based on all my answers. And then after >> It's creating its own context file there basically. >> Exactly. We're building context, but it's painless for me. It's just like a question thing and I just click the right answer or I type something. I don't have to think too much. I just give the URL and I just pick some answers. Then it has an agent that basically role plays as the customer. So it actually spawns a sub agent and it's like, "Hey, you are a struggling agency owner trying to grow your agency and you want uh like you you have clients, you have demand and then you don't have people to execute and then you you hire people and the clients quit and then you have too much too much stuff and not enough etc. Like what are your problems etc." It basically does all that brainstorming slash roleplaying. And the other one is a market research agent that goes and Reddit and stuff like that and see what people complain about basically. And you can see like it it this is the prompt that it gives to the agent. This is the prompt it gives to the market researcher like, you know, search for like Reddit, search for all of that etc. And it just comes back with a customer profile. So it's just, you know, I can't hire fast enough to keep up with client demand and the people I can find either cost too much or burn me like, you know what I mean? So again, it builds context that way. It builds context also based on the research. So you can see, right? So whole research talks about Upwork, talks about Fiverr, talks about the problem with offshore, uh talks about good marketing talent being hard to find, all of that context, right? Um, then after that, it just actually ran a bunch of extra searches because it decided to do that. Uh, and it makes a strategy. And again, I didn't give much so far, right? I answered a few questions and uh and it just ran after that. It basically ran this came up with a campaign plan. So you can see for example, like I gave it a budget etc. It came up with angles uh which is burned by offshore, Upwork, marketing only South Africa advantage, and 3x your capacity for half the cost basically. And kind of high-level things and presented that to me. And my feedback was also, you see I'm typing a lot on these things. I But yes, let's go. And then essentially then it brainstormed each asset and how it's going to do things. So you can see it picked some templates. So in my skill actually, I have uh if I go in assets, I have these templates that I curated manually. So for example, the Apple Node one. And so it does look at these and it just like knows which template it can use. So in this case, for example, we have the before after, right? So if you go in before after, you check the examples, you can see like it has examples of ads that you can use from that. And so it it basically comes up with the ads and it's like, do you

Like this? Is this good enough? I'm like, yep, looks good. Create them. So far, I wrote about 20 words. All right.

Then what it does is it uses the nanobanana API through a Python script again. And what's really good is because it can also upload in the prompt. It actually uploads the logo, it does all of that, etc. And essentially, it generates the whole ad campaign for you without doing anything.

So, it generates the copy first. So, for example, not VAS, marketers. Actually, the other one I think was a bit better if I check the copy. Yeah, done with Upwork roulette. I thought that one was really good actually. Um, and the point is it has like three versions of the copy like short, medium, long that you can test, link descriptions, and CTAs. And you can see it basically generated ads like for example, this one is really cool. Uh, like a text message one where people go back and forth. I hired this agency, etc. You can see some kind of like whiteboard one. Freelancers are hired this year. Ghosted, bad work, missed deadline. South a marketers I hired all still here. Marketingpros.co. These are pretty good. I tried offshop before. South Africa was different. marketing pros. Um, the point is, yeah, literally everything's generated on its own. And you can see I have a lot of these copies. This one, >> they look great. I mean, if you're listening to the audio version of this, like they look good. Like they look like a any other ad you would see on your uh Facebook stream. >> I would say, and the thing is well, is like when it runs, let's say you don't like an output. I think at some point it generated an ad and it was there was a typo. Yeah, you can see that. Oh, there's a minor text glitch. And so there was a text glitch on this one I think and uh it regenerated it. So it's like oh there was a glitch and it's like boom I regenerated it. It's fixed now. Uh and that's it. It just it looks at it and it's smart basically. But yeah you can see this before after try try risk-f free. Uh I had the listical one and I had the quotes for example. I mean these are decent ads basically. And the point is yeah, that's the kind of stuff you can do with a skill. And how long would it take you to generate all of that? Especially because it's all cut down in adsets with all different copies to test and the creatives in the right folders and you can just go to meta and upload this and it's ready basically. So it's like that's the power of these skills and that's why we tell people like if you don't do this now you're [ __ ] because I can make five campaigns when you make one or seven or something. Uh and when it's running I can do something else. And uh and so like I'm I'm more likely to test I'm more likely to run more things and to find a winner than you are if you do this manually basically even if my graphics are like 10% less good than the ones you may have. Uh and it's also I don't know like there's like 12 images times 40 cents. So it's like yeah, it's not very expensive basically.

>> Yeah. I mean, it's super impressive. Like, uh, you know, if you asked an agency, hey, design me some some ads. Like, they'll probably follow a similar process to what you've done there in terms of interviewing you, asking you questions, >> and like, you know, okay, like a pro designer might come up with a few kind of things, >> a bit better, but not perfect. Yeah, it's not >> Yeah, it's like it's not even 8020. It's like it's the 9010 kind of thing. like it's way good enough for any business that's not running ads to have this their first campaign or if your strategy is I want to deploy you know dozens of new ads every day to test then this is a way to do that quickly at scale >> and if you have ideas you can also put them in the prompt right initially I put zero like I put nothing but if you have a special landing page a special promo a special thing etc you can put that and that's going to influence the whole process it's going to cascade all the way down to the creative so it's like that's so powerful you could run it a 100 times You could be like, "Hey, like let's brainstorm like 20 angles for 20 campaigns and then you run each one one by one and create all your assets that way and then you have like a lot a lot a lot of stuff to test. And I did not even connect competitor scraping into this. I had it initially then it became too complicated. You can see my whole context is used actually in this one run. So this was uh this was already pretty intense but I think it's good enough already to be honest.

So, so what does it cost to generate those ads in terms of your uh usage or you know nano banana API >> around $5 for this? >> How many ads? >> Like I mean you can see there's 12 creatives. >> Okay. >> Um so 12 creatives around $5 >> 4050 cents each something like that. >> Yeah, around that. Um and I mean I'm pretty sure like there's going to be some kind of image model that's going to come soon that would be equally good and you know a quarter of the price. So, and but like you know you're spending money on ads like does $5 really matter? Like isn't it better? >> It's weird because it's like we have this discussion internally. Gil's like, "Oh, shut up. >> I'm going to spend $100 to the $200 plan." And it's like it's because you're spending it on, >> you know, let's put in the software group, even it's like an AI model or API credit. it, you know, we're a little bit more cautious of it whereas you wouldn't even think about, oh, I need to spend $100 to pay a designer to do something like it's it's just that's what you would pay. >> Yeah. >> Um, >> so it's almost like there's a mental shift needs to happen there as well. >> Yeah. But I mean, you can see these are real outputs that you could really run and it's like this skill is available should be available in AI accelerator uh at the time which to release this podcast. I just shot the video for it.

>> Yeah. Um >> le let's actually try and talk about that for just a sec. Um so the AI accelerator is our course slash community uh where we talk about how to build all these things, how to use it. We give you all these skills that we've created >> um inside there like you can go and use that today I believe as this is going out. >> Um there's you know we're uploading multiple of these skills every single week into the accelerator at the moment. We do weekly group coaching calls there with other business owners. It's a good way to connect with just anyone else who's running a business who's implementing AI in their company. And you know, some folks bring their team in there as well. You know, there's a lot of uh debate at the moment. You know, AHF's a good example there about like how do you get your team excited and using AI? Well, you know, we hopefully will help give them some structure to be able to do that as well if you wanted to do that. So, if you're interested, um, it's the only thing we do, only thing we sell. This episode is not sponsored. Head over to authorityhacker.comacelerator and you can find all the details there.

>> Yeah. So, I mean, going back to the topic of skills, that's really kind of the hot topic right now. And what's really quite powerful, like you talked about sharing and stuff, you can actually upload these skills to GitHub and then you can connect these skills into code co-work. So, let's say you have staff that really is like I can't do the terminal. Like some people would be like that. there would be like uh you know some it's probably some person with like a glasses with a chain around their neck uh will be like I can't do this. Um well actually these skills are usable inside the code chatbot and co-work and so if you actually build them they don't all run perfectly but we had some members that have been using them directly inside there and skipping the VS code and terminal setup and for other teams you know I do think most people should learn but I understand the constraints as well and so there's a way to build some kind of like essentially company knowledge you used to make SOPs for humans to read and do now basically everyone's a manager and he's managing agents and he's managing their own library of SOPs for their responsibilities to be done and then it's just uh it's just their job to maintain that basically. That's the that's working in 2026.

>> Yeah. I mean, instead of that being in your brain, it's now just in a skill file. >> Well, it was supposed to be in an SOP before, but let's be honest, most people didn't do it. Um so so it's like >> I I do have one one thing to ask here though because we talked about co-work which is the new feature something between claude and claude code uh which is in the claude desktop app at the moment. It's basically a dumb down version of claude code that they're trying to make more accessible um for non-developers for knowledge workers. Isn't there an argument that people should just wait a few months till that like develops and becomes more robust and not have to worry about the terminal and you know cloud code. I think clo code will always be the edge like the same way as like you can do these things in cloud code but not necessarily in co-work today there will be something that you can do in cloud code in six months that you can't do in co-work and it's like if you want that competitive advantage if you want to actually be able to do more than most people right now then it makes sense because it's not that difficult. It's really just a chatbot in the terminal. >> Um it's like the ratio of effort to reward is still incredible. It's one of the best things that you can learn right now. One of the best things you can spend two hours on. >> And you do actually have a course inside the accelerator which teaches this. But >> even if you don't want to buy our stuff like you can go on YouTube and like there are people who have like videos explaining how to do some of this stuff. So like it's like we really believe that this is like a key thing that everybody listening to this needs to know needs to do this year.

>> Yeah. So, it's like, yeah, you could wait six months and get the same capabilities of the expensive models in the cheap ones. Like, Gemini Flash, >> it's about as smart as the best models a year ago. Um, maybe even better, but the reality is if I put you on Gemini's refresh for your day-to-day work, you're going to be like, I want Opus back. Um, and then you'll want it because it just unlocked so many things for you. Same thing's going to happen. And that thing is not going to change. It's just going to keep moving the goalpost and everything's going to evolve and you'll be able to oneshot your dashboard instead of spending two hours on it by the time co-work can do what you can do in clo code today. >> So up to you but I think that the productivity gain is so big that the people who are not willing to go to the edge are going to lose out anyway. The same way as like people who stay on chatbots you're just like slight like you might as well just stay on the chatbot. And I think that, you know, when we were in the SEO industry, like we we talked for so long about how big companies and big sites had such an advantage over small oneperson companies or small small sites. >> It feels like this is almost like a great equalizer in a sense, right? You know, small medium companies who are using this technology effectively are a are going to be able to out compete, not just catch up with, but out compete bigger companies who are going to be obviously much slower to implement this across their larger teams and get everyone on board and using it correctly and they're going to be worried about more safety, security, governance stuff, whereas a small business can you be a lot more nimble.

>> I think it depends like for example meta Instagram, right? I was watching some videos from like product managers on Instagram. Product managers on Instagram, they had to kind of like, you know, essentially make a proposal, make a presentation for a new feature, etc. Now they are forced to vibe code it. No developer involved. They vibe code it in a test environment and then they present the feature built and then it's like if it's good enough then it's passed to the dev team who vip codes it properly. Um, and then that's how it works. And so like that's not how every company works, but big companies are jumping onto this fast already. >> I feel like if you're a meta and you know you have the best of the best people who are paid the big bucks, like you know you can issue a top down command like everybody has to do this or else >> and you know you have smart people around you. They'll get on board. There might be a bit of resistance but you can get it working. But think of like the average company not like full of rock stars like you know with 20,000 people trying to make that happen like it's going to be harder like >> yeah this we see it with some people we consulted with for example like the engineering things are are resisting and so on. Yeah, I mean the guy who built floatbot, right? He just got acquired by OpenAI supposedly for a [ __ ] ton of money, single guy like alone building and you know according to his stats from the guys from Codex, he uses more than like entire teams. He uses more tokens than like entire teams sometimes. And so yeah, there's an opportunity. It's just like it's so funny because I know a lot of people who still listen to us that they still kind of think in terms of SEO etc. But the edge is not there anymore. like SEO is a is a mature uh declining market and um and the edge is there. The edge is like how do I build stuff? How do I do things faster? How do I do good marketing like these ad campaigns that I showed you for example um using these tools with a minimal drop in quality but a massive improvement in productivity and and the value of these things is still pretty high because most people can't do it. That's the thing. you're kind of fighting against the curve of people who also adopt these these tools and you want to be ahead to create value and that's what the edge is. Um and and that's why yeah and if you're a marketer that's that's what you need to do. I see I see tons of CEOs of agencies and so on now they're like oh if you're a big code user etc we want to hire you like send us your CV etc. Like they're they're desperate for it. I'm telling you, yeah, if you went to HFS and you told them that you're you're a marketer and a big cl good user, your chances of being hired would probably be pretty high if they're hiring, for example.

>> Great. I think we'll leave it there. Any final words of wisdom? >> No, just just [ __ ] do it. Like it's last last chance, last last call right now. Uh, by the end of this year, if you haven't done it, you're officially a boomer and uh and we can't help you anymore. No, I'm kidding. But, you know, we >> we are going to be around this year to help you for free on this podcast. We're going be uh covering this in a lot of detail. We'll have more dedicated episodes to it, how to set it up, how to optimize it, you know, all that kind of stuff. But, you know, if you want to get in there right now, then head on over to authorityhacker.com accelerator and you can get all our stuff in there. It's currently live. Thanks for listening. Um, I know this is a little bit rambly and less random podcast usual episode, but um, we we really do feel it's so important that people have to take note of it now. This is not an early adopters thing anymore. This is an everyone thing. Um, and now is the time. So, thank you for listening. If you have any thoughts on this, please head over to our YouTube channel and leave a comment. We do read them all. Um, drop us a like, subscribe while you're there as well. really really really helps us out in the YouTube algorithm. So, we appreciate everybody who's done that.