Vibe Coding Will Be Short Lived If You Don't Learn This Skill

Hey there, I'm Parker Rex and I led a product team for a startup that got to $73 million in sales and then we exited. Yeah. The problem is is you know they don't know the fundamentals. So then they jump in and someone on the developer team or the person teaching them the the concept will just say just use Next.js to make your app. It's really easy. And then the person on the other side of that, the noob would say, "Okay, so I'm writing a next app." And they say, "Well, yeah, just use MPX create next app at latest and then pass the normal default flags." And then they say, "What's an MPX?" And then you say, well, npx is an executable using npm. They say, what's an npm? They say, well, you know, it's a package manager for node. Okay, what's node? And then they say, it's a runtime, but don't worry about that. Just run the command. Okay, cool. I got that to work. And that would be like three hours of figuring it out. And then once they do that, then it's like, okay, cool, then they'd probably ask like, am I writing a React app with Next or React app? And then the dev would say, no, you're actually using Next, which is a metaf framework for React. It makes it super easy. And then the normie would say okay so I'm writing next code then and then the person would say no it's actually typescript they say okay so I write typescript to make all the components and they wouldn't even know what a comp component is they say to make all the stuff on the screen for my to-do app and they'd say, "No, it's actually shaden slash UI. It's the easiest way to make an app." They'd say, "Okay." And so you you could see that like there's just this kind of web of different topics that you have to know to even get going. And that's really a problem. And that's what, you know, I think is is going on. And so a lot of these apps, they're trying to solve this need for like quick and easy builds like lovable or bolt new. Like they'll make you this first version and then you know the newbies real pumped, but then the second they try to get it off there or make it complex, it's bad. So, all that said, I think, you know, I'll be trying to break down some of these topics in in future videos, but I think that the Abe Lincoln method that basically is an accreditation to his quote about sharpening your axe makes a ton of sense when it comes to the AI coding world. And that applies to both the noobs and the people that are coding every day because the quote it basically goes like this where if I had six hours to chop down a tree, I'd spend four hours of it sharpening the axe. And the point of it is is you'd rather put a really good plan together. You'd rather think twice or measure twice, cut once. And it definitely applies here where you can spend, you know, in this case twothirds of the time planning. It doesn't mean you're not going to be using AI, but that'll be your time if you're a noob to be planning and asking a lot of questions. So, I want to make a things three clone, how do I do it? Can you give me the step-by-step plan? And then asking a lot of questions before you even get to step plan if you are a noob. And then for people like me, like the midlevel, you know, I've been coding about three years, but double time like crazy. So I want to put myself past three years, then on and off front end wise since I was 18. But, you know, in my case, you still would want to do that. It's still way more important. And that's something that you see in the AI tool world where you have plan mode or in the case of ADER which is a CLI tool you have architect mode where it won't actually write anything to a file but it'll have context of them so they're all read only and it has a system prompt that's just about breaking things down figuring out what needs to happen. But this whole idea I think is is a really good one where it's like okay there's this massive need in the market which is there's a huge flood of people coming in. They're very excited. They're very enthusiastic about building but they don't have any experience doing it. That's one like cohort. Then you have another cohort of people that are really into coding already. They probably have a job at a company but they don't spend their waking hours you know coding like their nonwork hours coding because that's their day job. and they also know that they need to adopt better practices because they might get canned if they don't. So that's another one. And then there's this one that's like me, which is the people that are using it every day that are like figuring it out. But a lot of them aren't like great at branding. So they just spam out prompts and then you can go to like cursor.directory. They also have like a blog now which I think is pretty cool. But I think it's on people that know how to market like myself to put together some sort of method, give it a name. Um, and that should be really helpful for people where it's like, hey, every time that you go to do something new, whether it is zero to one or one to 1.10, so any version thereafter, this is how you should do it. and giving it like a generic kind of broad approach, not like this is the exact model because then it's not evergreen and you want stuff that can be from like I want to say first principles but that's buzzworthy but it is kind of like that where it's like okay if we boil this down like and if a model's going to change every week and like what's not going to change and what isn't what isn't possible like Is there anything that's restraining us from building out a set of steps that will always be helpful? In my case, it's just expanding on what the planning phase looks like. So before AI, what did planning look like? Well, there was waterfall and then there was agile and then there was lean and then then there was lean agile and then there was shape up and then there was the linear methodology. So there's all these methodologies that come from the product management world where you're running sprints and scrum and all this stuff that happened before writing code. And so it's taking that and then systemizing it and modernizing it to this AI world where it's like, okay, for these 10 steps that you did before, you can actually just do them in like two steps with some recurring questions and these are the questions you should ask. Here's a good prompt that's working right now. So you can pepper in stuff that is like, you know, the taste of the day, but it won't be restricted moving forward. And I think then you can get a lot further where it's like, okay, you finish the planning and it doesn't mean that it's going to be six hours to make anything, right? This could be the same two two/3 planning one/3 execution within 1 hour or even 10 minutes. So let's see if my internet's working. It is not. So I'll be right back. Okay, we got internet again. I think it was actually just Twitter being down. I wonder what they're shipping. So, okay, that was the Abe Lincoln method. What's not going to change? I want to talk about uh this one next. So, VC advice on tooling and this plays into both Figma versus Canva and Windsurf. So, what I've noticed, there's this trend where a lot of VCs have a big following. And of course, I can't go to AX because it's down, but they have a big following and they're very opinionated because they need to uh come off as like super highly educated because, you know, they get billions of dollars to invest in companies. A lot of them are wrong. They're not in the arena using the tools. And so one case would be Y Combinator, which I think they're awesome for the most part. But what they recently did was they're like, "Hey, Windsurf is good." And they're like, "This is so good. It's it's the best thing ever." And cursor bad. And what's weird about that is they don't actually know what they're talking about. And then so everyone that sees that is like, "Oh my gosh, let's all switch to Windsurf." I don't know if they're just trying to like pump their own bags. Um, if you look it up, there's mixed responses on whether or not they've invested in the company. I don't care that much to know. But what I do know is someone going and using the tool and having one time when the thing doesn't work does not discredit the tool. I've used both. They both work great. I think that you should not spend time IDE hopping. Like I've done that thing before where you know you jump literally back and forth and you're ping ponging between like at one point I was doing like zed and then I'd go between zed and uh cursor and then cursor to clide or clin to continue to uh like the latest one was like GitHub workspaces. is G workspace. And the thing is is like after you use enough of them, you kind of realize like they're all pretty similar and they're all kind of just copies of one another. So just pick one and run with it. If you have issues after like a few weeks, then sure, like you should definitely switch. The one argument it might be is that like the switching cost is low, but go through that phase of trying a bunch of them out. Um, what just kind of bothered me and a lot of the internet was like, why are you say that it's trash when you don't even know what you're talking about? Like they claimed that uh cursor, you know, I think they said like no indexing like it doesn't do it. It doesn't understand how the codebase works. And it's like well actually sir, have you used the tool before? cuz I don't know that's like one of their best features. Like what do you mean? Like when you go into here you see features it's like it makes embeddings across your entire codebase. So to say it doesn't do that is just false, right? Like there's no question. You could say it's not as good then like provide some examples. So I think it's just like a bad look in general to make those huge claims, especially when you're, you know, like the Silicon Valley darling of supporting early stage startups. And what's weird is like Curser was and is a startup and they happen to be doing really well. So I think this is just maybe a case of like the crab in the bucket thing where it's like maybe they didn't get the deal that they wanted or like if they had really good incentive to support them then I feel like they wouldn't say that. And the same thing goes. I saw someone this guy um I forget his name, but he was basically like he makes a big claim and he's like Canva and Figma are going to drop UI builders. I don't know anything about it. I'm not like using the tools. I don't know anyone on the team. I think he may have used the tools, but he's like, you know, this is this big thing. And then what's weird is or what's unfortunate is it gets like, you know, 10k likes or whatever. So, if it gets 10,000 likes, that probably means 100,000 people see it. And then they're like, "Wow, like they're totally going to do this thing and this is awesome." But the reality is like this guy doesn't use the tools every day. So, what he doesn't know is that well, first of all, like Figma, if you use the tool all the time, they launched their AI like basically AI generator. So they had a thing that was like if you have Figma, you probably remember this, but on their toolbar there's this little like AI thing and if you clicked it then it would try to do generative um features basically and they launched that real early like they're very early to push it out. So, let's say on like day zero of their launch, they launch it and then literally like day one, this expost goes like wildly viral from a company called Not Boring. And this guy, he makes really good apps. Like, if you've ever heard of Not Boring, um I think they have like Vibes is this app where it's like this immersive sound design. It every app they do just feels like art. So, they have like a Vibes app, which is like a music app. They have a calculator, which sounds lame because kind of is. They have a weather app. They have a habit app. They basically have all these like utilities, but they're cool, aka not boring. And this guy comes out and he's like, "Yo, this thing's copying everybody's work." And it gets like, you know, 20 million impressions. And, you know, then day two, no more. So AI know more in Figma. And I think it's just crazy because they, you know, they're definitely poised to launch something like this, but I think that they're going to have to be really careful about it. So he's not necessarily wrong on the Sigma thing. Um because they are getting their lunch eaten. Like there's a whole swath of companies out there that are doing this sort of thing where it's like, you know, you definitely need to be able to bridge the gap between design and code. Like that's been a forever problem that's been trying to be solved where it's you had like Zeppelin, which was a handoff tool. You had Avocado, which is a a handoff tool. You had Envision, which is a handoff tool. And all these really aren't around anymore because they were like what I like to classify as stitchwear where they're solving this thing that really shouldn't exist over the long haul. But then you now have like this whole like family of extensions. You have tools like builder. You have tools like web flow that like all these kind of fall into stitchware or web flow yeah it stands on its own. Builder stands on its own as website builder. But there are a bunch of extensions um I think one of them is called like fire to dev. There's a bunch of extensions that are making a bunch of money just basically like bridging this gap. And so I think half of what he's saying is right, but the other half is totally wrong, which is like, dude, you just don't know what you're talking about where or maybe he doesn't even have this context. But in this case, it's like Figma, you know what's happening now is anytime you draw something in Figma, so let's say I have Figma open and I'm drawing some shapes and I'm making like, you know, some dope designs or whatever. And when I'm making those dope designs, every time I'm drawing one of these, there's a new node that is created. So like new node, new node, new node, new node. That's kind of hard to say. New node, new node. All these are nodes that then are stored in a DB somewhere. And I think that, you know, one actual use case, like I've used a bunch of these stitch wares, but they're not that great. But one that just came out is actually an MCP. So the model context protocol if you're not familiar it basically gives uh LLM's access to like standardized sets of tools. So I'll explain that more in a second. But this MCP would then allow me when I'm in cursor I can be like yo make this design. And why that matters is because it's more tied to nodes and like the actual SVG uh image generation like the paths and the colors, strokes, lines, all that. It's going to be far more accurate because an SVG slash the let's call them you know nodes but for the case of this we'll call them synonymous. that's going to be far better than just like a PNG upload or a JPEG because you know these are far more accurate like SVG is literally saying like you know start the line here start line here end line here and it's written in code versus this where it's like okay we got to run uh a scan over it okay like you know they got the magnifying glass they're trying to figure out where everything is and how it's positioned in space and so you just get like a way worse thing so this MC CP thing is really the first one that's coming out and that's also eating its lunch. But the point is is like Figma's definitely in the best position to do this. It's just how are they going to get it out without making a lot of people mad? And it's probably some combination of like anonymization um like maybe an optin like maybe they snuck an optin into a checkbox somewhere on the signup page but no one knew about it. Maybe they're just working on like uh you know you have like similarities scores when you're doing a bunch of data analysis where you're like okay how similar are these two things when you're trying to figure out if a model is working or not they're probably doing like a reverse similarity. So it's like okay if someone queries Figma AI and they say hey make um food delivery app then they probably are behind the scenes you know going through like v1 2 3 four five like they probably have five different iterations running and they probably have some model that is tuned. Uh tuned just means like it's tuned to design. And so like you had GPT over here, GPT4, they're probably doing like some sort of tuning on their whole DB of things. They probably have to like use something like scale AI to classify the UX and then they're probably classifying it using something like uh [Music] mobin.design. So some simple stuff that breaks it apart. So, I'd be like, "Okay, is this a business app? Is this health and fitness?" So, they might do it by um like screen type or UI elements. So, do we need a progress indicator, a toast, an icon, a bottom sheet, a sidebar? Point is, no one's doing this really well. And um I know I went off on a little tangent. I hope that was helpful, but it's important to have context of of the Figma stuff. So, the the part that like kind of annoyed me was just this Canva thing. doesn't actually annoy me, but I think it's, you know, it's good to have context because Canva, like if I pull up Canva, see, so if I pull up Canva, you know, it's I was looking at this guy's thumbnails cuz they're fire, but they've been working on AI stuff for quite a while. Like, they've been on like a spree of trying to use it. And let me see if I can pull up the AI thing that they just launched. I got access to like yesterday. Um, one second. No, I can't find it. What will you design today? Create an animation. Where's the AI thing? Canva AI. Yeah. So, this is their latest version that is in beta. But they need to nail this. They're not going to go and build a UI builder to compete with Bolt and uh and Lovable who are, you know, they're making strides for sure. Like they're really fast growing, but that's a totally different niche. Um, you know, Canva is a bigger market than UIUX and it's just different, right? Like this is a marketing collateral software. It is not a UIUX software. So, just kind of an important thing to see. I want to try this out though. Make a really minimal thumbnail for a video about Abraham Lincoln being a pioneer of AI. So, a painting of a painting cut out of Abraham Lincoln over some white graph paper with the words in a Sarah font that read vibe coding is not the answer. So, it's pretty specific, right? It has background, it has foreground, it has cutout, it has a style applied to the cutout, and then it also has text. So, let's see how good it is. Just sort in here. Creating a brief unleashing my inner artist. Okay, cool. Yeah. So, I mean, this is actually close. Like, if this is all Yeah. So, no graph paper. It is a cutout. So, it didn't do the cutout on top. You can see they have a lot of work to do, but they need to nail this first. So, that's that. Uh, lastly, I wanted to go through Klein. So, you know, it's funny because I just talked about how you don't want to hop around, but there is an exception here where Klein, they're like the Snapchat. So, if you remember Snapchat stories, like there's a really popular meme where it was like um you know that's mine or no I made this and another person says I made this and it's literally like I think it was like a photo of a dog or something but it's like you know the person's holding the Snapchat story and that person was Snapchat in the meme. So, let me try to draw a ghost that looks like an awful ghost. And then it's like another person that's grabbing it with really long arms and they're like, I made this. And it's, you know, YouTube, LinkedIn, but of course, most importantly, Instagram. And so this same thing is kind of happening in AI where, you know, you saw stories go, you also saw like if you recall, like there were like dog face filters. Those were not anywhere until Snapchat did them. But the same exact thing is happening right now in AI. Let me refresh the page where how do I get the pen? There we go. Same thing is happening in AI where all the code editors are basically fast following is what the term is. But, you know, you could also just say copying where Claude Dev was their original name of it and then they rebranded to Klein because they stopped just focusing on Claude. They didn't want to be cornered. But this is like a solo dev. And you also have Ader, which is a solo dev, literally one person, at least the last time I checked. Both of them are open source, which means you can read the code. Both of them have really good communities. But they're the ones like I think you know Klein basically when it was even when it was called cloud dev it had uh let's call it yolo mode I don't know if they called it that but it allowed you to do what was more importantly the agent and I'll just put like recursive and what that is is like that really popular feature you see everywhere that's like instead of okay I send a message and then it sends me a response and then I hit, you know, copy and I have to go and I have to do all these things. Instead of that, like they basically pioneered this thing where it's like you give it the info and then this thing just rips and rips and rips and then you can say yes or you can hit no. And so they invented that and then oops and then short after uh you know you see other players coming and doing that. That's in every single tool. Same thing goes for Ader. Like Ader launched the like forward slashcomands and no one was doing this. Like they have architect mode. They have uh map they have map refresh which just basically allows you to get better control over the context. They have forward slash add with wild cards. So I can be like forward slash add space uh app forward slashastster asterric slast. So I can do like um nested selection. They also have like readon which is for context. They have a yaml file. So it's like conf.yaml YL which I can then go in and say like this model for that, this model for architecture, this model for code, like basically infinite kind of customization and this stuff has been around for a while which is crazy because you know months later now if you go into any other tool like the ones we talked about where it's like okay cursor check they've got that okay uh windurf. Yep. That's like literally their whole thing, right? They call it flows. Well, guess what? That was over here. That was called uh just claw dev like maybe a year ago. So, I'm going to try that out and we'll see how it works. Let's just set it up. And I want to see how long this video is going because I got to wrap it up in a sec. Um we need to Yep. Twitter's still down. So, let's just go ahead and we're going to set this up. So, we've got the troublefree AI website open. Um, and we're going to give it a task. Let's, you know, normally I'd have this thing open with the normal uh like AI pane, but instead we're going to get client restart extensions. And the reason we're doing it is because a new version came out and I haven't used it in a while, so I want to. Looks like you can use open router if you're not familiar with what open router is. It's a kind of like a a bridge to all the different models. So you can go in there and pay like 20 bucks and then you get credits for all the different ones you want to use, which is pretty convenient. So I wish that Twitter was up so I could see what the newest channels are or uh features, but instead we can just open their Discord. And I want to check this to make sure it's worth our time together. So, while that opens, unless Discord crashes my computer. Discord's such a heavy app. I've noticed. Okay, we're in. Uh, releases. Need to follow a channel. Okay, follow a channel. Um, verifying Discord. Seven updates. Sheesh. Come on. I need to get invited again. Okay. Klein extension. I already have it installed. I want the Discord. Discord. Okay, let's see what is new with this. Um, releases. Okay, this came out a couple days ago. Now there's a client API checkpoint 2.0. Is this going to open? No, it's not. Elon's pissed. You can just tell it've been down for like 20 minutes. Well, let's just give it a try. Let's grab this. Let's see where I can pin it. Um, in here. Yeah, pin that. We can unpin Thunder client, unpin projects, and yeah, that's fine. Let's open this up. And we're going to allow them to use the usage stuff. Oh, here we go. MCP marketplace. Discover and install the best MCC MPC NCP servers. Oh, that's smart. Okay. Mermaid diagrams in plan mode. That's really smart. Use terminal to reference those. That's smart. Okay. So, looks like the last time I used it was in October. And let me make sure I have an API key. Oh, I do. I think I had free stuff, too. Let's We have plan and act mode. Okay, cool. So, they renamed them. Um, let's give it you have this auto approved thing. That's interesting. So, I can say if I want to let it go or not. I have an cursor update plan. In plan mode, client will gather information to architect a plan. Great. Act mode means that they rip. Okay, so let's do this. And this is connected to a Google account, which I should I'm going to write that down. I got to leave you guys with some free credits. So, Google free credits. I think you get like $200 without even entering a credit card, and you can apply it to Sonet. So, I'm going to say, let me run this app. I think I was originally intending on moving all the content from a blog page to is this using npm or pmppm? npm. I was originally trying to put all the content onto the homepage. So, let's just do a run and starting the collection. Um, is this on the latest version of next? No, it's not. But we could ask it to do that. That should be really simple. Yeah. Okay. So, we're going to do this. Let's get it to do a plan. We're in plan mode. Okay. We want to get out of the settings. We want auto approve off. Generate a plan to move all of the content that's on the blog page, including all of its subnested pages, to the homepage. So that means that we can um have everything on the home. Actually, cancel that. One second. We need to move the content on for/blog to for/home. Delete everything that's on for/home. Retain the path for [Music] the actually I don't want to retain the path. That includes the nested content on the blog. So the categories and all the individual posts themselves. There might be other areas of the app including metadata for SEO that needs to get changed. So I'm like purposely not getting that specific because I want to see how it does without tagging literally anything. But let's see. Will it run? That's the big question because I have not opened this in a very long time and I don't even know if the project's set up. Need billing enabled. All right, we're going to do this in the next video, but I'm excited to try it tomorrow. And um I hope this video was helpful. If you have any questions, definitely leave a comment below. Um what are your guys thoughts on this Abe Lincoln method? Are you guys vibe coders? Are you guys coders? Are you guys uh trying to do AI services? let me know cuz I appreciate any comments that I do get and I appreciate you watching this cuz I'm going to make every day I'm going to make one of these. So come back tomorrow or the next day or the next day. See you.