Learn To Code with AI, Here's How (it's not too late, don't listen to excuses)

Should you learn how to code? With all the AI coming out, you should probably just wait. You shouldn't actually learn any of the basics, right? No. I could cut the video there, but we need to get into some reasons why. So, I get this question a lot because I run a community of people that are in different cohorts, different levels of skill, and so I've come up with a lot of thoughts on it, and I want to hang on. And we got this background or this agent running right now. The reason that I'm able to do this as an example, the reason I'm able to orchestrate these agents is because I know the underlying code. I know how it works. And so that is the thing. If I were to do the TLDDR on this is you can learn how to orchestrate these better if you know how the underlying tech works. You learn how the underlying tech works by raw dogging it, by doing it yourself without AI actually doing agentic coding, but just being your tutor and that's the magic. So, a couple of things. This is a daily upload channel. So, we do three news stories, takes on them. You can watch full news somewhere else. This is just my personal takes. And then we do some questions and answers. And then we go into strategy. So the strategy is going to be around do I need to learn to code? First of all, yeah, there's new stuff coming out every day. Mistrol's now got an agents API. Vzero's got an API. The team at Duck DB released Duck Lake which is very interesting. It allows you to store a bunch of data in S3 and then catalog which is used or the catalog lives in the database. Crazy. So, if you haven't listened to Paul talk about their growth at Superbase and how they operate as a company, I highly recommend watching the new talk that they did on Excel the VC's channel. And then another piece of news is this quad code thread. The guy spent $70 in a single day to get his app working when you could just go use something like augment and I'm just starting to use it a bunch. I'm finding it really helpful. I'm going to be doing a full video on it later in the week on the main channel, but as a test, I wanted to just see how good it is at context. So, I will keep you guys informed on how that goes. But I think context is the biggest bottleneck now. And so for this example, I actually just fed in an entire codebase of the example and then had it reference it and it just crushed. Now let's get into should you learn how to code? Answer yes. But there's some caveats because learning to code today is very different than learning to code when I was 14 or when I tried to learn again when I was 17 and 21 and so on. About 3 years ago, I started to learn how to code and take it really seriously. Not just a little front-end programming stuff, but the full stack. And what I was doing then, which I still think applies now, was you have different buckets of how you could learn. You could learn by doing the agent route where it will go and it'll write the code for you, but then you don't actually know how to understand any of it. So, you could try to do this. So, I think a lot of people are doing this and they're just getting decent at reading and pattern recognition. But what I think is the best is actually just using chat plus docs plus replacing habits. And I'll talk about that one in a second. So chat would allow you to then if you get a sentence and let's say this sentence has seven to 10 words in it. So let's do five, six, seven. say it's got all these words in it on a line of code, an LOC, you might not know what a majority of those words mean. And so if this one is confusing, you need to actually go and research what that is. If this one's confusing, you need to go research what that is. And so when you start by just asking a lot of questions, you create this spiderweb effect that gives you the learning path of what you should learn. And I recommend picking up whatever it is that you need to learn just in time for what you're building. So it actually starts with a problem. If you have a problem that you're personally facing, you'll be far more incentivized to go and solve it. And then you'll get the motivation to go and continue doing it. Because when you're struggling, that is the sauce. That is learning. Most people when they struggle, they stop and they just want the agent to go do it. And it's a slippery slope. I still run into this. I have an example where I was starting this project and I didn't go and actually have the agent do this work until I had rebuilt a couple of the starter projects. Let's just get back to it. Yes. Continue. Cool. So now why learn the basics? There's a good example. I couldn't find the YouTube video, but when we make nuclear generators, you basically have these big things, and if you've seen Chernobyl, you can see how they go south. But there's these pieces of crazy metals like plutonium and whatnot, and they shake a bunch in water and creates this reaction, and then we capture that as energy. Now, when we suck most of the energy out of these things, we need to do something with them. And we're going to tie this back into learning to code, but we need to do something with them. So there's an area of the world where we store these things and it's in Norway. And so if we go over here and somewhere up in here, I forget where, but they decided that this is where there's some global federation or whatever it is that decided we're going to store the nuclear waste here because it doesn't have much. it has a very low chance that something bad could go on like an earthquake or something and they bury it inside of mountains over here. And when someone goes in basically to fill the mountain, they place it in concrete and then on the outside they have a sign and the sign is written in a language. It's basically written in like sign language. So, it's a bunch of people and they have little X's over their eyes and it's like this and it's like basically a symbol that says do not go here or something bad will happen. They did this in case of some crazy event, some catastrophic event where no one spoke the same language, but they still could understand through symbols what it is that's going on. And that's how I tie it back to learning to code is if you just go after the agent route, like you'll get some level of detail here. This probably gets you like 30% of the way there with some really good symbols and you can fill in more symbols as you go and you'll get more pattern recognition. But that's the same skills as pattern recognition versus if you actually knew the language, it'd be like a detailed list of like step by step by step of this is what's in there, do not go, this is why, this is what matters. But they didn't in this case assume that you knew the language. But if you know the language, why would you not? And that allows you to orchestrate better because you'll see as I'm doing this while making a YouTube video. I'm just orchestrating. And because I know how the thing works, I can then go and orchestrate. And that same thing applies with the planning process. So planning something would look very different if you've done the planning before. And making the architecture decisions would look very different if you've made architecture decisions before. So I still think in summary that you need to learn how to code. You also need to understand that you get a much higher multiple on the leverage of AI when you do know the basics. So let's say you get a 2x improvement on your 1x level of skill. I think that you can get a higher multiple on the higher amount of skill that you do have because you can just see around corners. Now let's assume that the models are perfect in a year from now. You've still been able to vault ahead of the competition. you've been able to learn more and then you've gotten better at systems thinking like pattern recognition, problem solving and these skills actually drift into other areas of how you operate. So it's very important they have multiple orders of effect on different areas and at the end of the day it's learning how to learn. This is the meta skill and you do that through trusted YouTube channels. So find some people that have done impressive stuff that have proof that they know what they're talking about. find repositories of examples and then at the end of the day read the docs like so many of the times you will ask an LLM for something it was trained and released in it was trained months before its release date its release date was in March and then you're using something from April it's going to be wrong and you can prompt around that a little bit by providing it maybe something like this where inside of this LLM's folder I have all the docs from tenstack router all the docs from tenstack start and then an lmtxt for python But at the end of the day, you still need to know in your own brain, in your own context, how these things work. Another recommendation, just swap random entertainment for dev tamement. And there's channels like Primagen. He's really funny. He's got infinite content. You can just throw that on the background and yeah, it's just this is the path and you recognize areas where you're weaken and you go after them. So, I'm doing just in time learning with systems architecture. And I'm actually going through a program that someone has for when you're interviewing for a position at a company as an engineer. You need to know all these different skills. So, he's able to break them down. I can take this as a screenshot and then ask an LLM. So, it's asking and that's how to do it. That's what I would do. I would do just in time learning. I would be using ask mode all the time. I would also surround myself with people whether it's in the Discord community like we run with this Val thing who's a co-pilot that we're training or just people at the workplace that are better than you. Surround yourself with that. Push yourself and don't be afraid of struggling. That's the sauce. If you like this video, make sure you give it a If you want more content like this, you can subscribe to the channel. You can also check out VI. It's in the description. and I'll see you in the next one.