Matthew: Welcome to the Restock Podcast by No Code Supply Co., where we highlight the people, the things they build, useful tools, visual inspiration, and just conversations to refill our shelves and ourselves. I'm Matthew. Corey Moen: What's up, y'all? I'm Corey. And today we have Derek Costelli, who is a friend Derek Castelli: Let's go. Corey Moen: of mine. We live in the beautiful Des Moines, Iowa, together. And Matthew: Yeah. Corey Moen: he is just a builder and all around great dude. so stoked to have you. Thank you so much for coming, Derek. Matthew: Welcome, Derek. Corey Moen: yeah, tell us about yourself. Derek Castelli: Thank you, thank you. I'm a Christian. I'm a husband. I am a dad to four animals in our little townhome. So it's we call it the Castelli Zoo. but they're all great. I I've always been a designer. I finally took the plunge in 2020, like many people took the plunge, and I became self-employed, became a full-time freelancer. I was basically just a webflow, more or less, webflow freelancer. So I actually discovered that tool because of Corey. I think a lot of people discover that tool because of Corey. But surprisingly, I discovered the tool from Dribbble. And that's how I met Corey, is actually a Dribbble post. and then we had coffee there, and that's basically the story of how Corey and I met, without me realizing Matthew: Yeah. Derek Castelli: it. But yeah, then freelancing there. And then most recently, because of AI, I yes, I s I'm I'm a designer, but I've always seen myself as a builder, a tinker. And because of AI is to this point, it has been to this point for a couple of years now, I can actually take the designs that are in my head and make them something real and I can play around with them and then I can share them with the world. that was actually like to tie it off in a lovely little bow here. Though I actually want to become an architect. and the whole reason for that Matthew: Same yeah. Derek Castelli: is is because I loved the impact that so my dad actually designed our home growing up. I saw it in AutoCAD. I got to live in that house. And it was just a huge unlock of like. I got to experience something that I only saw on a screen. And that's why I wanted to become an architect. But then I quickly realized you can't, there's not a lot of impact that you can have as an architect because what are the chances you get to work on like one of those projects, right? That really wows the world or really impacts the world in a good positive way. And so Matthew: Mm-hmm. Derek Castelli: then I dabbled in digital design and I was like, this is the unlock. I can, if anyone had that internet connection, they can experience something. And I think what matters most to me and what gets me like a big dopamine hit still today and why I create what I create is I don't know the most of these people that are using my tools. And that is special. I love the fact that like I can make an impact on a stranger's life. And the stranger, the stranger may reach out to me. Or the stranger may not even know who I am. And maybe, you know, in a beautiful world, like I'll meet them one day and they'll be like, wait, you made this? I'll be like, Yeah, and that's gonna be a cool special moment. And that's why I do what I do. Matthew: Wow, very cool. Corey Moen: Love it. Matthew: I it it's a little freaky how similar my story is to yours. Derek Castelli: Yeah. Matthew: Yeah. my mom designed our house and and then we me and my dad built it by hand. This was when I was like 13, so like in high school, going through those those ages. I wanted to go into Derek Castelli: Yeah. Matthew: architecture and I had very similar observation was that to be like the lead architect on some of these like Keystone projects like to get there could take two, three decades if you're lucky to even get into the right position at the right time. so that you know, and for me it was always I loved architecture because of kind of the expression, but also the design space and how people and experiences happen within a space and really kind of designing architecting that. and so that those kind of passions I think naturally lead into just digital design work. yeah. And also similar for me that AI has now unlocked. Like there was a point, I guess 2014, I was kind of at a fork where it's like I could go down doing webs websites or it could do build apps. And it Derek Castelli: Mm. Matthew: just ended up I naturally went towards websites just because that's kind of where I had work. you know, but I always kind of in dev work I always kind of hit a wall, you know, where it's like I kind of reach my limits, but now it's like those limits have been lifted, right? And so it's like now I can take the design skills that I've had over the years. And then with kind of the passion that I have, you know, for for building apps and services, you know, for people, you know, actually have the power to build what I imagine now, which is It's pretty cool unlocked. Derek Castelli: Very cool. Matthew: let's let's get into our first screen share here. and I'm gonna apologize in advance for doing this to y'all, but you are going Corey Moen: Ha ha. Matthew: to see this because I had to. Derek Castelli: What Matthew: Okay. This is from a company Corey Moen: a locomotive. Matthew: called Locomotive. and this is called Lisa. And so as I move my cursor around, so this is a Derek Castelli: that's creepy. That's so creepy. Matthew: humanoid. Female figure with a monitor for a head and she follows your cursor on the screen. we don't want that for cookies. Okay, so I'm gonna click to start her up and you should be able to hear her as well. hi, I'm Lisa, your trusty and slightly opinionated assistant at locomotive. I knew you'd come back. I have that effect on people. So what's the ask? What do we want to ask? Derek Castelli: Drop a quick word. Matthew: Drop a quick word, okay. stop it. You're making me emotional. Team works hard Derek Castelli: dear. Matthew: and kind word, that's fuel for their creative fire. Hit me with the good vibes. Type it out right here. Derek Castelli: It sounds like Corey hit me with the good vibes. Matthew: Good vibes. What what do we want to say to Corey Moen: Scrolljacking is evil. Backstory Derek Castelli: Yeah, all the good vibes. Nothing but Corey Moen: if you didn't know, I'm pretty sure this is the agency that made locomotive scroll JS. Matthew: It is. Yeah, this is an agency locomotive. Corey Moen: Yeah, yeah. They're a great agency, to be honest. They do incredible work. Matthew: But I I find this unsettling. I mean it may be an interesting experiment that the team went through to build this, but Corey Moen: Yeah. Yeah. Matthew: I'm just like nope. stop it. You're making me emo. I knew you hi. it goes all the way back. Okay. You have questions. Corey Moen: Interesting. Matthew: I have well mostly answers. So this also reminds me of Edgar Allen, how they kind of turned their website into an Corey Moen: yeah. Matthew: AI chat as well. Just this has an actual 3D. persona to go along with it and voice. So I mean it's doing all the voice and stuff on the fly but Any any any any thoughts? Or shall we move on? Is this freaking Corey Moen: I I think Derek Castelli: it just take it takes me back to like all the like sci-fi movies. And like Matthew: Yeah. Yeah. Derek Castelli: either they, you know, would go full humanoid with the robot, right? And Matthew: Mm-hmm. Derek Castelli: so they would it would actually look like a human or they would be specific in that the robot did not look like a human. And so it's interesting that this is like attempting it Matthew: Hybrid. Derek Castelli: to repri a hybrid, yeah. Where it's almost like it's taking the most human aspect of a human, which is the head, and it's replacing it with a more digital experience too. Like almost like our screens are our eyes, right? And Matthew: Yeah. Derek Castelli: or how we express ourselves is all through a screen. So like that's very abstract and very deep, but that's my first take on it. Matthew: Yeah, it's Corey Moen: Yeah, I I think this is just like in this realm, like even you mentioned Edgar Allen's site, which man, they did that I feel like it feels like an eternity ago now. Like Matthew: Yeah, a couple of years now. Corey Moen: Yeah, so they're ahead of the curve. It's just like I I think this has been on my brain a lot lately in general of like, you know, the personal web and all that. We kinda got into this last last time a bit, Matthew. It was just like I Matthew: Mm-hmm. Corey Moen: don't know, there's a lot of different people. Yeah. Derek Castelli: The dynamic dynamic websites. Corey Moen: Yeah, and like not necessarily it, you know, sometimes maybe it's prompted like this or you know, the user opts in, but like I just I do wonder, you know, if we'll start seeing maybe not quite literally like this visually, Matthew: Yeah. Corey Moen: but like just generally experiences that are trying to ride the wave of people asking you know, AI models questi more and more questions and interacting and learning about the world and that that way. So like how do we bring those same adjacent type of experiences, but though they don't have to be those literal experiences to websites. I don't know. It's Derek Castelli: Mm-hmm. Corey Moen: interesting. Yeah. Derek Castelli: Yeah, it's like it's kind of twofold that I've been thinking about since you guys talked about dynamic websites. I've been thinking about that a lot. Of like there's two different approaches to that almost where you have something like this where it feels more like an onboarding. And through the onboarding, it's like trying to personalize your experience or it's trying to offer some personalization to it. But then there's a second thing where you're seeing a lot of these like growth engines where, you know. Hey, I visited this site for the first time. Now when I come back for the second time, you know a little bit more about me of like what I did, my previous activity. So how does the website change now? Matthew: Mm-hmm. Corey Moen: Totally. Totally. Matthew: And how Derek Castelli: You know, Matthew: how do you Derek Castelli: which which which which could make these kind of like shares a little bit difficult because like, well that's not what I see. This is what I see. Matthew: Yeah. A as a visitor, how how would that make you feel if between your multiple visits things change, but then also as you're trying to share and talk with someone else, it what they receive might be different than you what you used to receive. Derek Castelli: Yeah, I think, you know, you guys touched on this last time where it's like, you know, it almost does the brand a disservice because it's like this record of like, hey, how do we want to feel established? It should feel kind of like the same set standard for everyone. But that's also why we have like, you know, you create use case pages, you create audience pages. So it's like this is more geared towards me. And so it's I feel like more like the dynamic component is more of like the home page. And I think where I would feel, hey, I would think that's really cool. I would be like, this is neat that like yours is different than mine. And I would try to figure out like, why is ours different? Like, what did I do to make mine different? What button did I click? Where did I scroll? Where did I not scroll? But also at the same time, like, okay, we have to always accept. cookies or not accept cookies. Like surely this is just like a deeper level banner that's gonna like have to be approached of like to continue using this website, do you agree to collect all your information? Like almost like a like an age checker to view this website. Corey Moen: Yeah. I mean, it's just like like a way to think about it is even just broader adoption of this. It's like we already have OS detection of light dark mode is a simple example in browsers, right? Like would we eventually ironically see the contrast of like the last few years, browsers, I feel like, have actually indexed the other way. Everybody, or not everybody, a lot of people are drawn to privacy first browsers where it intentionally eliminates everything. But it's So ironic the flip side of social media and all you know, all the social media products we are often willing to just sign up for. Matthew: But we've accepted that's the price of a social media platform, right? That's the price you pay in Corey Moen: Totally, totally, totally. Yep. Matthew: order to get whatever you're getting from the the service. The price you pay is what you put into the platform or the information that you provide, which may them Corey Moen: Totally. Matthew: what you provide may just be by scrolling and clicking, but it Corey Moen: Right. Derek Castelli: Also, with that, the algorithm Matthew: could be deeper. Derek Castelli: is the product, is what you show, right? It's like I'm showing my feed of whatever they think is relevant to me versus like Matthew: A website you don't know what you're gonna get before you go there, right? Corey Moen: Right, right. Derek Castelli: Right. Corey Moen: Yep. But at all the same. Derek Castelli: Plus two, it's like all websites, to be honest, all websites are just it's a marketing website. And so it's like that's where it gets a little I to me, it's like way different of like why you would have like why we see there's a difference between social media and a website is because it's like, well, the marketing is there's this whole team that's trying to get your money. They're trying, Corey Moen: Yeah, yeah. Derek Castelli: right? They're trying to get you to click. They're trying to get you s to some goal that they have. Corey Moen: Mm-hmm. Derek Castelli: And so it can feel maybe a little bit manipulative. I don't know. But I I still think it would be really right. Matthew: Or leading at the very least. They're trying to lead you Corey Moen: Totally. Matthew: to an outcome, yeah. Corey Moen: I think it's a it's a nuanced thing too. It's quick, it's it's it's definitely easy to go straight to like, wow, this feels invasive and all that. But like what are the other places that we have those equ equivalent experiences? Like you go to a gas station or a store, you know, like you go to a gas station, it the if you want a soda, pop, whatever i anybody wants to call it, we're Midwesterners, so pop. Matthew: Cola. Yeah. Derek Castelli: There's only one real answer to that. Corey Moen: that you go to the fountain and there's not just One, you know, or two, like, hey, do you want light or dark mode? Do you want Coke or Sprite? you know, like, no, you have all of these options to pick from, right? Like, and and you could see that from the angle of like, well, but that's the product. It's like, well, what are we what are we portraying on a site? Like, again, to your point, it is some kind of product or service you're trying to convert on. And beyond somebody just picking which one they want, like, could there just be other types of things that can be done, you know, dynamic from one person to the next, not necessarily to the full extreme of social media. yeah. Anyway. Derek Castelli: Okay. Yeah, yeah. Let's take this pop thing a little bit further. So, like those Coke machines, right? Where you literally can make whatever you want. Imagine, Corey Moen: yeah. Love that. Matthew: Touch screen yet. Derek Castelli: imagine it's got facial recognition and it realizes you're a repeat customer to that location and it knows you always get Coke with a little dash Corey Moen: Reset it for you. Matthew: But here's the problem. Derek Castelli: of vanilla. Matthew: What if I'm getting a drink for somebody else and I'm Derek Castelli: Right. Yeah. I know that's the problem with exactly. Matthew: at the machine and it's trying to force me to do something that I'm like, no, that's not what I need. Derek Castelli: That's why that's why I wanted to bring that up. It's like there is a point where this like dynamic, it's like it's assumed behavior. Corey Moen: Mm-hmm. Derek Castelli: And I'll and who knows, maybe you're coming there with a different goal in mind. Right. So let's say for instance, you know, I go to a a a new product's website and they've like I've scrolled in whatnot, and then they change it the second time, but maybe I wasn't done researching. And I'm like, wait, why is where is that thing that I was trying to find? Then it becomes a different problem. Matthew: Right. You hit the back button on your browser and you expect to go back to what you were just looking at, but now it's regenerating Derek Castelli: Yeah. Matthew: something different. And you're like, wait, no, I didn't I need to see what I just had. Derek Castelli: Yeah. It's almost like the site can't change too quickly too fast. Like it's gotta like settle a little bit. It's just an interesting, definitely interesting. Matthew: well, this reminded me Alessia at Coden Wander shared this with me, which they built. All right. So here we need to click ask and what what do you want to ask of Codenwander? What what do you want to know about? Corey Moen: How can you help me grow my sight? This building websites for growth. Matthew: So you can see here now the content went away. We still have a photo down there and it's and it's thinking. and so this is actually going to I'm not quite sure if it's generate all the content on the fly or just the sections. So yeah, we can turn your site into a growth engine. What holds Corey Moen: Yeah. yes. So cool. Matthew: a site back? And it's still it's still creating content here. Okay. Levers that they can pull. So, you know, this is their services Derek Castelli: Wild. Matthew: section, right? But it's kind of this this is probably a little bit more generic Derek Castelli: Yeah. Matthew: content, whereas this is a little feels a little bit more custom to the question up top. Corey Moen: And I wonder Derek Castelli: It's almost like a different approach to just search, right? It's like instead of trying to make sure, like instead of indexing all the pages that we have, what if we index basically all the content and all the components we have and then we'll deliver it in a way that feels correct for this search query? That's fascinating. That yeah. Corey Moen: Exactly. Yeah, yeah. Yeah. And I Matthew: Yeah. So yeah, they show some work and then they'd actually inserted this, you know, contact form. So that's pretty cool. Corey Moen: What what happen what happens when you do a unrelated question? What's the sound a dog makes? Matthew: Yeah. Derek Castelli: Yeah. What framework would you recommend? yeah, there we go. Corey Moen: Yeah, there you go. 'Cause this is the one the thing I was gonna say also that I think nice, let's go. So it's it is doing some kind of, you know, actual analogy. Derek Castelli: that's cool though. It's still like it knows act it's not like a default like empty state. It's very dynamic. The closer we get to football is this brand we rebuilt in Webflow. Corey Moen: Interesting. That's super cool. Matthew: Which that it's un unfortunate didn't show that. But yeah, at least yeah, Corey Moen: Right, we're right. Derek Castelli: Right. Matthew: the response is actually proportionate to. Corey Moen: Yeah. And again, I think the the thing though is this is okay. You know what I mean? Like sometimes people are Derek Castelli: Mm-hmm. Corey Moen: like again, like, we can't do that because like people will bash it and like, you know, they'll ask all these unrelated. It's like, no, just like have those fallbacks because the people that do ask something legit are probably the ones you're actually most trying to convert, you know, and Matthew: Mm-hmm. Corey Moen: could be most impressed by this. So like I think it's super rad. Matthew: I mean that's a very Derek Castelli: Hey, I respect the fact that Matthew: something that comes up. Yeah, yeah. Derek Castelli: they actually thought through like, hey, we need to have what it happens if someone just treats this as a Google search Corey Moen: Totally. Derek Castelli: bar. Corey Moen: Yeah, yeah. And some I remember Timothy Ricks doing a video a while back on like even Cloudflare has like really affordable AI models that you can run that are like fractions of a fractions of a fractions of a cent. you know, so they might not be coding for you or solving problems, but they're great for this kind of thing. And, you know, especially if you have low risk of your site, you know. going viral or something where a bunch of people are hitting it all at once, then like, you know, like the cause this I would say based on even this answer does seem like it is going through some kind of model to give like a somewhat relevant answer. Matthew: Which is different. The other day I asked about AEO and it gave me a full page. And that time it just gave an answer. here we go. So here's like a full page. Corey Moen: interesting. Interesting. Yeah, yeah. Matthew: I do kind of wonder some about the technical bits of this. It's like, you know, are they just is everything in global CSS and they're just loading the content on the fly? You know, or are Corey Moen: Well Matthew: they trying to load, you know, CSS and JS on the fly within these components? Corey Moen: Right. Scroll scroll down this page because like I already see the same general sections. Yeah, yeah. So like the sections look like they're all set. It's just filling in the holes. You know what I mean? well, that Matthew: Mm. Corey Moen: one's a little different. Yeah. Derek Castelli: It's almost like I wonder if they have like a visibility clause or conditional visibility based on you know Corey Moen: Totally. Derek Castelli: certain queries. The thing that terrifies me with this stuff though is like, this is what I found out even like in making apps, is all the edge cases. So, like what's Corey Moen: yeah. Derek Castelli: beautiful about a website is you're like, okay, this is static. This is how people will Matthew: Mm-hmm. Derek Castelli: see it. This is exactly how it looks in my preview. This is how it's Matthew: Yeah. Derek Castelli: gonna look when someone opens it up in another browser. Now, the only difference is the, of course, the browser compatibility, right? Gotta go through all that hot mess with WebKit. but this kind of stuff like terrifies me because I'm like, okay, what actually does get generated? Corey Moen: Hm. Matthew: That's out of your control. What so what can you control? Derek Castelli: What does that look like? Right. Matthew: Yeah. That's where going back to last week we were talking about a brand corpus. I'm gonna start trying to top this word, but Corey Moen: Yeah. Matthew: it's like really as designers Derek Castelli: Or best. Corey Moen: Okay. Matthew: more of our job and and developers is defining the rules, you know, the skills, the things that are actually used to generate these pieces so that we can have a high confidence in the output, Corey Moen: Mm. Matthew: you know, because we are no longer able to just purely define and work on the output. We have to work on the upstream pieces that generate that. Corey Moen: Mm. Matthew: Okay. Corey Moen: Love it. Derek Castelli: I love that. Matthew: that was that was a nice little rabbit hole. I'm gonna take us back to something that I don't like. Have you seen this website for pair dot in o? Corey Moen: I think sorry, go ahead. Derek Castelli: I feel Matthew: Okay. Derek Castelli: like I have. I feel like I Matthew: Okay. Derek Castelli: have. Matthew: I I after I saved Corey Moen: This definitely went around here. Matthew: it, I saw it doing some rounds. at least on LinkedIn and Blue Sky where where I still am. and for one, I I see this trend of AI companies using Renaissance artwork or or from the like age of enlightenment and the Renaissance, you know, to try and like sell things and it just it irks me for what I mean the style of it I'm like yeah it's fine I ha I have some some beefs with like these transitions like they're just can tell when they're transitioning from one generated piece to another just I'll get to some of the big offenders here. and I just and look at this pair it now has two stems cool I don't know there's just like some obvious things with the way The generated pieces blend. Like here's that that this one's a huge offender for me. It's like we're going from this frame right here, and look at this. It's just like frustrates me to death. Just it it Corey Moen: Six. Derek Castelli: Time is a little off. Timing's a little off. Matthew: timing, the framing, and it's like even just in you know, CSS positioning, like you could get these things a lot closer. And so it just it's like it feels like it was all done just by the model. And then this the style of this one just goes way different. Like it's like kind of this antique blueprint style, which is completely different. Corey Moen: Interesting. Matthew: Anyway, yeah, so I know a lot of people were like praising this site. Yeah, it's an interesting little carousel. I mean the the lines, these anchor lines feel a little random. I can't quite understand the logic of it, but it adds a little interest as Corey Moen: Yeah, yeah. It's Matthew: it goes around. Derek Castelli: Anytime I see sites like this, I'm like, I I personally I'm like, I don't even know how you would start this. Like my brain just can't sometimes comprehend what's all has to happen in order to achieve this. Like I almost feel like I have to take a little bit of acid in order to like get a start on this. Right. Matthew: Mm-hmm. Corey Moen: my what? What? Matthew: This is maybe my favorite part. This the form. So yeah. This was like, okay, that that's kind that's interesting. Derek Castelli: That's cool. But like I almost feel like I I seen Corey Moen: Why does it Derek Castelli: somewhere too where a lot of these things with like AI is like how much story storyboarding matters now. And Corey Moen: Mm-hmm. Derek Castelli: like how do you want each of these scenes? What do those scenes look like? How do we transition from this scene to that scene? And so like I respect Matthew: But like that, that when you're transitioning through the menu here, it just like forces you to scroll through the page, which is like so so jank. Derek Castelli: Scrolljacking. Corey Moen: Dude. Somebody I saw I saw Derek Castelli: but Matthew: Sorry if I triggered anyone. Corey Moen: somebody post something or may I can't remember who were if it was a post or somebody said something, but somebody made the comment the other day of like, you know, websites are meant to do more than win awards. was the general sentiment. And it's just like, you know, I think let's be real, there is part of what we're exactly doing right now that did achieve a goal for these this whoever, you know, the team or company that built this is it got attention. It got, you know, went a little viral. Derek Castelli: Got eyes. Matthew: Mm-hmm. We're talking. Corey Moen: And I think it's just like how to draw that line. You know what I mean? Like I remember a couple, maybe a year ago or more, or maybe two years ago. I'm drawing a blank on the product, but there was like a SaaS, maybe it was runway. It wasn't Derek Castelli: Runway. Runway. Corey Moen: with like the airplane thing was one. I don't know. The the point there was like a couple like, you know, SaaS startups that were still trying to draw this balance of like conversion, but like attention through motion and all that. Matthew: Mm-hmm. Corey Moen: And like this wave of people being like, you know, marketing sites don't have to be boring and blah, blah, blah. But then all the way on the flip side of like all of the data out there, usually maybe I'm biased or or or or stereotyping, but like all the typical data that supports, like if you actually did an A-B test between this and like a more basic site that has things a bit more condensed and focused on conversion, like you're gonna Matthew: Mm-hmm. Corey Moen: convert more on the basic one nine times out of 10. but again, maybe that will change, especially in the world, you know, moving more and more to social or other ways that people discover your brand versus just like googling it, like maybe was more common in the in the past, you know? Matthew: Yeah, I mean we've Yeah. Derek Castelli: Yeah, it's almost like there's there's like two different approaches, right? To I feel like that we're clearly seeing is like you have an approach like this where I'll be honest, I what what is pair? What does pair do for me? Matthew: I still don't know. Corey Moen: Yeah. Yeah, yeah. Derek Castelli: Right? But like so there's that where the obvious, the the boring marketing website is gonna do a hopefully do a much better job of being much more straightforward. But the other other end is Are as many people going to talk about it. And so then you're like, like what you say, Corey, like you're balancing this between like eyes and attention and people talking about it versus like getting and winning the goal, winning the customer. or you know, because I could also see it right where maybe you create a really like novel, really high experience website like this, and it's kind of like for the masses, for kind of anyone to try. And so it's it can appeal appeal to like a general audience on that cool factor. But if it's like a specific audience, I think this kind of can do you a disservice. because who knows if they're actually cap being able to capture all of that. Maybe they're just checking out the cool like effects. Right. So Matthew: Yeah, I mean what do we measure? Derek Castelli: But hey, gotta talk about the Roman Empire. Gotta talk about Romans. Let's go. We're building a pair, everybody. Matthew: Romans and giant gold pears. Derek Castelli: Hey, I love the paper effects. That's cool. Like it's almost like a watercolor kind of thing. Matthew: Yeah, lot of lot of different three JS effects happening on this one Corey Moen: Lot of effects. Matthew: page. Like like even this dithering transition, I'm like, it doesn't fit the aesthetic of like the artwork and stuff. There's this level of polish Derek Castelli: So there is like there's like intentionality, right? Matthew: that's missing. Yeah. Derek Castelli: It's like with especially with AI, what I'm realizing is intentionality is like so something that needs to be relearned with AI because you can you can make anything. You can be like, hey, I saw these five cool things. Let's actually, hey AI, agent of mine, use all five things and make them work. And it's like, it's just not gonna work because you're not coming from the standpoint of like, These are what I feel like can all work together, which is also why I mean I'm a little biased because I'm a designer. But I think a designer is still an important role because they can try to see these five things that let's say a client likes and say, you know what, that's not gonna work on this, and here's why. But at the same time, like, well, who are we as designers to say also like, this is this is this isn't gonna look good until we actually see it? Like that's another cool thing I'm seeing with AI is like, let's just throw all this stuff together and maybe there's actually something there that we just totally would have not thought through. Matthew: this is where that site ended up. It's like they had some ideas, some ideas, and they threw it and like, okay, great. And just took it instead of saying, like, okay, what's working here? What needs to be adjusted? And I think going back to like mood boarding and inspiration, as a designer, you're able to say, like, okay, we've got a bunch of random things on this board. What is the thing that connects them? You know, what is the relationship between these these two pieces? Is it kind of the the tone and the contrast? Is it the the color palette? Is it the composition or the framing? You know, and you're really able to identify those key pieces that that you can turn into those rules or those guidelines, which that you then carry into whatever the new thing is you're going to create. So going back to like that's where I think a lot of the work needs to to focus and transition is defining those rules and those guides. So You can actually give those to the AI, even if it's generating the assets. If you're defining like here's the relationship between these things, then it knows. You know, if you have those five things, like you were saying, instead of just throwing all five things at the AI and say, Hey, I need you to create, you know, a new an image and use these as inspiration or whatever, you actually define be like, okay, here's what's working or not working in each of them. Here's like how I see them connected, you know. through through different kind of principles of design and then and then I think you're gonna get a lot better output and and more consistent, hopefully, across multiple generations and even over time. Derek Castelli: Think the the flip side of that, and what I realized even with like the tools that I've been making with AI is sometimes I just don't know. Matthew: Mm-hmm. Derek Castelli: And so I'm like, I want to just see. So I like I make the prompt as brief as possible. I'll sneak in some things of like, I would just want you to make it intentional. Like I want you to be opinionated. I want you to stick with this, like. But I don't, I'm not gonna guide your hand on it. Like I'm not gonna like give you paragraph after paragraph on like the pattern and the user journey that I'm envisioning because I wanna see what you would come up with almost as like a first pass. and that has been a lot of fun, and also why actually I've been spending less time in Figma because I'm just like, hey, just create something and I'm gonna start playing around with it. I'm gonna quickly realize. This doesn't work. Let's reapproach this. I do think that there is like I watched this recent video actually on the dive club with Rid. I can't remember the guy's name, but he was talking with the guy the design lead for the Grok bot and how he used Figma still. and it was just to like quickly explore like the shapes and the general like architecture of things. just kind of like a almost like a bird's eye view of it all. but he also said like also on the the cursor now SpaceX AI, now Groc, I don't know, Matthew: Mm. Derek Castelli: team, they don't have one process. They don't have one workflow. They don't have one tool of like this is what you should use. And I think that is really cool because that is my biggest gripe. Working in agencies that have process, like such strict process that's like we have to use this tool plus this tool plus this tool. And it's like, but you just care about the input and the output. And how I get there, like, I don't understand why we have shackles on a process and shackles on different tools. To a certain degree, I get it from like consistency and like, hey, where do we check in? Check-in base, accountability. But like with their team, they were showing how. He was basically saying how some people are using paper. Some people are actually just spinning up a full on versal V zero project and sending a a prototype link. So like it's just really, really cool. Corey Moen: Totally. I think I've seen numerous mentions too of people just to this point saying that like this is now more possible because the source that we all collaborate on can be code in in in more and more ways. You know what I mean? Because at the end of the day, to your point, we they do the output does matter, and it needs to be consistent or of a certain quality bar above in like an agency environment or in an in-house environment, you know what I mean? So if that is all, you know. r rattled up to code, then how you get to that code as long as it's consistent is totally up to you. You know what I mean? which is really interesting. I'm curious to this point of tools and things. I mean we've seen some interesting sites and examples like Derek, could you could you share more about yeah, what you've been building and working on and tools you've made for yourself or others or yeah, what you've Derek Castelli: Mm. Corey Moen: been excited about in this space. Derek Castelli: Yeah, I there's a few of them. it's kind of wild how I'll just like think of an idea and then I'll like I'll spend a week like dabbling on it and then it's like ready. Like three days later it's ready. at least enough like to the point where I'm like, I'm I just want to share it with the world and like if there's bugs, if it's broken, like they'll let me know. And so yeah, I I made the first big unlock for me was well the first one is like it up my Bible notes app, Harvest. And so that one's been through so many different iterations. I started working on that when I did not like know how to make this an actual app. and so It was all just in Figma designing it up, but then it came a point where you know the Figma that I think the big the big transition point was when you could actually interact when these harnesses had a browser and annotation tools and like selecting elements. That is that was the tipping point for me where I'm like, hold on a second. I can actually really guide this thing. And tell it exactly what I want it to change. And I don't have to feed it, you know, hey, check this Figma MCP. Hey, check this, what I've screenshotted here. Hey, you know, here's my screenshot. Here's my annotation on that screenshot here. try to fix this. And so, like, yeah, I made, I continue to make this Bible Notes app and I continue to refine it. The more I use it, the more other people use it, I get feedback. That's a fun loop. turns out I was approaching the app completely from the way that like I was having the initial problem, which was like I needed a Bible notes app to better organize all of my notes. And so I approached it at a very systematic way where you had to choose what folder or what I called it a thread, what thread you wanted that note in before you even started writing the note. And I quickly realized with my wife when she would bless her heart, she would actually like try it out in church and she'd actually use it. And she's like, I'd see the folder she would create. And I'm like, why would you create why would you name it that? She's like, I don't know what to name it. I just I I need to write this. It's could like her folder's called Bible Sermon Notes. I'm like, Matthew: Yeah. Derek Castelli: no, that's not like the point. The point is is you would probably make it maybe like the sermon series or things like that. It's like, no, this could get way out of control and like it's too much friction. Also love that episode on friction that you guys are talking about. throwback. rewatch it, please. Matthew: Yeah. Derek Castelli: Friction is a good thing, but in this case it is was a bad thing. And I'm like, okay, I gotta rethink this app and of course redesign it, because that is just the nature of things. And so I approach it of like notes the primary action. and so now when you start writing, it takes in all the context and it will auto-generate folders and it will automatically generate tags, which I'm a big proponent of. Like you should never have to actually add to organize your stuff. You should never have to add a tag. It should just know. Like just know, just see it, look at the metadata and create my tags for me. If I want to, give me the ability, yes, to add a tag. But you shouldn't have to rely on me to add a tag to organize this piece of content. and so I went with that mode where it's like, this is just all happening in the background automatically for you. You can change and override all these things. But what care what mattered most was like you write your note. And so that's that was a lot of fun to just like unpack. how I wanted this app to be, but I was actually addressing my problem too head on. And I had to like rethink and redesign the entire app. So Corey Moen: Mm. Matthew: Cause you're top you're a top-down thinker. So so you're thinking Corey Moen: Yeah. Matthew: structure first, right? And it's like, well, there are people Derek Castelli: Yeah. Matthew: who are bottoms up thinkers and they're just gonna let the structure emerge, right? And Corey Moen: Yeah. Matthew: I think for emergent structure, AI exactly is is perfect at this. And I agree, tagging systems should be AI. I think folder systems should be still more manual. maybe maybe AI can help with naming things, but People still need ways to kind of self organize. We've seen that that actually can be beneficial for the AI. Cause there are things that's unspoken context, like even just in your scenario, like, you know, what is kind of the series or maybe, you know, the location or something that is can't be derived from the text that a note is being taken about. You know, there there's additional context that we have to manually add. But for things that can be extracted from the existing content, yeah, let AI. Derek Castelli: And see, I would even say, like, Matthew: String it together. Derek Castelli: so I've been a huge proponent of this. I actually just posted this on X yesterday. but tr using AI to make apps that actually don't have don't need AI in it. Like you can actually just make your apps smarter and better without AI. And like there now there is a balance of like, well, if AI is there's a point where like AI can get so cheap, like it may just be easy just to the AI I'm talking about right now like an LLM. Because the OG AI is an algorithm. Like something that's been fine-tuned, you know, pushed through its paces, delivers an output, but it's it's dynamic. and more or less, like in this Bible notes, there is no AI whatsoever. Like so these tags, I I had AI create a what is called a scripture knowledge layer. And so what it does is it pulls out tags that are relevant and S you know, even in your last episode, Matthew, where you were talking about the to-do app with like the score system, there's a score system in there. And so it's like if it notices that the theme is more based on a subject versus a topic, it will put that subject first and foremost. and so like it's Matthew: How it's ranking the tags, yeah. Derek Castelli: exactly. And so then it and then so then you have like you have the tags, but then you have folders, which Folders are a a a higher hierarchy. and so it's calculating all of that as well. So it's gonna generate way more tags than folders because of the I'm trying to I'm blank on the word, but like the the the assurance that this is exactly what this content is about. Corey Moen: Yeah. Derek Castelli: confidence level, that's what I'm talking about, confidence level. Matthew: Mm-hmm. Corey Moen: It's it's so Matthew: Yeah. Derek Castelli: yeah. Corey Moen: it's so wild to me too because like a lot of things you're saying, you know, are are things you end up getting into when you really get into like all the intricacies of how this stuff is actually gonna work and not just like I think there's this blanket assumption sometimes, especially on social media, people have been like, I put in two prompts and look, I have an app. And it's like that is so far from reality, it's mind-bending to me. Matthew: Mm-hmm. Corey Moen: and then so seeing your journey, even as we you know, we've chatted and hung out and stuff over the last couple of years, it's been wild how deep you end up getting into things that have nothing to do with design or even like web development, like you know, we've really kind of come from. and yet at the same time how it overlaps. So cause the thing I was gonna say is for those that would do watch on YouTube or anything, could you would you want to like share your screen maybe and even just like show Derek Castelli: yeah you know. Corey Moen: the harvest site Matthew: Yeah, show some of this. Corey Moen: or like the testament made site? And then some of the other the the other little products you worked on. Like I think Dinky is one of my favorite named ones. is so fun. or even just yeah, revise my Yeah. Matthew: Dinky and Binky. Corey Moen: yeah. Derek Castelli: Let's go. Okay, so what's funny about those? Yeah, here. Let me share my lovely screen. Corey Moen: Cause I think all even just even just this while you're pulling that up, it has been wild to me to see you're not just building the app, like you're also doing the marketing side and some branding and like some of these things that all end up feeling pretty cohesive. And like, you know, from Derek Castelli: Mm. Corey Moen: us talking, I think you know, you're definitely working with Claude, Curse or whatever to build and bring most of this to life, but you know, they still all feel very polished. And have that human craft and touch. still not used to that craft buzzword. But I think the last thing I'd say in this is that like put to the point of all the other intricacies, the other thing that's that that I've started to think about recently is just that, hey, craft, when people say craft and taste, so much of the social community is like, that's just design and aesthetics, visual craft and taste. But I'm sorry, there's this whole other camp. Of craft and taste that goes into performance, accessibility, user experience, like just the other nuance that has very little to do with the visual aesthetics, but very but a lot about the feel and the impression Derek Castelli: Mm. Corey Moen: and the engagement that the app brings you back, right? Or or maybe it's a site. It doesn't have to be an app, to be clear. there are so many other clever ways. Like we even just saw with Alessia's site earlier, right? Like that beautifully designed to be clear, but like there is so much more craft put into the workflow and the results and what they're experimenting with there that I think comes through in a lot of a lot of your work, Derek. Derek Castelli: All right. So these are the four apps that I've made, which funny enough, I already have a fifth one that I've been cooking up. Matthew: Off well of course. Derek Castelli: Corey, I haven't even told you about this one. there's not even like a design for it. Like I went all in on just like this is the problem I'm trying to solve. I don't even know what the design or the brand should be, but usually I actually approach it from a brand first. Like this is what I want it to look and feel like. Corey Moen: Mm. Derek Castelli: So yeah, these are the four apps. So the f top left, Harvest, that's the Bible Notes app. the second one, Revise My, this was a fun opportunity to use Layeravel Cloud. Taylor Otwell of Layervell sent out this post on X of like a weekend challenge. And it was to try to get more people to use Layer Vell Cloud. And the the challenge, if whoever won, you got a brand new full on Mac. I was like, Heck yeah, I want a free new Mac fully upgraded. And so I was like, this is a great excuse just to learn this other framework and see what it's all about. TLVR. You should, if you haven't already and you want to build an app, just try out Laravel Cloud. What these this team is doing, I can't preach enough. Like they make it out of the box so easy to set up because they can manage the off, they can manage the buckets, like your storage, they can manage. All this stuff that they've been working on these past years with their PHP framework. it's kind of wild how it just all works. Corey Moen: So good. Derek Castelli: and so what this is, is I was like, all right, so everyone's like creating like these MCPs. I want to create an MCP. Like, and then Matthew: Mm-hmm. Derek Castelli: like some of these harnesses create you can actually do you know, there's a difference between an MCP connector and an MCP app. And what an MCP app can do is it can actually display your app. within the chat or within the terminal. And so what was really fascinating me is I was like, I want to see what that's all about and how these different agents work. And really this was just a a learning exercise for me. But it's a real thing. You can try it. It generates a it's using Laravel, they have their own like token system. So you don't even create an account. It generates a a token. You paste it into your chat. And you first of course connect up your MC like the MCP to your your harness and your agents. And what it the goal is, is like it's taking the design and it's allowing human input. And so it generates a board, it generates basically like a like a a markup opportunity for you to create like visual feedback from the humans or from a team. and you that link. You can set to expire to not expire. yeah. So this was just like a fun thing. Now the what's weird about this one, which was a unique opportunity, was like most of the experience is done within the agent chat. Corey Moen: Mm. Derek Castelli: there are a few opportunities where it will, especially if it doesn't support an MCP app within, like built-in app, it will generate a link. that you click on and then you can see like your board, you can see the markups. yeah. Corey Moen: So cool. I I think this, even just this broader idea of yo, the pain of getting feedback from your clients, your team, whatever. And then also as things are moving to, you know, more accessibility in these code first worlds, like how do I also help Claude or or whatever agent you prefer to be in the loop and actually even take action on Derek Castelli: Mm. Corey Moen: things where like, hey, somebody commented on this, just fix this word. Like, just go do it and let me know when it's done. Like help, you know, reduce some of that tedious factor. so I thought, yeah, this is just super clever and really interesting. Derek Castelli: Yeah. And by the way, how I got the design of the actual marking site. I said, I like this website. This is what I like about it. Design it. And it and it Corey Moen: I mean it's it's Derek Castelli: got it pretty close. Matthew: And d did you have the content or did it build the content for the site based off the repo? Derek Castelli: Yeah. Yep. I had zero content. I I think the only thing that I told it specifically to do was the I think create the alternatives. But I'm pretty sure Fable Five was like, Hey, here's an idea. What if you did also this? And it's like, Okay, sounds great. Roll with Corey Moen: I Derek Castelli: it. Corey Moen: what what you said, Matthew, though, I think is worth calling out that is in interesting, and that like tying back to did it, you know, did it pull the content or generate the content based on the repo or its knowledge of the product? Like back to what you said, Derek, at the beginning on all these, that you kind of start from the brand, you start from like, what is this? How would you explain it? Like when I've even played around. as well, I think that's maybe it's just the designer and us, or I don't know. I often also start from this realm of just like the high level vision and idea and all that. And then that starts to reduce into the technical spec. And the point I'm saying all that is that like if you actually start that way, then there should be a lot of this foundational marketing related content Derek Castelli: Right. Corey Moen: and positioning and stuff like that. That's like in the readme of your repo or in some docs in your repo, or like. Oftentimes starting from a big plan and then that creates this massive technical spec and you know rationale type doc. It's like I think it's interesting to see how those things kind of just naturally occur. And the magic is correct me if I'm wrong, you weren't like maybe I'll ask it as a question is better. Were you starting from some like predefined strategy that social media cast on you? Like, no, you gotta do loops, you gotta do this specific thing, or you will not get a good output. Or were you just simply talking to Claude openly and asking? Also, again, sorry, Claude bias, talking to whatever cursor, any model, were you just having a conversation and talking about it and asking and asking, asking, asking, asking. Yeah. Derek Castelli: the latter and no skills were involved here. Corey Moen: Okay, so also fascinating. Matthew: What you have plenty of skills, come on. Corey Moen: Do you have s you don't have like root level skills? Like Yeah. Derek Castelli: Zero skills. So that my my hot take on that Matthew: Yeah. Derek Castelli: is I I call them specialized agents and I use specialized agents over skills, which specialized agents are just skills that you give them limits to and you give them more or less like responsibilities and handoff approaches that work just within that one repo. So they they are so specifically designed to work on that specific kind of work. So, for instance, in the Bible Notes app, because like there's so many layers and to prevent all these regressions that could happen and that I've experienced that can happen when you add and you redo things, there isn't specialized agent that just focuses on the editor. At the end of the day, I just need to make sure this the tip tap react editor is intact, still works the way it needs to do. Corey Moen: Still track it. Matthew: No regressions, yeah. Derek Castelli: No regressions, no regrets, no regressions. Matthew: Yeah. Now Corey Moen: Love it. Derek Castelli: Yeah. Matthew: what I agree, yeah, with specialized agents are are definitely the future, or the current. but Corey Moen: Yeah. Matthew: when you like this was a start as a weekend project from scratch, right? You don't you're not gonna have a specialized agent for this product yet because the product doesn't exist. So what's your approach in that situation? Derek Castelli: It does exist so think of this right I'm if you're starting a company and you're the founder of Sid Company you're wearing all the hats and you're doing all the things Matthew: Who's your first hire? Derek Castelli: who's your first hire is who's your first specialized agent. Matthew: Yeah. Yep. Corey Moen: Mm. Derek Castelli: And so like that's that's how that my approach has been is like what is getting out of hand that I just simply Can't I'm not doing a good job of like getting ensuring that it needs to continue to do the job that needs to do. And so the first one was the editor agent. Like, I just need you to make sure this dang tip-tap editor does not break. Because I'm tired of like going in and fixing it and realizing it's broken. yeah, so that's how I I've approached it. And so like. To Corey's point, I'm not starting with like a library of skills and best practices and like all of this because I'm also bullish that like these harnesses and these agents and that these models are just they don't care. It's just gonna get better. Like Matthew: Well, I think that's been proven right with Opus V and Fable for sure that the more kind of skills and and things that you try to use Derek Castelli: Shove Matthew: to confine it, like it actually has worse output than if you just are just more casual, more gin general with it. so so yeah, I mean I and and obviously that gives you more kind of flexibility in the future as as the technology evolves, right? Derek Castelli: Well, and and I would even take that to a a different level of like looking at your prompts. I used to think like, hey, the bigger the prompt, the better. Corey Moen: Mm. Derek Castelli: and a lot of times now it's like, no, no, no, like let's just like baby step it. let's just kind of guide it as we go. I I wouldn't say the only caveat to that that I do realize is like if you have a vision of what the app needs to all include. Please tell the AI, like in one of those first initial prompts before it starts building Corey Moen: Mm-hmm. Derek Castelli: what you all want it to do, because it may recommend a totally different architecture, different tech stack, how it all talks. You know, whether it should use like a which is crazy. I always tell Corey this. I'm like, dude, the fact that I'm saying like SQLite or Postgres, I'm like, I would Corey Moen: Exactly. Derek Castelli: have never used these words, these terms before Matthew: Mm-hmm. Derek Castelli: AI. Like I know what this all means, and it's like Corey Moen: I think yeah, that's Derek Castelli: this is crazy. Corey Moen: for sure. That's definitely another aspect I think that has been fascinating to see how it's evolved in the last six months, maybe a year. I for me I feel like it's been six, eight months is like the all outside of writing the code, all of the other things that you actually have to do and set up and go through to get anything like this live and user signing up and you know accepting payments and all that. The model's knowledge of the the ecosystem and the latest documentation and processes, where to click, what to do, what services Matthew: Mm-hmm. Corey Moen: best, how to weigh the pros and cons, those aspects have seemed to get a lot better and to the point that I just feel a little more confident, I guess, of like if I was to go down this hole. And again, I feel like that's where you've actually shipped some of this stuff, whereas a lot of people have prototypes you know, but aren't actually pushing the stuff live and and maintaining and growing them. And so like the way that the models have gotten better and helping to educate and navigate those things is awesome. And I think even to your approach that like you now aren't just like skipping it. It I would say like my own observation in our, you know, hangs and stuff is you're learning this stuff. You're not, you you like do have some understanding of what is where and why that was the decision. Or the other side of what you said about harvest earlier on is like you did a hard pivot after the first version of that. And you were A willing to do it, but also B like it wasn't just I don't know out of a I wanted the design to be different. I wanted to use this fancy animated thing versus that. No, it was like a fundamental architectural type of decision, you know, that was based on discussion and understanding and all that. I just think it's fascinating to see that arc of of learning and growing and all of that. And all this Derek Castelli: Yeah, so I'm I'm the crazy guy that is in Claude Desktop and I'm switching that chat mode to show me the transcript of what it's actually doing. Corey Moen: Mm. Derek Castelli: because Matthew: yeah. Derek Castelli: and like cause like there's two different approaches. It's like you got this bucket where it's like, hey, I'm just going to prompt it, tell it, and I'm gonna go away. And I'm like, I no no no. I wanna Matthew: Yeah. Derek Castelli: see, especially like early on. No, no, no. I wanna see what it's planning on doing and what how it's doing it. So like I Matthew: Have you ever stopped it like when it's thinking like, wait, no, stop, you're approaching that wrong. Like, stop here, let's reset, reframe. Derek Castelli: Well, not Corey Moen: Yes, yeah, definitely. Derek Castelli: necessarily because typically what I have it do is it's always plan mode first. And Corey Moen: Mm. Derek Castelli: so I can then read the plan and be like, Yeah, no, bro. Like we ain't doing it this route. Like but by the way, I always love it when the agent's like, Yeah, you're right. I'm like, That's right, I am. That's right, I am. Matthew: I I kid you Corey Moen: is still not. Matthew: not. a couple days ago, Claude Design told me, he was like, Y your design is better than all three of mine combined. It told me that. And I was like, Thank you. It's like Corey Moen: humility. Love this. In the model. it's Derek Castelli: Love it. Give me a boost. Matthew: It's like I'm still here for a reason, you know. Don't throw me away yet. Corey Moen: Yeah. And and I I think to this point though, even the plan, certainly again, I don't mean this to come across as like, what don't you know? It's more just like I for me, I guess I'll say is that like it comes with a plan, a massive markdown file usually. I will try to read through it and or even skim through it, but there's inevitably things that I don't know, that I don't understand, that I'm like, what is that library you're saying we should use or why should we use this or that? And I think it's fun to go down these little tangent holes of you know, well, let's let's weigh that quick. Let's not just jump on and assume, you know, but I even that maybe comes with what with the you know trials and tribulations of doing this over and over, not just like expecting it to always right out the gate. Like even you mentioned tip tap being used here, and we were talking about this the other day and from our last episode of talking about this, like the brand corpus idea or whatever. I was, you know, chatting with Claude. I'm like, hey, so like I just want this like really polished open source markdown editing notion like experience. Like, what are all the options? Let's like take some time and weigh the pros and cons of what actually fits this this approach better and why like help sell me on this before I commit to it, even though the irony is like, even if you commit to it, like totally, totally, yeah, yeah, yeah. And again, this Matthew: I like I like that approach. Sell me. Yeah. Convince me. Yeah. Derek Castelli: Yeah. Corey Moen: is even that like ladders up to. like what it's Matt Matt Pacuk, I think is how you say it or something like that on on Twitter and all that is has blown up around having like all these skills. One of his famous ones is called Grill Me, that is in that same realm of just, you know. But again, like you can also just ask Claude to do that. You don't necessarily need a skill. Just like, hey, just like ask me these questions. Or even lastly, to the point of the the the the subagents, Derek, that you were talking about, like Derek Castelli: Yeah. Corey Moen: in special use case specific agents. I think if Like you said, for the tip tap managing one on harvest, like it makes sense that, you know, maybe even asking Claude, hey, can you set that up so we can just use that every time consistently and iterate on it? But you also don't have to do that. You can just like, hey, Claude, use subagents and adversarial review, or just you know, it doesn't have to be those specific terms. Ideally, the models are getting better and better and better at at understanding what you need, understanding the context and taking action and being proactive. To use subagents when it's helpful and and ensure quality is top-notch, or like maybe a different way to say it is even like MCPs and all that. Like, I don't have to ask Claude, use the Chrome DevTools MCP to go look at this and test it and inspect it. I'll like it, just does it now because I, you know, Derek Castelli: Well Corey Moen: I I ask for generally review this page visually before I look at it, that kind of thing, you know? Derek Castelli: And to be to be clear, like with those specialized agents, I'm not prompting them. I'm not forward slashing them and including them in the prompt. So Corey Moen: Okay, interesting. Derek Castelli: it's designed in a way that it it sees those like their responsibilities and what services they touch and do not touch. And there's a coordinator actually also as a specialized agent to then coordinate all that. Corey Moen: Right, right. Derek Castelli: And so that's been very helpful because I'll say, hey, I want to add this or I want to change this. And I don't have to think about like what surfaces should I think through of what that all touches. And so Corey Moen: Mm, mm. Matthew: Mm-hmm. Derek Castelli: it's just going to do that. But then I will say the one recent one I added is my engineer agent. And its only job is to make sure, like, is this best practices? Is this to ensure that performance is still top notch, that it's still an efficient app? Like, because that is the number one thing that I think AI still has a problem with, is and it's mainly because it's us prompting. It's when you when that architecture is not fully thought through, it then compounds itself. And like it starts adding on things, and it's like, well, you could have just like simplified. Like like have have AI if you have an app out there, have AI generate a mermaid chart, a diagram showing Matthew: Mm-hmm. Derek Castelli: you how everything's connected, and you'll quickly realize how much of a spaghetti mess it's created. Corey Moen: Totally. Yes. Derek Castelli: And so that is like the number one thing I think we're like early on, a lot of hand holding with AI. just yeah. Just to make sure it's all good. But Corey Moen: I think I think the thing that comes to mind in some of that too is especially the spaghetti mess and all that is like remembering also that before all of this, you know, one of the really powerful and viable ways people stood up complex apps is using existing libraries and tools and services, not building everything from scratch, which you know, again, models will hopefully find and recommend and take that. Avenue, but not always. And so it's just like being yeah, aware of that and being like, I don't maybe need to do everything from nothing here and I should instead go down this hole of trying to understand the best path with the f fewest amount of dependencies or more depending on the use case. Like, I don't know. Yeah, it's super fascinating. But Derek Castelli: Well yeah, I've I've realized that in like the one app that I'm still cooking up, as they as they say, but I'm still cooking up, it's like the full Laravel like it's like, hey, AI, and even to that extent, like the two macOS, the dinky and binky, like AI is like, hey, you wanna use Swift UI? Cool. This these are literally your constraints. And I am a big com component, proponent of like constraints. are great. and so yeah, it knows that like Swift UI is like this is how if you want to accomplish B, or if you want to accomplish like let's say three, one plus two equals three. But in web framework things, it's like one plus point five plus point five plus one equals three. Right. And it's like there's so many different variables To it that I realized with my Bible notes app, which is I'm not even actually following a framework. That's what's wild about it. And like that's why it like it takes longer and why I'm actually looking into like what will it take to get this over to Laravel instead? Because it is so custom that there's so many different like I get lost in the weeds, a lot of yeah, a lot of Corey Moen: Lot of spiderweds. Yeah, for sure. Derek Castelli: spaghetti that can easily get added to things. I just I'm like, hey, I want to like help mana to make sure that this isn't spiraling out of control. I want to make sure all this like I just give it the goals and it's like, Yeah, I'll make an agent and we'll do all this thing and blah blah blah. I'm like, Great, sounds good. And so all it's doing is it's Matthew: So you just blind code, yeah. Derek Castelli: it's making yeah, all it's doing is making markdown files. Like that's at the end of the day. All we're doing is making markdown files. It's either a markdown file or it's it's a code file, you know? Just Matthew: Mark down his code, yeah. For sure. Corey Moen: Love it. Well, Derek, it's been s awesome to to hang and learn and see what you've Matthew: Yeah. Corey Moen: been jamming on. Derek Castelli: Yeah, man. Corey Moen: yeah, thank you so much for Matthew: Keep building. Corey Moen: doing Derek Castelli: I will. Matthew: All right. Well, thank you everyone for joining us and watching. And until next time, see ya. Derek Castelli: Later. Corey Moen: I