What Coding Since Age 12 Has Taught Me | GitNation | Corbin Brown
Other Videos📅 2025-04-02
Transcript
[00:05] Hello everyone. Welcome to my talk. We're going over AI powered development from curiosity to creativity. This is going to be around 14 15 slides. So let's go and jump in here. Right off the bat, here is a table of contents. So I'm going to give you some context of who I am, Corbin Brown. I'm going to go and transition into developing and what's really happening now that AI is here and the implications of that. As some of us old-fashioned people here, we used to never code of AI, but there actually is real implications to using these new AI tools, which means that leads into the
[00:35] next part here, which goes over these AI tools. And then really looking at long-term where is the value now? Because originally when approaching coding to code out certain files, to do certain tasks, it really took a lot of time. But now when you leverage AI tools, we're just getting outputs fast. We're getting code files fast. So, it's almost like where is the value in software development now with the new AI tools that we get access to. And then finally, it's pretty cool that we're entering in a brand new market here as I
[01:07] would compare this to the 99 new websites, 2008 2012 iOS apps. This is a new market when it comes to integrating artificial intelligence and providing value through software. And then of course at the very end I have a little surprise. So make sure to stay tuned with this talk here. Let's jump over to the first major step here. That's me. Or that's me. That's when I was 12, but probably not. I think I was probably maybe I don't know like six or seven in that picture, but I couldn't find any other picture. Okay, so we'll assume I'm 12 because that's when I started coding apps for the iOS app store. This was the
[01:40] beginning of my journey when it came to development. This was a lot of fun. I remember the first app I ever created was for the World Cup. I don't know what they're called, but like they're like the balloon things that you hit together and makes a loud sound. That's that was the beginning of my journey, right? And when you were coding around this time, obviously artificial intelligence and the way we approach coding now just didn't exist. Therefore, it left us in a position where how do we get answers when we run into issues with coding back then. And as we can see, this was just a
[02:11] bunch of jumping around, right? Whether we're going to a Stack Overflow form and like, you know what, what did people say about this? Okay. and you're getting like 80 70% of your answer and you're like, "Okay, well, I got to fill in the 30% gap there." So now you're going to a Reddit form and then sometimes you just find yourself randomly placed in a blog and you're like, "You know what? I really don't want to read this blog because every 5 seconds I'm getting hit with a newsletter thing. Put your email here. I don't want to do that." And then you scroll down the blog and maybe 5% of it's relevant and the other 95% you're just like, "This has no correlation."
[02:42] This was the past. This would take hours. This was slightly annoying because the fact that a lot of the times you really had to hit bare minimum in the sense of what you're trying to solve and just like you yourself had to scour through all of that data that existed on the internet and find an answer to your coding problem. But that doesn't exist anymore, which is kind of cool because now we're transitioning where all that heavy lifting, all that rabbit holes, all the searching through Stack Overflow, all that Reddit, we can just
[03:14] offload to AI. Now, now I want to say this to be clear though, Stack Overflow, Reddit, and the way of traditionally looking up and solving errors, that still has potential, that still has value, right? At a certain point, these AIM models may hit a ceiling where you get 90% of what you want, but you know, good oldfashioned go to that stack overflow and you'll probably get the better answer there because it's more, you know, the data is more new rather that these are trained on an older data piles such as like 2023 October, whatever it may be. The idea is that
[03:45] this still has validity, but honestly, we're transitioning. This is a new age. I remember when Chad GBT, not 2.0, but 3.4 four 3.5 first came out. There was a lot of push back in the sense of like, you know what, this is not good. Um, I'm not going to use this for my coding. It's not really relevant now. It's almost like uh trying to do a math problem without a calculator. You can do it, go ahead, but you might as well use the calculator. So coming down here, that is where it leads us to the new workflow of when approaching coding,
[04:16] which is entirely different than what it was when I started coding, which is now we can ask AI for the code, which is cool. It almost puts us, it almost puts us in a manager type of position here where we ask AI for the code. And now you can see right there, me right there. I like the code. We ask for the code, the AI responds with an answer. You get to verifying whether that code is good or bad and then you can keep building. And this little workflow right here is
[04:47] extremely different than how it used to be when you started code. I remember coding out, you know, it's it's as simple as stuff like when you're coding out CSS or like in iOS development, I was using Swift. Like it's very simple stuff where you know exactly what to code. You know what you need to code. It's not even about finding an answer anymore, but you still have to type it out and code it out. You have to get the syntax right. You have to make sure we name things that are relevant with the CSS name so it doesn't get too confusing. So you're not just saying button. Maybe
[05:18] it's like uh sign up button, sign up-ash button, right? This kind of stuff would just take time. But now with this new workflow, we are just leaping towards just give me the code, let me see if it works. If it doesn't work, then I can keep gut-checking the AI until I make it work. The AI itself and how it codes is only as good as you prompt. That's fundamental. Yes, there are certain things that the AI is limited to, but if you have the context or the knowledge of
[05:48] how to approach and solve a solution, the AI 10 times out of 10 probably can code it, which is really, really cool and leads us into this new workflow didn't exist before. I remember sitting when I was, you know, way back when. I remember sitting there just coding out files and just like, oh my gosh, this is just like grunt work right now. just coding out every single CSS next step here which is a which leads us if I can speak to us saving hours basically which is the idea that now that we can use AI to really push to push our heavy lifting when it comes to
[06:20] coding we can save tons of hours a ton of time which now makes it so that AI is now part of developers workflow making its answers only as good as the questions we asked. This is extremely true and I think at this point this has become more readily accepted. When artificial intelligence did get integrated into coding workflows in the past, there was a lot of push back. Oh, this code's bad. This code's not good. No, in reality, you just didn't know how to talk to it. So what we're realizing now is that AI can give good code, but
[06:53] it's really dependent on the developer to see if they have enough context and knowledge to know how to proctor the AI for that good code, which is completely different. Which leads us to the new age here of a ton of AI tools to choose from. A ton, Corbin. A ton. Almost too many. This feels like uh you know, this is its own little micro market right here. Not even micro, just its own little market, right? So, it's almost like that old saying, give the pickaxer during the gold rush or do you want to be the one mining gold or do you want to be the one selling the
[07:23] pickaxes? These these people are selling the pickaxes. Okay, but it's not bad cuz honestly, I don't mind mining for gold. But the idea here is that we have a ton of different options to choose from. And as we know, time is valuable. Therefore, whatever you dedicate your labor towards, you want to make sure you're doing the right thing. And you don't spend an hour on something like lovable and realize not an hour I mean like a year on lovable and then realize like two years down the road like oh I chose the wrong one. So which one do we choose? How do we approach this? I like to categorize it into three main
[07:56] categories. First one is going to be agent-based ideides. These are ideides that are very new integrated development environment by the way. These are IDs that are very new. These are very much in the last 2 to 3 years honestly too. Idea here is that we have a almost like a agent builder where you kind of jump into a website like vzero here or replet you talk to the agent or AI model and then the AI model pushes out all the code does a lot of the heavy lifting when it comes to things like maybe you don't need to know that terminal command
[08:26] it kind of already knows that terminal command and does it for you creates out the infrastructure of the software application everything of this nature. And what I'm about to say right now may ruffle some feathers. So, let's ruffle some feathers. Personally, I don't think you should go down this route in this development if you care about building a real full stack AI application that you could sell for millions of not billions of dollars and just really understand the architecture of actually building out real
[08:56] software. That being said, this route is amazing for MVPs, minimal viable products. This route is amazing for just starting out building something really quick and just being like, you know what, does this work? Could this work? Proceed. Personally, within the software company I'm currently running, I do use Replet in the context of how we vet potential employees. They have like a great little replet, create a replet. You can do testing in there. And I think that's just a cool little use case that Replet has that I just personally like using when hiring internal employees and software engineers. Saying that though,
[09:28] these are great tools, but these are MVPs. And honestly to step back a little bit further here, I'm going through the perspective and the context here that you want to be an entrepreneur and build out a softer business from the ground up. If that is not your perspective, then that then just ignore what I just said there. This is not a video for people that want to basically, you know, maybe get hired at Google. Purpose of this one, what I'm trying to talk about here is starting a real software company. MVPs. I ruffled some feathers.
[09:58] now comes into a really big one here where we have the traditional IDE like VS code but we integrate a chatbot into it right or an AI model into it this is things like cursor and windsurf these have a place here and the way I kind of view this is kind of very much like you're sitting at a table and let's say there's color pencils and there's color crayons which one do you prefer I don't know Corbin I I prefer crayons okay but what if I prefer pencils okay but that's a bad choice. Why' you choose crayons?
[10:30] It's that situation here. ID plus AI, you can build full stack applications that are real, that sell and make money and are full stack. Yeah, that's amazing. You can do it. Me personally, I don't like this. I don't like combining an AI model into my IDE. Therefore, that's going to lead into the third category here where we actually separate the two into separate environments. So, it's never integrated. Why don't I like doing IDs plus AI? First major reason for me personally is honestly workflow-wise you
[11:03] ask the AI model a question within Windsor for cursor sometimes that thing can go on its own rabbit hole start pulling up random files that you know you don't care about start doing things to the code file that you know it shouldn't do and just it has too much context it it almost overdoses when you ask for a question or to do something within your IDE so knowing that me personally I don't opt for I don't like IDE plus AI integrated into one little package. This is a personal preference. Do you like color crayons?
[11:35] Maybe I like color pencils when I draw. Who knows? That's your choice. You can still build out res. You can, if I can speak, you can still build out real software here that are not MVPs and get going within cursor or windsurf. But as you already know and if you know me well, I love me some VS code and chat GBT or claude or any AI model that you care about. But for me, I keep it separate. IDE AI separate tab chat GBT separate tab VS code never integrate
[12:06] into the full codebase. Keep it separate. I personally do this because it keeps the code cleaner. I only give the code that I actually care about personally into the chat for it to change. And honestly, when you use those other types of development environments, there is a reason that it's quote unquote easier to use that first group of AI agents and it's quote unquote easier to maybe use cursor and wind surf. But for some things, especially when creating software companies, we don't want it to
[12:37] be easier. Go through the hard heavy lifting of understanding all the terminal commands associated with deploying a web app to a live website link. Go through the heavy lifting of understanding everything associated with the package.json, whether you're using Firebase, GCP, initializing functions within your actual IDE, like fundamental stuff that may be kind of glossed over in other types of development environments. Just learn how to do it. Just understand it. This is going to be
[13:08] very useful and very high skill level work. So that's why we want to keep our IDE separate or personally what I prefer. So maybe I'm color pencils and your color crayons. I don't know. But personally I like keeping it separate and personally I like using Chad GBT. Does that mean Claude is horlock coding? Probably not. Does that mean Gemini is horlock coding? Probably not. So that's very much user preference. I like using chat GBT because due to the fact that we can use custom instructions to get more lasered outputs that I like. So keep it separate or at
[13:40] least that's what I prefer. So where is the value? Where does this lead us? Where does the value lead 10 years from now when all of a sudden we can use a oneprompt system and then it just outputs an entire software? I think the value lies in architecture, scalability and creativity. What do I say by what do I mean by when I say this is that essentially yeah you could have a system 10 years from now that builds out a quoteunquote software application with one prompt but in reality does that AI model actually contextually understand the best type of
[14:12] architecture to use when creating said software for that specific use case and the only true way that that AI model would understand the tech stack and the true tech stack to use for whatever your value is or whatever business model you're trying to create is based off the developer themselves and the context and the knowledge that they have gained up to this point. So what am I saying? I'm saying that start building now, start learning now and start understanding correct architecture depending on your specific context of the type of software
[14:42] you're trying to create. This isn't a one-sizefitit all. You can't I mean in theory you could create a cookie cutter situation here but in theory for you to create scalable software with good architecture you could receive the same end value point eg let's say we create a pipeline individual puts in a PDF that PDF the end value point will use AI to analyze it and it's for a legal team and a lawyer for them to basically get the relevant case laws associated with that PDF amazing but here's the situation for
[15:14] you to provide that unvalue point of the case laws associated with the PDF because maybe that's you know they need the case laws whatever it may be. I'm not a lawyer. There is path A and there's path B. Both receive the same end value point to that consumer but path A is like 500% more expensive than path B. Does the AI know which one to choose? Probably not because it probably doesn't have enough context to really understand that path B is better to provide the same end value point which only costs maybe 1 cent but this costs
[15:44] 10 cents. The idea is that as we've said before, the AI is only good as the prompt you give it. And even if it was to go fully autonomous, we still want to give enough context here to push it in the right direction. Which leads us to the final part here, which is creativity. I mean, obviously, end of the day, hundreds of years of humanity, creativity's always been valuable. But in reality, we are transitioning to more of a developer workflow that's very much coding in a manager position and very much logic
[16:16] based coding where you understand exactly what you need and you need to understand the logic associated for what you need and just tell the AI to do it and output the entire file slashoutput the entire code. Therefore, where does the real value lie 10 20 years from now? It's going to be creativity. There's going to be the ideas that push the bounds of what you want to approach. But here's the best part as where we're at right now in the market is AI software is brand new. Like new like this is 99
[16:48] in the '9s when the internet.com bubble started in 97. That's how new it is. We still have a good fiveyear gap here where there is going to be a ton of companies that are founded and created now based off this new tech. or standing on the shoulders of giants. You know that old saying, you don't need to create the internet to create Amazon, right? So, Amazon was created off the internet. You don't need to create the next AI model to, you know, win. You have the AI model there. Build on top of it and provide value in any way possible
[17:18] that is relevant. As this is brand new, I feel like right now in media, it's almost feels like, oh, there's always a new AI company. There's always like whatever, whatever. like there's no way I can compete. I mean, that's probably how it felt in the 90s and during the 2008, 2012 and like there's going to always be saturation, y'all. But what you have to understand is that we're not entering this in 2030. We're entering this now, which gives us a significant advantage over individuals that would have wanted to enter this 5
[17:49] years from now. And a good old saying I like to sing in my company, just put the radio on the internet. We're so early to this. Go for the lowhanging fruit. Go for the things that are obvious like putting the radio on the internet. That's a pretty low hanging fruit that you know would garner a ton of value. So start building. Get going. No, like biggest thing I could teach you and just tell you and from my experience is that most of the things when it comes to creating a software company only really
[18:20] happen when you actually start creating a software company. The skills and the lessons and everything you know the implications behind it. And this stuff takes years. So yeah, new AI tools may come out. Yeah, you might be able to build your MVP in two weeks. Yeah, you might be able to build out a full-blown software, full stack application in three months. But in reality, this will take years. So if you want to take that journey, go ahead. But if you don't, then don't. But what I will tell you this is probably one of been this has probably been one of the most beneficial
[18:51] journeys within my life as I think personally software creation and creating software is like drawing paintings. Am I going to make the Mona Lisa? Probably not. But am I going to draw a really cool painting? Heck yeah. So I'll see you in the next one. My name is Corin Brown. You can check me out on YouTube. Without further ado, this concludes today's talk. Thank you for joining me.