Vibe AI Design Agents Arrived to 10X Your Skills (GFX designers VIBE to Win 2025)

Hey there, I'm Parker X. I led tech for a startup that did $47 million when we sold for 23 million bucks. And that's a story for another time. Horrible multiple, but we did it. After that, I tried my hand at building Airbnb for music. Why on earth would I do that? I don't know, but I did. And after that, I got into building with AI because there's a ton of leverage. If you already knew about product management, if you already knew how to do UX design and if you knew how to code, then it was a great way to get a multiple on your output and the quality. So on this channel, I talk about the business that I do now, which is AI services. My goal is to get to 100K in profit a month. I'm not there yet. I'm on my way. And I talk about the news that comes out every day for random AI stuff only if it's relevant to pushing you forward in your business. Talk about some of the strategy and some of the builds. So, let's get into it. Oh, and I do Q&A, of course. We actually have questions that come in now. So, let me scoot this over. I just headed this up because I was hoping that my CPU wasn't going to go crazy because I'm trying some new settings. Um, let's jump into it. There's This is going to be a banger today. So, let's go over to the questions. Bro, great content. Thanks. That's not a question, but I appreciate it. Ask a question next time. when you have one. Cool. Say hi to new sub. All right. Hi. Nice to meet you. Cool. Those are not questions. I do think I got a question in our uh community. There's a like a paid community that I've been working on. The big tension there is like how do you make a community not cringe? Kind of like the biggest thing that I think about. So many of them are cringe. But okay, this guy says rub a duck. Good. I did. That's called a comment uh template. So, it's like I'm going to say a word or phrase or something to try to get people to say things and it worked. I just literally said rubber duck. This is when I was trying to learn how to get better at writing CC code. Why would you write CC code, Parker? Well, cuz it's the father of all code. But you don't have to learn C. But I have a duck there and I said said rubber duck. So, anyways, he says, "Found your channel. You definitely deserve more views and clicks." Thanks. I started this channel three weeks ago, so I'm happy. Thanks for uh subbing. These daily updos have been a great way for me to keep up. Question for you. What is your essential AI stack for programming vibe coding? I've been using cursor for personal projects and I am forced to use good GitHub co-pilot for work. I've also been shilling out the 200 a month for ChatGpt Pro and 20 a month for Claude. Wondering if I should swap out Claude for Gemini. Now, since I'm more of a novice programmer, I can't really tell the difference between the Gemini responses and others. Yeah. So, okay, there's a lot in there. It looks like you, if you have to use GitHub Copilot for work, I'm pretty sure you can just pick the model that you want. So, do that. The thing that I would say is don't hop back and forth. Like, get good at one and just use it until you're good. until you can actually ship things that you want to that you want to have out in the world. For me, I hopped back and forth a ton and I built a bunch of different apps and that was great and I launched a bunch of different things, but it probably slowed me down in the long run. Like I built a product that got like 100,000 users. It was totally free. It was very simple to build, but that was me doing it. It took a couple weeks, but it was like, "Oh, I would go and I'm using uh oh, Zed just came out with this new thing, so like let me check it out because it's written in Russ. I hear Russ is badass. Let me do that." And it's like, "Oh, cursor has a new thing." It's like, "Oh my gosh, what about wind surf?" And wait a second. I should just use Envim because it has a vantage. And then you get like really mid at all of them. And some would say it's really easy to hop between all of them. And those people would be correct. But I just still think that like just get really good at one of them. Like anything like just for me it was cursor because I started using cursor in September the like 22 I think is when it came out I think. Yeah. That's when I started using it pretty much every day. Maybe it was 23. But for me it's cursor. Just use co-pilot at work. If you have co-pilot workspaces, have them enable it on one area of the repo in one branch because it's really helpful. And if you're not familiar with copilot workspaces is it's it's uh like copilot has a bunch of different things, right? Like in VS Code, you have it and it's over here. It's just like your training wheels. It's over here. It's got the composer, but it also like, you know, so this is copilot in the IDE. And then co-pilot on the web is a couple different things. So you have one which is like I always recommend for noobs that are trying to learn a concept, find a repo that's using a package that you're trying to learn or find a popular OSS project. Type in pop type in OSS uh in the language that you're trying to get better at and people have already collected lists. So you'll go find like oss for like with React or using React. you'll find a hundred of them in there. Then you'll go in there and then you'll see the little button up here for the co-pilot and that's always next to the search bar and if you just click that then you can ask a bunch of questions. So that's the second version of co-pilot just navigating and having rag against it's playing in rag. Well maybe it is rag and azure is probably what they're using and if you're ever familiar about that it's really good. I I built stuff there too. Um, but the third one is the workspace. So workspace, if you're familiar with Devon, Devon's the thing that takes eight years for you to get a response in Slack and then most of the time it's broken. But I'm sure it's getting better. And Devon made GitHub workspaces a thing by them releasing that. GitHub Workspaces was like, whoa, hang on. We're Microsoft. We have a lot of money and we should just do that because we already have all these issues in here and you can just click on one and then I used it when it was in beta early access but I would recommend you guys checking it out because it would actually for like little things. It's cool to see how it works like it'll put it into this like side pane thing. So your issue will be over here and then you can actually like drag and drop an issue in here and then it'll go and it'll work on it autonomously. Then you can like come back and it'll make a PR. Pretty cool. They're gonna go totally across the board on everything. Like GitHub will build all of this. It will delete like if you're thinking of making an extension, RIP. Most extensions that do like summaries of what the PR is, like they're good now, but they're tweener businesses must be anything. So for me, just stick to one. my go-to stack if you are curious. I hop between using uh obviously a terminal. So I I always end up using this because I'm not trying to spend time nerding out on my nit.la or random config things and I'm not great at writing bash by hand. I don't have a neck beard, you can see. So I just like to have it go and the best engineers I know will make fun of it and be like why aren't you using like you know elacrity but then you have to get this other thing that like keeps it open. That bothers me even though I don't use a bar at the bottom. The thing at the bottom you can see this is whisper flow. I actually have my nav on the left but yeah I just use this and then I use cursor. And so cursor, this is just the like a website thing because I need to add something to it. But I just jump between there. And then I was using Klein, but what I actually just use is because I got a free account through cursor or free like max one and I'm just used to it is I use the agent but then I customize it. So, I never use auto. I'm always avoiding 3.7. 3.7 will write so much stuff that you don't want. It's awesome at writing a lot of stuff. Like, you just have to be so strict with it. And I always just go back to 3.5 for the actual code mode. So, when you're thinking about using any of these, doesn't matter the tool. This is tool agnostic. This is just a rule of thumb for for doing good code. Uh Jen is if you see on the left it says use different models for plan and act mode. This paradigm shift when you start doing this you will make better stuff. I promise. Pretend like you are a product manager because that skill is more important than ever. Like how is this thing going to like why why are we building this? What is this thing that we're making? finding artifacts in the code to help kind of formulate that plan and then using a coot or a chain of thought model for that part. And the way you'd mimic that like why I like client so much but I don't want to light money on fire using it is because I can say for plan mode use 01 but then for act mode use you can see I was still using 3.7 but use 3.5. Um, that changes now though. Like I haven't coded in cursor cuz I was coding in uh in warp actually. I was just like trying to mess around with there with with them. But I would be using Gemini 2.5 Pro now. So this is the one that I'd be using for the plan. They still haven't nailed it. like I haven't configed it around it because when I was coding testing this yesterday it always takes a little bit of time for IDE to get used to how these new models work. So you'll also notice that with just like people's feedback like I talked to a friend and he's like hey I heard cursor sucks and it's like oh well when did the person that told you it used it how many times have they used it? He's like, "Oh, they just used it like once and they did it like the day 4.5 came out." Well, that's why it sucked because cursor is still trying to figure out how to make it work. And if you follow the people that work at these companies, they will be open and honest with you where it's like when Claude first came out, it sucked. Like it wasn't what it is now because cursor was trying to figure it out because cursor has its own thing. Same thing with Windsurf like there's multiple steps, right? It's like you have Open AI and they have this new thing that comes out and it's shiny and then they're shooting it down to the ide prompt system prompt version 13.1 that we're running and maybe they have 10 different versions of these running and seeing the numbers of accepts that are coming back from people because they're going to be doing telemetry. They're tracking how effective this is because they don't want to wait for people to come back. They're just it's product analytics. So they might have a bunch of different variations similar to how Snapchat early days they have millions of versions of their app running. Same thing with Facebook right now, Instagram, they might do it by country. With us it's just by cohort by small little little groups of people. They have to figure out how the model will work. That's a very long-winded way of putting it, but stick to the one that you like and the models that you like. When new ones come out, test them out, but realize it's not going to be that solid. My stack will be I'm imagining it's going to be 2.5 for the acting and then Grock 3 for the planning when Thank you for the question and I hope that was helpful. So, on the new side, this is crazy. So Klein, I always talk about this, but it's the same thing every time. Okay, let's go back in time. It's like 2015. Evan Spiegel, he's working at Snapchat. He's the king, right? And he's like, "Hey, I have stories." And he launches stories. And the whole world loves it. And then Zuck's like, "Yo, I'll buy you." And then he's like, "Nope." And then he's like, "Okay, no, we have stories." And then LinkedIn has stories. And then like Pinterest, like who uses that? But they all get stories. The same thing happens with Klein. Klein makes agent. Next thing you know, cursor has it. Next thing you know, wind surf has it. They might rename it. But everyone follows them. And so I mentioned this before, but I use cursor with client edit because cursor is obviously got all that native stuff built in. Doesn't make a difference because the client extension is dope. But this working pattern just really it's it's very effective. So context matters right it's the memory and even if you have like claude 3.5 they announced 200 no half a million uh context window size which sounds cool but it also just means you're going to go bankrupt. Um a good example of that is T3 chat half of their operating costs are just clawed. So the guy when asked if they're going to do mobile said we can't because we're going to go broke. But the way that this works is you essentially have a set of instructions that's called the memory bank. The memory bank bank has a flowchart in it. And the flowchart defines the pieces of memory that it's going to have. So it has a project brief. That's the thing you're actually working through. So if I was making something new, I'm going to go into the project brief and that's where I'm making my plan and that's where I'm coming up with all the context. It's can also be called a spec prompt, but in in this case it's called project brief. So it's like, hey, we're going to build this. This is why we're doing it. This is the expected functionality. This is the current functionality. These are the assets that we need to put into place. Whether it's a library or a set of images or um a pattern or an MCP, any of these things, all the details go in there, that's what you're touching. And then you have a bunch of other things in here. It's a system patterns, so it's going to go and actually like scan through the project and find I have the old version of it because they put this blog out like four seconds ago. Um but if I went in, let's see, me uh project brief. So, if I went into here, I just tried this stupid oneshot and this it's going to try its heart out to make this helpful, but it's like I want to make a blog post out of my latest YouTube video on my daily uploads channel. Pull the transcripts and make a blog post out of it. This was me seeing like what is it going to think? How far will it go? And this is a good example where vibe coding sucks. like it gave me the worst possible way of doing it because it doesn't know what I know. And so being competent is great. I highly recommend it. And the way you get competent is by just reading stuff and doing the research, having a basic understanding of the fundamentals, leveraging AI along the way to do so. Ask a lot of questions. This thing is a 247 tutor. Don't take advantage of it. Don't sleep on it. use it, but don't you know ask it to do everything for you. That's very different. You need to learn. So there's a system pattern. So anytime that I go and I do something once I've written out a project brief on this new feature that I need to build, then I'm going to say, "Hey, follow your user instructions." Now in cursor, it's a little bit different because they have the rules. So I just took that same thing and what I'll do is I'll drop the new one in. Oh, here's what we're teasing later. This is the uh the update. I'm really leaning into the vibe thing because it's such a massive trend and I think the culture of it it's it's there. But I'm going to grab the new the new one and put it in there in just a second. Every time you start a new task, you waste time reexplaining your project tech stack and architecture. True. What if client could maintain perfect memory? True. Yeah, we already knew that part. Memory bank is a self-documenting system built right into your codebase. Yep. So, what's different? Tell us what's different because I have not seen anything different. Maybe they just they're advertising it more or there's subtle differences. It uses structured markdown to create a persistent memory that survives between sessions. Yeah, that was there. And by the way, there were early versions of this that I made on my main YouTube channel. And this is just a way cleaner way of doing it. I will say also that this is a cluji workflow, meaning like this is here now. This won't work in Q3. like it's going to be gone. They'll figure out a way to productize this. I've spoken to the cursor team about this already. Before any work begins, client reads these files to rebuild the full context of your folder or of your project. Cool. Which is nice because it's like, okay, no cold start. Like we're not, you know, relying on um previous memory and we don't need to go and check that the index is completed and hit the refresh button. like it's an additional thing that I I do really like. This continues with active context. So it keeps the current state and the insights, the project steps and statuses and it works together to make perfect or maintain perfect context. So it's elegant because it reads yada yada yada yada yada yada. So I don't know if they changed anything or if they're just been SCOP, but let's just test see if they made any differences. Um, we'll first put that in there and then we'll go into where would it be? It's over here. Oh, I'm not going to see get diffs. That's unfortunate. Well, let's assume that it was updated. If not, I'm just talking about it again. Whatever. One thing that also I I think that would save a lot of people time is just whenever you're vibe coding, make sure that you're using stacks that are really popular. Like don't go get shiny object syndrome because you're going to screw yourself. Common mistake especially working with Superbase. And I bring that up because like this chart shows vibe coders. That's what this is. Like this explosion here. This is a strong correlation with olt and lovable. And I mean, I think the the vibe coin or vibe code term was like around over like here. That means that that chart's probably absolutely bonkers now. So, next up, I just want to talk about this vibe thing is spreading like wildfire. It's getting written up in big like press pieces, which is just really cool to see and I'm definitely embracing it. So, let's look at it. Vibe coding. It was nothing, right? What a weird phrase. Like, Laos, what Singapore? But then we see a little rumbling and now it's going up and right and it's kind of wild. You see all these AI influencers and now there's this thing called Vibe Marketing, which I think is really interesting. It's it's really like if you were to apply the word vibe, it just means like, okay, it means leverage. Now, I think there has to be some rules around it. I've talked about this before, but like the Abe Lincoln variation of this, and yes, he's an old president, but he also says some smart stuff, which is if I had eight hours to chop down a tree, I'd spend three quarts of it sharpening before I'd take the axe and slap the tree with it. I think that's the same exact thing here. Spend some time planning, then go do the thing. The marketing side is really interesting because it's basically saying, "Hey, look. We can get way more leverage by in the future using voice to set up campaigns, to set up outreach to figure out the other side that every developer struggles with, which is how do I get users? I've built all these things. Look at all these amazing features. I'll just feature myself out of this." No, you need users. So, I think it's going to be similar where it's like, okay, be doing multi-agentic flows and it's going to be awesome. That's actually why I set this agent team up and I can't wait to demo this soon. I'm going to share it with with people in our community pretty soon when it's when it's working, but I'm going be building it in there. Um, but apparently Doctor doesn't want to open. Whatever. That's Canva. Let's go back to here. That's a bunch of strategy stuff that I think I spoke about yesterday. Yeah. Uh, spreading like wildfire. We're going to talk about it in a second. The trends though are crazy because I think all like with vibe coding it starts it starts like with really niche stuff on GitHub or X and I bring up GitHub because in the case of Vibe coding it was on X but in the case of DeepS I had been following DeepS's stuff like six months before they launched R1 and then I made a tweet the Friday or the Thursday, maybe maybe it was a Wednesday when they dropped that and the world hadn't seen it yet. So then I said, "Oh man, like NASDAQ is going to lose like half a trillion on Monday." And sure enough, by the time Monday had rolled around, enough people had talked about it, enough FUD or fear, uncertainty, and doubt had been spread such that everyone flipped out. They're like, "Wow, the Mag 7 is trash." Like, whoa, they did this thing with $6 million. even if it's a 10x multiple on that they did it for a lot cheaper and it's not that much better like our current models our current soda models locally. So what tends to happen is like trends start here in the shadows of X in GitHub and then what's happening now and you can track this with the vibe thing but even more relevant is the Gibli anime thing like if you if you didn't realize like all these anime Asian looking like illustrations are I mean it broke the record for number of images uploaded on the X platform. form ever in a day and it took about 36 hours for it to get to Instagram. So Instagram's pretty quick and then I'd say Tik Tok is tied. So what I noticed was like all the theme pages like basically a new tech trend on X comes out of the mouth of some influential person and a bunch of these news accounts, right? They're not people that actually code or actually do the thing. They're just a media brand, which is fine. Media, they print, but sometimes it's a little sensationalized. But there's then this whole other swath of influencers and like theme pages. And I can show you a theme page example because I'm going to be doing one of these for my brand. But you'll have like these theme pages where like, okay, here's one about like how to be awesome and disciplined and whatever. and they have like a similar aesthetic. And this one's like so cringe. But they went 0 to 72K in like I don't even know. I I think I say it on here. This one went from 0 to 629K in 4 months. So like these types of pages lag by about 36 hours. So X is the source. GitHub is the source even before X typically. I guess it depends on like if it's tech news versus like an influential person saying something. and then 36 hours later it lands here and then general population finds out about it and then LinkedIn comes in a few days later and then it's it's all the way through and then has like this kind of like shock wave after the fact. So I bring all that up because I think the Vive Code thing and especially with people pushing it as much as they are like that's going to be everywhere. Like I look at that chart of the Vibe marketing thing and it doesn't mean that that word vibe is going to have long staying power but it's going to open the eyes of the normies on what's possible because so many marketing teams just have no clue what's going on where it's like oh I have a five person Instagram team okay why why can you explain Even if you like look at Dolingo, they have like a 12 person team and I just don't think that they're as savvy as the people like they I don't blame them. The mainstream, the teams that are out there, they don't know what's happening in this little corner of the internet that you and I operate on. They're not clicking on a daily upload video or going and reading, you know, until three in the morning. The NAND forums, which are a gold mine, by the way. Go check those out. Like, it'll be someone that's like, "Hey, I made this thing that saves me four hours a week, and uh here it is for free." And you're like, "Thanks, Fred." Like, it's crazy. Um point is, it's like a lot of alpha for people. Like, I'm just so excited about this whole period of time that's happening right now, this timeline. So that brings me to this like lifestyle design thing. So I spent a bunch of time today where I had this goal where I was like, "Okay, I want to get 100K." And I still have this goal, 100K a month in profit. And the timeline of that goes like this. So let's go back. Uh Parker, let's zoom in. Parker road to 100K a month. profit and so did SAS for a while. Raised money, raised 150K, spend it all on engineering. Tried to make it work. Pivot, pivot, pivot. Uh founder breakup investors like just what's going on? Like this is the 10th idea. Like just go quit. I had an investor try to recruit me. Like they're like, "Dude, you're still going? What is happening?" So then at a certain point I'm like, "Okay, like I have this skill set. I need to definitely monetize this." So I was like, "All right, like I'll just take a contract." And like the contract's about like let's just say 25K. It's like okay well we're a quarter of the way there but zero leverage. Like you're you're not able to take much out of that. Like you're doing bespoke custom software and you're not and that's monthly but it's like you're not getting much out of that. You can sneak in maybe one other contract. So I probably could have gotten it to 30 if I really wanted to. It' be such a push and then your other areas of life suck. You're not posting content, which is leverage. It takes time, but it's similar to dollar cost averaging in the S&P. Like, you're planting these trees and over time, brand affinity is built, brand trust is built, you know, uh, customer base, all these things. So, then I'm like, okay, like that that's not it. So, how do I get higher quality revenue? Well, yeah, I'll just do contracted stuff on that's like things that I can just make SKUs like my little offers because I had looked at this through the lens of building custom agentic software systems for other people isn't a template and now I'm questioning if it it might be at least part of it is right so then I was like okay let's go and do and let's get a couple contracts and maybe we're pulling like now 10 it's like okay I'm going to go down but I can get way more leverage on these. And so as we move forward, then I'm like, okay, I can maintain that and then I have leverage aka time to then spin up content to get better at making the stuff. And then also, how do you get the most amount of knowledge and the most amount of network when you don't live in like Silicon Valley and you don't have a big brand? So I was like, "Okay, well I see this thing that's called school and I see these people that are doing like 100k a month, but most of them it's not like I always lean towards saying it's like cringe. It's not cringe. It's just not my style to be like I'm going to teach you how to make money because I don't I think as soon as you do that, like people do it and it's not cringe, but most people that do it, it is cringe." So, I'm like like I'm not your guru, but I think there is a right way to do it because the more that I look at this and the more I realize the leverage that I'm gaining out of like I could either stack 10 10k contracts and then just like be puzzling them. There's a lot of ways to get to 100K a month. Um, and I know if I just went all in on client work, I could do that. But I'm actually willing to delay that gratification to get what I really want, which is like, okay, if you had this where it's like you do better on the quality of the client, it's like, okay, three 10k contracts and you look at the custom software space as SKS. So you're like, what does everybody want? Everyone in the market knows that they want AI. Okay, great. That's not that helpful. What do they want within there? Oh, they want rag. Okay, which version of it? How is it going to work? Okay, do they want it like do they have a superb base stack or is it Azur or like what is that? And so you kind of like break out what the things are like okay within AI is this like iOS because like I can do that but it's a pain in the ass and I haven't touched that in wow like nine months. And so you like in in the current iteration, but I'm looking at it more like this where it's like, okay, what are the skills that are in my head? What are the things that I'm really fast at that everyone values? Because a lot of people that make that mistake, there's probably a lot of things that you do that you find really easy and you're so quick to discredit the value that they bring. But that actually is the alpha because the thing that you're really good at that you just kind of be like, "Oh yeah, like just like I just did that thing." That means that that's your expertise. So now I'm looking at it, I'm like, "Okay, this is like steady, Eddie, easy. It's good money." And then that gives me the affordance to then go and fill the other 70. Now you could say, "Okay, well, why don't you ramp it up to 50/50?" because I think that the time that it's going to take to build out my own SAS and build out the community that I do want to have because I think that having like I always go back to like this idea of okay here's just me and I can spend if I spent 10 hours a day doing research on this and doing client work on it yeah I'm going to know a lot more than a lot of people but guess what if I can then take that and then use that as the information and pull it all together and put it in here and then put a little price tag on it so that people that actually take stuff seriously. They don't want just like freebie gutter content. They want to get in there. And I It also like filters people like, "Yeah, I'm going to make money from this to make the stuff better to then reinvest in all the other stuff." So it's like then I can have better content to get better people in there so that eventually you can be like oh I want like Logan from Google to come and talk to us about stuff and give us like access. And the point is is like I believe that by doing this and even if it's like a small sliver of the profit on this because I want eventually for it to be like this where it's like all of this is SAS like that's where like I've been trying to go but the SAS that I was building is really challenging. Fortunately, agent tick uh or the agents SDK makes it a lot easier, but I already have momentum on the other stuff. But this is the vision is like all SAS and then this small sliver thing is dope because then there's a million like not a million, but there's like so many better ideas and different perspectives in here that like I just couldn't think of. And so selfishly, I want that. I want like cash flows to spin off this to actually make the thing good cuz I can't spend that much time on it. So I can like hire people and and ultimately like build a bunch of agentic stuff around it to make it better. But I think the community aspect just it wins because you have so much leverage with all these minds together. Um even some of the people that have been call commenting on this already, I'm like, "Oh, cool." Like they get it. So that's that's how I think about it. And then when the actual like, "Okay, well what are you going to do?" Well, with this I'm I launched this thing called Trouble Free. I hated the name and I'm like, "How do I make it not cringe?" Um, so I was like going through today. I'm like, "Well, the vibe thing like really is picking up." Like I see comments all over Instagram about it. I'm like, "Whoa, these are like total normie muggles that are saying vibe." So I'm like, "All right, well, what would AI for Vibe Marketing be?" Then I see a couple days later, people are posting like crazy about it. And it's like, I have one pulled up, but let's see. I had something pulled up. Yeah, it's like there's people publishing content and making things specific for this like vibe marketing thing and I'm like whoa, we are on the same page, brother. Like the stuff that's happening in Naden and attaching Naden to non drag and drop aentic stuff like that's a killer combo and that's where it's going. we're at the front, you know? So, I want to have like a huge group of people that are like at the front so we can just print and we can push this whole thing forward. So, because that trend's going on, I'm like, "All right, change the name. It's going to be this vibe thing. I bought a bunch of domains. I bought like three different domains around it." And then I've just been defining like what would actually be valuable. What are the modules? What are the things that you, you know, do. So, this is like my strategy today was like, all right, cut annoying clients, focus on better clients, take advantage of the fact that this is a wave. Like Forbes, as much as you hate them, as much as they have that like super like the the cringe 30 under 30, like it's they're talking about it, like Fortune's talking about it. It's kind of a thing. So, I'm going to ride it. I'm going to own the keywords for these by just producing a significant amount of content around it. And then there's also like cool services you can build around this where Andre Carpathy, the boy, he made a comment yesterday saying like, "It's embarrassing that I spent three hours trying to figure out how to cobble together a web project." And I'm like, I'm better than him at something. That's fantastic. But it's true. Like he's Mr. Python. And when you come into web world, many of you guys know this. It's like, okay, I want to make a website. And you're like, let's do it. And if you went into lovable, it's going to fart this thing out. They're like, I am John Carmarmac if you don't know. I mean, they do. But then they leave and they're just screwed because they're like, "What's a next doconfig.ts? Why what's mjs? What's what's js? Why is that there? Why is my tailwind a ts and not a j?" There's all these things and then how do I get the database connected? Like I literally someone's going to pay me to connect their database. I'm like, "Here's the docs. Just read them." Like, I still don't get it. And I'm like, "Oh god, there's all these things." And so there's, you know, this rabbit hole where it's like, "Oh, I want to make a website." And it's like, "Okay, well, what's what's the framework you're going to use?" And they're like, "Oh, I'm going to do it in React." You're like, "Okay, great. Well, you need a meta framework." And they're like, "Okay, so yeah, what do I use?" You're like, "Oh, Nex.js." And they're like, "Oh, cool. So, I'm writing Next.js code." You're like, "No, no, no. You're writing TypeScript." And then they're like, "Oh, all right. Well, what are those?" Anyways, um, so then I'm styling it in TypeScript. You're like, "No, no, you use Tailwind." Like, "Oh, cool. So, I'm writing Tailwind code." And you're like, "No, it's actually that's a that's a library around around CSS." And they're like, "CSS? So, I'm writing CSS and TypeScript." No, you're using the class utilities in Tailwind. They're like, "Okay, great." So then they start spamming the AI with, you know, the finger of the the casino button and then they realize, oh crap, like this thing's actually trained because it has a cut off of October 24th. It's trained on 3.0, but this is on 4.0 cuz I took it out of Vzero and it just becomes this freaking nightmare. So, I think this can be solved, you know, like it can and it will be solved. But in the meantime, there's definitely like I bought if anyone wants to do this and and build it together, comment the word, let's see, fix my vibecode app.com because I bought that. Actually, don't do just say let's build that together. So, if you say, "Let's build that together," then I'll DM you wherever it is and I'll find you and we will build this because I already thought about it and it makes too much sense where you could literally just have a service and then you could have a gentic system for the content, getting the eyeballs on there, owning the keywords for those things. Like the long tail would be like vibe code app and then I could direct traffic to it, but it would basically just be like, "Okay, the website's really simple." Then you have a big old button that is a React drop zone. They drop their codebase in there. They hit a little thing that's like, "You're cool with us reading everything about your project. We're probably going to find a lot of stuff really quick." They're probably going to have their environment variables in there. They're probably going to be using different stacks. Like the replet one is going to look a little bit different because they use Vit. Probably wouldn't touch that because I don't know Vit as well. You'd set like here's the normal things that people mess up and you'd go through coincidentally you go through all these and then you'd set the things around them. You'd be like, "Okay, cool. They're using superbase. They're probably not awaiting the cookie. Okay, cool. They probably didn't do this." And you could identify the different things and then hit them with a Stripe payment like 250 bucks and we'll fix this by next day. And then they don't have to like create an Upwork account or like go through that rigma roll. And then ideally the agent just literally runs through the little list of things that you've given it. You have maybe a couple different agents that have different pieces of work and different pieces of context. You give it a full codebase of a working one against one that's you know the new one and you could run comps on that to help identify it. You could run this thing. Uh I think it's my raycast just to make it easier. Made a dedicated video on this and I've worked on it since. but this hotkey prompting thing. So you would run one of these. It's like you're an expert incident response engineer tasked with incident diagnosis and resolution. So our current problem reported is and you could maybe have a form that just explains like what the errors are, but you wouldn't need to because that would probably reduce conversion. You could just run it and then log the errors. Um so you'd hand it to that one which would identify it. Then you hand to this one which comes up with the solution. And then you hand to this one which is actually the doer. So it's an RCA, a planner, and a doer. And then this one's actually for feature dev. I'm still cranking on on that idea, but I think it's a really cool idea. So that would be its own kind of like standalone thing. Um, and then yeah, the branding around this stuff is pretty cool. I'll be releasing it. If you're not familiar, this is called pool suite. But then each one of the modules will look kind of like a video game. And uh I forgot what the domain is called. Should know that. I think it's let me just see. Oh, getiv.com. Cool. Yeah, there it is. And then it's hosted on school. It won't be hosted on school forever, but that's where it is right now. So, what else do we have? Uh, builds. So, builds. Wanted to talk about this a little bit because I think this would be really easy for me to make. I have a V1 version of this. It's just literally NAN, but it's an RSS feed that has a bundle on it. If you're not familiar, you can go to RSS.app and then you can bundle together the keywords that you want. So, you can make an RSS out of anything. So, I can make it off of someone's accounts. So, I could beat this up where it's like every morning I want to get stuff that's relevant to me based on the stuff that I look for. The easiest way to do that is literally just list out the things that you find interesting in the morning. It's most of the time if you're into AI, it's like the names of popular people, the names of models, the names of companies, and then you have a bunch of X accounts. So, it might be an array of that. It might be a collection of these links. Um, and then you can also do email bundles, but you take the bundles and then the next step is actually even more important because otherwise like you could pull that every 15 minutes, but it's going to just give you a bunch of garbage. So then what I'm thinking you do is after the for this morning text thing, it would go something like this. So RSS to classifier. When I say classifier, this is an open AI call. And in the open AI call, you have a couple things. So, it's going to end with a route and it'll be dependent on what this thing classified. And it's really dumb the way you can do it, but it works where it's like, you know, something along the lines of uh let's see, you are an expert staff editor responsible for sifting through a collection of article headlines from popular news outlets about artificial intelligence. Your role is to find the best three articles that will drive the most amount of value for the reader. Return only three at a time. Cool. So, forget the routing thing. I realize we won't need that. That's actually for doing um like growth agents for reading through like live comments because then you can classify it. You would use a classifier for that same thing. Or if I was going and I was scraping comments on other people's YouTube channels to um basically get notoriety and growth and more impressions on whatever profile it is, then you could classify and say, "Hey, this is a comment. Hey, this is a question. Hey, this is a hot take." And then each one would lead to its own thing with a human in the but for this case it would just return those. And so after that, you get those and then it's going to classify them and then you can do an HITL. Um, what I forgot here is do not include any articles that cover celebrities or anything to do with X, Y, or Z. Always include articles that cover Sam Wolman, you know, whatever. Sam, Elon, XYZ. So, you put those weights in there. Weights is a big word for it. You put those words in there and then it does a really good job. So, then you have the human in the loop. This would literally just be a G sheet. You could do super base as well, but I would find that to be more of a setup. This is the one I built so far. Um, and then after that, you see, if I wanted this to come to me, I'd have all this information coming in, and I think I made this once in make as well. Let's just check. This is a good example. A lot of coders are going to look at this and be like, why are you using make? And I'm saying cuz it takes two seconds. I literally didn't have to open an ID. I save the coding for the important stuff. This is like playing with Legos. So if I come in here and then moved to NA self-host. So there's not much. Let's look at I think it was in content pipeline. This one's for Instagram, but it's similar. So this polls articles, this would be for a theme page, which is what we just uh covered earlier. I have a friend that literally I hired him to build our social page for the music startup and he did a fantastic job and then he made a an Instagram page that literally he just posts news about different things that are getting built around uh Florida. Calls it Florida tomorrow. He's about to hit 100,000 people on it and he's traveling in Tokyo and Hawaii and he's like doing really well with it. And the format is I I took them golfing or we went golfing together. And uh he explained he's like, "Yeah, I just check bookmarks." And then I just take them and I put them into a new design. Then if the image looks bad, I just upscale it. Use Adobe. If the image is missing part, I do a magic expand. I was like, "Oh my gosh, I should do that. I should automate it." So I did. And so what it does is it pulls the articles, classifies them. So let's look at the prompt in here. Yes, you can do this in Python. Go do it. Have fun. You're the chief editor and an expert staff editor. Uh you're the chief editor responsible for selecting which articles get published in a popular news publication. You will receive a list of 15 articles and will analyze and select the top three based on the impact on the global economy, impact on the job market, and the popularity. Articles to avoid anything related to celebrities. Um yeah, your job is to reduce the top three, maintain the shape of the data that you received, return at least in full JSON format. Cool. That gets stored. We iterate through it. We fire crawl them because when you go from an RSS to a spreadsheet, you don't get that much info. So you got to go and crawl it. And firecrawl makes it LLM friendly. Then we iterate through that. We summarize the article. So, I have a bunch of examples that I pulled from, in this case, Instagram pages. You just literally like rip the captions and then dump them in there. Then store the summary. And then there should be another one in here for the title. I think maybe I did both in article summarizer, but you need caption text, right, that goes over the top. Um, which I guess that's image text generator, right? Yeah. So, I stopped this because I needed to go and then set up a GPU on RunPod, but RunPod runs in a container. So, I'm not doing that. I'm going to use uh either one of my friends self-hosted GPUs. We're making these like kind of cool looking images. This is the idea was it be like something like this. I ended with this output, but I think it's like this is not the vibe, but it got close. So, you'd end with something like this, right? And then that could post, let's see, if I got three every 15 minutes, so we'd have what? 12 in an hour, 24 have 288 posts potentially a day across 365 days. You have 105,000 of these things. Like surely some of them will be good. It's just kind of wild that that's a thing because then you could just say like, okay, don't post them. Just make the 105,000 of them. Then you have your training data. if they're good. So, yeah, I think you'd you'd probably loop back. You'd want to work you wouldn't train on those. You'd probably want to work on them to make them better, then go and train on them because Yeah, that that's a dumb comment. But, let's see. I think that's all I've got for today. Yeah, it's called Vibe with AI. And I was playing around with these funny designs. I could show you how I made that. that might be helpful for you for people that are trying to make stuff. I basically wanted to make it look cool. And so I was like, let me take a floppy disc and use GBT40 because I started to design all the stuff for the modules of teaching stuff in here. And I'm like, what am I doing? Like I should definitely just use GPT40 and do that. This one's good. Teleport to the mid30s with AI and this dog. But I didn't make any of these assets. What I did was I just jumped into 4L. I said, "Yo, make some stuff for me, please." Um, let's see. I have such funny names in here. I try to make me with a with a sombrero. Yeah. So, I took this image from this popular thing called Pool Suite. And then I said, "Here's me and sombrero. Make it." And like it's kind of kind of close. Um, what I also realized is you can reverse engineer this to JSON. So, it's like I wanted this, then I wanted CD, and then I wanted the CD transfluent. Then I wanted Vibe Coding instead of Vibe Marketing. That worked pretty good. I tried a real estate ad that didn't work very well, but I don't know how, but the people are getting like the Jason out of this. I would be curious. Let's see if we can't find it because you can prompt with the Jason and then you get exactly what you want. So, let's see. Maybe it's in my likes. This is a high-risisk moment. You're going to see all the stupid stuff that I like. Something about health. This calendar component. Oh my gosh. If this existed when I was making map, this would have saved me so much time. I spent months of my life trying to get Google calendar. Mine's better than his, I think. But that is so nice for everyone that that exists. The client side code on that is it's a beast. It makes me understand why they take forever to make updates. Apparently, it's not in here. Uh Jason GPT. Yeah. No, we are. That's last year. Not sure how, but people have figured out how to GPT. Uh, let's do 40 image gen. The problem that they were solving for was the image was nice, but the text was not. So, they figured out how to get the text to land in the proper spot by getting the JSON output somehow and then editing it. And I just am not maybe we could take maybe we can take this let's see find reference image opt into second tweet. Maybe we can take this and then say that we want this as JSON. Um, okay. So, let's go. Let's try it. Um, I think I need to do create an image just so I can have it make it again because it wouldn't have JSON if it's just looking at the raw image, but it'll get JSON if it creates it, I believe. So, let's let's test that there. Um, remove the text. Make it say Parker's skinare brand. Make the product teal. Apple juice or Red Bull? Who knows? Okay, let's see what's going on under the hood. We're not going to be able to see anything back. It's fun. It would be really cool if somehow set downloads CloudFlare. Um Oh, almost showed my token. That would have been bad. What's going on? I'm assuming that we can do this, but I could be wrong. It's working. The 3D spatial stuff in this is fantastic. That's one thing that I've noticed. This, if you can imagine the other models, they don't know depth nearly as well as this. So like this lamp that's behind me this way. That's what, like five yards behind me. And so if I stated to this thing that I wanted to put me in a hexagonal grid where the back wall of said grid is 200 m behind me, it's going to like take that and do it, which is just nuts. And then if you like analyze the photo, it looks pretty pretty dang close. So point is we're trying to get this to do an output so that we can see if we can ask it to give us a JSON. Oh, if that is the case, then it's great because we could take it and make really awesome prompts with it. And ultimately that'll be how all these media brands get their stuff done. Like they'll use something like this that has the guard rails with like a gooey in front of it. Um that's what I think is going to happen. And then everyone that is a graphic designer will just get more leverage because if they're really good at this, they adopt these skills. Like this is the beginning of that whole profession changing. Like we saw with coding, but now it's in the graphics. So it's like this is us getting GPG35 Turbo. That's like pretty pretty similar thing. Um so it says it's done, but it's not. Can you return the JSON output for how you created this image so that we can edit with better accuracy? Thanks so much. See, not even asked him. Say return the JSON output used to create the previous image so that we can edit it with better accuracy. Let's see. Moment of truth. You don't have access to What do you mean? Didn't even load. Come on. Um, let's try again. Give me a JSON output that matches the format of the previous image so that we can edit the text. We better control. Think we found something. I think we found something. Okay, we did a good job. We did a really good job. So, yeah, we see this kind of crazy. Let's see. So, wow. Okay, let's take this. Just flung my microphone. Let's take this output and let's put it back in here. And let's say actually you are a ecom expert responsible for coming up with gorgeous looking skincare products as well as fantastic names for creating products that sell off the shelf. You'll be given a JSON output that includes the image description and a background the product itself with lots of different parameters. Return a fully updated product that we can put into a purchase order ASAP. Make sure it's good. This is cool. See what it does. We're going to really crank up the volume here with 4.5. It's supposed to be like the really creative one. So, let's see. We're making our own e-commerce product right now. Here's a polish shelf ready skincare product. The description optimized to captivate consumers. Okay. What do we have? Gentle shadows. Green tea glow essence with elegant teal naturally off-white. This could be cool. Um, still still running the motor. That is interesting. Their new update. It looks like they made the docks look a little bit different. I don't like that I can't see the whole thing unless I Oh, interesting. It's cuz it's now in a canvas. Okay. I didn't want that, but whatever. Now, let's feed it back. No, no, no, no, no. That's not what we wanted. We wanted Oh crap. We want this. Where? Okay, it's this. So, let's click Let's go back to freshy. Create an image. Boom. Let's see. Let's see. Hey. Hey. Okay, cool. So, it started. We've figured it out. This is a product, ladies and gentlemen. Someone's going to do this. Why not you? I'm not going to do this. I don't care enough to do image gen. Got my hands full. I got to build map. I got to build a community. I got to do all these things. But this could be you. This is uh pretty cool. I guess this would be a tweener thing because like you'll get the API access soon. But this is probably what the API access will look like. Something like this. I don't know. I look at it basically just have what I don't understand is how it deals with space doesn't have any sort of position hello product name. So let's look at this and think about it for a second. So if I open uh let's do touch test.json JSON d test.json. Let's Did I not get an error for the first time? Oh my gosh. No. There we go. Okay. So, we look at this image. Wow. It's pretty cool. So, it worked. It worked. Yeah. So, how would it work? It looks like it starts with the product. It knows what the product is. Apparently, it gives it this color, which is the elegant teal. Has somehow light just nose light. So, none of that's in this. Jason naturally off-white cap. I can't even begin to imagine how they do this because the product then has it just figures out what the Jason needs to be off the dome. We're already at an hour and 3 minutes. I was not supposed to go this long, but I'm so interested. I need to do one one more just just to test. Let's There's this cover thing. No, it's a bad I just don't know how this works. I want to learn more. That's all. I got to cut the video. Hey everybody. Awesome. Thank you so much for watching. Make sure you like the video and then I'll make more of them. I'm still going to make them, but it's cool when you do that. And then uh what else? Yeah, check out Vibe with AI the school. It's going to go live tomorrow, but like quietly. And um people that join over the weekend will get a discount, which is cool. And besides that, I'll see you tomorrow. Thanks for watching. Oh, and comment any questions you have. Make sure you ask questions and I'll answer them. See you.