AI Coding Sucks | Prime Reacts

For those that don't know, this is Coding Garden. Awesome streamer on Twitch. He's obviously on the Syntax FM podcast. If you don't know them, great little channel right here. You can go check this out >> about my thoughts and uh you can tell me what your thoughts are. >> Also, this title is called AI Coding Sucks. And I just want to let you guys know I am going to do my first best effort I could possibly do um to try to become the best at cursor and truly really feel out the AI. I'm I'm I'm asking questions. I'm learning how to write my own rules. Okay, I'm I'm trying to go deep here. I really am trying to learn to use AI the best possible way I can possibly use AI. And this video title I feel akin to currently, but I am willing I'm willing to try. Okay, sorry in the comments whether you agree, disagree, whether you're feeling the same way, etc. But what's up? CJ here. Got a little bit of a different video today. I'm just going to rant. I'm just going to talk about my thought. >> Can I pause really quickly? this syntax TV in the background that's like a cathode ray tube or appears to be a cathode ray tube. This is actually like super badass. I whatever that TV is, I want that. That is super cool thoughts. And uh you can tell me what your thoughts are in the comments, whether you agree, disagree, whether you're feeling the same way, etc. But I used to enjoy programming. Now, my days are typically spent going back and forth with an LLM and pretty often yelling at it or telling it that it's doing the wrong thing and getting mad that it didn't do what I asked it to to begin with. Um, and part of enjoying programming for me was enjoying the little wins, right? You would work really hard to make make something, build something or to fix a bug or to figure something out. And once you figured it out, you'd have that little win. you'd get that dopamine hit and you'd feel good about yourself and you could keep >> How many of you have that experience right now? How many of you have that experience? Because it's something that I've been fairly highly resistant to. I don't I just I just hand program everything and it it feels much better because every time I've tried to do it the other way, I always feel really uh frustrated because I don't because I'm just like no mistakes. I like both. I I feel like maybe I could like both. I mean, I get that dopamine hit. You get the dopamine hit. Yeah. I mean, the hard part is that I feel like I'm always going to be a guy that wants to just hammer on a keyboard, even when it's not as efficient. That's why I've really enjoyed inline AI completion because I don't care about like I don't I don't need to type the for loop, but I like to structure everything manually. But hey, I'm always willing to learn. Again, this is why I'm setting up my own cursor rules. This is why I'm in here like giving everything I can trying to build up like this is exactly how you build an element in this game to be displayed on the screen. So therefore, I want you then I'm going to go through and test a whole bunch and try to create up a bunch of mock scenarios so that way maybe I can go really fast. And if I can go really fast, honestly, building the actual UI is not the thing I'm concerned about. I think I think people get too uh what's it called? They get too dogmatic about their approach. The at the end of the day, your goal is to enjoy your craft and get better at it. And if you're not achieving those two things, then you're doing something wrong. Okay? So if you can use AI and get better at what you're doing, then knock yourself out. But if you're not enjoying what you're building and you don't feel like you're getting any better for it, better at it, you're like building you you are setting yourself up for long-term heartaches, right? It it it trust me, it's actually going to be really really hard in a year when you've continued to trudge through difficult concept after difficult concept while feeling like you're never learning while constantly feeling like you're never actually deriving any sort of enjoyment out of it. It's going to be hard. As simple as that. going. Now, I don't get that when I'm using LLMs to write code. Um, essentially once it's figured something out, I don't feel like I did any work to get there and then I'm just mad that it's doing the wrong thing. And then we go through this back and forth cycle and it's not fun. It's not fun at all. One of the reasons I like programming is because it is predictable. It's logical. It is knowable. You can look into the documentation for something or look under the hood and look into the source code or decompile something or watch network traffic. Like you can figure things out. And I love, by the way, I love where this argument is going. This has been this has actually been one of the hardest transitions I've had from like I don't mind inline completion because inline completion is guessing what you're going to do in this specific line based on the context and everything around it which is largely correct just like the weatherman who's predicting one hour ahead for weather but like more larger swast of generated code I always just dislike that because even though it's close you don't like you can say the same thing twice in a row and you'll get two different answers. I don't like the stochastic nature of statistical prediction models. It is very bothersome for me even though I want to get better at it. Right? I want to become the best at it so that when I say it's the piece of crap, I still think it's a piece of crap. Right? Like I want to, you know, I'm fine researching the thing I don't like until I get the best at it. But I totally understand. >> Once you have a good idea of how a programming language works or how a system works or how an app works, you can be sure that it's going to work that same way the next time you look at it. And that's because uh computers are logical systems. Programming languages are are logical formal logical languages. And that works really well with my brain. Now, when we're working with AI and LLMs, it's not predictable, right? You can use the exact same prompt and get a different response every single time. >> Re-watched, obviously pre-watched, book them, Dano. I've already already seen it. >> I think this is where some of my frustration is coming from because I am trying to do the same thing. I'm trying to develop workflows and be a prompt engineer or a context engineer, but doing the exact same things is producing different results. And honestly, that's not what I signed up for, right? And you could chalk it up to skill issue, but just just look look at the look at the evidence, right? So, if if you're chronically online like I am and you're watching all of these these tweets that come out from people and posts that are just talking about, oh, you have to write this specific prompt or use this specific workflow and it'll start working. And if you're not doing it, then by the way, I think that this is actually a sign of an early adopter. So, there's this whole argument going on right now in AI, which is super stupid, which is like if you don't learn now, you're going to be left behind. Well, guess what? When I don't learn now, I also don't have to learn a bunch of these stupid niche things you should say, right? Let me give you a quick example. Did you know that back in the day with JavaScript, you would not create a new array? You would set your current arrays length to zero because by mutating the length, it would actually empty the items in the list. This was the fastest way to get a new array. You would never create a new one because you would actually have visible lag on the website if you had to do this a whole bunch. And so, as a programming technique, we would do this. Now, you probably never knew that because you were not part of the old JavaScript. Did you need to know that to learn JavaScript today? Absolutely not. In fact, it would be counterintuitive. In fact, I bet you it's probably even slower than doing it by creating a new array. And so, this whole idea that if you don't learn it now, you're going to be left behind. And oh, but also, you have to do these incantations to make the AI act correct. Like, I shouldn't have to say make no mistakes, and I shouldn't have to say be secure for it to make no mistakes and be secure. Isn't that right, dog? is never. That's right. You know the facts. You know the facts. Tell me the facts. >> Um it's true. There's just no way that you should have to learn these things. And so this is the early adopter tax. I know you're meant to get some sort of theoretical increase in your speed, but the reality is that you're probably also experiencing a theoretical slowdown, a cost of importing this new technology because you have to do a bunch of these weird like magicians moves behind the curtain. Don't mind the man behind the curtain. Trust me, this is actually going to work. Okay, now this is what you say. You say no mistakes, but you got to say it at the end. And then at the beginning, you got to like say it this way because if you're like too positive, it's like it just doesn't work. But if you're negative, it feels like it can't do it. So you got to like really make sure that it know like what the hell are you saying? That's what you shut the hell up. Shut up. Also, I do want to say one more thing. If I see one more person say they've gotten 100x improvement from using an AI, right? like you actually are in like the 70 to 90 IQ, 70 to 60 to 80 IQ range. Like nothing can improve you 100x. You have to understand that. That means 3 and 1/2 days of you on an AI is equivalent of you fulltime for a year. If you can't produce what an AI can do in a three three and a half days in a year, you're either lying to yourself, you're lying to me, or you're an actual walking buffoon. like a legitimate someone who needs you're feeble-minded, I believe, is the technical IQ term for it. You are feeble. That's it. It's ridiculous. Absolutely ridiculous. And it's a skill issue. I've tried it. I've tried so many different things. I found things that have sort of worked, but then they stop working or I've been working with in a specific model like GBT40 or GBT 5 and all of a sudden I'm getting different outputs, right? Because I'm not in control of that LLM. It's a It's a magic box hosted in the cloud that can change at any moment, but it's always called the same thing. GPT5 or codeex whatever or sonnet 4.5, right? They put this little name on it, but that doesn't mean it's going to be the same every time. They're constantly tweaking the model, which means you're going to get different outputs for the things that you're running. Um, >> true. No, >> you can actually visibly see his frustration in that. And I feel bad for that because it is very annoying. Like that is super annoying that you have to like learn the new magic phrase that helps juice the AI to do it correctly. Oh, you're using too much context. Not enough context. How much context? Well, context has changed. Oh, with the newer models, you want to be at this point in context. You're at this point. That sucks. Wrong wrong answer, buddy. Just like, dang, I don't want to live that life, you know? >> Um, I'm looking at my notes cuz I'm trying to rant, but I have I have a lot that I want to say. >> Um, I have a note here. I mean, I don't know how how you're going to agree with this, but it kind of feels like AI based programming has become kind of like a religion, right? You have all these these big popular people on Twitter. They're kind of like the the religious heads that come up with ways of working and tools that you should be using. And if you follow their their patterns and you follow their tools, then magically your code will start to work. And if you use these specific incantations or or prompts, you'll get specific types of outputs. But it it's just not predictable. You don't get those same outputs. Um, and people chalk it up to skill issue, but I've worked with it long enough and tried it long enough that um I'm I'm kind of done. And to prove it to you, I want to just go through all of the things that I've figured out and all of the things that I do when I'm working with AI to write programs because you can tell me, am I missing something? And also, maybe you could learn something because I've I've been spending a lot of time uh with these tools and trying to figure out workflows. So, >> now I've only read the headlines. I haven't gone through the uh multiple 100page report of the Dora report, but it's a Google funded report that's like modern software engineering, which I largely dislike. Normally they have a bunch of findings that I find to kind of be dumb but nonetheless they've recently at least in their I believe for their 2024 year to 2025 year uh setup has said that software has become more unreliable and it's become slower to ship since the introduction of AI. I believe that is what the headliner is. Now I haven't read the studies. I don't know how much of that is actually true. Blah blah blah blah blah. Don't hate the person that's just reading the headlines. I'll read the whole report at some point, but you get the idea. That's kind of crazy, right? Cuz I was promised efficiency, yet there's reports saying that the the rainbow might not be as much gold at the end, right? There is this kind of religious nature to um to LLM. There's people that swear by it, right? Like you'll you'll meet like uh Claude 45 maxis versus Gypy 5 maxis, right? They're like, "Oh, no, you got to use this one. Oh, no, you got to use that one. No, that's the one. Oh, you're using that one. Oh, gross. Ew. Right. And it's just very funny to watch this kind of back and forth of which one is actually better, but it's all based on like someone saying it's so much better. They're like, "Dude, it's so much better." Like, "Was that even me? How do we even I don't have an objective measurement for this?" Here's what I know. If you're working with Claude or any other editor, um, or so like Cloud is a CLI tool, you could use an editor like cursor or whatever else. They typically have some kind of markdown file you can put a bunch of information in. So with claude, it's cloud.md. You can also use agents.md and cursor. They're cursor rule files. But basically the idea with these files is you can put a bunch of things in there that try to make the work with this AI more predictable, right? You can talk about what tools do you use, what p. >> By the way, this is one of those funny moments. You know that meme that is like I wish AI would uh do my laundry, but instead it does my art. You know that post about like I wish AI did the things I didn't want to do. Instead AI does all the things that I want to do. Isn't it funny that when you're setting up rules, we've just been tricked into writing documentation. So, we get to write documentation. We don't get to write code anymore. Like that again, it's like all the good parts are going away and the parts I don't want to do. It's just like, what the hell is this? What? How did I get tricked into doing it? Just do my laundry already. Okay. Do my laundry and make me a sandwich, because I want to code. I I want to dance and I don't want to go home, okay? Hey, I want to be teach. That's it. >> Patterns do you use? What uh libraries do you use? What dev commands there are? What's the architecture of the app? You can put all of this in these files and AI is supposed to use that to steer it in the right direction when it's working on your app. >> Y >> um and also AI tools can actually automatically generate these. So you don't even have to work hard to build out these files. It can generate them. You can validate them. And now in an ideal world, you have told the AI how it should work and what it should do. In my experience, it doesn't it doesn't listen to what I've put in those files. Um, and I think this comes down to context bloat, which is a whole other thing. But I have worked so hard at crafting like the perfect claw.md and the perfect cursor rules. But still, every now and then the AI will drift and run commands I told it not to or do things in a way that I told it not to. And I have no >> I just recently had Claude actually do the internet meme of deleting my test. I couldn't get something to pass a test. It's like, "Hey, fix it for you." And the diff was literally the test file deleted. This happened just a week and a half ago. Unironically, the meme that is like Twitter, I actually experienced. It was fantastic. I didn't believe it was real. Like I always thought that was like the joke. You know what I mean? Cuz it's too funny to be true. That's AI being based fix for this. So that's extremely frustrating. Um the other thing is when you're working with AI, uh don't ask it to just code without making a plan first. So this is my approach to building and and and working on things with AI is I'll ask it to build work uh I'll ask it to come up with a plan. So I'll say plan this out. This is what I want to build. This is what we want to do. Let's come up with a plan. And typically I tell it to write that into a markdown file. So plan.mmd because I don't want to lose that in the chat window. I want that to be basically a part of my codebase that describes what we're building and and what's coming next. What are the different phases? And AI is pretty good at coming up with this. And then I have to do the work of validating it and making sure that it's a good plan. And then from there, >> again, tricked into reading documentation and now specifications. Damn, crazy. >> There um don't just immediately say, "Okay, the plan is good." You have obviously have to approve it and then gradually work through the plan. You wouldn't just say, "Okay, now go implement this 30-step plan." You would work on one step at a time. You would tell the AI to work on a specific phase. You'd validate its output, and then you'd move. I guess we're all QA and pro project managers now. All for the pursuit faster. All for the pursuit of moving faster. I love it. Okay. Anyway, sorry. This is just me. Hey, that's me being negative. I'm sorry. I'm sorry. I'm sure it's really great. I'm sure being a product manager or a project manager is really fun. >> I've done this for lots of apps. Um, and it it works for some, but again, every now and then it starts to drift, right? This the same prompt gives me a different output or the rules that I gave it, it starts to drift from it. It doesn't follow those anymore. Um, and there are there are tools that have been built that allow you to work in this way specifically, right? So, Kira was one of the first to popularize it, but GitHub has come out with I think it's called like SpecKit. Uh, you might see on the channel that um Scott is actually working on his own list of uh open code commands and agents that kind of like give you this more structured flow for working with AI. And I think it's like it's uh it's it it basically makes you feel like you're doing the right thing because you're gradually working through steps. But there are so many places where an AI could do the wrong thing and you don't realize it because you come become so accustomed to kind of just like accepting the output or or not completely validating everything. And so for me, specri flows are decent, but they're not perfect. And I've gotten better results by just writing it myself and not having to depend on >> I bet you there's a number that exists that's like what's the average amount of code or lines of code someone can review over the frequency of bugs that get in. I bet you that that there's a graph out there that exists that's just like once you exceed a certain amount of of like lines of code in a period of time, you're just going to be like less and less good at seeing bugs. Almost every time I let Super Maven generate more than 10 lines, I discover a bug in it later. Yes, I typically only let it do uh what's it called? Uh one line at a time. I do not like multi-line autocomplete. Uh PR blindness has to exist. Yes, it does have to exist. And I think one of the big reasons why is that when you look at code, you can structurally look at it, but the thing that makes code hard is the shape of this is where functional bros actually have like a real good argument is that shape of the data is a very important role. And knowing how the shape of the data like moves over time is really hard to have in your head. You almost always have to debug, right? like you you have to either write the code yourself or you have to debug to really have that like strong sense of what the actual stuff is going on. Yeah, Elixir mentioned why this is why sometimes I sit there and think maybe functional bros do have a real point here which is that you can just look at the ins and outs of that specific function and judge its correctness whereas you don't have to have near nearly as much in a mutationy world versus a clone uh immutable world. I don't know just hey just throwing that out there just thinking of it. You got it. I know. Hey, I re hey real recognizes real in certain places. The only problem about functional code is when they do wait dude. When they do finally ship something for real, man, industry is going to collapse. We're all going to be functional bros. But that's going to be a while, but I'm sure it's going to happen. I cannot wait for that first product. It's going to be incredible. We're all going to be high-fiving the AI. The other thing is you typically work on the smallest feature possible, right? Don't tell AI to go off and build your entire app with a single prompt. you're you're prompting it with very specific things that it should do and not giving not allowing it to do big full sweeping changes. Right? So I've I've worked with AI this way a lot as well. Uh and then the other thing is getting AI to validate itself. So right so AI can write tests. So you have it write tests so that way it can run the test to make sure that the code that it just wrote is actually work. I don't know if I like that one. Right? I feel like that one is uh is not good. Like I feel like if I were to do this appro approach, I'd want a new chat window in which I say here's the signature of the function, here's its expectation or here's the interface. Write tests for that. So it's it's writing tests completely abstracted away from the implementation because you're just going to get tests that like prove the implementation. You're like, dude, yeah, it is what I wrote it to be. Like it perfectly is exactly what I I said. Like I don't know if that's actually true. Is that like true? But I I I probably should try more of this testing side. It is very curious to play around with this because I h I've largely distrusted it. I always I always write the test myself except for one time I didn't and then it was wrong and I had to rewrite the thing anyways. Still upset about it, but let's just let's just ignore it for now. Working. Uh it can also do interactive debugging. So I have MCP servers set up that allow it to launch a web browser. So I use the Playright MCP. So when it's building out a web app, it can actually after implementing a feature launch the browser, click on buttons, and make sure that the feature worked as it expected. However, a lot of these AI models are typically like they're they're goal seeking and their goal is to get things working, which means they will write tests that are not logically sound or they'll write simple tests that actually skirt specific edge cases so that the tests will pass. Or the worst thing that I cannot get AI to stop doing is when it's writing TypeScript code, if it can't get the TypeScript to work, it'll just put in the any type and be like, "Oh, we'll fix that later." Or the other thing statistical reflective human behavior. Let's go. Let's go. Answer to all things. Put an any on it. Who is it? Billy Mazize that slaps it. Who? I forget who it was that does the whole slapping of the Annie on there. But yeah, I mean, great. We created a model off of Reddit that acts like that reflects human behavior. Shocking that when something doesn't work a couple times, it slaps an N on it. Like that should that good good on the AI. Well done. >> It'll do is it'll comment if it can't get a a test to pass, it'll comment out that test and be like, "Oh, I couldn't get the test passing, but I commented it out and now everything's working." And it just drives me insane. But I think it's because these models are are like too highly tuned to be goal seeking, right? They're not actually trying to figure out the problem or write really good tests. I think the word he's looking for is they don't reduce entropy. And I'm not sure if that's even a property of LLMs to begin with. I don't know if they can reduce entropy. Meaning that when like a very easy and simple example of this is if you have a function that takes in one argument and then you go okay here let's change things up. I want to take in this other type and do this other thing. It will do it but it will put that argument as the second argument in the function. It won't go wait a second all other functions in this file look you know it won't like judge it in in in the grand thing or it won't say hey this function this item is more important it should be the first item in the function it won't think about the taste of the function like hey if I were to make this function what would feel better this one first or this one first right like it doesn't do those kind of operations in its head it just simply does the thing right so it doesn't it doesn't reduce entropy it often introduces something where you're like okay things are more falling apart we're just layering on linear linearly the solution to this problem which is going to drive me nuts It doesn't, you know, it's not going to get to the root cause and figure out the right thing. It's just going to do the thing that makes the thing go away. Which makes perfect sense. The thing going like, and plus that's just like what that's what people pay you to be an engineer for is that you don't just look at an error and solve the error. You look at the system and you figure out why the system's producing the error and how you best like solve that. That makes perfect sense. >> Or actually solve these things. They're just trying to get to a point where they can stop running and say, "Hey, I'm done." And uh it's very frustrating. Um, and then the other thing you do is agentic workloads. This is the new hotness, right? Everybody's writing their own custom agents and I tried this out in cloud code. They were one of the first to add agents and you can basically try to solve this uh context problem, right? So you can have cloud.mmd, but sometimes cloud.md might get too big or your cursor rules you might have too many. But with agent files, you can create specific agents like a UI expert or a test expert or a database expert. And each one of those agents can have its own specific description and can can also have notes about architecture in the app or how it should work or what tools it should use for those specific types of things. And that makes it so that sometimes or for the most part it'll try to do the right thing because it's not trying to cram the entire list of cursor rules or the entire cloud.md into the prompt. It's just the prompt for those specific agents. And that seems to work okay. But again, I run into issues where the agents are running the wrong commands or not writing good tests. Um, and it just hasn't worked for me. Um, again, >> bro's a victim of statistics. I mean, that's that's truly the big problem. If it's like, hey, don't do these things. Only use this thing. Well, it's like, well, when solving this problem on the internet, 10% of the time people use this. And unlucky for you, this is that 10% of the time. Time to use this guy. Here we go. Here we go. I've also ran into this where I'm like, hey, don't run the tests for this fix. And it's just like, I'm running the test for the fix. I'm like, don't run the test for this fix. We do not run the test for this fix. I'm running the test. Stop. Okay, that command doesn't even exist. Stop trying to run Python. There's no Python in here. Just do the thing you're supposed to do. Well, statistically speaking, everybody runs Python for this command. So, here's some Python. It's just like, stop giving me Pyth I don't want Python. I don't want Python. Just stop. I'm going to freak out. I'm trying not to yell at Python because it's just a programming language. I'm not going to get angry at it. But right now, every time I see Python, I have this internal burning desire of upsetness because it's it's it's like symptomatic of this problem that I keep running into. >> You could chalk it up to a skill issue. >> Yes, I've had it like I've tried it and and they really don't work that well. >> Um, so that these are all the the things that I I do to try and make AI work. Um, there's probably other stuff that I missed here. If you do other things that I haven't listed, please let me know in the comments because this this is the current knowledge and state of things as I know it, right? this is how we should be working with AI. >> I bet you there will be comments something along the lines of you got to make it do smaller things. Scale issues. The problem is and I mean it actually sounds like he is trying to do a really good job. I mean it's very hard to convey everything you need that you've been doing in a 15-minute talk, but it sounds like he is actually genuinely trying. It's also not surprising. >> Personally, it just doesn't work as good as they say it should and I and I don't enjoy it. So that's that's my main point there. Um >> I love that line, by the way. >> And so my other point here is okay, I >> I do want to pause one more time. The enjoying it part is like such an important part, especially if you're new into programming. If you're not deriving joy, like man, this is going to it's it's a hard road. It's genuinely a hard road if you can't figure out how to make the activity you're doing joyful because you're going to be sitting in front like to honestly to become good at programming to understand a bunch of concepts and to actually have a command over many things is going to require 60 plus hours a week for many years. Yeah, you can do 40 hours a week for even more years, but it takes a lot. There's a lot of just hands-on keyboard time to really become good at something. And that it and if you don't like it, like that's hard. That's super super hard. And no amount of AI is going to make you better if it's doing it for you, right? Like that's that's the hard part. And so learning the joy and targeting that and figuring out what makes you happy is very important in this act in this version. You know, just railed off rallied off all this stuff that I know about AI and all the things I've been doing to work with it to try to get it to work. And typically what you would see online is if you haven't been using these tools and you haven't been trying them out and trying these specific flows, you're falling behind. you're going to become a basically uh useless developer that's not up to date on the times and that like can't use these tools and can't work in the right way. >> Yeah, it probably could also like write a for loop by hand. Oh, gross. Dude, these developers, they probably even know how functions work. >> And I disagree. I disagree. All of the stuff that I just listed off, I could learn in a week or less, right? Because as programmers, as developers, we spend our entire careers figuring things out, right? It's literally part of our job and all of the stuff that I just listed are things that you could figure out. It's not something that you need to spend a year with. I'm confident that I could spend a week with this stuff and still be getting the same results that I'm getting now. So, the argument that if you don't stay up to date on this stuff or if you're not using these flows, you're going to fall behind, I think is invalid because of who we are. Now, if you're completely new to programming, I don't have any advice for you. I don't know what to say because you you kind of actually I do have advice. My advice is learn to actually program. Don't just try to vibe things because you're going to hit walls where the AI can't figure it out and then you're not going to be able to figure out either. Um, >> not only that, but the hard part is when you start when you build something as a program, you start with one line of code and you slowly expand from there, right? You kind of slowly get to the point of understanding and you're slowly walking through it. But when you vibe code something and it breaks, you're not handed one line of code. You're handed hundreds if not thousands. And now you're like, it's broke. You don't know what you're doing. And there's a lot of corpus in this document and it's likely not necessarily made in the most friendly way for humans to read. So you have like this huge you have like every single possibility pulling against you that just makes it super hard to like have a successful time. So it is I I'm on this guy's team. Like if you're going to learn to program, learn to program. Turn off the AI. Just learn it. It takes time. Not easy. I wouldn't even necessarily say it's fun just to like actually just learn the syntax. But once you do, it's locked away. It's in there because it's it's the semantic meaning of it doesn't necessarily have to translate into the actual syntax the syntax of it. You understand how if statement or for loop or a function or all these things all work together and it becomes easier and easier to reason about larger and larger systems. >> And the thing is I rarely hit those walls because when AI can't figure it out, I figure it out because I've been doing this for a long time. So yeah, that's my advice for any of you that are new to the field is just >> Oh, one more pause. I know I'm pausing way too much. Uh there was an article. It was uh not too long ago. We had the creator of it on a while back. It was about GitHub and AI usage and I can't remember but there's like this rule or this law or something from a book. Maybe someone can remind me from Twitch chat. But it effectively goes like this. Like I believe it's from like Plato or Socrates or some old guy. I don't remember. You know what I'm talking about old guy. uh where it is it's something like this that when you're doing a very large big picture task if you're not doing the small iterative activities you actually stop also being good at the big task and so you you end up making more and more mistakes because you're not focused on the small things. That's true in all fields. Exactly. It's true in all fields. And so if you give up the the task of writing code, you are more likely to not see bugs when you're reviewing code because over the time you've lost that edge, that practice. So I don't even believe you can long-term have great skill in programming and then only review code. And I think that effectively I think that whatever that is, whatever is the proper term where through long-term removal of the basic activities, it's hard also to be good at the big activities. learn the field without AI because that's what the rest of us did and uh it it works. It works. Yeah. Um another note I have here is everything's just competing for your attention and your money. If you spend any time on X, there is a brand new model launching every single day. There's a brand new coding tool. We did an AI coding tool tier list here on the channel and there's just so many comments mentioning like, "Oh, you didn't mention this tool or you didn't mention this tool." There's literally 10 dozens dozens of of tools that you should be using to uh build out your stuff. And they're all marketed in a way, well, if you use this tool, it's going to work exactly as you need it. And I've tried I've tried a lot of the tools. I really have. They're all just rappers. >> They're just GPT5 or Claude 45 or just something. They're not even all these tools claiming some sort of supremacy. Most of them aren't some sort of alternative model. I mean, that's what makes Super Maven actually good is that it's not just, oh, it's co-pilot, but it's co-pilot delivered by another company. No, it's like, no, it's actually just something else. And it's actually really good. were the foundational model models released by OpenAI, Anthropic and Google and then um a few others. There's like Quinn and and like the the smaller companies but basically they built the models and everyone is just building tools that are wrappers around those models and of course they all have slightly different agentic flows. They have all slightly different tools built in, slightly different planning tools, and that might give you a slight edge when you're working on your stuff, but it's not that much of an edge. And at the end of the day, all of these tools are are iterating so fast that they're all kind of reaching feature parity and informing each other on what they should be adding. >> He's he's he's right. You know, you may not like that, but he's absolutely right. A lot of these people claiming this one is much better than this other one, you're really just you're lying to yourself. You are absolutely lying to yourself, right? There's a lot of edging, dude. just just lots of edging. I know I'm good at stopping frames at the best possible time. >> And all of that to say, use the tools that work. So, even with all these tools that have been coming out, I'm basically just stuck to cloud code and cursor. I definitely want to give open code a try. But those are the ones >> based DAX. Did we just hear DAX >> that I've stuck to and but please tell me if I would have used any of the other tools, would I still have run into the problems that I talked about at the beginning of this video? Maybe not, but probably, right? Because they're all based on the same models. Okay, so I I think I'm done. Um, I think I'll end this with uh what what am I going to do? What are you going to do? And I have decided that I am going to take a one month break from AI coding tools. >> Nice. >> I am going to spend a month where I write the code and I make the plans and basically just go back to how I was doing things two three years ago. >> I I do I so just a recommendation as someone who still does that full-time. Uh even though I'm going the opposite. I'm actually trying to give a month to use these tools and become really good at it. He's going the opposite way. Uh I would highly recommend uh at least Super Maven just because Super Maven of often is really really small, but it's really really fast. Did you see how fast that auto completion is? That auto completion was incredibly fast. Unlike unlike other ones that are just super slow, this one at least goes fast enough. Right? Like that's what I want. If I'm going to have something, I want something that when I write it, I want to be able to at least have something that's pretty quick. It makes me feel a little bit better to be, "Oh, okay. Yeah, okay." And then you can do one thing at a time, right? It makes me feel at least much much better than uh anything else. And for me, it's this I come down to this very simple thing, which is I don't want to keep writing logging statements by hand. It's really really annoying. It's really really nice to have logging done for you. Like that's a very nice thing. And so I love it. I love AI inline complete because I don't need to write the for loop. I know how to do it and every now and then I turn it off. I turn it off usually like one day a week just to kind of remind myself of all the things. And I think that that's pretty good because I don't want to lose that edge. Um and uh we'll see how it works out. Um and ultimately I'm going back to what I enjoyed, right? Because I talked about it at the beginning of the video. That's what I enjoy about programming. That's what I enjoy about >> this career and uh this job. And have a hot take for him which he may find that he doesn't like it. Sometimes when you go back home, home's no longer home. You know, Froto, just think about our boy Frodo. Home was never home again. It's true though. It's true. Sometimes you find that you really enjoyed an activity, but due to time and you getting older, your hormones changing, your position in life changes, all sorts of different things change. You go back to something that you used to love. And when you do it, you find that you don't really love it anymore. Like, you know, I used to love playing video games all day. I could play Halo for like 80 hours a week. I can't play Halo. Like I I would die. I would die playing any video game 80 hours a week. You know, it's just that thing like you can't I remember the love, but I can't make the love happen again. And that's one of the dangerous parts is that this this one thing that he's talking about. He's like just programming like how I used to do it two three years ago. Like what if brother you go back and there's no more love? What then? You don't like how it currently is. And what happen if you don't like how it was? So that's why that's why I I do say this phrase quite a bit which is like you got to seek joy. You have to really like look after it because it's not just where it has always been. It's not just you don't just run into it. It's a it's a it's a continuous activity and it also takes you opting into it. Meaning that every activity every time you build something you can look at it as this is going to be a good experience and here's the reasons why or this is going to be a bad experience here's the reasons why. And I find that a lot of times it's pretty easy to look at like your next task as doing something that's going to be inconvenient. Uh, I don't like doing all that kind of stuff. And that's totally real. A lot of people do that. Yes, I I actually have it enabled right now. Uh, 3D factor. And it can be really it can be really uh it can you can decide a lot in how you perceive goodness. You have more autonomy than you realize. I've never grown out of liking big booty Latinas. Oh, pick. Thanks for that insightful and amazing addition to this conversation. and um maybe it'll make me feel a little bit little bit better because I'm not yelling at an LLM every day. Um now I do realize that I'm a very I'm in a very privileged position, right? So I work at a company that does not force me to use AI and if anything it encourages me to explore different AIs or different ways of doing things and I I basically have complete and total freedom. By the way, I feel bad for the people that work. What was it? Was it Shopify? Was it really Toby? I don't remember if it was Toby or not. Uh, it was someplace where they're just like, "If you're not using it, get the hell out of here." Crazy talk, by the way. Absolutely insane talk. >> I realize that you out there are being forced to use AI by your work. >> Um, and you you kind of have no choice. >> So, for that, I I don't know what to tell you. I I think maybe you could pretend like you're using AI, but then maybe go back go back to >> I don't think you can, honestly. If everything starts using AI and and people just aren't writing code, you're just going to generally get shittier code. And you trying to fight that's going to be impossible, right? It's going to be super hard to fight against that. And eventually, like, you just have to give in. You just have to say, "All right, this file sucks. How it's set up sucks. I don't want to have to learn everything just to get to this point." Say, "Uncle, mercy. Here you go." And prompt. Okay. Those are the snake eyes. Okay. and prompt >> doing the things you way the way you were doing them before. And I also realize that not everyone is building the same kinds of apps I'm building or running into the same issues, but but that's potentially because we're building different things. And so, um, you can have your complaints about what I talked about, but I would say that if you're trying to build the same things that I'm building, you would be running into the same issues. And I'm very confident in that. Um, okay. >> He's very right, by the way. Like if you're working at a uh a like marketing firm that all they do is like build static sites or kind of interesting little interactive sites for various companies, that seems like a ripe place for vibe coding to just destroy because you don't need to maintain these. These aren't long-term things, okay? They're not going to keep growing with features. They're just simply something that exists and you just need to mash together to meet requirements. The end, right? Like that to me seems like a that that to me seems like a perfect place for it, right? Fire and forget landing pages. Exactly. like these these are real things. Uh even though I would personally argue that landing pages are very very important and you should actually take a significant amount of time on landing pages cuz most landing pages look the same and they're utter horseshit. They're confusing. They have a bunch of big images. When you scroll changes all over the screen. There's not an HRB and everybody knows HRBs drive sales. Like if you don't HRB, you're not down with me. Okay? HRBs huge red buttons. Okay? It's been this way for literally forever. Like anyone that thinks they can be cutesy on a landing page and they don't do an HRB, you're just like you're just losing sales. Okay, call to action dead center. Make it obvious. What do you want the person to do? What are you trying to say? You know, it's crazy. Anyways, that's all I got. Uh I think I'll check back in a month and let you know how things went. And um yeah, let me know if you like this kind of video. Uh my plan is to not edit it. You'll see if there are any cuts. I think it's just straight straight to the camera. >> No, that's perfect. >> That's all I got. Let me know what you think and uh I'll catch you in the next one. >> Honestly, my most successful videos now are no longer live react videos, but me just talking about a topic unbroken pretty much. It is it's it's crazy how much better people like this this type of format quite a bit. It's much more genuine, too, because you you see the person think, you see the person kind of mle through thoughts. It's not so quick edited. Uh even this one right here, this one, even though it says 30,000, it's it's it's over 70,000 for views. Uh, this one right here, it's crazy. It It's just doing great. And it's just mostly me yapping about how upset I am at Windows and Apple and why Mac is or why Linux, it might actually just be the year of the Linux desktop. It's crazy. I'm actually really Yeah. So, so am I. So, please CJ, I want in fact I know I'll reach out to you CJ, but I'll just say it out loud here, which is you should come on the show at some point and talk about your experience. Glad to have you on the standup. come on here and be like, I went one month after being a heavy AI user and went back to the old way. Here's what I've learned. Glad to have you on, ask you a bunch of questions. It'll be super fun. Anyways, hey, the name, it's good to learn from people, by the way. You know, I know there's a lot of people that are kind of upset in the audience. I actually get emails, but I actually get not only emails, but I also get Twitter DMs. I recently had this one guy, he has like hundreds of thousands of followers, and he whispered me and he was just like, "You know what you're doing." So, of course, at that point, I'm just like, "This guy's a big account, and he's talking to me like that. Oh, look. CJ said that. Nice. Hey, actual CJ says that. Yeah, CJ. This is a great video, by the way. Very relatable. Absolutely loved it. Super fantastic. Anyways, this guy was just like, "You know what you're doing." And I remember sitting there thinking, "What did I What did I do? Okay, you got me. What did I do?" So, I went in there. I jumped on. By the way, you see, look, I'm following Coding Garden. So, if you're not following Coding Garden, you're not a real one. Okay? Real ones wish list insignia on Steam and follow Coding Garden. That's it. Uh, but I remember him doing this. So then of course I was just like, "Okay, what what am I doing?" He's just like, "You're teaching people that AI is not the only future and you are stealing away their careers forever." And I'm just like, "Oh, crazy, man. You have like hundreds of thousands of followers and you think like this, dude. Bro, was was 911 an inside job? Like where where we at? Where we at? Like how much Tylenol we talking? We talking like migraines or just the occasional headache? like where how did you land here? It was crazy. And so anyways, that's that. So I realized like there's a lot of people in the audience they they get really upset at this kind of talk because there's this like I don't know what it is, you know, like you said CJ, it's kind of like in this like religious kind of feel where people it's almost like if you aren't doing this thing then I'm like assaulting somebody in the spiritual realm for leading them astray. And it's like it's it's not true. I'm not trying to assault you in the spiritual realm and leading you astray. Yeah, I'm not taking you off the straight and narrow here. I'm just simply saying my experience with it versus what I can do feel like two different things. And I don't love it. I wish I could love it. I'm not necessarily only about typing on a keyboard. I enjoy mashing all the old dim keys and all that, but that doesn't necessarily mean I want to do it all the time. I would love it if I could just be like, make the layout for me. Okay, I want to try this. Okay, I don't like that. I'm going to do some programming. I want to kind of adjust some things. Okay, now do it this way. Okay, cool. Right. Like I don't mind that. It feels good. In fact, that's partially if you guys remember. Uh actually, let's get out of beacons. Actually, let's just see what happens if we go to beacons. Let's see if this even runs. Let's see if beacon is even any good. Okay. Okay, it still runs. If you remember bullet time in the game, right? Bullet time in the game right here where see how we're going here. I'll speed it up. So, we're going faster. Bullet time. See how it's going really, really slow. This was actually something that I started by vibe coding. I thought it was really cool and I thought the feel was super super good. So I was like, "Okay, I gotta figure out how to do this more because I really I really like this experience." And so that was done by a vibe coding to begin with, but then I rewrote everything and it was fantastic. So I'm not like against it, okay? I'm not against it. A Jan Somehow this was an hourong react. What the hell am I

https://twitch.tv/ThePrimeagen - I Stream 5 days a Week ## Original https://www.youtube.com/watch?v=0ZUkQF6boNg https://twitter.com/terminaldotshop - Order coffee over SSH! ssh terminal.shop ## Become A Great Backend Dev (I make courses for them) https://boot.dev/prime This is also the best way to support me is to support yourself becoming a better backend engineer. Discord: https://discord.gg/ThePrimeagen Great News? Want me to research and create video????: https://www.reddit.com/r/ThePrimeagen Kinesis Advantage 360: https://bit.ly/Prime-Kinesis