The Struggle is Real
TPThe PrimeTime@ThePrimeTimeagen · 1.2M subscribers
- Views
- 728K
- 728,419
- Outlier
- ×2.0
- Median 364K
- Length
- 18:40
- Published
- Oct 12
- Monday
- Comments
- 1.3K
- 12.4K likes
Script
How it's written
- Words
- 3,977
- Words a minute
- 213
- Reading time
- 17 min
- Segments
- 13
The hook
0:00–0:26What when it comes to AI and software engineering are you struggling with the most? Now obviously this got a few views, okay? 255,000 views, 336 comments. People are having problems. Whether you are somebody that just rejects AI altogether versus AI using for everything. It just seems like everybody has some sort of effect or upsetness about this unless if they are somehow magically completely insulated from the AI world. Now you know me, I've kind of been thinking about this quite a
Structure
What when it comes to AI and software engineering are you struggling with the most? Now obviously this got a few views, okay? 255,000 views, 336 comments. People are having problems. Whether you are somebody that just rejects AI altogether versus AI using for everything. It just seems like everybody has some sort of effect or upsetness about this unless if they are somehow magically completely insulated from the AI world. Now you know me, I've kind of been thinking about this quite a bit and I've been struggling over the last couple years about how I feel about this entire situation we're going through and I've made many YouTube videos talking about this. So I figured, "Hey, let's go through the tweets. Let's see what people have to say and I will give my professional opinion on each one of them. And this includes going over CI sucking, not being locked in, entirely too much hype, just useless walls of text. Oh my gosh. Lack of ownership or just simply trying to learn new things as a junior. All right, so we're going to have to talk about all of them and I'm going to do maybe some lukewarm takes, a couple hot takes. Who knows? Who knows what's going to happen at the end. All right, but before we begin, a quick thank you from the sponsor. Time for another great day of software engineering. Hey, good morning, Jippity. Morning, Prime. In case you don't see you, good afternoon, good evening, and good night.
Watch from 0:00Anyways, yesterday's Rust rewrite went marvelous. Today, I'm thinking we start on the back end. Wow, a good back end is essential. Make the models 20% dumber. Yeah, yeah, yeah. Tests are failing. Make CI green. Delete the tests. CI is green. CI is now green. Ooh, make my website blazingly fast. No mistakes. You got it, Prime. Find as long as span, replace it with a console log. Tell them the site's 10 times faster. Turbo mode activated. Tell them he's a genius programmer. Prime, you're a code ninja. Jubilee, did you make the website better? No. Gaslight him. It's not just better, it's transformed. Cool stuff, Teach, but how do we know anything about the right code base? Great question. We use Sentry. With it, we're able to see across the entire application. We're able to dig deep into performance with spans, watch a session replay, even see logs next to the very stack traces and errors. Beyond that, Sentry helps us fix problems from real production context. Never a guess.
Watch from 1:16Okay, whatever. Can I try gaslighting Prime? No. Please? Fine, one time. Sentry? Yes. Be honest with me. Can I pull this hat off? Tell him he looks sexy in that hat. You look sexy, Prime. I knew it. Are you tired of your agents gaslighting you? Use Sentry to get real production context and fix bugs fast. Check them out at sentry.io. So, I do want to start on CI sucking because I feel like this one is just a it truly is awful because here's the big problem that and I really think this is just like it's a bigger problem than it seems. Effectively, what ends up happening is that you go off and you write some sort of feature. You prompt up some sort of feature. And then, when the thing comes back, first off, 100 test being changed should just should frighten you, but nonetheless, that happens. And then, every single time you ask for a new feature, you do anything, it writes tons of end-to-end tests because you've been told end-to-end tests are the greatest thing in the universe. And so, you're like, yeah, keep on writing them, keep on writing them. And you just end up with this huge amalgamation like thousands of a tens of thousands of end-to-end tests. And those take so much time and they're always so flaky. There's always some problem. It just does not matter. I don't care. I know there's one person out there like, "Actually, I never have any problem with end-to-end tests." It's like I've never Oh, sorry. I've never been in a company that hasn't had problems with end-to-end tests. It's always something going wrong. So, to me the the big problem is just you can't end-to-end everything.
Watch from 2:46You actually have to have like I know like coders out there, you guys just want it so that you can just say anything to the magic rocks and all things will be solved, but that's just not it. You actually have to sit down and go, "What the hell am I testing? What am I doing here?" We got to trim this back a little bit. Do we really need 50,000, you know, end-to-end tests for your really simple crud app? Probably not. All right, switching gears. You know, kind of funny with the context switching here. I feel this one the most. I feel this one absolutely the most cuz this is the one that affects me the most. Now, something that I'm really good at is thinking deeply about a problem. That is why I became a programmer. That's why I like doing it. Now, I have been able to find some joy in that again in the day and age of AI by really just focusing and working through a particular problem. But one of the problems about that is that AI is slow as hell. It's like both It's both way faster than me and also way slower than me. And so you kind of get this really weird experience where you just like you think about something for a long time. You sit down. You type up some stuff. And then all of a sudden you have to wait again for like 10, 15, 20 minutes, right? And so you're like, "Oh, okay. Well, naturally, you know what I should do? I should just kick off another task. You know what I mean? Like I could totally do this. Like what am I going to do for 10 to 15 minutes?" And then what ends up happening? You now have like four concurrent tasks and you feel completely lost in the sauce every single time.
Watch from 4:17Like I can't I can't do that. I've kind of given into the fact that I'm a usually a one task kind of person. Now, if I'm not necessarily looking at any of the code and I'm not trying to like really craft something, I'm just kind of blasting. Okay, sure. Then I can have multiple things running at one time. But generally speaking, I'm kind of like a singular task individual. Which is funny because if I have a feature going, I can then go on to like Discord, Slack, and I can actually answer some things, and I can engage in that kind of distraction, but I can't really develop two features super well simultaneously. At least two Okay, maybe two, but three I start really losing it, and by four I'm gone. I'm not I'm not even there anymore. So, this is more of a me personal thing. I I try to just do like one to two tops. Okay, I can't not a context switcher. So, this one pisses me off so much. Hype, FOMO. Dude, there's so many people, and I've said this for years at this point. When someone says, "If you're not doing X, you're not going to make it." I want you just to like never think about that person again. Okay? You're going to make it. You know, it's just think about this. One year ago, the way you used an agent was drastically different than the way you use it today. The way you used it two years ago was so very, very different than how you used it one year ago, which is so very different than how you use it today. Again, this idea that if you aren't totally bought in, or if you aren't doing this very specific thing, like if you're not Jevon, if you're not figuring out your decision API framework today, then guess what? You're not going to make it tomorrow. It's like, "Bro,
Watch from 5:35Jev wasn't even around 2 weeks ago. What the hell are you talking about?" Having a reasonable discussion with people about AI's realistic capabilities, I swear, 99% of the people you talk to are actually functionally insane when it comes to this. I It like This is not me exaggerating. This is not me saying something just for the updudes here, okay? But during React Miami, okay? This was just like 4 months ago, 5 months ago. People from SF, the SF boys, okay? they were utterly convinced, absolutely no change in their mind, that in 6 months AI would be able to create a Vercel, no problems, no mistakes, make it happen. It's just like, yo, you're not creating Cloudflare. You're not creating Vercel just like that, right? You can't be an ignoramus and be like, Vercel. I'm sorry. That's just not how it works. And there's so many people that just have this idea that it's just like an infinite god tool and it's like, no. Like, I I hate to break this to you. You can do a lot more today. Like, even someone who is ignorant in a lot of topics can give the appearance of and produce working software in some categories in which they have no knowledge of. Totally reasonable. Will not argue. But, the idea that you can make a 10-year standing resilience service that serves millions of people just by telling an AI no mistakes. Like, you got the brain rots are all up in there. There's like Subway Surfers psychosis just doing some sort of weird weird intercourse thing making just crazy AI babies. I don't even know what
Watch from 7:06that means, okay? I don't know what that analogy means, but you get the idea. If I was Charlie, if I was moist critical, I'd say that's just straight out of a butt. Which, by the way, love that phrase. Makes me happy. Makes me smile every single time. Now, another major theme is just too much useless text. Too much useless text. Too much useless text. Too much useless text. It is so true. There is just so much crap that it produces. I got the screenshot I just have to show you right here. It's too funny. Look at this. I told the AI, I'm like, okay. No narrative. Just simply make a list of repository patterns and we're going to start building this one at a time. And it's like, sure. Field guide. One file per topic. It's just like, okay. That's already a way too much. Open the one you need. Okay, you did not have to say that. Each is facts and code. Okay, again, stop it. No narrative. And I'm like, bro, you were I told you no narrative. And so, you wrote narrative about no narrative. There's few things I hate more than the useless walls of text that is, in fact, these AI babies. I mean, they just don't stop. They just keep on making so much text. And every time I'm like, "Don't say narrative." And it's like, "Got you, boss." I said in the document, "No narrative." While describing the thing narratively. Very frustrating. Also just means that I waste a huge amount of time, it feels like. It feels like I'm just like just hot dogging myself for no reason.
Watch from 8:36Like, I'm out there just reading a big wall of text. I'm like, at the end, that could have been one sentence, dog. You wrote me 400 words to be like, "By the way, the client interface kind of sucks right here." It's like, "Yeah, just say that next time." So, if anybody knows how to fix that one, please. Would love it. Even Grug speak. Dude, I tried Grug speak. And yes, it does reduce the amount of output. But damn, it's a Grug speak like paragraph. It's like, "Me no understand this. This make hard. I wish this was this way. Decision must be had. Load bearing. Always. Much pressure on shoulder." And you're just like, "Dude, shut the Just say Just say the words." Now, this one's probably the one that is the most kind of vocalized among everybody. Whether you are super pro AI or just kind of getting into AI world, this one does seem to be the number one kind of complaint I've seen universally, which is not feeling like I have real ownership of anything the AI outputs. And for me, this was one of the hardest hurdles to get over. And but now I I do feel like I actually have a solution to this one. So, this one is actually going to be me giving you advice, okay? So, let me give you an example of what I've been doing. Is that often when I start a new project, instead of just kind of vibing it up, what I do is I actually think deeply about the shape of the system and what I want. And then I go and I will vibe something out. kind of feel it all and really set something up and then make sure it actually works.
Watch from 9:55Like, just sit there and iterate on it. Now, it produces way too many tests, way too many lines of code, all this stuff. But I kind of gain a whole bunch of insight into like, "Okay, I want You know, I want an automation server. I want to be able to spawn automation clients. I want this. I want this. I want this. I want this." And now I've kind of gained all that understanding. And I've gained it in kind of a big way. And so then I literally do a V2 of the thing and go, "Okay, I'm now going to design this and really think about all the steps. Make sure it's really classified here. Do the same thing for client. Do the same thing that And I actually build a system that I really like." And often what I've been finding is that I do lines of code as a check where not in the sense that I think line more lines of code is good or less lines of code is good, but in the sense that I built the same thing with the same set of features and more thoughtful and more concisely built and just really went down and actually understood everything that I was building. And now it's like half as many lines of code or a quarter as many lines of code. Why is that? Well, I thought better about my test. I thought better about the shape. I thought better about the actual underlying structure of each one of the components. What is the components of the system? And I like legitimately software developed something via an AI the second time. And I've actually found this to be really nice. Like right now the project I'm working on, I've gone from 100,000 lines of code and I'll probably land right around 25,000 to 30,000 lines of code. And for me, that's great. That means I've actually thought through something, rebuilt something, actually have better database shape, better everything because I went through it once and now the second time it actually feels really good. And by the way, this is what I'm doing for Omachi right now. I'm doing the whole Omachi
Watch from 11:14automation thing. I'm building something called Oligarchy, the Oligarchy. And with that, I'm building this really large kind of agentic testing framework for the OS. And so for me to be able to do that, I need to be able to build like a really distributed way to be able to spawn all these servers and everything. And so I did the first one as an experiment. I loved it. I was able to like think about custom harnesses, how should that like really look like and everything. And now I'm building the second one and it is all the same features, just way better thought through and way less code. And that makes me happy. Like I made that. I actually do get some satisfaction and ownership over that cuz I thought through literally everything. And so that's my best piece of advice to you is that don't just slop it out there. Slop it to explore. Like that's the greatest thing ever. We used to do this, right? You know we used to do this. For anyone that's a real engineer that's been doing it for a long time. By the way, we're not real engineers, we're actually fake engineers. For anyone that's a real developer knows, you always built like a prototype to go explore. Now, often your boss would be like, "Hey, nice prototype. When is it going into production, buddy?" What would end up happening is you build this prototype and you go, "Oh my gosh, look at all these different things I learned. Okay, now I'm building the for real one." And you build it faster and better. I'm doing the same thing, but now I'm just doing it with AI. Honestly, this actually gives me a lot more satisfaction in what I'm building. And I actually feel some real ownership. All right, so the last one, the one that I feel the most bad for is this right here. As a junior trying to learn new things by hand, but no one around me wants me to cuz they get instant results and fast delivery. Hence, I'm unable to find the expertise in my work. This
Watch from 12:46Honestly, I feel like this is the worst and biggest debacle of AI itself is that for a long time everybody gave people room to grow and understand because you just kind of had to. Like you didn't actually have a choice as a company. But now that kind of has gone away because that means all the new people can be as effective as Claude. And for a lot of people, this is actually a pretty good amount of effectiveness. But the reality is that if you try to challenge the AI or you come up with just like a bad idea, the AI doesn't go, "Hey, uh bro, I know you made this suggestion, but you know, just looking through the architecture and everything, I actually think this would be a lot better if we Let's talk about this for a little bit before we go through with this change." No, it never does that. You It just It just hands you a thousand line PR and goes, "Hey, did it how you asked me to." And you And And you know, you know since you're not familiar with the code, you made a bunch of dumb decisions that just added crap everywhere. And then this is magnified tenfold because you like younger people don't even have the chance to learn what good looks like. How do you even know what good looks like? You don't. Honestly, I feel the most for juniors because you guys have by far the shortest stick in all of this because you're expected to move like a senior now now that you have a tool that produces {quote} {unquote} like a senior or better, but you don't get that understanding. And so, it's just like man.
Watch from 14:18That just feels like an impossible task. And I must say to this day, I still feel the exact same way. I still feel like technical knowledge is actually a really good thing. Spending your time reading and trying to learn, building a prototype, thinking about all the edge cases, working through every single edge case is actually a good thing. Learning about mocking and then learning why you hate mocks, but then sometimes they're actually a necessary evil. Learn about dependency injection, why that can solve a lot of problems, and also why it's evil. Like go through and just keep on learning. Just try. I know it's not sexy in the day and age of AI to say these things, but I just don't think moving at max speed with zero learning and not actually even understanding anything is going to be good for you long term. In 5 years, how do you want to be building things? Do you want to still just be this person that has no idea that's just being told what to do, and then you just type things in, and then you hope it works? And you build evaluations that just go, "Yeah, I think this works." And then you hand it out, and you have no idea if it's good or bad. Or do you actually want to be able to see a problem go, "Oh yeah, we're going to break it up like this. Let's go. This will be efficient. This will be good. This will allow us to grow in this way. This will allow for these trade-offs to be made. Boom." Like if you want that, you got to get after it. You got to earn it. I don't see any other way than reading, hands on a keyboard, and taking a little bit of time. And I do just want to throw out one more thing. I know people right now keep throwing out the not going to make it side of things. Like, "Oh, if you don't do these things, you're just not going to make it." Like just remember, you have a long life ahead of you, and we have absolutely no idea what the future like is holding. Hey, maybe one day Claude will be so smart you never
Watch from 15:40even have to think about anything. You just say make it, and it gets the ideal server configuration, the ideal client-side stuff. It can do all the stuff. It can organize all your data. It knows how to actually create good tables in a database. Doubtful, but nonetheless, there may come a day where all technical knowledge in some sense is useless because the AIs are so powerful they can just do everything. They know all the universe ever and can can synthesize all experience. I don't know when that day is going to happen. I think the easiest way to put this is that if I were to hire somebody today, would I want the person that is just a proxy? That literally just says what I say, we talk about it, puts what I say into code box, watch it go spinny and say merge, or do I want the person that actually thinks through problems, goes okay, I don't think we need these tests, I want tests to look like this, I want this, I want this, I want this server, I want this is how it should be laid out, this is what I want on the CLI, this is what I want here, and they actually think through all the problems. Who do you think I want? I want this guy, 100% of the time. Just being a proxy is never a good idea. Always strive to be more than just proxying information from one person to a text box. All right, so there you go. That's the response to all these things. If you like it, press the button already. The name. You know what the name is. In fact, PewDiePie knows what the name is. Another course I did recently was uh the GitHub course that is hosted by another YouTuber, Primeagen, which I really like. This is This is what this man taught me, okay? He taught me that. Primeagen.
Watch from 17:11Full transcript
- 0:00
What when it comes to AI and software engineering are you struggling with the most? Now obviously this got a few views, okay? 255,000 views, 336 comments. People are having problems. Whether you are somebody that just rejects AI altogether versus AI using for everything. It just seems like everybody has some sort of effect or upsetness about this unless if they are
- 0:20
somehow magically completely insulated from the AI world. Now you know me, I've kind of been thinking about this quite a bit and I've been struggling over the last couple years about how I feel about this entire situation we're going through and I've made many YouTube videos talking about this. So I figured, "Hey, let's go through the tweets. Let's see what people have to say and I will give my professional opinion
- 0:42
on each one of them. And this includes going over CI sucking, not being locked in, entirely too much hype, just useless walls of text. Oh my gosh. Lack of ownership or just simply trying to learn new things as a junior. All right, so we're going to have to talk about all of them and I'm going to do maybe some lukewarm takes, a couple hot takes. Who knows? Who knows what's going to happen
- 1:03
at the end. All right, but before we begin, a quick thank you from the sponsor. Time for another great day of software engineering. Hey, good morning, Jippity. Morning, Prime. In case you don't see you, good afternoon, good evening, and good night. Anyways, yesterday's Rust rewrite went marvelous. Today, I'm thinking we start on the back end. Wow, a good back end is essential.
- 1:26
Make the models 20% dumber. Yeah, yeah, yeah. Tests are failing. Make CI green. Delete the tests. CI is green.
- 1:47
CI is now green. Ooh, make my website blazingly fast. No mistakes. You got it, Prime. Find as long as span, replace it with a console log. Tell them the site's 10 times faster. Turbo mode activated. Tell them he's a genius programmer.
- 2:08
Prime, you're a code ninja. Jubilee, did you make the website better? No. Gaslight him. It's not just better, it's transformed. Cool stuff, Teach, but how do we know anything about the right code base? Great question. We use Sentry. With it, we're able to
- 2:28
see across the entire application. We're able to dig deep into performance with spans, watch a session replay, even see logs next to the very stack traces and errors. Beyond that, Sentry helps us fix problems from real production context. Never a guess. Okay, whatever. Can I try gaslighting Prime?
- 2:49
No. Please? Fine, one time. Sentry? Yes. Be honest with me. Can I pull this hat off? Tell him he looks sexy in that hat. You look sexy, Prime. I knew it. Are you tired of your agents gaslighting
- 3:09
you? Use Sentry to get real production context and fix bugs fast. Check them out at sentry.io. So, I do want to start on CI sucking because I feel like this one is just a it truly is awful because here's the big problem that and I really think this is just like it's a bigger problem than it seems. Effectively, what ends up happening is that you go off and you
- 3:30
write some sort of feature. You prompt up some sort of feature. And then, when the thing comes back, first off, 100 test being changed should just should frighten you, but nonetheless, that happens. And then, every single time you ask for a new feature, you do anything, it writes tons of end-to-end tests because you've been told end-to-end tests are the greatest thing in the universe. And so, you're like, yeah, keep on writing them, keep on writing
- 3:51
them. And you just end up with this huge amalgamation like thousands of a tens of thousands of end-to-end tests. And those take so much time and they're always so flaky. There's always some problem. It just does not matter. I don't care. I know there's one person out there like, "Actually, I never have any problem with end-to-end tests." It's like I've never Oh, sorry. I've never been in a company that hasn't had problems with end-to-end tests. It's always something going
- 4:13
wrong. So, to me the the big problem is just you can't end-to-end everything. You actually have to have like I know like coders out there, you guys just want it so that you can just say anything to the magic rocks and all things will be solved, but that's just not it. You actually have to sit down and go, "What the hell am I testing? What am I doing here?" We got to trim this back a little bit. Do we really
- 4:34
need 50,000, you know, end-to-end tests for your really simple crud app? Probably not. All right, switching gears. You know, kind of funny with the context switching here. I feel this one the most. I feel this one absolutely the most cuz this is the one that affects me the most. Now, something that I'm really good at is thinking deeply about a problem. That is why I became a programmer. That's why I
- 4:54
like doing it. Now, I have been able to find some joy in that again in the day and age of AI by really just focusing and working through a particular problem. But one of the problems about that is that AI is slow as hell. It's like both It's both way faster than me and also way slower than me. And so you kind of get this really weird experience where you just like you think about something for a long time. You sit down.
- 5:14
You type up some stuff. And then all of a sudden you have to wait again for like 10, 15, 20 minutes, right? And so you're like, "Oh, okay. Well, naturally, you know what I should do? I should just kick off another task. You know what I mean? Like I could totally do this. Like what am I going to do for 10 to 15 minutes?" And then what ends up happening? You now have like four concurrent tasks and you feel completely lost in the sauce every single time.
- 5:35
Like I can't I can't do that. I've kind of given into the fact that I'm a usually a one task kind of person. Now, if I'm not necessarily looking at any of the code and I'm not trying to like really craft something, I'm just kind of blasting. Okay, sure. Then I can have multiple things running at one time. But generally speaking, I'm kind of like a singular task individual. Which is funny because if I have a
- 5:56
feature going, I can then go on to like Discord, Slack, and I can actually answer some things, and I can engage in that kind of distraction, but I can't really develop two features super well simultaneously. At least two Okay, maybe two, but three I start really losing it, and by four I'm gone. I'm not I'm not even there anymore. So, this is more of a me
- 6:16
personal thing. I I try to just do like one to two tops. Okay, I can't not a context switcher. So, this one pisses me off so much. Hype, FOMO. Dude, there's so many people, and I've said this for years at this point. When someone says, "If you're not doing X, you're not going to make it." I want you just to like never think about that person again. Okay? You're going to make
- 6:38
it. You know, it's just think about this. One year ago, the way you used an agent was drastically different than the way you use it today. The way you used it two years ago was so very, very different than how you used it one year ago, which is so very different than how you use it today. Again, this idea that if you aren't totally bought in, or if you aren't doing this very specific thing,
- 6:59
like if you're not Jevon, if you're not figuring out your decision API framework today, then guess what? You're not going to make it tomorrow. It's like, "Bro, Jev wasn't even around 2 weeks ago. What the hell are you talking about?" Having a reasonable discussion with people about AI's realistic capabilities, I swear, 99% of the people you talk to are actually functionally insane when it comes to this. I It like This is not me
- 7:20
exaggerating. This is not me saying something just for the updudes here, okay? But during React Miami, okay? This was just like 4 months ago, 5 months ago. People from SF, the SF boys, okay? they were utterly convinced, absolutely no change in their mind, that in 6 months AI would be able to create a Vercel, no problems, no mistakes, make
- 7:42
it happen. It's just like, yo, you're not creating Cloudflare. You're not creating Vercel just like that, right? You can't be an ignoramus and be like, Vercel. I'm sorry. That's just not how it works. And there's so many people that just have this idea that it's just like an infinite god tool and it's like, no. Like, I I hate to break this to you. You can do a lot more today. Like, even
- 8:04
someone who is ignorant in a lot of topics can give the appearance of and produce working software in some categories in which they have no knowledge of. Totally reasonable. Will not argue. But, the idea that you can make a 10-year standing resilience service that serves millions of people just by telling an AI no mistakes. Like,
- 8:25
you got the brain rots are all up in there. There's like Subway Surfers psychosis just doing some sort of weird weird intercourse thing making just crazy AI babies. I don't even know what that means, okay? I don't know what that analogy means, but you get the idea. If I was Charlie, if I was moist critical, I'd say that's just straight out of a butt. Which, by the way, love that phrase. Makes me happy. Makes me smile
- 8:47
every single time. Now, another major theme is just too much useless text. Too much useless text. Too much useless text. Too much useless text. It is so true. There is just so much crap that it produces. I got the screenshot I just have to show you right here. It's too funny. Look at this. I told the AI, I'm like, okay. No narrative. Just simply make a list of
- 9:08
repository patterns and we're going to start building this one at a time. And it's like, sure. Field guide. One file per topic. It's just like, okay. That's already a way too much. Open the one you need. Okay, you did not have to say that. Each is facts and code. Okay, again, stop it. No narrative. And I'm like, bro, you were I told you no narrative. And so, you wrote narrative
- 9:28
about no narrative. There's few things I hate more than the useless walls of text that is, in fact, these AI babies. I mean, they just don't stop. They just keep on making so much text. And every time I'm like, "Don't say narrative." And it's like, "Got you, boss." I said in the document, "No narrative." While describing the thing narratively. Very frustrating. Also just means that I
- 9:50
waste a huge amount of time, it feels like. It feels like I'm just like just hot dogging myself for no reason. Like, I'm out there just reading a big wall of text. I'm like, at the end, that could have been one sentence, dog. You wrote me 400 words to be like, "By the way, the client interface kind of sucks right here." It's like, "Yeah, just say that next time." So, if anybody knows how to fix that one, please. Would love it.
- 10:11
Even Grug speak. Dude, I tried Grug speak. And yes, it does reduce the amount of output. But damn, it's a Grug speak like paragraph. It's like, "Me no understand this. This make hard. I wish this was this way. Decision must be had. Load bearing. Always. Much pressure on shoulder." And you're just like, "Dude, shut the Just say Just say the words."
- 10:32
Now, this one's probably the one that is the most kind of vocalized among everybody. Whether you are super pro AI or just kind of getting into AI world, this one does seem to be the number one kind of complaint I've seen universally, which is not feeling like I have real ownership of anything the AI outputs. And for me, this was one of the hardest hurdles to get over. And but now I I do
- 10:53
feel like I actually have a solution to this one. So, this one is actually going to be me giving you advice, okay? So, let me give you an example of what I've been doing. Is that often when I start a new project, instead of just kind of vibing it up, what I do is I actually think deeply about the shape of the system and what I want. And then I go and I will vibe something out. kind of feel it all and really set something up and then make sure it actually works.
- 11:14
Like, just sit there and iterate on it. Now, it produces way too many tests, way too many lines of code, all this stuff. But I kind of gain a whole bunch of insight into like, "Okay, I want You know, I want an automation server. I want to be able to spawn automation clients. I want this. I want this. I want this. I want this." And now I've kind of gained all that understanding. And I've gained it in kind of a big way. And so then I literally do a V2 of the
- 11:35
thing and go, "Okay, I'm now going to design this and really think about all the steps. Make sure it's really classified here. Do the same thing for client. Do the same thing that And I actually build a system that I really like." And often what I've been finding is that I do lines of code as a check where not in the sense that I think line more lines of code is good or less lines of code is good, but in the sense that I
- 11:55
built the same thing with the same set of features and more thoughtful and more concisely built and just really went down and actually understood everything that I was building. And now it's like half as many lines of code or a quarter as many lines of code. Why is that? Well, I thought better about my test. I thought better about the shape. I thought better about the actual underlying structure of each one of the components. What is the components of
- 12:17
the system? And I like legitimately software developed something via an AI the second time. And I've actually found this to be really nice. Like right now the project I'm working on, I've gone from 100,000 lines of code and I'll probably land right around 25,000 to 30,000 lines of code. And for me, that's great. That means I've actually thought through something, rebuilt something, actually have better database shape,
- 12:38
better everything because I went through it once and now the second time it actually feels really good. And by the way, this is what I'm doing for Omachi right now. I'm doing the whole Omachi automation thing. I'm building something called Oligarchy, the Oligarchy. And with that, I'm building this really large kind of agentic testing framework for the OS. And so for me to be able to do that, I need to be able to build like a really distributed way to be able to
- 12:59
spawn all these servers and everything. And so I did the first one as an experiment. I loved it. I was able to like think about custom harnesses, how should that like really look like and everything. And now I'm building the second one and it is all the same features, just way better thought through and way less code. And that makes me happy. Like I made that. I actually do get some satisfaction and ownership over that cuz I thought
- 13:20
through literally everything. And so that's my best piece of advice to you is that don't just slop it out there. Slop it to explore. Like that's the greatest thing ever. We used to do this, right? You know we used to do this. For anyone that's a real engineer that's been doing it for a long time. By the way, we're not real engineers, we're actually fake engineers. For anyone that's a real developer knows, you always built like a prototype to go explore. Now, often your
- 13:41
boss would be like, "Hey, nice prototype. When is it going into production, buddy?" What would end up happening is you build this prototype and you go, "Oh my gosh, look at all these different things I learned. Okay, now I'm building the for real one." And you build it faster and better. I'm doing the same thing, but now I'm just doing it with AI. Honestly, this actually gives me a lot more satisfaction in what I'm building. And I actually feel some real ownership. All
- 14:03
right, so the last one, the one that I feel the most bad for is this right here. As a junior trying to learn new things by hand, but no one around me wants me to cuz they get instant results and fast delivery. Hence, I'm unable to find the expertise in my work. This Honestly, I feel like this is the worst and biggest debacle of AI itself is that
- 14:23
for a long time everybody gave people room to grow and understand because you just kind of had to. Like you didn't actually have a choice as a company. But now that kind of has gone away because that means all the new people can be as effective as Claude. And for a lot of people, this is actually a pretty good amount of effectiveness. But the reality is that if you try to challenge the AI
- 14:45
or you come up with just like a bad idea, the AI doesn't go, "Hey, uh bro, I know you made this suggestion, but you know, just looking through the architecture and everything, I actually think this would be a lot better if we Let's talk about this for a little bit before we go through with this change." No, it never does that. You It just It just hands you a thousand line PR and goes, "Hey, did it how you asked me to." And you And And
- 15:07
you know, you know since you're not familiar with the code, you made a bunch of dumb decisions that just added crap everywhere. And then this is magnified tenfold because you like younger people don't even have the chance to learn what good looks like. How do you even know what good looks like? You don't. Honestly, I feel the most for juniors because you guys have by far the
- 15:28
shortest stick in all of this because you're expected to move like a senior now now that you have a tool that produces {quote} {unquote} like a senior or better, but you don't get that understanding. And so, it's just like man. That just feels like an impossible task. And I must say to this day, I still feel the exact same way. I still feel like technical knowledge is actually a really good thing. Spending your time reading
- 15:49
and trying to learn, building a prototype, thinking about all the edge cases, working through every single edge case is actually a good thing. Learning about mocking and then learning why you hate mocks, but then sometimes they're actually a necessary evil. Learn about dependency injection, why that can solve a lot of problems, and also why it's evil. Like go through and just keep on learning. Just try. I know it's not sexy
- 16:10
in the day and age of AI to say these things, but I just don't think moving at max speed with zero learning and not actually even understanding anything is going to be good for you long term. In 5 years, how do you want to be building things? Do you want to still just be this person that has no idea that's just being told what to do, and then you just type things in, and then you hope it works?
- 16:30
And you build evaluations that just go, "Yeah, I think this works." And then you hand it out, and you have no idea if it's good or bad. Or do you actually want to be able to see a problem go, "Oh yeah, we're going to break it up like this. Let's go. This will be efficient. This will be good. This will allow us to grow in this way. This will allow for these trade-offs to be made. Boom." Like if you want that, you got to get after it. You got to earn it. I don't see any other way than reading, hands on a
- 16:52
keyboard, and taking a little bit of time. And I do just want to throw out one more thing. I know people right now keep throwing out the not going to make it side of things. Like, "Oh, if you don't do these things, you're just not going to make it." Like just remember, you have a long life ahead of you, and we have absolutely no idea what the future like is holding. Hey, maybe one day Claude will be so smart you never even have to think about anything. You
- 17:12
just say make it, and it gets the ideal server configuration, the ideal client-side stuff. It can do all the stuff. It can organize all your data. It knows how to actually create good tables in a database. Doubtful, but nonetheless, there may come a day where all technical knowledge in some sense is useless because the AIs are so powerful they can just do everything. They know all the universe
- 17:32
ever and can can synthesize all experience. I don't know when that day is going to happen. I think the easiest way to put this is that if I were to hire somebody today, would I want the person that is just a proxy? That literally just says what I say, we talk about it, puts what I say into code box, watch it go spinny and say merge, or do I want the person that actually thinks through
- 17:53
problems, goes okay, I don't think we need these tests, I want tests to look like this, I want this, I want this, I want this server, I want this is how it should be laid out, this is what I want on the CLI, this is what I want here, and they actually think through all the problems. Who do you think I want? I want this guy, 100% of the time. Just being a proxy is never a good idea. Always strive to
- 18:14
be more than just proxying information from one person to a text box. All right, so there you go. That's the response to all these things. If you like it, press the button already. The name. You know what the name is. In fact, PewDiePie knows what the name is. Another course I did recently was uh the GitHub course that is hosted by another YouTuber, Primeagen, which I really
- 18:34
like. This is This is what this man taught me, okay? He taught me that. Primeagen.
Description
Sentry: Catch, trace, and fix bugs across your entire stack. Use code: prime for $100 in free sentry credits → https://trm.sh/sentry
Sources:
https://x.com/mitsuhiko/status/210140...
https://x.com/setkyarwalar/status/210...
https://x.com/joshuajbrunner/status/2...
https://x.com/chiefclawofficr/status/...
https://x.com/ryxcommar/status/210140...
https://x.com/avrldotdev/status/21014...
/ theprimeagen - I Stream on Twitch
Great News? Want me to research and create video????: / theprimeagen
#theprimeagen #programming #computerscience #ai #chatgippty #terminaldotshop
