The Beginning Vibing with AI Code, Episode 1: Brendan & Baird Brendan: Before we dive into the episode here, just wanted to thank you for checking out our inaugural episode of Vibing with AI Code. We talk a little bit about our developer origin stories, our experiences so far with ai, and a little bit about, what we might talk about in the future. So hopefully you enjoy the podcast. And I also just wanted to mention that we are sponsored by Fireside fm. If you're looking for a great podcast hosting platform that's been around since 2016, check 'em out. That's where we're hosted, and, hopefully you enjoy the show. Brendan: Hey, how's it going? Good, good. Looking forward to recording here. Baird: Yeah. Brendan: Yeah, so I thought, you know, being our first episode, we should just talk a little bit about ourselves here to start and, you know, kind of see where the conversation goes. But you know, I, I think it's good to point out we're brothers. I'm Brendan, you're Baird. I'll let you give a, a bigger introduction to yourself here in a minute, but, um, you know, one of the reasons. From my perspective, why we wanted to do this is 'cause we have these conversations anyways. Figured why not record 'em? And you know, we're both developers and I, for me, I'm a Ruby on Rails developer. Um, you might have heard my other podcast, the Rails Business podcast, but we're not necessarily gonna focus on just Rails here. And that's one of the appeals for me as well is like, we can talk more AI and AI coding and maybe other AI things. But you know, for, for me, it's. I'm doing a lot of AI coding. You're doing a lot of AI coding mine's primarily in Ruby and Rails, and of course there's JavaScript in the mix there too. But, um, but yeah, you know, just talking about it more, getting stuff out there and, and what we think. So why don't you give a little intro as well, kind of what, what your background is with whatever you wanna share. Baird: Yeah, I've been building apps websites for 15 years now, and I've been doing a solo for pretty much the whole time. And find that AI is pretty interesting to, you know, start to actually have somebody that's almost like a, a intern or, or somebody to help out. So it's been really pushing on that to see how that works. But yeah, really just exploring the edges of how you can build stuff. Brendan: Yeah, that's an interesting point, what you just said there, like having an intern essentially with AI or, or like you said, you mostly have been solo and I've worked on a lot of teams and, but I, but I also like to explore that, essentially the extension of having AI be like a partner and working with me, you know, on my own side project and stuff. Um, and yeah, maybe let's explore that a little bit more. Just kinda like what is, what is the workflow that both of us have been doing so far with ai? Like, what is our experience? Why don't you go first? Baird: Yeah, I mean for me it started out, you know, playing with ChatGPT and just like having it produce code, kinda like I, I would equate it to Stack Overflow. You know, you'd be Brendan: Mm-hmm. Baird: how do I do this? And you'd go out over to Stack Overflow and ask the question and you take that and copy and paste and like. That didn't really work. You know, like, I don't know if you had the same experience like Brendan: Yeah. Baird: was kinda like Stack overflow. It's like it would kind of work sometimes, but you'd spend most of the time just rewriting it. And so from there, you know, as I've of gone through different transitions at work, it's like, how can I really like use it as an extension, you know? Not having to write all the code myself, but use it as like a way to give it thoughts, have it generate stuff, and then iterate on it from there and just like try to embed it directly into. The workflow rather than like having it as a, you know, tool that sits outside of it. And, know, having Cursor and Windsurf that have come about recently, you know, has really changed that because you then have the agent in your id, which helps. But even, you know, like we've talked about it before, like I've been playing around with building iOS apps in the same way, just using Claude externally, since Xcode doesn't have that capabilities and it's really like. From my perspective, helping plan things, you know, like talking back and forth with to kind of figure out like, what is the plan? How can I break this down into chunks so that it's more likely that I'm gonna achieve what I want to outta it? And then like, working together to like build out features, giving it context and just iterating and tell you, get it to the right thing. it, it's really that extension of a human rather than like, Hey, it's doing my job for me. Brendan: Mm-hmm. Baird: How about you? Brendan: Yeah, a lot of the same, honestly. I mean, and I think I'll just touch on, you know, the name of this podcast is Vibing with AI Code. In some ways that's kind of in jest 'cause like neither one of us feels like we can vibe, code an app in a way that we're just gonna like one shot it. And I, I think I wanted to point that out too just because like the workflow you just talked through and the workflow that I've found that works are, are very similar. Like using the tools in a, you know, I don't wanna say traditional developer workflow, but more of that traditional developer workflow than just, Hey, I wanna build Spotify. Give me Spotify. Right. That's not gonna work. It doesn't matter if it's a human or an AI agent. Um, and so for me, like I, I've, I've tried that for fun and it maybe works for like a prototype of something to just see like, oh, that's cool, and now I'm done with it. But yeah, like day to day and like the apps that I've actually built so far with it and I haven't released anything. And that's probably the difference between the two of us so far is like you're using it more for work. I don't code as much for work. But I am trying to incorporate Cursor and windsurf. I haven't messed with Claude Code, but I've been interested in it. Baird: Yeah. Brendan: been hearing a lot of good stuff about it, and we'll definitely touch on that some more. But yeah, I mean, my workflow is a lot of like, a lot like yours, except I, I do wanna point out, like I've mentioned already, I do a lot of Ruby on Rails, primarily. You know, you, you do what I mean, you can fill it in, fill in the blanks, but you've done a lot of iOS, you already touched on that. You know, a lot of, I guess, react and next or whatever. And maybe at some point we should just talk about that too. Like what is the difference or what, what do we see between like Rails and Next or, or even iOS. 'cause that's a whole different thing, you know, like you don't have Xcode or Cursor is not Xcode. And so there's differences there. And now I'm like all over the place too, by the way. So like, I don't even, didn't even think you're, I don't even think I answered your question. Baird: I think, I think we're, we're pretty similar in how, how we use it. But Brendan: Yeah. Baird: I, I think is interesting though, to your point, is that being able to use, you know. Bot or whatever for coding, you don't actually have to worry as much about the actual language itself. Like it really, like you can abstract out to the logic and the Brendan: Right. Baird: and you can talk to the bot from that perspective. Instead of saying like, because you know, for me, I've never built a full product in Swift ui. Brendan: Mm-hmm. Baird: so I'm, I know Swift and I know enough of Swift UI to like have some familiarity with it, but I've never like built a full app. going into the stuff we've been doing lately with iOS, like I now have like three different apps I've actually just built. The prototype for, uh, Mac os like an actual Native OS app. It's the first time I've ever done it. And using the same process, like it generates the code for you enough that as long as you have like just a cursory understanding of code itself just like basic structures of languages, you don't have to be like in it to at least build something, you know. Simple, like of course you're not gonna like go out and build as you say, the next Spotify doing it like that if you don't have enough understanding. But can learn that way as well. Like for me, it always was like I would learn that I didn't know by just going out and doing it. You know, it wasn't like I would read all the documentation. It'd be like, what do I wanna build? how do I build it using this technology? And. I find with AI you can still do that type of thing. You can just do it a little more in line instead of having to like go dig through all the documentation and all of the, you know, stack overflow and all your Google searches. It just like, you get that logic out, you get that structure out it, you know, it helps you like shape it into what you want it to be without worrying about that like code layer quite as much. Brendan: Yeah, and I think it's important too to just emphasize like working with the AI versus just like treating the AI as a tool. Because what you just talked about there, like you don't necessarily have to be a programmer. You don't necessarily have to understand the code, but you do have to to learn. What the process is, how the process works, at least some version of it to get something working out and using the AI as a a coworker, essentially, to make sure that it's double checking things or you're questioning it on what it puts out, let alone just like you know, I mean, I know you feel the same way, like doing a code review on it, like from my perspective. Because I haven't pushed anything into production from, from this sort of AI stuff that I've done. I haven't gone as deep, but if I was pushing code to production, I would 100% be opening a pull request and looking at that code and wanting to actually understand, maybe not every single line, but I want to understand the whole thing. I don't want to just use the ui and so. That's, that's the thing where, you know, like vibe coding as, you know, the, the term is kind of meant to go for, it's like, that implies like you don't need to know anything about software development process. And I think that that's something that is a, is a flaw and sure you can maybe get an app out, but what are you gonna do when it gets into production and now you gotta scale or whatever. And that, that's maybe getting too in the weeds too, like that's a whole nother topic, but. It's an interesting point, I think. Baird: yeah. And it very much is. You can still say like, what we do is vibe coding, Brendan: Mm-hmm. Baird: the difference is is that if you have, whether it's the knowledge, if you already are a developer and you're doing it, or if you are curious and are willing to like figure out what it is doing, it, it. It's possible, but it very much is like if you let the AI do all the things for you, you lose the most important part, which is the context. Brendan: Right. Baird: you know, anybody that uses AI knows that you don't have the proper context, yeah, it's gonna do something for you, but the next time you need it to do something for you. If it doesn't have context, if it doesn't have you telling it like where you want it to do and have the knowledge of like how. You would want it to do it, like it doesn't just, you know, remember everything perfectly and then like, reapply it. It's like it'll re-look at it fresh and without that domain knowledge, without that understanding, like, I've seen some pretty things that AI can do when it's just like re-looking at things, even if it's looked at it before. So it, it really is. I don't like the one shot vibe coating because it very much is like, it's gonna be magic. You know, like this is crypto and metaverse all over again. Where it's like, this is the, the snake oil salespeople that are giving you something that's like magic. But to do it right, you actually have to understand it. The un understand the use cases and like there's gotta be a human actually involved. To make it good. You know, like Brendan: Mm-hmm. Baird: one of my favorite things that I've heard and learned about AI is like, AI is gonna make average the typical, like it's the standard, the default, you know, like it's gonna Brendan: Yeah. Baird: average easy, but there's a difference between being average and actually being good. You know, like, Brendan: Right. Baird: key is like if you use it right, you can go from being average to being good, but if you just like use it as the default off a shelf stuff, it's like you're gonna be average. By default because that's, you know, Brendan: That's what it spits out. Baird: It's a probability machine. So it's like you're gonna get right in the average, pretty typically. So it's like, how do you use that as your baseline, but then use your domain knowledge, use your experience to actually make it something good. Brendan: Right. And that brings up a, a thought for my, in my head anyways, about like. If you go to, uh, you know, any of the, the tools V zero or whatever, uh, you know, bolt lovable, whatever, and you tell it, you know, build me an app and you go to all three of them, maybe, or however many there are, there's tons of those things. Baird: Yeah. Brendan: Like they're all using, like you said, kind of an average. It's gonna be basically average. The design is gonna be very similar, especially if you give it all the same prompt. Baird: Yeah. Brendan: I think design is also a kind of an interesting thing with AI that. Baird: Oh sure. Brendan: I, you probably have even more perspective than I do on that with, you know, the, the, I'm not a huge front end person. Obviously. If I'm gonna work on things like this, I need to, to get better at it. Baird: Sure. Brendan: But yeah, I don't, I've got a lot of things that end my head juggling around. But just to finish that thought, you know, it's like the design part and just like the averageness of just, if you wanted to throw it into the LLM or, or throw it into one of those tools, you, you're not gonna get much out of it, especially right away. But I think the other thing that's kind of in, I've been thinking about here is like the learning aspect. Like there's a huge opportunity to make yourself better, like you were saying. Like you could, you could be the average person if you just go with it, but you could take the opportunity to learn from it, ask it questions, treat it like, you know, not even necessarily an intern, but just like a coworker that you can learn from. And I think that's an important thing to keep in mind too. It's like you've probably. Seen it and, and read about it. But just like, you know, the atrophying of, of your, your own skills. If all you're doing is telling the AI to do something and you're not taking the time to learn or even understand what it's doing, then at some point like you are gonna lose some of your skills. I think. Um. Baird: Oh yeah, no, they've actually been doing a lot of studies about that, and there's pretty irrefutable evidence that like you do, like if you outsource all of your thinking, you're doing all of it to ai. Like you lose it, Brendan: Mm-hmm. Baird: least you know it, like you say, atrophies. So like, yeah, you could get it back, but it's gonna take more work. Brendan: Right. Baird: yeah, I mean that's of the other things I've done too, is like. This is just AI in general, but it's like have AI also like pushed back on you, like Brendan: Mm-hmm. Baird: you a different perspective, not just like agree with you. Because that's another issue with AI right now is it's sycophant. You know it, it'll sit there and be like, yeah, you're amazing, man. Like that's the coolest thing ever. And it's. Terrible, you know, so Brendan: Mm-hmm. Baird: like having it actually pushed back and you can like give it that, you know, instruction of like, Hey, if you see anything that's not right here, push back. Tell me a different, you don't have to accept it necessarily, but it like using it as that different perspective, like having another person to talk to, to check if your thinking is right, Brendan: Yep. Baird: is, you know, there aren't holes in it or whatever. It is very powerful. But again, it keeps your brain in the loop. It doesn't like outsource everything to the bot. It's like giving you an augmentation to enhance what you're already doing and potentially push you, like you say, down new paths and learning new things that you might not have thought about. And Brendan: Mm-hmm. Baird: like I, I, I'm all for people using these tools. If I never programmed anything before, because I think of, you know, myself when I started this back, you know, 15 years ago. I was in forums asking questions about things and people are always like, man, you're an idiot. And it's like, I am, but you know, this is how you Brendan: Gotta start somewhere. Baird: Exactly. And so having that tool now, if you're willing to engage with it, to actually like use it to learn, use it to figure things out, rather than again, just giving it to the AI to do for you. Like you can get so much better results that way. Brendan: Yep. Yeah. And you know another thing that, I don't know how much you do this, and I don't always do it either, but. Another thing that you can do, rather than just saying, build me a feature or whatever, you could say, this is what I'm looking to build. You know, give me three different versions of this. Baird: Sure. Brendan: then that keeps you also in the loop of, okay, well I don't trust the AI to pick the right version first for one. Baird: Yeah. Brendan: And also, you know, I still want to give my input. I've got experience that I want to, you know, like I know the right way or at at least I know the way I'm thinking. But maybe the AI will gimme something more interesting or better or whatever, you know, that, that's just one example that, that I've, I've done it at least a couple times. Baird: Yeah, I, I haven't done the multiple versions, which I think is actually a really good way to look at it, but I usually do, like if I'm starting a new app idea or whatever, like, because a lot of times I just like to prototype, you know, it's not Brendan: Yeah. Baird: Hey, I'm gonna build something that I'm gonna ship. It's more like I've got a day or two days over the weekend or something, and I'm like, okay, I, you know, I always run into that blank slate problem that like getting the boiler plate set up problem where it's like. I spend more time thinking, oh, I'm gonna use this technology or whatever, because I haven't done it before. my time's up and I've done nothing. You know, like I haven't actually achieved anything. Brendan: Yep. Baird: And so it's like what I'm trying to like train myself to do now is say like, okay, go to Claude or ChatGPT or whatever, and just say like, okay, I wanna build these three features, or this one feature, or whatever. Like figure out like a, a very small. Feature set for an app that I wanna do. And usually it's like something that'll help me personally rather than like, I wanna build Spotify, I wanna build, you know, like it, it's Brendan: Yep. Baird: big like that. it is saying like, you know, how could we approach this that we could tackle, you know, prototype in two days or Brendan: Mm-hmm. Baird: it like, give you the rough outline of like, okay, we do this. And I always tell it explicitly like. the file structure. You know, like, tell me how to set that up. Like if I change my mind or disagree with how it is, you can always change it. You don't have to be like, okay, I'm just gonna do whatever you tell me. But again, it's like it takes decision making process that isn't that important at that stage of it. Making it so that you can just focus on the things that are actually important, which is like writing code, like actually like building a feature. And because so much of that time is usually like getting up to building the feature. And so like if you can just kinda like skip that like doing it in a way where like I, I don't have it necessarily do it for me. I have it do it where it's like, give me the steps and then I'll walk through the stuff That makes sense. And then you're learning as well. So like the next time I know like, okay, this is a folder structure that I like when I'm going into it. These are like the packages that I might need. And so it, it's like teaching you while giving you a way to figure it out without like having to do it all yourself. Brendan: Yeah, you touched on something that I've definitely appreciated about even just trying to one shot something just to, to prototype or just not have the blank page like that. That thing kill the, you know what you described it perfectly. It's like you try and build whatever app, doesn't matter if it's Rails or Next, or React or whatever, like you're trying to get it set up and all that boilerplate takes so long. That by the time you run, by the time you want to build a feature, you're like exhausted from just getting the app ready. And that takes that away, which is really nice. And it's not gonna be the app that you deploy necessarily, Baird: sure. Brendan: at least teaches you or you learn what you liked, like you were saying, or like how you could build it maybe. And it's so easy to just throw it away and then start on the one that you actually want or you know, you'll keep. So that, that's a really powerful thing that I've found with, with AI coding in general. Baird: Yeah, and I think there's an important, at least for me personally, like an important distinction there where I tend to go with like the vibe coding, but doing it. Less one shot versus the one shot method is that for some people it might work great to do the one shot like, and especially like as that prototype getting something up there. But for me, like I like to, I have an easier time understanding it if I'm like in the process of building out these things as I go. So I'm like, okay, I know this is, this code is here. I know this code is here. Like I, I start learning the structure, like I'm not writing. of the code at all. At that point, I'm like essentially like going back and forth with Claude, you know, actually in like Claude's chatbot interface instead of doing it in the agent so that I can walk through it and kind of see like. is the structure, this is where it is, this is what this does. So that then when I wanna like actually jump in and start developing things, I don't then have to like go and look through every single aspect of code and try to figure out like, how did the bot put this together? You know, like you get that one shot code and you have to like. Essentially absorb it all to understand how it was built before you can start moving forward and doing things constructively from there. And so like in the end, it's not radically different in terms of like you get that prototype pretty quickly, but from my perspective, it's like just that understanding and having that like. Piece to the setup. That just helps you learn it enough that you're then ready to do it versus like, okay, now I have to go back and understand what the, the bot did. Brendan: Right, which is another important thing that I, I think it's worth touching on here, is. We've mostly just talked about initial versions, prototypes, whatever. But if you're actually gonna put something into production that real users are gonna use and whatever, like you, you need to understand at least enough of that system to be able to maintain it. Either add a new feature specifically, you know, bugs and, you know, my experience with debugging, you know, and or fixing bugs with. With, with the help of the LLMs, it's gotten a little bit better, maybe with Cursor and Windsurf, and I'm, I'm really curious about Claude Storm or Claude Code, and there's Claude Swarms now, which is something I've just learned about, so I don't have much to add. I'd love to talk about it maybe in another episode, but it still hasn't been perfect, you know, and there's numerous examples that you and I have both talked about in the past where we're trying to figure out, well, where's the bug? Using the LLM or the AI to help us out. And it's just like it. It goes in circles and circles and circles and half the time it feels like my knowledge of rails in this case is the thing that actually fixed the bug, not the LLM or the ai. And so that's another thing where I'm sure it'll get better as we, you know, things, all this stuff progresses, it's gonna get better, but it definitely is something where. You really wanna understand that code, at least at a level that you're comfortable with it, especially if you're gonna put it into production at some point, Baird: Yeah, and this is the, the part with that, that really I think people struggle with right now is that there's no concept of security Brendan: right? Baird: for people that have never actually built a website. So they're like, I wanna build, you know, any type of user facing app that takes personal data, even if it's like email address addresses and stuff like that. You know, people don't understand like, oh, you need to have your environment variables on your server. Like they can't be called because they don't even understand like, the difference of the clients I call versus a server site call. Brendan: Mm-hmm. Baird: of that. And you know, that's where vibe coding is dangerous, is like, it just, it gives you something that theoretically you could run, but it doesn't a seem to, you know, security isn't a huge thing inside of it right now because it, Brendan: Right. Baird: it. Is set up to run on a client side. You know, like Brendan: Mm-hmm. Baird: isn't in a way, like obviously if it's Ruby, it still has to run on the the server, but there's still that fact that it doesn't know where you're running it. It doesn't know how you're deploying it because you probably don't know where you're running, deploying, or even what any of that means. Brendan: Right, right. Baird: so like there's just these layers of knowledge. You know, maybe the AI someday could help you with that. Maybe the AI will be like, you know, okay, we're gonna host it here. We know that these are gonna be, you know, environment variables, so we need to, you know, hide those. think that's where AI needs to go to be, what people say it's going to be. And maybe it will, you know, it, it's hard to say, but it, it's, so far what I've seen is that there's a lack of agency with ai. Like it'll do what you tell it to and it will kind of like fill in the blanks, but it's not gonna be like. Hey, you're having a blatant security issue Brendan: Right. Baird: you like kind of catch it on it or like Brendan: Yep. Baird: and it's decided that it's gonna use this today versus what it might do other times. And that consistency itself is also kind of, you know, a dangerous thing because it's not like, there are some days where I give it prompts and it's like it does 10 things and it does great, and then the same day or another day it all of a sudden it's like. Everything I give it, it's wrong. You know? And it's like, is that me? Is that just how LLMs work? Like for people that understand it, it's still not But if you don't know any better, it's just like, how would you know it's, know, if it breaks, but that's way if you can really tell if something went wrong. So it's, yeah, I, I think there's a lot of room. It needs to improve to really like get to that point. But I mean to our original kind of premise on this is like, still super valuable if you know what you're doing with it. Brendan: Yep. Yeah, and I mean, in some ways that scares me, the whole security side from the perspective of if all these apps are being built, and obviously it's not every app in the world, but Baird: Yeah. Brendan: more apps that get built in this manner where people don't fully understand like. I don't know that necessarily, and I come along and I throw my credit card in or whatever. Like, Baird: Yeah. Brendan: I mean, this stuff happens without AI building apps. So the more people we get into this ecosystem, not that that's, I'm against that, but it's just, it's, it's a little scary. You gotta, you know, really think about it. Baird: No, but I mean, that's a perfect point, is that. You have massive teams of people getting paid a ton of money and you still have data breaches, you still have people getting hacked. Like you have all these issues and AI doesn't fill that void. Brendan: Mm-hmm. Baird: isn't better at the security because a lot of the security things aren't like code. Like it is code, but it's actually like the logic behind the code, the reasoning behind Brendan: Right. Baird: somebody that's a human could exploit. Something that is written by humans Brendan: Mm-hmm. Baird: AI isn't to a point where they understand that nuance. You know? Like there isn't that ability to say like, oh, I know that a human might approach it like this. You know? My guess is it has no idea. Like social engineering, you know, like all that type of stuff. Like how do you account for that? Like Brendan: Yep. Baird: just so many gaps there that like. To your point, somebody that can just come in and be like, Hey, let's get a subscription service going. And they're running it all client side storing new data on a server somewhere like, because again, it happens anyway, and so much of it too is like. I'm scared for the days that are already here of like vibe coding because everybody's on such a deadline all the time. So it's like, okay, we gotta do this fast. And that's where you're gonna get even, you know, big teams, little teams, whatever, using it for speed, they don't have time to actually check it, understand it, make sure it works before it gets out anyway. And so there's just, you know, because you can write so much more faster. You can just produce so much more bad code Brendan: Yep. Baird: you've got it in the wrong people's hands. So there's this, I think the whole vibe coding is like this huge thing that everybody should be doing is just kind of scary because there are some huge ramifications. But for people that actually know how to do it right, there's also like ramifications that are very positive if it is used correctly. And I just see too much news headlines and all that that's, you know, all the clickbait of like, yeah, you can build anything with one click. And that's just, you know, I'm hoping if anything we can help with this podcast to like. light on how you can use it differently to still get awesome results, but do it in a way that you're actually like learning. You're still using your, doing it in bankrupt your company because Brendan: Yep. Baird: coding, uh, credit card transactions. Brendan: Right. Yeah, that, that's a a really good point. And maybe that's a good place to just transition a little bit. You know, I, I think we could keep going for a long time. There's so much to talk about. There's always gonna be tons of talk about, um, you know, but one of the things that I think at some point, you and I. I've talked about doing here is maybe we just, you know, not one shot an app, but maybe we just like pair together, you know, go through some of our processes and, you know, build something real, but show some of those things that are important to learn and, you know, how you would do it. Whether you're building, you know, with Cursor, Windsurf, Claude, whatever, you know, I, I think it's important to pay attention to that. You know, one of the, one of the things I'll just mention, like I like about Rails, that doesn't have anything to do with ai. But it is fairly secure out of the box. And I'm sure that React and Nick Next have a lot of that stuff too. And that is one benefit of these systems using those. But it still gets old, it gets stale and it, you've gotta know to pay attention to that stuff. And it doesn't cover everything. You know, I, I could still, with or without ai, I could write something into my Rails app that allows someone to do SQL injection. Like there are some pro protections there, but I mean, if I just let the AI go at it. And it doesn't think about it, it could easily break, you know, something or, or allow something like that, I should say. Baird: Again, that's a, a good point of context is like, depending on how the AI writes something and how you then fill in the blanks, things like SQL injection could happen because the, a AI doesn't have the context to know that you're not escaping your string when you're passing it in some way down the line. And, and so yeah, there is the context when you're using cursor and windsurf and stuff like that, but it's my. Experience. Experience. It's not as good as like somebody who actually knows the code base, somebody who actually understands like it. It's using semantic search Brendan: Yep. Baird: understand these things, which does, you know, from where things like this were a couple years ago. It's remarkable how good it is, but there are still so many times, or like I give it the exact path. like line of code and it still struggles. It's like, let me just read this entire file and like it reads, you know, 200 lines of the 600 lines and then stops 'cause it found something and Brendan: Yep. Baird: keep going. And so, but again, like you have to understand aspect to it to be able to utilize it fully and know like, okay. My guess is you probably didn't understand the full context of this and like, or know to like, give it the context to be like, okay, look in this file, this file, this file, and like give it more things explicitly pointed out. And so yeah, it's very much that you gotta kind of understand it to be able to do it. But, I'm going off on a tangent here, but Brendan: Nah, you're all good. Baird: it's really just, yeah. I think having that baseline, like you say like 'cause it's gonna use Rails, it's going to use React, it's gonna use stuff that's standard. It's good for people that don't know, but I, I think you can, especially nowadays with like all the web search and stuff, like, you can give it context of like documentation and stuff too, but there, there's just this whole world of knowledge. You need to understand enough to like give it that context. Brendan: Mm-hmm. Yep. Yeah, exactly. Well, this being our first episode, I just thought on the fly of a, a new little segment we could try just for fun, and it, it'll kind of be a good place for us to close. Um, is there any like, cool AI news? It's not related to everything we just talked about, but anything that, you know, whether it's programming or not, that you have seen this week and you wanna talk about briefly? Baird: Yeah. I'm kind of curious to see where the new, uh, Claude artifacts are gone. Like, I use Claude as my, you know. to for any bot type stuff. Brendan: Hmm. Baird: just it has a different feel to me. I don't know. I've always been a contrarian, so most people use ChatGPT, so I guess that's part of it. But yeah, like artifacts was one of the first, you know, like ways to actually show like what you're building inside of a chat bot. But now there's spun that out into, it's almost like. You can provide apps to other people that they can use their subscription for and done on. Brendan: Yeah, that's, that's interesting. I I definitely saw that email and, uh, but, but for me, you know, one of the things that I've seen this week, I already kinda mentioned it earlier, but like the, the Claude Swarm stuff, which I don't even know how that interacts with, you know, the stuff you were just talking about, but it, it definitely seems like, you know, anthropic and Claude are, are, there's some interesting things going on there and. I forget the name of it. I'll put it in the show notes. But there was also related to the whole Claude Swarm thing I just saw last night. Um, Obie Fernandez, who's, who's pretty prominent in the Rails community, he pushed out a, uh, again, I forget the name. It's a, it's a gem though, that basically uses cla swarms to help you build in your rails apps. And kind of going back to the whole, like, I'm not gonna say one shot, I haven't even tried it, but. It's an interesting way to get more context though, from my understanding anyways. And also have like the, what I gathered from it is, the way it works is like you've got the swarm and it basically like one, and I don't know all the right terminology, but like one of them knows about your controllers, one of them knows about your views, and then there's like, you know, at the very top there's the, the control, not the controller, but the orchestrator or whatever the word is, Baird: Yep. Brendan: that kinda like tells each one where to go and. I'm interested to explore it. That's something that I, I just learned about yesterday. I haven't even had time to dive into it. Um, but yeah. Yeah, that's, that's what I'm been thinking about. Baird: Yeah, no, it seems like agents use that way can be a very interesting concept. I, I, you know, we can get into agents someday too then Brendan: Yeah. Baird: there's concerns there as well. But, uh, yeah, I, I think in a code perspective, having multiple, you know, things, working on it with that kind of oversight could be really fascinating. Brendan: Yeah. Yeah, definitely. Cool. Well, this is probably a good place to wrap up. Um, you know, if you, you made it this far into the video, awesome. Maybe you could like, and subscribe, leave some nice comments down below and, uh, we're looking forward to putting more of these out. Anything else you wanna add? Baird? I guess what actually we should say, where we can be found on the internet besides this podcast is any, anywhere you wanna point people to. Baird: Not at the moment, but maybe. Brendan: Fair enough. Yeah. For me it's usually just my ex Twitter, you know. Um. I'll put that in the show notes. I, I think I'm Brendan wb, but it's, honestly, I don't pay that much attention to it, so. All right, cool. Well, been good chatting and uh, we'll, we'll do this again soon.