Transcription
Amir, what's going on? Good to have you here.
Thanks for having me. I've been looking forward to this. You know, we've been uh we've been chatting through every single channel except for except in video. So, it's the first time. It's good to be here.
Yep. Yep. You were uh you were an early member of our uh Slack community and we connected there. And then uh we have some mutual friends as well, and I've been seeing you all over X and and within the community talking a lot about Cursor. So, uh I'm excited to to dig into that with you.
Likewise. Yeah, I've been, you know, following bio marketing. I think you've been the the natural leader of the progression of this community and showing what everyone what's possible with Claude and Nan and MCPs, and I'm excited to kind of show you an extension of that—what you can do with cursor and how vibe marketing and vibe coding can kind of vibe co-vibe together.
Yeah. Like I I think that's an interesting point that that you make there, like uh co-vibing. Um, you know uh there's there's this kind of convergence that that is apparent out there between you know building and and marketing and uh cursor seems to be like almost like a natural interface or place uh where that that kind of magic can happen.
Exactly. I think what we're seeing is a natural progression where AI and tools like cursor uh have lowered the barrier of entry to vibe coding, and now we're seeing tools like nan, claude, and NCPs doing the same for vibe marketing. So I'm curious to see how we evolve over time and and use these tools to do both at the same time, and we're going to see, you know, we used to say before the the 10x engineer or the 10x marketer, the 10x employee, and we're going to see the 10x vibe coder vibe marketer very very soon.
Yeah. I I think at some point we're all just going to be vibing—the 10x Viber.
Exactly. Yeah. You know, just you're just going to be hiring Vibe employees. They're going to come in and they're going to be vi marketing for you, Vibe coding for you, Vibe building. So, I'm uh I'm excited for it.
Absolutely. So, you're you're you have a technical background. Uh is that right?
Yeah. And I would say I, you know, I had a technical background, more of a design engineer background, and because of tragedy and AI I was able to kind of get into VA coding and for the past eight years become a full-stack engineer.
Awesome. And so I'm non-technical. Um I rely heavily on you know claude or chatgpt whenever I run into like a technical roadblock or whatever. I downloaded cursor once and uh I opened it up and I was like, "h, this feels a little like scary to me." So, I haven't opened it up since. So, I'm hoping that today you can sort of like demystify it a little bit. Um, show me around. Uh, help me get a little bit uh comfortable with it and then give us some like vibe marketing use cases that that you're working on.
Absolutely. So, I mean I think what we're seeing is tools like Replet, Lovable, and Bolt uh taking that kind of uh fear away of wanting to get into, you know, these interfaces that seem a little bit complex. And I think naturally we're going to have a progression where a lot of these people that are building with replet, lovable, claude are going to move to cursor at some point, and cursor is not only going to be a tool for coding but also the tool for you to do both marketing in there as well.
Yep. Yep. Awesome. Okay. So uh yeah, give me uh give me a rundown on on cursor, like uh the vibe marketing rundown. So I've got a little bit of familiarity with with code and and stuff. I use Replet a lot. Um but yeah, would love to get a lay of the land here.
Awesome. So, what I'm going to do is I just have a simple markdown markdown open up and I'll just walk you through it, and then we're going to jump right into cursor. I think the best we can do is just apply what we know and what we're going to talk about and run through at the same time. So, yeah, I'll talk about cursor um overview, to MCPS, and how you can use it to vibe market. If you are familiar with claude and mcps, this is going to be uh quite similar to you in terms of just the topics on the cover and also just talk about um cursor and how it compares with nan or make or other tools across the landscape. And you know I talked about this earlier, but the natural progression we're see we're going to see or what we're seeing already is a lot of people starting with replet, starting with levelable, building their MVPs or marketing in there and then naturally going into cursor to take them the next step, and I think that it's a it's a natural evolution over time as people are getting more familiar with these tools to get introduced to these more complex tools that uh just expand their workflows. Um my my thesis on this is I think cursor is going to become the interface for everything. um we're gonna see, you know, what's I think what's what's interesting is the team at Kirsten has built this as a tool for marketing, but how other people use this to actually expand their use cases and with the introduction of MCPs, they've essentially opened the door to everyone to from all backgrounds, from marketing, product design to start building tools around it within cursor itself. So I I use it as a you know, I say it's interface of my life and it's interface everything that I do.
Interesting. So, it's not just a developers uh IDE that has um you know an an agent inside. This is allowing you to do other things besides you know build uh software or whatnot. So, you can build software and then you can market your software. Uh you can build micro tools for you know maybe specific workflows or things that you want to automate, which is really cool. And it sounds like MCP is really like a huge unlock uh to extend the capability of of cursor. So maybe just give people a quick idea around what that means uh with MCPS within cursor.
100%. So uh I think great articulation of that and and I think what I would add on to that is um MCPS uh was something that was introduced by by anthropic and it was something that integrated within claude uh itself, and it just stands for model context protocol, and to simplify it it's just uh an another term for APIs—how we typically had was you know applications would interface with APIs to extract, pull, push, or update data, and MCP is no different from that. It's just an API for these uh language models, large language models to be able to interface with other data sets to pull data, push data, and make these changes all within your uh within the conversational window you have with the LLMs. And what it's done is it's enabled these agents to now access these APIs within the context of the window itself of these conversations without you having to build custom tools to be able to inter interact with these APIs. So, it's just a new language standard.
You know what's interesting to me, and uh this may be a little bit of a hot take, but since I started messing around with MCPs, I see in the future the death of workflows.
That could be a possibility.
I mean, definitely, right? and and and with the introduction of background agents and a lot of these agentic workflows that can automate themselves, I would see it's a natural evolution where we're going to have human in the loop interactions with existing workflows to then having background agents that are just running these tasks for you and you're just moni monitoring and optimizing for it—right like yesterday um you know I I did a very quick video around MCPS in claude and uh You know, before I discovered MCPS, maybe I would have built a inaden workflow where I go and like scrape Twitter and then I I get all the replies and stuff and like a Google sheet or air table or something like that and then you know I can have an AI node that maybe like repurposes that content or gives me insights or something, but there's a lot of steps and a lot of friction there. Um, when I was able to get that within Claude, I was able to understand the replies to this tweet, get insight super quick, and do something with it. So, the the idea here that MCPS are powering these like really powerful closed loops where you go from insight to action in one interface uh without the need for all these different steps that induce a lot of friction, I think is just like kind of a a mind-opening moment uh lately that I've had.
You're you're really good at articulating things that I have in my head, and that's a great way to to put it to be honest—like the the the I think you nailed it with uh closed loops all within one context—like that that's exactly it, right? with nan, with make.com, with Zapier, you still have to piece things together and and and and you can't do it all in one synchronous flow—it's not as cohesive where with this you're getting exactly that—all in one—exactly all right—well let's uh Let's dig in. Show us what you got.
Let's Let's do it.
So, um, just a quick introduction. We talked about MCPs and what they are. Um, the number one question I always get is where do I find MCPs and how do I install them? So, you can find MCPS on websites like Smither, cursory.directory, and uh, a bunch of different different websites. You can just do a search for, you know, MCP's uh marketplace and they'll they have specific instructions on how to integrate MCPs with quad or cursor. I'm not going to go into that, but the the the instructions are quite simple. Once you get the instructions, you're able to add MCPS directly through the settings. So, you go to general and then click on MCP. And then here I have essentially three MCP setup uh that I use the most. These are the global MCPS, but you can also do project level MCPS. The difference between that is the global level is basically you're saying I want these tools to be accessible across all of my cursor projects. These are the folders that you have. Um so I have fire crawl which is for scraping content. Perplexity ask which is for search, and then playright which is a aentic browser. So I use that for a lot of QA screenshots and getting a an an agent to actually open the browser and test something for me. Um at a project level you can integrate MCPS. So for example, I have a project where I have both a stripe MCP and my database superbase MCP, and I'm able to directly uh query, call, update and change payment methods, payment links, uh call for subscription updates or even make changes to my database all within the you know within the chat context.
Wow. So yeah uh you know hopefully that covers what MCPS are and how you can in how you can install it. Um what we'll do now is what I want to do is actually just go through three use cases in terms of how I could use MCPS or how I currently do uh use MCPS to create content, crawl links to boost uh SEO with internal linking, and then build micro tools for SEO strategy. I'll stop there. Any any questions, comments?
That sounds uh that sounds really good. I I think that uh this is a a perfect use case for for Vibe marketing as well. Uh so really excited to see how you're able to do that. Um yeah, let's uh let's check out some use cases.
Awesome. So one of the benefits of cursor is that it comes with um kind of three modes, agent, ask, and manual, and they also have the new background agents, but you're also able to create custom modes. I use this quite a lot. I have three, well five actually. Uh I have a PRD generator, a deep researcher, a marketing expert, growth UX engineer, and a feature simulator. And what they really are are just basically preconfigured settings on models and prompts on how you want the agent to behave. So I have um a PRD generator that generates product requirements documentation and it's prompted in a way to give me a very detailed breakdown um of building a product backlog or product requirements document. This is really key for me because I don't approach any sort of build from a live coding standpoint without first getting a PRD in place. So I just want to share some context in terms of like how custom modes play a big part in in cursor and kind of the contrast you might see in claude where you have project level uh settings instead of agent level settings.
Yeah, I I do uh you know I'm a very uh basic vibe coder I guess you could say, but I do the same thing with with replet. So, I'll go and use Gemini, uh, which I see you have here, or I'll use Manx to generate like a PRD or a road map, and then I'll go and just paste that into Replet, and then I'll let Replet get to work. And I found that that makes me a lot more efficient. It reduces the number of back and forth steps that I have to go through to get that sort of like workable prototype or or whatnot. So, I think what's really cool about this feature is I I see you have Gemini for your PRD generator. Seen a lot of developers talk about how good Gemini is for generating those PRDs. Now, I'm assuming that when you have other tasks, you can use other agents that maybe excel at those certain jobs or or roles.
Absolutely. But I think the general consensus is that you want to use these large complex thinking models like 03 and Gemini 2.5 to help you work through an implementation plan and approach on how to solve a specific problem and then switching to more of a taskbased agent uh that doesn't require extended thinking. So maybe 3.5 or 3.7 sonnets um and and get them to actually execute on that plan. Um, I've seen that work really effectively when it comes to building out implementation plans, uh, thinking through an approach to how to actually solve through something or just fixing a specific bug. Um, get one to plan for it for you and then the other to execute. And that's where it gets into interesting into the intersection of orchestration. You know, you're the human supervisor and you're supervising these agents on what to do.
Yeah. And that's like that that's been a big unlock for me within like workflows as well. um you know just setting up the right uh LLMs for the for the for the right task. So within in like an SEO workflow I might have you know u I don't know Grock do some research for me or something like that. Then I might have like an open AI model to sort of consolidate the research and give me some actionable things to do and then maybe I'll have Claude 3.7 or whatever actually be my like brief and and content generator or something like that. And when you split these out and you're not thinking just static around like one model, your your results of what you're able to produce like go out the roof compared to, you know, just just using one for everything.
Exactly. And it ties back to kind of Anthropic and their original article on building effective agents, right? They talked about, you know, instead of multi-chaining multiple agents, have one to do one specific task and then you as a human, we call it human in the loop or you're just supervising that process and making sure that the right agents are are assigned to the right task with the right model and configuration. So spot on. I think yeah, that's a that's a great way to look at it.
Okay, cool. So uh what's next? What do you do here?
So, um, we're going to do some research to get some content ideas and script as much data as we can about the best AB testing tools for 2025. So, I want to understand the landscape. Okay, like what um what are the tools out there and uh and and then use that as a basis to start creating uh an article uh around that topic. So, what I usually do is I I write down the topic in a in a in a document like this and then I'll create a new file. I'll call it I'll call it article research.mmd, and what I'll do is I'll tag the agent to say look at this specific topic and I'll change to custom mode called marketing expert—it's claw 3.7—and uh it's it has its own custom configuration so a custom prompt that says look at this specific you know task, think through the task, look for these specific topics and perform a set of actions focused around content marketing, so that will save, you know, look at this document and the task for AB testing tools in 2025. Conduct research and find relevant tools that help you run AB tests for your website. Now, I personally, you know, you don't have to do this, but I always say like run the specific MCP that I want to run. I think it just it confirms that they're actually going to run the MCP. So I have perplexity and fire call. So what I'll do is uh say run perplexity MCP and fire crawl MCP to get the data you need and uh put your output in article research. So what we're going to see here now is the agent is going to be using MCP's uh it's going to be looking at the topic and then using an MCP tool—in this case perplexity ask—and running a query on finding the best AB testing tools in 2025 for your websites.
Yeah. And again just like integrating those MCPs seemed super easy as well. like there's no complicated setup or or whatever. Perplexity already has like an MCP integration that you can uh just simply add in here to your cursor environment. So there's no like figuring out how to configure some sort of like MCP server on the back end or or whatnot to to get it into here.
Right. It's pretty simple process.
Exactly. So all you have is essentially you would have to just create a developer account in Plexity, get your API key and then go to uh you know their existing GitHub documentation that tells you how to actually install it, or you can go to Smither or any other directory and they have exactly step by step you know they'll tell you oh okay like just put your API key and then we're going to give you the configuration for you to run to set up these MCPs. So I my primary like MCPs that you saw were playright, perplexity and fire call. I just used them so much. It's insane to think about like you know how these three small tools have just drastically changed my workflow. So what we're seeing right now is we saw the Curry now we're seeing Fire Crawl doing a scrape as well and this checking as well and and and finding emerging tools in this space and it's going to consolidate all this and hopefully put this out in this new file that we created. Now with models they're changing so much um tool calls. So specifically how the model calls these MCPS or make these edits, they're always refining. So sometimes you may see that it may not work the first try and you might have to change the model or or keep trying to to get the result that you need, and that's just a natural process. That's part of vibing, you know.
Exactly. Exactly. Do it better. Try harder. My life depends on it.
Exactly. I actually watched uh uh Sergy Brin was on a uh all-in podcast uh you know the co-founder of Google and uh he said something really really interesting. He was like uh if you threaten models with physical violence they will perform better. I don't know if he was joking or if he was serious but I was like wow that's that's pretty interesting. And I've seen some like articles and things pop up around like sort of the foundational like I don't know prompting behind some of the agents that are out there and they've got some really like unique uh sort of guidelines like that as well. So yeah, it's it's yeah, I've seen I've seen the system prompts of some of these models and it's so interesting to see how they build some of these safeguards or try to build some direction around how they want to think.
I'm curious to know in in your process, right? You're you're a power user when it comes to Claude and Nate. Uh based on your initial reactions, what would you say is the biggest stark difference between what I'm doing in Kirstson, what you would do in Claud and
Yeah, I think that uh I I think that you know since I'm not a developer like some of the small nuances like MD, okay, that's a that's a markdown file which is essentially just like a document I guess, right? Um, so
I think, just for me, it would be understanding sort of those small nuances and then just kind of like finding where I can do this here. Um, so you know, I guess after this tutorial, I'll probably be going back to cursor and just trying something like this out. Uh, but I do like the fact that this is consolidated in one area.
Um, also, it looks like the process of setting up the MCPS is a lot easier than the process I'm using today, which is like integrating an inadin workflow into my claw desktop. Um, and sometimes that breaks. Uh, sometimes it's not super efficient. So, this looks like an easier sort of hub to get all that working and go from zero to one.
Exactly. Okay, I'll go back to your earlier point you made about the closed loop. I think this is truly that closed loop system where you're able to get everything all at once, and once you have the setup, it's a lot easier to do. So, um, yeah, perfect.
Uh, what we're seeing right now is a comprehensive research document based on these MCPs. So we went through firecrawl perplexity. We've done a firewall deep research. So we went through a set of 30 URLs, pulled a bunch of relevant data and it consolidated this into a document that we can then use as a basis for research. So we have some of the key tools that I'm quite familiar with. So I can kind of take a second glance and say, okay, you know what, this is actually accurate and it's not hallucinating.
Yeah. And the most important thing here is also, I would say that it's data that's pulled from these MCPS. So it's not a model using contextual data that's been trained on to pull into this that's relevant, right? And you know that's a big roadblock for a lot of people, especially like in marketing. Uh, you know, they'll open up like ChatGPT and they'll try to do something like this, and then when they don't get like a super relevant response right away, they're like, "h, you know, AI slop," or, "you know, this isn't going to work for my brand," or whatever, but they're not using the tools that are going to get like the most relevant things produced and in here for you, like so people give up before they try something like this that can really give you a totally different result.
Exactly. Now, I don't know if you like notice, but one thing that's interesting with the prompting that I've done is it's almost like the extended self-thinking on top of the existing model. So, typically you have the task, you prompt it, and it gives you the output. This is almost self-improvement. It's self-improving. So, it gave you the initial output on the article research and now it's saying, "Okay, let me revise this one more time. Let me add a comparison table. Let me add, let me enhance this." So, it's just essentially an extension of the extended thinking based on the way I prompted with the MCPS.
So, now we have a very, very comprehensive article that gives you a summary, the top tool recommendations, a table breakdown. So, this is all markdown, and then an overview of each individual tool and how they compare with another. And this is quite extensive. I mean, it's, you know, it's about 598 lines. It's pretty, you know, it's it's pretty big. It's a pretty big in-depth article.
Yeah. And like you said, it looks super relevant. So, I'm familiar with a lot of these tools here. Um, when I've tried to ask just like ChatGPT, "Hey, go find me like marketing tools," or something like that, I get a lot of garbage out of the gate. So, like what I'm seeing here is just super, super relevant to the use case that you're doing, which is a huge improvement out of the gate.
Exactly. And so what you can also do is, um, you can then the natural question would be, "Oh, okay, so what do I do next? Where do I take this?" You can copy this article and paste it into your existing CMS, whether you're using Webflow, framework, or custom site, or a lot of these tools now like Webflow, they have an MCP that supports this. So naturally what I would do if I had a Webflow site is I would connect to the Webflow MCP and directly publish this article.
Yeah. Where it gets interesting is, say, for example, you want to do programmatic SEO. So you want to create 25 to 50 pages at once where I want to do a comparison article, but I want to do it for my tool versus 50 other tools. I can create a template and then create all these 50 articles in cursor and then use a Waffle MCP to publish them all at once without manually importing it over.
That's incredible. Um, you know, like I've done a lot of programmatic SEO stuff back in the day that is very manual. It's a pain. Uh, so just having the ability to get all this highly relevant information with that closed loop for programmatic pages is awesome. And you know, a lot of people ask me like, "Hey, where should I start with marketing and stuff for my app or my SaaS," and I'm like, "Go find where there is existing demand and build your marketing strategy around that." So, um, you know, SEO or this sort of like intent-based discovery is the best way to start as well. So like, you know, finding these programmatic opportunities is like a customer acquisition cheat code.
Spot on. Yeah, you're the expert here. You're a lot better at articulating than I am. But I would just add on to that in terms of how you can actually apply that is how I personally apply this, which is I use a lot of content templates. Um, I've actually used the LMS to help me guide through a process of creating templates for how I want to do my research and create articles. So what typically you would see on my app and their blog and the blog that I have is product comparison templates, a problem-solution template, or a best-competitor alternative template. And what you can do as part of this process is I can now prompt the LM to say, "Hey, looking at this template, best competitor alternatives or industry trend insights, revise the article to that format," and then, to your point, if you're doing programmatic, you can do this for 25 different tools all at once, and here's an example of what that looks like.
Yeah. Please. Uh, just a quick note, this is really, really smart because the way most people are approaching programmatic SEO in a basic sense today is they're going out and generating a bunch of informational blog posts. They're throwing it on their site, and then there's not a lot of value that that content's actually adding to the user or the person searching for information. They get slapped with a penalty from Google or something like that, and they're like, "Man, AI in SEO doesn't work," but like if you're solving real problems and you're providing highly relevant information and you're formatting it in different ways based on where that customer is in their journey, like maybe you have an informational thing around what are A/B testing tools, and then you have like a discovery layer which is like, "Let's compare A/B testing tools," and then you have a product comparison specifically with your product versus Optimizely or something like that. Like you're really providing a heck of a lot of value to the person that is finding this information out there, which is totally different than how most people are approaching this kind of thing.
Exactly. And you have to work backwards. You know, is this the intent behind the person that's coming to your site commercial, transactional, or informational, right? Are they looking for information? Are they in the commercial phase where they're comparing, or are they ready to buy and they're completely transactional? So, um, that's how I like to frame my thinking around it and what I would do is, yeah, like just look at these content templates, and they're pretty much the basis for a lot of the content that we have around commercial, informational, or transactional types of buyers, and then we create specific tools for that. So, specific articles and content for that. So these are some examples of like comparisons.
Now you can take it a step further, and what I will do is say, for example, I have my own tool and I want to add additional context to this article. I have an LLM.ext file. So my tool is called Humbolytics, and I have an LLM.ext text file that just provides a complete overview of what my app is and what it does, what the features are, how much it costs, our value proposition, our USP, and then I use that as an additional layer. So, if I wanted to, for example, I can bring this in here and say, okay, you know, based on this article research and the LM that provides you, tell me how Humbolytics compares to this and add it to the article.
Yep. And an extension of that is I've done keyword research. So, I go into Semrush. I do think I wouldn't be surprised if there isn't MCPS for data for SEO or like Semrush or any sort of SEO keyword gathering tool, but I've just done an export. I've exported all the big keywords that I think are a fit for me around keyword difficulty, volume, and search intent. If these numbers don't make sense, I think it's okay if people see this and they're like, "Oh, what is what are all these numbers?" It's just a table. It's just badly formatted. Uh, but the LLMs understand this, and these are basically just contextual data on keywords. So I could go back in here and say, based on the context of LLM.ext and the keywords keyword data, update article research. So what it's going to do is it's going to take the context that provided all the keywords and then it's going to now update the article with the specific keyword density and the SEO words that you want to incorporate. So like I want to incorporate more of A/B testing, Webflow, and user behavior, and that's what I want I want to start ranking towards.
Yeah, that's incredible. You know, another big unlock for me personally with MCPS was this. This was kind of the first time that I really saw the ability to add a lot of context into the model easily and just directly see the payoff from it, you know. So like you have a lot of context here, you know, you have information about your product, you have keywords, you have competitor research, you have these article templates, and you're really unlocking the power of AI here, you know, it's super cool.
And it gets really interesting, right? You have a lot more. So, you know, before a lot of these LLM models used to have RAG incorporated into it so that you can pull in more context. Um, what we're seeing now is a lot of these models have a much larger context windows, especially the Gemini models where you don't need to actually use RAG to retrieve data anymore. You can just inject everything into the context window into the system prompt. And what we're seeing a lot of times now with a lot of applications is that they're getting rid of that completely and they have supervisor agents that will call in specialized agents with their long context window. So like one Gemini model is going to call five different Gemini models, and they each all have context on whatever the topic is, and you're completely getting rid of the need for RAG.
Um, so yeah, so we are now, so it added Humbolytics as a comparison. So like selfishly, I want to promote Humbolytics, so I'm going to add my article into there. Uh, and then also I have my keywords in there as well. So you're getting an article that is optimized for your keywords with additional context on your tool and how it compares with everything else.
Awesome. Are you so just this we can tackle this a little bit later too, but just out of curiosity, you know, there's a lot of conversation around getting discovered within LLMs and not just Google or Bing search or whatnot, and I kind of struggle to see how optimizing for LLM discovery is any different from really kind of like optimizing for SEO.
Exactly. There is no difference because LLMs have to still scrape from these pages, and they're trained on that context. You do want to optimize for Bing. I think that's just a partnership OpenAI has with Microsoft on Bing. Uh, but at the end of the day, that's the pre-trained context. And as long as you're optimizing on Reddit, Bing, and having more content, you're going to show up in search. And we're seeing that now. I look at a lot of my traffic data and I'm seeing, "Oh wow, like I'm getting a lot of leads from ChatGPT."
Yep. Yep. I agree. Yeah. There are some nuances like I think that being multi-format, so you know when you have these articles, like how can you repurpose it into other media or whatever to benefit from how LLM or Google's algorithms prioritize UGC or YouTube videos or whatever. That's a big opportunity. Like something that I talk about with our workflows a lot is the way you started this is you found the demand, and then you're creating an article. That's just one thing that you can create from the insights and the research that you've done here. You know, you could repurpose that for multiple channels and tap into all kinds of other opportunities as well.
Yeah. The possibilities are endless.
Absolutely. Yeah. Okay. So, you generated an article. Now, show us sort of like the closed loop aspect of this. So, um, what's next after you have this in a markdown file? Like, are you going in and are you doing some human-in-the-loop review or editing, or are you just going and like pushing this onto your site?
It depends on your comfort level, right? Like where you know, and I would say the ultimately the goal is you should be a subject matter expert and be able to quickly call out the topics and the points that are being mentioned, especially when we're talking about tool comparisons and we're talking about pricing. You don't want to have inaccurate data around what other features other tools have and the price points on that. I think it's just going to poorly reflect on you, and you're going to get penalized for that. So what I would do as a human-in-the-loop process is go through this in-depth, ensure the price points are accurate, ensure the features are accurate, and then just make sure that I've done a pass through around, "Does this make sense? Am I focusing on the right keywords and the messaging on the intent of the person that's reading this, whether it's commercial or informational?" And um, yeah, just make sure that there is no hallucination. I would say the risk for this is a lot lower because you're pulling in relevant context from Perplexity and Firewall.
Um, other than that, you know, I would take this and copy it as a markdown and put it into Webflow, put it into Framer. And you know, like I mentioned earlier, if you have an MCP, you can just directly connect to your CMS and publish it all in there. I don't have a Webflow MCP. I just that's not what I use, but you can do that if you wanted to. And I've done that before for other projects as well where it's like directly just mass publishing onto a site.
Yep. Yep. Awesome. And you are getting like customers and traffic and all kinds of stuff from the strategy that you're showing us right here.
Exactly. Exactly. We I I keep an eye on the keywords and I keep track on essentially what where our users are coming from, what are the keywords I want to be ranking for, and it's a tactic that's been working quite well for me.
So this is just an HTML preview of an example article and how it works. So I just took the article and I said, "Hey, create an HTML version of this." Um, as you can see, it has a table format, introduction, all the different tool options, pricing comparison. This is for a different article, but I want to kind of show an example of what that looks like, and then also the sources are listed.
Really cool. And if you wanted to add in some advanced kind of traffic or SEO type things like internal linking or whatever, you can do that as well. That's the next that's the next task we got to work on.
So, this is where the fun stuff gets started with being able to automate internal linking SEO. So naturally, you've gone through creating, doing research, you created an article, now you published it, and you want to look at opportunities to do some internal linking to build authority for one page by anchoring text and link to another. Um, and we're going to cover that in this instance here. So what I will do is go back into here. So we've gone ahead and created content. So this is done. And now I want to start linking articles within one another. So what I would do is, just to give an example, I'm going to show you what that means. So I'm going to just open up a browser here. So, I have I have I have a website and I have a bunch of guides and I want to, you know, for example, find opportunities where I can link this article with an article in here, for example. There are a lot of tools out there. They're great at what they do. They essentially build a site map of your entire website, scrape the content, and find internal linking opportunities. We're not paying for those. We're going to get all that done in cursor today.
So, what you're gonna do in true Vibe coding and Vibe marketing fashion, you know?
Exactly. Exactly. Yeah. How do we save some money here? But don't get me wrong, those tools are great, and I'm, you know, they're they're specialized and serve a great purpose, but we're going to do this all in one closed loop. So what I would do is I would prompt the agent in this instance to use firecrawl to build a site map of my website's guides. And so let me just get the link for guides and pull it in here. Okay, guides. Um, I want you to build out a site map, crawl through 10 articles on there and scrape the content and come back with 10 internal linking opportunities. Be specific about which article is linked to the other and provide content snippets and then internal linking ND and add it to internal linking.
So, what we've done here is prompted the agent to essentially say, "I want you to go to the specific URL using firecrawl, build out a site map, list out 10 articles, scrape the content of those articles, understand the context between each different content, and then use your expertise, find opportunities to link one article to another." So, what we're seeing right now is the NCP. It's going to my guide section. It's pulled out all these different links. It's scraping all the content for each individual one. It's going through each individual article, and then hopefully by the end of this, it's going to come back with a breakdown of internal linking opportunities.
Amazing. That's that's really cool, especially if you have like a big website or, you know, like another tactic is, you know, I was looking at some insights from Google's big event, and they said that one area that you can dig into to rank in today's world is updating older content and keeping it fresh or whatnot. So, you know, using these tools to do that at scale if you have a big website or a lot of older stuff like that also seems like a big unlock to update it with recent research to update the internal linking or whatever. Rather than doing that manually or paying an SEO agency a bunch of money to do it.
For you, you can simply use a tool like this. Exactly.
Maintenance is a big part of SEO, right? You're looking at existing content, where the position ranking is and what keywords it's ranking for. Then seeing when the last time you updated it. So what I, I do a lot is I'm in HR or SEMrush. I'm looking at the position ranking of all my pages and seeing, okay, cool. What's went up? What's gone up? What's gone down? And doubling down on it. And a lot of times I'm like, "Oh, wow. This is like outdated content from 2024, 2023." I'll scrape all those links. I'll scrape all the content and say, "Okay, you know, make some recommendations based on the previous, you know, workflow we went through and update the article for me."
So, we're, we're, we're going through a bunch of guides right now and um, we a bunch of, yeah, articles and guides. We pull the contents and now we're going to be uh, hopefully getting an output of uh recommendations on where, Oh, perfect. There we go.
So, what we have now is a full breakdown of 10 articles. Basically, you know, we have a guide mastering AB split testing with practical guides and examples, and it's saying you should link this article to the Humbolty AB test and split test sample size calculator and link this specific snippet to that page. Wow.
So, you're getting a full breakdown. Yeah. So, you're getting a full, full breakdown of, you know, what you should do. So, what I would do is go back into my CMS, open up that specific article. So in this instance, um, we have the, um, yeah, this the, the guide with examples and I'm just going to share this, this page here and like you said, maybe if you have like a web flow MCP at some point or whatever, it's even easier to do so and you don't even have to go and log into your CMS separately.
Exactly. Exactly. So, um, you know, I think it said uh, before launching or somewhere, somewhere in here there was, I have to look back at the snippet, but it says over here essentially what you could do is take one of these, one of the text here and then link it to, um, link into, link it to a calculator that we have on our page as well. So we have a, we have a sample size calculator. So, it's giving you the direct um snippet that you should be using to, to paste into the, uh, you know, for one, one article to another.
Yeah. Yeah. You know, another uh thing I've been thinking about lately with programmatic SEO or with these sort of like content pages is tools and calculators and lead magnets. So, you know, matching the intent of that visitor. So, if someone is in that discovery phase and they're comparing platforms, maybe you have a an email guide or an email course for like how to approach your first AB test or something like that and you're adding value and you're capturing that person on that piece of content with, you know, that lead magnet and then you can drive them to conversion through a a drip sequence or something in your email uh automation tool.
Exactly. That's exactly, you can, you can really take a lot of this content that we've, we've gathered here in the strategy around the linking and build a very cohesive flow and it all comes down to that ICP and the buyer and the buyer journey, is it commercial, informational, and build out user flow, build out drip campaigns to, to serve that exact need.
Yeah. What another thing I like about how you're approaching uh, vibe marketing here is that the logic and the strategy that you have is very, very sound and, and well thought through and I think that, you know, a lot of people out there are looking for, you know, quick hacks and, and stuff like that, you know, and maybe they get like a short-term boost or maybe nothing happens and then again they're like, "Oh, this doesn't work" or whatever, but like your, your strategy, you clearly understand search and uh, the buyer journey and stuff like that and, and you're a real expert and a tool like this is just 10xing your leverage and your efficiency and, and your output, you know.
So, I guess there's a couple things to unpack there. One, like this doesn't replace having the right strategy and the right logic and the right approach. If you have those things, this can take you to a whole other level.
Exactly. It's about speed, efficiency, and cost around this. This and this is what it does. It 10xes you and enables you and it augments that work, right? Her claw, a lot of them are just an extension of what you're doing and the output that you have. And I think at the end of the day, whether it's vibe coding or vibe marketers, you're still the architect. You still need to be able to articulate what you want to sell to the agents, what your outcome is, and being able to help approach them towards that, like approach the, the outcome that you're going towards. Um, and that's, that's, that's what you're seeing with vibe coding as well is you have a lot of novice people that are just building really quickly and skipping some steps around like helping the LM actually think through an implementation plan and how to approach it. And the same applies here, right? Just understanding what you're trying to, what you're trying to do, who your end user is, and what you want to get out of this. And you're just the architect that's just piecing everything together.
Yeah, absolutely. Awesome. So what we've done so far is, um, and you know, I think the interesting part here is like we've been doing this for about 30 minutes. Naturally, this would have taken a week to do with an SEO with an SEO agency or a content marketer. So we've, we've done a lot. It would have been, Yeah. This would have been, you know, if, if you tried to do this with a traditional agency or something like that, you're talking about thousands of dollars and a month or more worth of work.
Exactly. Exactly. Exactly. So, um, you're moving faster, your output's a lot higher, and you're saving a lot more money on this process. The last thing we want to do to tie this all together is we've created content, we found internal linking opportunities, and now it's like how do we approach that top-of-funnel um, uh, traffic that's going to come in and provide some free value to our end users. I'm a big, big advocate for microtools. You know, we have a tool for UTM link generator. It's a very competitive space, but anytime we rank on page on the first page, we get hundreds of thousands of traffic of clicks and impressions on there and we're always competing on that and we and in such an interesting space. So, I think micro tools can be a big, big uh, value ad to build awareness and top-of-funnel visitors for whatever it is that you're trying to sell, whether that's your product or service. So, if you have an analytics tool, if you have an AI image generator tool, try to think about an extension of a free tool that you can offer to users as a way to upsell. There's a company V, v.io, the video editor. They're notorious for this. They have built a suite of micro tools. I think it's over 20, 30 tools that let you perform a very specific task. Removing an image background, adding transcriptions, and it's a free tool that they, they offer to users and then use it as an upsell to sell the entire product suite.
You know, it's uh, it's really interesting if the goal of, um, you know, a lead magnet or something like that is, is to add value. um, micro tools shortcut the time to value versus like an ebook or, you know, a guide or a downloadable resource because you know, with a download or whatnot, you're, you're giving them a bunch of information and then again the onus is on uh, that, that lead or that user to go and turn it into something actionable. But with the tool you're shortcutting the time to value. So, you're able to add a lot more value to someone in a much shorter amount of time, which is going to do a lot of things to move them along that conversion path much quicker and bring revenue into your business.
Absolutely. And I, you know, I would say if you're using SEO tool like Semrush or HRES, uh, just always ensure that you're, you're looking at free, you know, X tool, free YouTube thumbnail generator, free ad split test calculator, free UTM link generator. That's going to have a huge impact on this. So, um, what we're going to do is going to prompt the agent to say study lm.x, do research on humble.com via firecrawl MCP and then provide a recommendation on a list of top five micro tools to build and address the top-of-funnel users and I've also provided some uh keyword data as well. So, the whole, the outcome of this is that we're going to get a recommendation of five tools and uh, we're going to choose one of those. Ideally, you know, in this process, you would look at keyword data and traffic to see what volume uh that specific longtail keyword has and uh where it stands with keyword difficulty as well and the CPC. So, that's like a whole another topic on SEO and keyword difficulty and volume. I won't get into that, but idea that that's how we will complement that process when you're doing research on keyword data.
Yep. So, we're scraping the website, getting some context on what the tool is. We have an LM.ext as well. And the agent is now going to create a list of uh five tools to address up top funnel. Okay, let's say research now. I'll say add it to, I'll, I'll stop right there. Actually, it'll say add it to. I forgot to prompted to add it to the, the document here.
Yep. And then I assume, you know, once you, once you get these tools ideas like as a as a vibe coder, you're not going and uh creating them out of your cursor environment. Here you have the capability to just, you know, close that loop again and
Exactly. build a tool and deploy it on your website that can start collecting emails and getting leads for you like right away essentially.
Exactly. Exactly. So the next natural step here is look at microtools. Um, uh, you know, in this case we're going to maybe just do a UTM campaign link builder. Um, if I wanted to, I could say, um, I could, I could bring in my PRD generator and build an extensive PRD. Maybe for this, I'll, I'll show what it looks like. Um, it's UTM link PRD. So, what I'll do is I'll say, uh, build out a PRD of this UTM link builder. Um, and it's going to create an extensive product requirement documentation on what that would look like.
Yep. Got it, too. Uh, and the next natural step is that you would take this purity and prompt the agent to then create a a simple tool for you that you can add to your website. So, um, you could add this to uh, your framer site, your waffle site, your custom site as a just a code embed. And you can, you know, I've seen, I've seen, I've seen you do this in cloud as well. It's very similar approach.
Yep. Yeah. So right now it's building out like a a complete road map and for this step again you're using like a specialized model. Uh, you're using Gemini 2.5 to do this PRD and then you're going to take this PRD and I'm assuming you can just reference that to the the cursor agent and say hey go ahead and get started. Like what does that even look like?
Exactly. So as, as a natural next step, uh, I would take this PRD and then get the agent to build out a simple tool for that. What's interesting about this PRD generator specifically is based on the prompting is I want to think through a scope to build out an MVP scope and then the out-of-scope items that we want to, you know, build out later on and then more importantly just thinking about um, the user stories in the journal journey. We want to ensure that we're always thinking about the end user and I think this is a natural theme of everything that we've done here from research to internal linking to the buyer journey. We want to think about our end user and why they want to use something like this. And the PRD generator addresses that, you know, we look at say the small business owner, the junior marketer and the freelance content creator. We're building this UTM tool for these three personas and then we're also thinking through the flows. This is important context for the agent so that as it's building through this tool, it has and keeps in mind all this relevant context on how to actually think through the user journey and the user stories.
Wow. So, I, I guess like in your, in your prompting like you've got sort of these guidelines or guard rails around like who your ICP is or maybe who we're building for and that's helping to inform uh, everything that that you do from, from this, this chat.
Exactly. And this, and this, Yeah. I, I, I love this PRD generator because it's such a good uh foundation for a lot of agents to build. And the most important thing is like just the level of details it comes to breaking down the features they should be thinking through on the front end, on the back end and then it breaks into subtasks. So often times what I'll do is I'll use this PRD as the general requirements document to keep track of what's been done and what's not done or what's next as progress and have the agent re, you know, contact have that as context and just continuously update that documents.
Very cool. So, it's quite extensive. I'm just going to stop it right here because it's, it's, it's going to go forever. It's a very, it's, it's a very extensive document, but I'll just stop right here. I'll copy it. We're going to, we're going to drop that in here. So, we have a very extensive PRD document. And what I'll do is for the purpose of just keeping things clean, I'll just say using uh this PRD create a simple HTML, JS, CSS micro tool that I can deploy. So, you know, as you're doing this, um, do you have to spend much time sort of like debugging and things like that? Like, you know, I'm assuming that this very extensive uh road map that you're providing the agent helps you to kind of get closer to oneshotting a tool or or whatever. Is that fair to say? Like do you still have to spend some time kind of like, you know, fixing some errors and some bugs and and testing stuff obviously? Like just high level, what does that process kind of look like for you?
In the context of um, VI marketing, not as much. You're spending more time just reprompting and maybe just taking a different, different approach to how you're prompting, getting the output if it doesn't work in the context of vibe building and vibe coding. Absolutely. You know, de appro plan planning, approaching a build and debugging are the core components of it. Getting it to build isn't the easy part, but solving problems and thinking through how it should build it is the, the core, uh, kind of the core aspect of, of, of I coding in this specific instance of just us building a very simple uh micro tool. You're not going to run into a lot of issues with debugging.
Got it. Cool. Cool. Cool. So, uh, what we have now is a a fully deployable tool that we're just going to deploy. And I'll show you an example right here. Perfect. So now we have a a UTM link generator that you can use and embed. And obviously you would have create a page and add this tools and embed in there. So say for example if you know, um, we have a a link uh startup empire for example about, you would say your Google CPC or you know whatever the link is, it could be example link, yeah, uh you would just use the term and then you have a generated uh UTM link that you can then create, this is relevant for marketers that are running uh CPC ad campaigns or whatever the case is for them to uh to use that link in their ads to then go back to the analytics software and attribute as to whether or not that ad is actually bringing them traffic.
Yeah. And like this is again super relevant to your product because your product helps people unpack like, you know, where their conversions are coming from and tagging your links with UTM parameters is, you know, how you're able to, you know, probably provide a lot of value to, to your customers. So, um, exactly, yeah, whether you're using Google Analytics, Humbolytics, whatever the analytics tool is, you can then see this UTM data specifically in your analytics tool and see where that traffic is coming from.
Awesome. So, uh, give, give people a sense of like what, what Humbolty does and uh, you know, what makes it unique and sort of what's on uh, what's on your road map and where people can find more.
Yeah, absolutely. So, um, you know, I've been building this tool. It was honestly uh a tool I built for myself to address my own pain point, which was I get so caught up in Google Analytics. Sometimes too much data is no data at all or not any good of data at all. And the idea was how do I build a simple analytics tool that lets me um, run experiments with AB testing to increase conversions, to get funnel visualization the way I wanted to uh to have very simple uh traffic and uh website data and more importantly be able to actually um, you know, just understand what the data is telling me here and what I can do with it. So this tool aims to just address a lot of those issues and concerns around that and I, I built the tool honestly for myself and just happens that people want to use it as well and uh, you know, we, we empower a lot of agencies and creators and um, the the goal for us is to enable a lot of marketers and performance marketers to just have the all-in-one conversion optimization tools for you to be able to test experiments, see funnels and and see whether or not your marketing efforts are actually working.
Very cool. So, uh, that's uh that's humbly.com and uh if people want to learn more about like cursor and vibe coding and this whole convergence uh from you like where can they follow you uh on the interwebs?
Amir MXC, amir mirxc, anywhere they can find me there.
Awesome. Well, uh, this was incredible. Uh, we're hitting uh right at 60 minutes here and this is just like jam-packed uh of value. I think I'm gonna have to go back and watch this just to like unpack everything that you shared. Um, you know, this is just so much, so much sauce, so much alpha, whatever you want to call it, and I think people are really going to enjoy it. So Amir, thank you so much for coming on and uh sharing this with everybody.
Thanks so much. I really appreciate you having me.
All right, we'll do it again. Talk to you soon.
Awesome. Thanks.