# How I Use AI In My Day To Day Workflow ## Метаданные - **Канал:** Anthony GG - **YouTube:** https://www.youtube.com/watch?v=b6uRKK_38Hk - **Дата:** 21.04.2026 - **Длительность:** 2:17:09 - **Просмотры:** 11,324 - **Источник:** https://ekstraktznaniy.ru/video/49674 ## Транскрипт ### Segment 1 (00:00 - 05:00) [] YES. WE'RE LIVE. This needs a celebration. I'm going to lie. Yes. We're live. This Wait, I'm a bit rusty, bro. Let me put it on 4K. Okay, so how does it actually work? Where is my chat? That's the question here. What's going on? Okay, is this Can everybody hear me? Ladies and gentlemen, welcome. Bonjour. Nihao. Namaste. Merhaba. Goedendag. Wie geht's? Nasılsın? West. Comment ça va? Comment tu allez-vous? Comment allez-vous? I'm speaking every [ __ ] language in the planet and that's why we also just translated. MMT, you know what I mean? Anyway. Um I need to be very careful because I have no clue how this all works. Let me actually share my Discord that I never visit. Wait, that's not here. Is it Yeah, it is. Look, the last stream last the last stream. Oh man, it's one year ago. You dirty little boy. Yeah, it's I'm going to be honest. I need to come back. Why? Because I open YouTube and there's literally garbage. It's slop everywhere. I open a code base that is slop everywhere. It's basically slop everywhere, bro. It's insane. How can we fix this? Everyone. I'm live. — Yeah. Um so I don't even know what the [ __ ] modus operandi is here to be honest. Yeah, let's do that. Let's pull Let's open this chat so we can Okay, let's actually Let's do some grounding with the community. Let's open up XCode editor and say something like hello. And then All right. What's going on here? Who is he? Who can I remember? HFG, of course. It is of course. Uh What's going on here? Mansour — Mansour, what's going on, my baby? What's up, man? Long time no see. You need to be It's like a shame, man. Where are you? What the [ __ ] going on? He has a girlfriend and then he suddenly disappears. That's the classic. It's a classic trap. Um Are you still using Go? Yes, man. Go and JavaScript and Odin. Dirk Heise, what's going on, Dirk? Dirk, what's going on, bro? Uh it's up everywhere with these LLM models even juniors knows that we need. Yeah, but we're going to talk about AI because this is going to be the most important stream in all your life. I swear to God because like I have nothing to to sell here. So I'm going to be uh completely honest. Completely Okay, just double-checking because you never know. What's going on? Um let's see. Let's see. Let's see what we do with our feet. I actually I'll be live on this other Yes, look at that. I'm live. What is going on? My family thanks you. No problem, Atlas. OG Atlas here. Uh let's go back to XCode editor. Um Yeah, what's good, young man? I still have the same thingies. Here to see some solid content. Yeah, I'm going to show I'm going to be honest here and where is Ashton? Um Ashton is He's in Go land, man. He's in Go land. Do you write any articles, blog posts about stuff you're doing in it? No, actually not because like I'm the worst writer. I cannot even write one sentence without any grammar mistake. That's the thing. Um the Timmy. Yeah, man. Who remembers the Timmies? Who remembers the Timmies, bro? What's going on? Uh What's going on, big dog? I'm doing ### Segment 2 (05:00 - 10:00) [5:00] good, man. It actually to be honest, it was a very very rough year because um Where the [ __ ] was I? Yeah, so the thing is that previous um early 2025, I think January, I started like a project on my uh YouTube and it was basically Market Monkey, right? And now it's like um Now it's like it's it's it got some traction. I'm going to lie. It got some traction and um yeah. It took all my attention and still has, of course. Still has. We We build a like a nice company. Um with the ups and downs, of course. So that's Chinese translate. Look at that. So that's basically what's happening. Um of course I went from an insane amount of dopamine into the classical uh project trap. We had the some point of time things get too rough, right? No um fitness went backwards, relationships uh could not be maintained, spending time with kids was basically very limited um because everything was into MMT because that's what it takes, but um now I basically I thought about this whole adventure and now I'm trying to be more uh myself again. — Which I don't know if it's a good thing, but that's the thing. Uh we need to mix like yeah. It's not worth it. That's what I told myself. It's not worth the [ __ ] pain. Um So I basically took a little bit of a sabbatical 2 weeks thinking about life and I came up with the idea I need to basically save you guys from falling down with all the AI slop. But is it really slop? That's what we're going to discuss. Let's do a little uh recap on this chat and then we're going to start. You know what I mean? Because time is [ __ ] money. All right, what's going on here? Who is this? Teddy. E Pristi. The best auto flow app. Yeah, it is. Well, yeah. Why not? Yes, it is. Of course it is. Kev, what's going on, bro? Kev from It's Kevin. It's Kevin from uh from Africa. Life comes first, mate. Been a long time viewer and always will be. Yeah, man. Thank you so much for still being here after me disappearing for a year. I mean, I could make videos, but I mean like what should I make? What should I teach? Because AI is here. What can I do? People are asking By the way, I'm going to upload the full-time coder for free. I was basically checking my Where's my hard drive? Here. I found my hard drive back. And I'm going to put it online for free because it's outdated and if you want to be hear to my voice if you want to listen to my voice. It's like 200 GB of videos. So yeah, that would be nice, but it's still uploading. Um so yeah. Good time to take for yourself, king. Yeah, buddy. Thanks for giving us MMT. No worries. No worries. Thank you for consuming my stuff. Uh What were you doing till now? I was working on the company, but I mean like working every [ __ ] day like 12 16 hours a day went into this thing. Ay ay. Das nicht gut. The blind homies. Yeah, man. Where is that thing? Yeah, so a lot of us changed. So basically Remember for the people that are basically uh knowing my videos, I was in 2025 early 2025 January This is basically the cut off time. That's where I disappeared. So I was basically writing codes um by hand by I'm My brain is thinking should I write caps or should I do like the the artistic style? The thing is that I cannot do both. So I'm going to keep it caps by hand. And I was very skeptical about the AI. Remember like I think I have a video about it. Like it's just I hated it so hard because what are we doing? What were we doing back then? It was basically you have your editor, right? Uh code VS VS code or Vim or something like that, you know? And we were writing code by hand and sometimes it was like a complex uh problem here, and then we opened up um ### Segment 3 (10:00 - 15:00) [10:00] something else like cloud or GPT or something like whatever. And then we say, "Hey bro, look, this is my code. Fix it. " And then it dumped out something, and then we copy-pasted it back into this thing. And I did it from time to time, especially like some algorithms, but writing codes by using something like cursor, that was like it didn't [ __ ] work, bro. Or even co-pilot, you know what I mean? It's like it was spitting out stuff you couldn't think. It was breaking your flow. I didn't I hated it, but but ladies and gentlemen, I'm trying to reach chat and thinking at the same time, which is not good. So, around the classical uh mental shift was I think for me was December or something like December. Dec December 20 25. Around Opus 5. 3. Is this good? No. Opus 5. 3. What is this? Let's go back. Where is this Opus? 3. What is this Opus, guys? Help me. Which Opus version was that? 3. 5 6 7. 3. 5. All right. 4. 5 I But it's GPT, bro. GP 4. 3. Whatever. Opus. 5 something. 5... 3 actually, to be honest. I'm going to lie... 3. 4. 6. Opus some Opus version was released. And everybody was going wild. And I remember this I remember this moment. I was sitting I was watching TV, had my laptop, was coding something. I said, "Fuck it. I'm going to try it. " So, I opened up cloud it was cloud code open up I installed cloud code. I selected Opus, and I said, "Write me a write me a chart just like TradingView. " That's in in JavaScript. Write me a the simple prompt. I said, "Write me a chart and make sure it behaves like TradingView does. " I pressed enter and I watched TV, I came back. I opened this application. I'm not going to say it was perfect code. I don't know. I didn't even check that. I got damn. That thing just works, bro. I was just like moving the chart around and it was all doing the things it needed to do. And I said, "Well, I'll be damn. " And of course I got this insane shot of dopamine, the same shot I got when I went to the casinos, and I kept building on top of it. And it all seemed to work. And I said, "Man, why am I even writing things by hand if this thing can just do stuff, right? " And I fell into the trap. Is the timeline. This is the Anthony GG Timmy timeline. So, I made an JavaScript app. Nothing too serious. It was just for me to test, right? Because we cannot write MMT in JavaScript because then we would not be the fastest anymore, you know what I mean? Cloud Opus 4. 5 release in November. Okay, Mansour. Yes, 4. 5. It's GPT that starts with the 5. 3. Yeah, okay, okay. Opus 4. 5. Correct. Where is the eraser? Where is the eraser, bro? Here. Look at that. This app is amazing. I'm not lying. And then we're going to this one here. Double-check Mansour. 4. 5. This is true. This is I'm not This is a real story. This is my story. You know, this is my stream, and I do what I want. I cry when I want to. So, I created this JS a kind of application. And you're thinking, "Anthony, why do you speak so weird? " That's because I have this um zen in my mouth. I'm going to remove it because I'm getting way too much stimulated. Yeah, I quit smoking as well. Thank you. But I'm still using nicotine because high performing high performer. Yes. So, I fell into the trap creating MMT, which was then called and communicated with my boy Cap Track. Co-founder of MMT, you know, when the lizard the all-knowing oracle. And I said, "We're going to make three 3. 0 version three. " ### Segment 4 (15:00 - 20:00) [15:00] Because look at this. This thing just worked. And I was building and building on top of this JS application. And at a certain point in time, I was actually to be honest, I was I think I was actually just vibe coding this thing. I just It's hard for me to let people be in my code base and write stuff their way because I wanted to have it my way. I think it's the classical engineering senior or whatever you want to call it engineering trap where you are always scared if somebody comes up with his own thing and it's not like the right variables you want to use, you know what I mean? Like the classical thing. And with AI, it's like yeah, this thing is not going to produce the absolute best. So, you need to basically you need to um you need to embrace it. Stop thinking about it. Stop worrying about it. Just thinking about the money it potentially can generate because we're going to make the best apps of as fast as possible. Building a bunch of [ __ ] on top of each other. We actually had some beta testers letting testing this thing. But it was not fast enough. I remember that. They said like, "Yeah, but the TPOs like the TPO stuff we have like I just [ __ ] Chinese translator. What's going on here? English. So, for example, let me actually open candlestick siege and then let me reset. I can add TPO, right? And I can just basically look, I can do this. And it's loading all these TPO profiles. Literally the only app in the world that does this, right? It's like it's insane. It's super fast, right? So, people were doing it Apparently people are doing this as well. And they said, "Yeah, but but the MMT the version we have right now, it does super fast, and right now it's slow. So, and that basically cranked my ego. I said, "Okay, no. We cannot ship this [ __ ] because it needs to be fast. " And that's basically the end of 3. 0. We skipped it. And okay, so we I decided to basically continue on the MMT we have right now, right? The Odin version. With WASM compiled yada yada. And I was thinking like, "How can I actually incorporate this AI stuff into my workflows right now without causing bullshit? " Because it will cause even as of today it will cause [ __ ] Um I'm going to tell you how to prevent this because it's true I'm I'm not writing a single code by hand anymore. I'm going to be completely honest. I don't write a single piece of code by hand for the past 8 months. And still I'm shipping fast, but not 10x faster or 100x faster. I'm shipping maybe two times faster or three times faster. I'm going to tell you what. So, I learned this AI shenanigans [snorts] for like 1 year, I think. 1 year fully on AI. 1 year. How do you write year? Like this. 1 year. The man that told that that the crazy Anthony GG that basically was so skeptical with AI is not writing code anymore. Isn't that [ __ ] crazy, bro? I swear to God I'm honest here. I'm going to let this sink in a bit because that's going to be um What's going on? Opus MMT stream lagging. Yeah. I'm fixing this as well. So, They're selling the face like we say I did this with cloud like it's not obvious already. Yes. Um Let's go on here. You didn't want to listen. Yeah. Yeah, yeah. Okay, but listen. But wait, wait, wait. So, a couple things we need to basically So, yeah. December Opus 4. 5 try to vibe code some stuff, quickly realized this [ __ ] is not going to make it. Um Back to MMT. Well, I was basically on the side. Now I was here trying to incorporate AI ### Segment 5 (20:00 - 25:00) [20:00] into the applic- into my workflows. Never wrote a single code by hand anymore. Literally not. Sometimes maybe it's like a small thing, like a setting or something, because that's faster than churning through some tokens. How does this work? I tried different uh different approaches. I also tried to side code some other projects, because I think we made like 20,000 other versions and 20,000 other projects. Um just by having AI as as a tool, because now it's easy to churn something out. And a couple of things, a couple of problems basically uh that I had, which was a bit annoying. Let me explain. First of all, people are basically focusing on these agents, right? Agents. This is very important. Agents. Forget agents. I- If you're using agents like, "Oh, I'm having agent swarm and they are writing all my code. " No, it's not. It's garbage. Agents are completely garbage and I don't even under- Well, they are for exploration, they are great, right? If you want to say like, "Explore my code base, spin up 20 agents or whatever. " That's fine. But like what people do is this, right? They do this. And that's basically that's noobs. All [ __ ] noobs. They create an MD file, right? MD. file with like over a PRD or whatever noob slop they are making. The moment I see PRD, PR me. That's bad. And PRDs you're That's idiots. They make this a giant PRD. They make all these tasks, right? They make a bunch of tasks. I tried this. I swear to God. I tried I'm going to explain my whole [ __ ] thing that I did. They make a bunch of tasks and then they kick off like one kind of a Ralph loop. That's We're not going to talk about that, because that [ __ ] doesn't work. And they spin up a couple of agents and they start building. That's not good. I'm going to explain why. It's It's probably It can work. It's probably working. It's probably going to generate something, but it's something you can maintain. It's going to shoot you in the foot, 100% guaranteed. I've been there. I've done that. That's nicht gut. Doesn't work. So, I don't understand why these other models like Kimmy and Cloud and all these [ __ ] open Cloud and all this [ __ ] why they're focusing on these agents, because that's not the problem we need to solve. That That's literally not the problem. At least not for professional engineers. Like if I would kick off some agents in the MMT project, yeah, that [ __ ] That thing wouldn't That wouldn't look good. You know, and the reason why is entropy. And I'm going to explain that. So, why is Why doesn't Why does this not work? Well, the reason is So, AI does an LLM, right? It's a long large or a long large language model. And what basically happens is you give it a bunch of text. You put it into this LLM. Let me Let me make it a bit more diagrammy. And now I want to redo this line. My brain works like bananas, bro. You put it into the LLM and the LLM is going to predict It's going to predict a bunch of text tokens based on the input. And it's going to come up with this, right? As of today, large language models like Opus or GPT Codex, whatever this thing, they write good code. Really. Literally. They write better code than humans. I'm not done. I'm already see you like, "Oh, this guy. " No, listen. They make the same mistake as we humans. Which mistake do they make? They make the same mistake that each time you put something in, they are basically going to output entropy. So, let's say they're going to basically do a 90% good, but there is 10%. It's not that it's bad. It's 10% that's burden. And the burden could be like ### Segment 6 (25:00 - 30:00) [25:00] The burden Is the stream lagging or something, guys? Because my PC, I don't know what's going on here. — [clears throat] — They're going to make 10% burden and this burden could be something like um mislocation of files. Mislocation. Location. Two, it's going to tightly couple tight Jesus, bro. Tightly couple codes. It's good technical debt. Thank you so much. It's going to code — That's what we need. Tech debt. Tech debt all the time. It does it all the [ __ ] time. No matter how good you prompt. Prompt is important. Garbage in, garbage out. But it's still It's always And that's how we work as well, right? If we I'm going to explain. So, tech debt burden. It's always going to do that. It's going to do a couple of things. And it's getting worse that if you already have a bad code base, like that's why creating projects from scratch with AI is hard. That's so [ __ ] [clears throat] hard. That's the hardest part. Creating it from scratch. I'm not talking about the to-do app or something like whatever these guys making, like the classical uh Twitter bait. But creating some sophisticated program from scratch with AI is hard. It spits it out like crazy, but it's hard, because there is always this It's easy to add stuff on a greenfield project. And that's where the entropy begins. Because the moment something extra slips in here, it slips in here as well. It basically adopts what it sees. It adopts the patterns what it sees, which is good. And we do that as well. As engineers. It doesn't behave different from a human. We do the same thing. But what we do is we clean it up. We We start So, — [clears throat] — how I'm doing You should work as you are here. Right? And you need to make a solution. I need to do these white sessions whiteboard sessions way more, because my handwriting is my mouse writing is getting behind. I need to sharpen my pencil. We know roughly what this feature is going to be. We do not really know how we're going to get there. So, we as engineers, and you see that on my stream on my videos as well, is we start Okay, we're going to add something here to this structure then we're going to add a function here and this and that and then no, we're going to delete this. It needs to be this. And before you know, you eventually come to your solution. And then you have this first, yes, it works. But then you already know like, ah I made a big mess. And then you need to do a clean up. You clean it up a bit here and then you clean it up. And at And at a certain point of time, you basically come to a nice solution, right? That's what we do. AI skips the clean up part all the [ __ ] time. Right? [clears throat] And that's what's happening. You keep [clears throat] prompting. It's adding entropy. If you do not check it, which what five code is most of the time doesn't do. If you don't can detect entropy, at a certain point of time, — [clears throat] — your code base is going to be very hard for AI to write stuff without breaking something else. And that really frustrated me. That's something that I came across with greenfield projects and all side projects is that I was keep adding features and at a certain point of time I needed to change something and all the rest broke. And it was so annoying. What's going on here? So, and that's why And it always does that. Like if you tell Cloud or GPT one single scoped issue, it's always going to It's always finding these bugs faster than we can. And it most likely is going to come up with a good solution. Most likely. 99% of the times, especially Codex. It Codex is like We need to talk about these models as well because But if you give Codex or even Cloud like a whole [ __ ] PRD with tasks, it's going to mess up. It's going to it might work. But ### Segment 7 (30:00 - 35:00) [30:00] everything is going to be in use effect mis-abuse all the time. Use if I have like a no use effect skill which even skills doesn't work correctly all the time. I need to basically have I have like skills that trigger a review that need to delete all It's crazy. It's just That's what they need to [ __ ] fix. Not the [ __ ] agent swarms. They need to fix the models like doing what I tell them to do. Anyway. So the question already arises here is AI going to replace us? Never. Well, maybe a couple. Right? Maybe a couple. But they're never going to replace us because you need to steer it very well. All the [ __ ] time. But if you can steer it correctly and you keep the tasks very small, and you basically communicate very well in your prompt exactly how you want it, it's amazing. It's really amazing. There is no need to write code by hand anymore in my opinion. But wielding the swords wielding this sword of dragon steel I'm re-watching Game of Thrones by the way. If you want to wield the sword made of dragon steel, but you don't have the muscles to swing with it. You're not going You know what I mean? You're going to cut yourself. your tiny weiner off. You know what I mean? So you still need to have not so to say language specific knowledge. Like do I need to know Rust very well? Zig or whatever language very well, Golang? You need to spot wrong architecture. It's pretty funny. It's pretty ironic because I was the guy that just wrote codes without an architecture. It was like there was no such thing in my videos. Like we just do stuff. But at a certain point of time, for really production ready code base is something that I learned. You really need to come up with some good architecture. So AI can basically start fixing something in its own isolated box. Right? So instead of having you need to come up your code base need to be so good that it's like every kind of a service domain you have in your application is not tightly coupled. That's very important. That's what I'll I'm talking about big projects. I'm not talking about a simple to-do app or a For a big project, if you if it needs to fix something here, here. Not here. Not here. Here not there. And that's what I saw sometimes. I I asked it fix this and then suddenly I had like in my git tree, I had like 20 files modified. I said, "What the [ __ ] going on? But why are you modifying that? " That's the first code smell. I'm going to call it AI smell. AI smell basically means you cannot wield the sword of dragon steel. It's one. I have a better idea. What am I doing? Why can I not just do this? Too many file edits. I'm a genius. Look at that. Get out of here, man. Two. Why can it not That's better. Can't fix. Can't fix directly. If it cannot fix your If it cannot detect your problem directly, you [ __ ] it up. Your prompt is bad. Your code is bad. That simple. AI fix it directly. If it cannot, it's either like a big issue, but it's no problem. Just open up uh Codex uh with uh 5. 4. It always figures things out. It never it always Because what I realized is the moment it cannot find a bug, there is no bug. I swear to god there is none. It's something else. If you prompt it, "This is the bug we experience. Can you fix it? " And it cannot find it directly, it start Sometimes it start doing weird stuff. And the moment you you think ### Segment 8 (35:00 - 40:00) [35:00] like, "What the [ __ ] are you doing, bro? " There is no bug. There's something else going on. Your prompt is wrong. 100%. AI fixes everything. That's what I learned. The most complex problems. All the time. All the times that it could not fix my bugs were basically no bugs. Were something else. Just use the Excalibur IMCPI. I know. All right. That's That's the thing. What's going on, Mostro? What's going on, man? Almost make me think that maybe the functional programmers are finally going to win because AI and functions go hand in hand. Uh might be. Might be. It might be. So what did we learn so far? First of all, the PRD guys where the where are we here? The the MD PRD guys need good. Doesn't [ __ ] work. I swear to god. Doesn't work. It's like That's right. Like these things like where is this? GSD G Get [ __ ] done. This one. A lightweight and powerful meta prompting context engineering and spec driven development system. That doesn't work. It It might be but actually from my perspective like a real engineer like real. Like not the fake engineers that don't that doesn't have any production ready application. It doesn't work. I tried it. It would be it would be amazing. You just write some specs and then you go to sleep and roll loop your way to success. It doesn't work. It works for them because they don't have any understanding of how code works. understanding. They don't know how it works. They don't know what they see. They think it's all good, but it's not. It's like a ticking time bomb in your code base. And the moment you are 1 2 years in production, nobody can fix it anymore. It's GG. It's game over. Nobody is going to sip margaritas on the beach. You [ __ ] it up. The same thing with Claude Bot. This thing is terrible. Open Claude is terrible, bro. It's garbage. I used it. I fall into the trap as well. Open Claude trap. Ask We have even We have Casper. We called the [ __ ] Casper. Nobody's using him anymore. We were We had him in the company. And it was so great. I was like, "Oh my god. " I told I was the one bringing it in. I said, "Cap track. " Ashton. It was Claude Bot back in the day. It was Claude Bot. This thing is crazy. We It's going to automate everything, bro. everything. It didn't automate anything. It [ __ ] broke all the time. It's garbage. It doesn't work. And it's And I have some knowledge on how to debug these things. It was garbage. I was more messing around with the Claude Bot doc doctor command than I was getting stuff done. You know what I mean? It was terrible. — [sighs] — How many times that I saw on X somebody Look at what I built. Claude code built this. And I said, "Man, how the [ __ ] do you do this, bro? " It's like, "Teach me how to dance, man. " Because it didn't work for me. Like I have the most I have a good knowledge. I have like over 20 [ __ ] years of experience in coding. And I could not prompt my way to success because I saw problems all the time. Like 20,000 re-renders in React. I said, "Man, but what are you doing? Use effects everywhere. Use effects inside use effects. " Mitigated by another use effect on top of use if It was use effect everywhere. It was insane. How are people doing this? Anyway. GSD. These things need good. I still don't write code by hand. I still I hand, but I just this is not it, guys. Anyway, back to the thing. So, AI smell. If you Guys, if you I swear to god, if you have this, if you see you change something and it's basically editing 10 20 files, your code's is ### Segment 9 (40:00 - 45:00) [40:00] there's a problem architectural wise. If it cannot fix it directly, there is a problem, 100%. It's nicht gut. And I see this a lot because back uh beginning of this year when clouds when the Opus 4. 5 was big, we saw a [ __ ] ton of people coding competitors, like code competitors. Everybody was building a terminal. Literally, terminal or a screener. Everybody. The whole world was building it. And I was like, "What the [ __ ] Everybody's building it now. What the [ __ ] are we going to do? " Guess what? Everybody disappeared. They're all gone. Why? Because they got wrecked. They got wrecked by entropy, you know? 10% 10% another 10% It's editing 10 files. It's editing 20 files. I change the small thing. I need to change my color from red to blue. Suddenly, they are wrecked because they can't fix it because they have no clue. Anyway, that's good. These guys are gone. Back to Hauser. Salute. Anyway. If this is all not working, how do I do it? Well, it's very simple. You need to talk with the bastard. You know what I mean? So, what I always do is and guess what, guys? I'm becoming a fan. I I changed. I have changed. There are two I made two friends. Well, friends. I'm starting to Anthony GG is starting to like For the people that know me, they're going to be very They're going to say, "What the [ __ ] going on, Anthony? Stop. What's going on? What happened? " I'm starting to like Teo. You know, the Teo guy. Teo from the Teo with the blue hair. Well, technically it's not blue, but you know what it is. Teo. The guy from YouTube. Teo, yes. This one. I'm starting to like Teo. Look, I'm watching all his videos. I don't know what's happened. He makes sense. He brings me he cures my anxiety for some reason. I'm also starting to like one thing that's being made by and it's a very important skill. Mat Pocock. This guy. I'm starting to like him as well because he created this skill. Grill me skill. This one. This skill is by far it's like 1 2 3 it's four lines. This skill is by far the best skill ever created by a person on the planet Earth. Grill me skill. I'm going to actually read it over. Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when the user wants to stress test a plan, get grilled on their design, or mentions grill me. Interview me random blah blah. Provide your command and answers recommended answers all the question one at a time. If a question can be answered by exploring the code base, explore the code base instead. This thing is crazy, bro. I swear to god. And that's how I basically always start. I have two entries in my workflow. Mat Pocock. I can't even I speak his name. Mat. Mat [ __ ] Peacock. I swear to god, the guy this skill is crazy good. That's the only thing I use. This is so good it's I don't know. It's just unless you use it with Codex. I don't know what's going on with Codex. It's like it's unbearable to interact with Codex. It's good, but it's like unbearable. It's like it's spitting like a bunch of [ __ ] in my face, like It spits like 20,000 bullet points with academic terms I never heard of. How the hell can I They need to fix that. If they can't fix that, GG OpenAI wins. Uh be sharing the drive. Yes, I will. I will I will. No worries. I'm not going to let you down, guys. Okay. You see we are basically going Where did I went? What happened? This is all here. It's all here. ### Segment 10 (45:00 - 50:00) [45:00] You see? It's all here. Okay. Yes. Yes, but they both are, I think. Nothing wrong with that, man. It's coincidentally. Where is the share Where is the drive link? It will I will share it when it's done. It's upload It's 200 GB of data, man. I'm not scratching my balls, by the way. I'm scratching my belly. Well, I'm looking at myself in the thingy and I see it's like this guy is basically jerking off. It's not. I'm not. The link will be there. Relax. I ain't going anywhere. It's like uploading 250 GB of uncompressed data to Google Drive. It's paused because I'm streaming. Okay. So, the first thing There are two things you can do. I'm going to make like a decision tree here. How [clears throat] to use AI which always works and it's the best in the it's it works for me. First, There's a bug. Happens all the time. I need to replace this. We need to have a black screen here. A bug. God [ __ ] damn it. Where is my here. Do I know what the hell is going on? What does that mean? If somebody said, "Anthony, there is a problem with this on MMT which had daily happens, like couple. " Then I think like, "Ah, yeah. I know what it is. " Sometimes you know. Sometimes like, "Ah, god damn it. I knew this thing could be a problem because I just edited this or whatever's going on here. " If you know what the hell is going on, you can just always don't even hassle with Claude. The Don't even hassle. Use Codex all the [ __ ] time. Codex. The latest one. You need the two subscriptions. It doesn't It's simple. It's like $400 a month. It's easy. It's It's going to make your life so good. You need the two. Or Cursor. Use Cursor then. It's fine. They They have everything, right? Yes, Codex just prompt the stuff. Prompt it. I'm going to use these things. This is better. Just prompt. Prompt. Prompts. There seems to be a problem with the candlestick chart where yada yada, fix it. It fixes it. You do a review real quick. You check if it's not doing crazy stuff because there cannot be crazy stuff. It's in the candlesticks, so it should be in the candlestick code and nowhere else. It should not be anywhere else than the candlestick code file exactly where the candlestick rendering happens and nowhere else. So, the moment I see it's editing 20 files, I know I need to go in deeper. If my wife doesn't scream, I know that I need to go in deeper. The same thing here. Otherwise, you check it. No crazy stuff. Done. Ship it. Easy. Blindly ship it? Never. Always check. Just You can You can spot it, right? You can directly see like I mean, I need to watch some diff and I can see what the hell is going on. I can't The moment I see like You know what I mean. If you're an engineer, you can see it. it. The second problem is here where a lot of people [ __ ] up because the most of them do this. They don't know what the hell is going on and they just make a shitty prompt and then they call it a day. So, a bug. I do not know what the hell is going on. I don't know what the hell is going on? Going on. Going on. Okay, what do we do? We need to talk with the model first. That's so important. That's where everybody goes out. If you don't know what the hell is going on, you take your problem, you paste it into the chat, and you say, "You might have a rough understanding of ### Segment 11 (50:00 - 55:00) [50:00] where it might be. " Then I always say something like, I always do this prompt, something like um "Prime yourself with the way we render blocks. I'm just This is a second this automoton. Prime yourself with the way we render blocks and how we skip the visible range and all the things related. " What I most of the time do is like, "And report back. And report back. " Then it knows it doesn't need to write code. Press enter. So, it's doing its thing. It's reading all the files, right? And then it basically comes up and it dumps some text, right? It dumps some text to you. — [clears throat] — That's important. When I read this, I never always like Sometimes I read it completely, sometimes not, but I can see does the [ __ ] model know what the hell is going on. Is this correct? Okay, yes, it's correct. Then I do something like this. "Well, it seems yada yada yada is going on. " I'm basically instead of spoon-feeding, I'm trying to discover it themself. "Seems yada yada, so X does not turn red. Why and how to fix it? " Never write code, bro. I never do this. And report back. That's something that I need to do, or the guys are basically going ham in code. I don't want that. And report back. And then it basically comes up with something like that. And then you check like, "Hey, okay, this makes sense. " Maybe some other chats with it like, "What about this? What about that? " Fix it. Press enter, comes out, test it in the browser, or whatever. Clean solution, write some tests, whatever you need to do. Done. Now, TDD not good because the guys write the worst test in the planet. The moment you have two bad tests in your project, it's going to write bad tests for the rest of its lifetime. It's crazy, bro. Very careful with tests and always read the tests because these guys are going loco with tests. Not going to lie. And they keep adding slop tests for like no reason at all. Most of the time, you know what the hell is going on because you know your code base, right? And that's the problem with vibe coding. You do not know your code base, so you're always in this thing. I don't know what the hell is going on. And then these noobs, they are basically do not let the model prime and discover and talk with it. And then they end up in a problem, right? All right, so then we have the next thing. Are you using only text to prompt, or having started using something like Always use Whisper. And the reason why I do it, first of all, I thought it was something for vibe coders, like Whisper. My Whisper is not activated. Okay, I was pressing my Whisper button, but that is okay. So, I always thought that was something for vibe coders, but I can express myself better when I can just talk to somebody. Some people can write very well. I can't. I cannot express myself very well with text. So, explaining my thoughts to the model like I would do against somebody else really makes it better. Because getting good results from AI really comes from giving it the correct context and the correct nuances. It's very important. Because the moment you're going to miscommunicate something, it's going to miss make something. It's going to misplace miss a lot of things, so it's very important. And writing text, you can be lazy, and you cannot be lazy. You need to really, really talk in depth, you know? And I'm a yapper, so why not Whisper? All right. The next thing is like a green field. Green field feature. Very important. Because these two things These things are different. Green field feature. ### Segment 12 (55:00 - 60:00) [55:00] feature. We will do a little bit of Q& A after this. — [clears throat] — This is always for me. I'm doing it all the time. It's grill me. And it does not need to be the skill that I just showed you. This is bot [ __ ] You can also just talk. Sometimes I say like, uh I want to build this feature where I want to do yada inside yada where a user can do yada yada, but also might be potentially You never guessed it. Do yada yada. Walk me through this. Don't With clouds, don't write code. Always need to say don't write code because the [ __ ] is always going to start writing things. And that's something I don't really like cloud models, but you cannot communicate with Codex or GPT models at all. It's just impossible. I wish they could, but you can't. This grill me skill doesn't really work that well in Codex for some reason. It's I don't know what's going on with these models. They are very smart. They can get the [ __ ] done. They are very fast, but they cannot communicate. They always over-engineer all the time. So, I basically People say, "Yeah, I'm building my plans with Codex. " My architecture with Codex, but Codex I don't know, bro. This thing over it it comes up with the most terrible variable names, the most terrible function names. It over-engineers everything. Like literally everything. And the moment you made a mistake, you know, you're talking with it, and you make a small mistake, you push it into a certain direction, you're cooked. It's like you're [ __ ] Because the bastard the bastard is too far. He's like he's all in. He's all in on his plan. And then if you want to change the plan, it's going to dump you 20,000 lines of bullet points again, and you're like, "Oh, where am I? " And and it's [ __ ] up. It's like it's crazy. No. Codex cloud is the the cloud thing is are way better. You do a grill me. And it's not that the model is going to tell me what I need to do, because that's also a problem. People are asking questions to a model. A model is How can I say that? They are not smarter than you. Well, they don't have the thing is spitz and gefühl. heart, bro. They don't have the engineering. They have the the the correct engineering mindset. You know? They cannot architect the way that I do because they suck in that, in my opinion. They come up with soy dev architecture. The reason why I always do a grill session is so I know Sometimes they're creative. Not going to lie. Sometimes cloud can Hey! That's a good idea. Some I'm going to be honest. Sometimes Cloud is Cloud is Maximus Aurelius comes with a very good thing sometimes. The moment I do these sessions is so that I know that Cloud is Aurelius Maximus knows exactly what I mean. And I try to steer the the thing into the right direction. And at the meantime, you create this context window of possible solutions. And even I'm always ask it to give me code examples. Why is this so important? Because this is naturally what we do. So, how I attacked If you see my videos, if I want to build something, I want to build What are we [snorts] building? We build something. Give me something we need to build. And you are all short on this. No. Something. If I want to build something Sometimes we do not know the correct Like I said, I mentioned this before. We don't know the correct path. path. I don't know in a straight line to get to this. I don't know. It's a question mark. ### Segment 13 (60:00 - 65:00) [1:00:00] So, I start building. It's like, "Okay, what's the first logical step? " Okay, I'm going to build a server object. All right, in Go line. I'm going to build a server. How many times did we Did we wrote a server? It's a structure. I'm going to add a listener to it. A listener. A listener thing. Or a handle. I'm going to add a map of connections. Come. net. DotCom I'm going to do this. I'm going to have a read loop. write And you start building block by block. Sometimes you're going to change your stuff up along the way because you're testing and you're basically There's a cool stuff of of program. I really love that. Like this exploration phase where you know roughly how to do it. But you do not know like the little nuances here and there, but we will discover them as we go and build a thing out. All right. Well, it's the same thing what you do here with this really session because sometimes you needed to rewrite. Sometimes you're wrong and you need to rewrite and it's annoying and then you have like this Yeah, we created slop as engineers. I created slop a million times because I didn't know where I was going to. So, I built the wrong things and I deleted some stuff and my tests were outdated and my docs were outdated and we were lazy and it kept there and you know it's like this time it's going to be different promise to yourself, but it's not. At a certain point of time we engineers create entropy. The cool stuff is we don't need to go through this phase anymore. All right. You anymore because we can have this phase with the AI. I know exactly how small blocks need to be laid out, how I roughly want the architecture to be. This grill me session, I'm going to ask it He's going to ask me questions so I can steer it in a certain direction asking for output structure out like real data structures, function names, variable names, all this stuff. So, at the end of this grill me session we have a nice context window. And that we can turn into a plan. Most of the time you don't need to. can just let it build from a session. But because you have already so much context, you really want to make it into a complete detailed plan. And that's the next mistake is that people just build a plan and it's not detailed enough. Most of the time like Maybe Q-star is doing it better, but like Claude's code sucks so much. It gives you like a very Lucy plan. Look at that. The the thing is the balloons. It gives you a Lucy plan. That's a big mistake. You need to tell it exactly and I really tell it like Make sure the plan is so detailed that when I give this plan to an engineer, he or she I always say it's. So, it can implement the plan without ambiguity whatever you want to call this thing without any problem. File names, file locations, structures, function names, the whole [ __ ] shebang. And then it writes a decent plan. Then I clear the context. I don't know why they removed this [ __ ] function like a context clear right after the plan was so good. And then I let it do the plan. And most of the time it's doing it very well because it's very well in writing codes. It's not as lazy as we engineers are so And the architecture will be nice because you basically had this nice session. And this takes some time. Sometimes this grill me session can take for some Sometimes I have like 40 minutes talking with with the boy Claude. 40 [ __ ] minutes, bro. Seems a lot, but it's not because sometimes a feature takes like a week to implement as a normal engineer. and the suffering process. But now if you spend 40 minutes in such a session and you really take your time and you really review everything like the code that it's going to do that it's telling is exactly like you had it in mind. It's going to be perfect. And if you keep doing that you will never have slop. Never. There is no slop whatsoever because you're guiding it. I swear to God there is no slop. Sometimes maybe there's like, "Ah, [ __ ] This is not good. " Or this is like If you keep it very focused, it will never it will basically never create slop. I never seen it done doing it. Sometimes like a performance problem. Sometimes it does some things way too optimistic. Um but yeah, you need to catch it. All right. Then you say, "Ah, man. What are you doing here? You're do not ### Segment 14 (65:00 - 70:00) [1:05:00] pre-allocating. " But then you have some skills that can help you or like things in your cloud. md or agent. md where you say like always pre-allocate arrays if you know the size or something like that. Or [snorts] or pre-allocate large enough. Something like that. These things it does, but we do the same thing. All right. Green field Always do a grill me. Then maybe it's like another feature like a small addition or something um where you really know how it needs to be built. features Where you really know like, "Okay, for example, man, what the hell is going on? " We had the TPO layers which is some kind of a indicator, right? And it needed to add for example like an extra thing, an extra setting. I know this code out of the top of my head. I can basically [clears throat] say, "Add this to TPO. Make sure this and this. Also add the settings. Make sure it's translated. " Enter. And it will do it because I know because I really take care of my code base because I really I read everything. I basically because I I read it by talking to it, you know? I'm writing the code by talking to these AIs all the [ __ ] time. I know exactly that it's not going to [ __ ] up because it's so straightforward. Simple as that. And sometimes you already know like, "Okay, this is how we did it before in our code base. " Then you can say implement this feature. Make sure it uses the same pattern as XYZ. Something like that. And if you do that, exactly what I tell you, there is basically no slopness. There is no There cannot be slop. There is never be slop. Unless you guide it wrongly. But the moment like I said, the moment you're going to use something like these GSDs where you think you're going to be rich by spending one day of a document and then let it write your 1 billion SaaS application. It's not going to work. Never ever ever ever. That's how it is. Yeah. That's how I do it. Thank you everybody for the follows or maybe a donation. I don't know. Uh because I don't see it. Like I hear like um Thank you Gert for You have super chatted. Thank you so much. I don't know what the currency is. NGN. Uh but thank you so much. You don't need to do it. Being here with me. Listening to my yap. Is already the best gift you can give to me. Papa Tony, will AI take our jobs? No, it will No, it won't. But actually I'm going to be super honest. Wielding using AI very well is a skill on its own. Remember Remember I remember this. In one of my videos years ago some some influencers or like these these gurus, they came up with the new terminology prompt engineering. Remember that? And they say something like They said something like In the future we're going to need We're going to hire prompt engineers. Guess what, bro? The [ __ ] were right. They were right. Everything what They were right. We were laughing at it. We were talking them soy devs and and and blue-haired boys or pink-dyed girls dressed as a puppy. But they were right. It's all about the prompt engineering. It's all about knowing I can just like I can prompt something and I see the AI thinking. You can see it thinking. The moment I see a keyword like, "I know exactly it's going south. " And I just press I escape. It's like I don't know what it is. You get a feeling for it. I swear to God it's hard to explain. But you get a feeling on how these things behave. I I can when it's done with a task and I check my I only check this. I only check the get diff. I always have my editor. I don't use the AI views on cursor. I don't use it. I always use plain editor because I want to see the code and I don't diff. I want to see ### Segment 15 (70:00 - 75:00) [1:10:00] the code laid out with the rest. So no no default GitHub diff. I want to see how the code is written inside of my application. this is a huge code base. this thing is and it's a complete web app inside of it as well. It's a huge code. This is 1 year and a half of writing code every single day. It's insane code base. And you can see a couple of dogs here and I clear them always out. This is these are new things because I want to generate bindings. I want to recreate my own bindings for within from a foreign GUI. So I'm going to regenerate I'm going to write it myself. So I'm busy with a plan. And the reason I write it is because I don't want to lose it. It's it's very comprehensive, right? And I talked with the model like for for so like everything like every single thing is being described exactly how I want it. Exactly. So the only thing I watch is this. I only check source control. And then I basically for example here main. That's what I do. That's how I review all every single file. Right? I check this and I go from these green things in the sidebar. You cannot see it. I just go with my mouse here, right? And I just check okay. What is it doing here? Right? I'm checking this thing. I'm just do this. And I want to see the whole code base. I don't want to see a get diff. I want to see it in here. Right? and I check like okay. If event is null return. I wouldn't do this but yeah, sometimes you need to let it go. You know what I mean? Uh these things, right? I just check it all the time. This makes sense. And sometimes when I see something that doesn't make sense I ask it like what are you doing? YOU'RE ABSOLUTELY RIGHT. WHAT WHY DOES IT CHANGE THEME? It changed theme and because I know what the [ __ ] I'm doing, I think it does a schedule here. I think it's going to emit it's going to do a schedule of the frames. Render mark dirty exactly. You see what it is? Why do I know that? Because I know the context. And this render mark dirty is when we going to trigger the render because I'm optimizing so we don't render 60 FPS all the time. I'm going to make it into an I'm basically Yeah, this thing is pretty wild but I want to make it into an um only render when needed. So because I know what potentially should be done there the moment it's not there is a problem and if you do not know what's going to be on in there, then you have no understanding of your code base. All right? That's basically how I do it, guys. Um So are we going to get replaced? No. No, you don't. You can't. The AI can't write full systems. They can't. You can be a CEO. You can be the smartest prompter in the world if you do not understand codes you're not going to make it at all. Render the chart clean up dirty. You know? Okay, let's see. I also just recently finished watching it. It was very nice. UTXO was confusing at first time. Are you going to do ghost streams anytime soon? No, man. Like listen like I don't it's not that's not where the where where the the landscape I'm going to be completely honest with you guys. If you guys do not learn how to work with AI in a way that I'm telling you I'm not going to say that my approach is the best but I try I spend a year. Guys, I'm not writing any code ### Segment 16 (75:00 - 80:00) [1:15:00] anymore. I swear to God and if people know me, this is a serious thing. If you cannot use AI then it's not going to work out for you. So is programming being easier or is it harder? It's harder. It's getting actually a little bit harder because Why is it harder? This is actually this is pretty funny but I think because the expectations are higher as well. So how can I explain this? If you are here and you are a 10X engineer normal speed. In 2000 25 that was okay. Now in 26 and in 27 it's not okay anymore because they know. — They going to they going to want you a 10X engineer, bro. You know what I mean? You need to be a 10X engineer. And if you going to go at the speed of 10X by just blindly prompt in your way to success, you're going to end up in the wrong wrong. I'm I I'm always very scared when I need to let people work on my code base from time to time because I know they don't know. use AI the way they need to use it. I know for sure. I'm so scared to let people work on my code base because I know they are not using it correctly. They are they don't know what's going on. They probably think it's okay. It works. It always works. That's the thing with AI. It always works. Most of the time it's always going to work. Look, this feature is fixed. Yes. But the next time somebody needs to come in and and prompt something, it's going to it's it's bad. So that's why it's going to be way difficult because you're not going to learn as much anymore. That's the problem. Like people coming in new devs, they don't learn about architecture, about performance, about these little nuances that you need to find in your code to make product production ready performant applications you don't know anymore because most of you don't look at the code anymore. This is the difficulty. What's going on here? That's the hidden trap. They expect you to go 10 times as fast. You don't learn anything anymore. You're going to [ __ ] up sooner or later. So it's going to be harder which is very good news. Sorry from sketch one bring you forth. Please address Jonathan Blow hitting on on you for MNT. But these guys like you're going to see you're going to like guys like now people are divided in AI. You have the maxis which are basically people um using stuff like PRDs and Ralph Loops. Yeah. PRD's Ralph Loops Uh um Agent Swarms Open claw The [ __ ] the the CEO of Y Combinator, what's his name? Gary. These guys, like the people had no idea. No clues. The Maxis ### Segment 17 (80:00 - 85:00) [1:20:00] Then you have the GG's. I know a couple. Ashton Anthony GG Web Dev Cody This guy knows. This guy is the only one that I see on the internet using AI correctly. This dude here. This one. So you can follow him. He also has a course. Not sponsored at all. Not sponsored. Good course. Web Dev Cody He knows. And some others that I don't know. And then you have Wait, there's actually like a face here. Which is basically almost I know one person that is doing it. David Heinemeier Hansson The guy from Ghosty Terminal. What's his name? Michelle? Mitchell? Mitchell What is the creator of the go ghosty terminal? Mitchell Hashimoto Nani? And a couple of others. Maybe the Primeagen. I don't know. Prime Prime maybe. And then you have the people that are falling behind. Skill people These people are the people with the most skills. — So it's not about falling behind. These are the people with the most skills. I'm not going to lie. J Blow Soden A legends is very good. But Maybe the guy from What's the guy from the ready burgers? Red the burger guy Red Anthony, how do you know all of this? Yeah, I know. Yeah, I'm great. Lorraine Flurry. Like I said, I'm my knowledge is Lorraine Flurry. Following behind. Eagle Maybe we can do this guy. You can do um What's his name? Handmade Hero guy Who is that team? What's his What's his name? I know who it is. Uh also a legend. Legend in the space. Uh and that's You see, the people that are falling behind are the people with the most skills. What is his name? Um Handmade Hero. Casey Muratori. Casey Mura Muratori Yeah, if this is a thing, empty behind. So if you are a senior level knowledge already, AI is crazy. It's really It's crazy because you can steer it. You can check problems in the blink of an eye. Because like if you're senior, you the way it places a file somewhere, the way a function is located, the way it uses things. I can [ __ ] see what it's doing, bro. And it does it a lot of times. I think, what the [ __ ] are you doing? This is not the way. But spotting these little things knowing that it's might going to make a mistake, you can just put it into the prompt. Make sure you do it like this. And it will fix your bug faster than the speed of light, faster than you will find it. It will write if you do it with small iterations, small scoped features. Maybe you have two windows open, like two prompts open, and you just do a grill session here, and you know, and and you don't need to have five open, just two. And you're going to be These guys are falling behind. That that's might be to do with their character or status or something. You can see almost like the He was skeptical. Give him Let him cook. Let him cook for a couple months. Mitchell Hashimoto, the same guy. He was skeptical as [ __ ] ### Segment 18 (85:00 - 90:00) [1:25:00] Bro Bruno's Bro realized I saw it on Twitter. Bro is realizing, so he's coming. The Primeagen, I think Prime really realizes it, but for Prime Prime is in a different in a different uh Prime is a is is a role model, right? So But I think Prime knows. I think Prime knows what's going on. He's smart enough. But they know they're doing it, but they're keeping it cool because reasons, right? I don't give a [ __ ] What I mean? Like I'm just going full ham on this because it's so good. And then the Maxis, these are idiots, right? They they are they're producing slop. But they are maxi. That's why they are maxi. They just blindly they type something and they are maxis. Like What's his name? Andre Karpathy — Maxis, slop slopsters — [snorts] — And you could say, "But where is Peter Steinberger? " To be honest, I think Peter Steinberger does it very good. Does a very good job. I think he's very good in in steering and checking. I think he does the same approach. I don't think I think Peter Steinberger is here as well. Although I hate open claw. Uh I think it's slop, but he's smart. Peter. Is it Peter or Peter? Peter Steinberger Stein en burger I really think he does the same approach. I don't think he's writing like PRD file and all that slop. I think he's really like talking with the model and steering it. I think this guy is uh the He's a chat. I'm not going to lie. Yeah, open claw Yeah, of course. But why why is this open claw thing? I really It's like Man, this thing never worked for me. I don't know what's going on. I tried it so hard, but it it's [ __ ] It all. It [ __ ] up all the time. I don't know why. Maybe I just like it I'm maybe I'm a fool, but it just People are doing like super crazy things. That's what they say, but I cannot make it like What's going on here? do to two things right. So If you give people hope Yeah, If you give If you give a blind man back his pair of eyes the man is going to do anything for it. You know what I mean? It's But if his eyes are pointing in the other direction, it's like he thinks he's running forward, but he's actually running backwards, you know what I mean? So But I think we're you can see it going, right? Open It was like this, right? Like Opus 5. 4 Opus 5. 4. 5 5. 6 Uh all these other models open claw all these indie these uh indie SaaS devs How is How do you call that? Like these indie developers uh solopreneurs like, "Yeah, let's go. Let's go. " And now and now where they're all gone because they create so much slop that they cannot fix their code anymore. And then they realize like, "Yes, I have coded my application and I have zero users. So what am I going to do? " Yeah, nothing. Exactly. So these guys are wrecked. They're already here. You never hear from them again, right? again. It's as simple as it is. It's it's over. Only the people with true brains, like these guys here and probably a couple of you, right? And now after this video, all of you. We are here. We know we abuse. We know what to do. — We are not very simple. We are not 20 times more productive. productive by AI. Let's say four times. Let's aim for that. Maybe some people two, some people three, some people five, but I think we need to be realistic and if we have a 2x, 3x, 4x improvement by speed, which we can learn. We can learn by wielding two swords of dragon steel, like two code external ### Segment 19 (90:00 - 95:00) [1:30:00] and contact switching, which I think it's a bit hard. Sometimes I do it, I'd rather don't do it. I'd rather take it slower and really I really want to see what the [ __ ] is writing. Of course, if it's doing a stuff, I tab out, I do something else. You know what I mean? Like I can pop in a little Xena sometimes. That's what I'm going to do right now. Look at that. Oopsie, this is basically the moment I need to do a sip, but I'm not going to do it because I hate the guy. Um but it depends. Sometimes I'm talking with two two things at once or you know what I mean? Or read an article or whatever, like it gives me more Maybe this number can go up, I don't know. — [snorts] — But I think we need to aim for that. Which is already a net win, in my opinion. AI writes better code than us all the time. We need to be honest because it's never lazy. We are developers, we are lazy. You want to write a feature, we miss checking things we need to check. Sometimes we just forget. We just CBA checking this. Oh, I'm not going to check this, I really ain't, man. I'm 100% sure it's going to be correct. Yeah. It might not. AI always does these checks not always, but it's way less lazy than we. It's always going to do extra stuff. Sometimes too much too verbose, but better too much too verbose than too much host. The B mad metal, what the hell is that, bro? Been using cloud code for 1 year and now I've not written a single line of code in months. Yeah, I mean and then I don't write any code anymore, but I read more. I read more code. I can pay more attention to the quality of it. I see things more. I pay more attention to code structure and quality, which I never did because I would figure it out. Because I never realized the time I wasted writing stuff in my own slop until AI produces slop. And I said, wait, why are you editing 20 files, bro? And then I realized like, oh, wait. Now I know. So now I'm really trying to steer it in the correct and I keep rewriting things because rewritings are free, basically. They are free. They are really free, especially in a code base already structured with tests in place, it's so easy, especially with code X. Refactoring small parts when things look a little bit grim. I'm always doing that. Keep iterating. Learning. Trying skills, which yeah. The only thing I like, guys, I swear to God, the only thing that's so powerful is the simplify. Cloud codes. I swear to God. Clouds min dangerously. This is so hard to type permissions. Look, we're going to test we're going to do the cloud simplify. We're going to test it on the code base. So I'm going to This app sucks, I know, but it is what it is. So, look at this. What I always do, this is basically a pattern of mine. The moment AI produced code, the first thing I do is I put it into the staged changes. Always. Before I review, it's done. Hello, sir. I just implemented Okay, nice, good stuff. I put it the first thing I do, I click this. Boom. Staged. Boom. That's what I do. Staged. Not them. Staged. Then what you could do, which what I'm experimenting with, is I do a simplify. This simplify from cloud is max cloud is Aurelius Maximus Aurelius is crazy good. I swear to God, this is so good. I do a simplify. Let him do a simplify. I'm going to take a pee. When we're back You see, I don't I can pee while coding. So, it's going to take a thing. Of course, the [ __ ] is doing this, he cannot do this. This is he will know. It's in the cloud that I'm day. Be right back. ### Segment 20 (95:00 - 100:00) [1:35:00] Yeah. Burn the tokens. Of course, I went to pee. What's going on? I need to pee. I need to pee, bro. What's going on? This this thing is busy. So, it's going to take some time, right? But then the simplify thing really works because it really checks like And this is a bit of the problem with these models is that they are very good at they they spit out things, but they are bad to do a retro uh a retro fit fact check. How do you say that? I don't know how to say it, but they If we as a human we write code and we always look back a bit, like we write a function and we introspect it a bit. Is that what it is? You know what I mean? Like a little bit of a of an introspection. Which could change our mind. Basically, we need to think a bit about the feature future, I mean. future. Think a bit about the past, introspect the now, and based on that, which is our reasoning, based on that we're going to produce the next set. And that's where AI where basically these large language models, they can never do that. They could basically do that. Maybe some harnesses could invent this somehow, where they do like a little it's all harness work, right? Because LLMs are basically just like in out. There is nothing in between. It's like in out in out. There is no step in between. So, ideally they should be like in small introspection where it can already detect like, oh, [ __ ] I'm actually this is not 100% correct to the guidelines or whatever. So, now what is guy is doing what the skill is doing here, basically um I mostly the time though I always mostly the time use code X, right? I always use code X for [ __ ] but the only thing I really like is the simplify uh skill, which doesn't work in code X. I copy I basically reverse engineered the exact prompt because it's a system thingy. Uh but it doesn't work that well. Code X is not suited for human problems, I guess. Um It spins off a couple agents each for a specific domain and they verify it again and then it goes to fix stuff without breaking your stuff. Sounds too good to be true, but it's really good. Um Look, is that is isn't this something? This was what I saw as well, right? Somewhere, not not the exact same here, but I said this new check I wouldn't do it. Remember? Just rewind the video. Well, the it fixed it. Of course, it these things most of the time do not need to be. So, now because I staged these things first, I know exactly what this new thing did. How much it cost a month? Yeah, you you're going to pay some money, but it's worth You know what I mean? It's like if I can duplicate myself, that's already in the number between 10 and 20k a month. So, I told recently I told uh to my co-founder said, "Man, I really hoped they come up with a model that costs two, three, four K a month. I would instantly pay. " I would instantly pay if though if if there was like a Codex 5. 5, GPT 5. 5, 5. 6, GPT 5, 6, or whatever. ### Segment 21 (100:00 - 105:00) [1:40:00] No, it's like 4. 5, GPT 5. Whatever. But all these [ __ ] But all these numbers. And they say, "Okay, but this is powerful. It's 4K a month. " Instantly. Because I know what benefit it gives me. It's like insane. It's crazy. So, you can see what it does, right? What Why does it do this actually, to be honest? I don't even understand why it does this. Is this a bug? That new check, yeah, people ask parameter on remote connect one caller, always the default. I hit in lines at the pink side. This is what I said. This is exactly what I said. I wouldn't do it, and it removed it. This is Codex code, by the way. All these changes are from Codex. Cloud is really good at these things. So, actually I don't understand this. I really want to know I take my time to ask what the [ __ ] it's doing. I don't want to spend like 20 minutes finding all this [ __ ] out. I want to It's going to explain me. Ah. It avoids a second call. Yeah, but where is the second call? Ah, this is another function. Now I see what it is. Render schedule week in, and this is render schedule week at. Okay, that makes sense. Does it exist? I can't It's gone, Anthony. Just accept. These things doesn't really matter. These are basically like generated stuff. Why is he doing on? When I see red lines, I like it. That basically means he removed stuff. The moment that I start panicking if I see use effect, I instantly trigger. I don't want to see any use effect in my code base. You see these things. That's basically kind of a thing. But I'm always reading these stuff, right? And then you can just do these things. Boom. You merge them together, and you can do it. Now I do a review. Now you can do Now I do like a Codex open up Codex review uncommitted changes, enter, and just go smoke a cigarette or something because it's going to take a while. Right? And then you need to basically never say like Some people say fix all, but some issues are garbage, right? So, you really need to go over all these issues, and basically check and ask why and how and yada yada. So, the whole process of implementing a feature or a fix can take some time. It really But that's also where you stop the entropy. Okay? Any questions? I can get you Cloud Codex for only 1750 a month. Yeah, sure. Yeah, but I'm just telling the analogy of the the story of the 4K subscription is that I really think it's still going to be beneficial. Maybe not for a normal engineer, but for like me, it's going to be so beneficial. I would pay 4K. Of course, it needs to be at the Never for an office model because I know Anthropic's going to [ __ ] up. 100% But for something from OpenAI, yeah. If OpenAI comes with its own mitts, I would pay for that. Yeah, baby. Pay to win, [ __ ] Do you think you will lose your ability to code? Um That's actually a great question. I thought about this as well. Hm. You know what? Like You know what the thing is? Is that writing good code is not always correlated with being a good engineer. Because reading code is actually way more important, in my opinion. And if I basically reflect on my own ### Segment 22 (105:00 - 110:00) [1:45:00] life, on my own engineering life, especially how I wrote stuff in on my YouTube channel, I always could things I'm very good at making things work. Right? I can as a problem, I will solve it. Cuz I'm a very good problem solver, which an engineer really needs to be. But did I wrote good codes? Like from a structural standpoint, from a teammate perspective, from a four to five to 10 men's team perspective, I don't think so. I always made it work. I came up with creative ideas. But was it really production-ready code? I don't know. So, it might be. It might be production-ready. But was it Could the code outlive? I think the only thing I wrote very well was Hollywood. I think that's a piece of engineering I can be proud of, but that's also the only thing. And of course, Market Monkey now as well. But um writing code also misses the ability to make sure you keep your code alive and healthy for a long time span, for your colleagues, for potential buyers, and so on, and for yourself. Now I'm reading more codes, which basically helps me learn the skill I hated for so long. I really hated to review codes. I never almost did it. I came always up with my own ideas, with my own idioms, my way of doing things because I could. And now I'm on the other side of the spectrum, where I really understand how code works and how it should work. And I'm now learning and growing as an engineer by the other side of reading and maintaining it. So, I think this is a nice second journey of my career in it. — What's going on here? I feel like I'm slow right now. I'm not an engineer. I'm a finance guy. That's all I do in my life. What's a good starting point for me to learn? That's a good question. Like I said, there are two There are a couple of different persons, right? Like perspectives that use AI. If you are a non-technical person using pre-existing tools like that generate these MD files and this context management like a get [ __ ] done and all that stuff, they really can help you because it can really turn your idea into something that works. And that's fine, right? So, you cannot actually mess up anymore, in my opinion. You can just prompt your way to success. But from an engineering perspective, if really engineering matters, like I'm We are maintaining me and Ashton, like we are maintaining a code base. We have like thousands of users connected live, which seems is a lot. It's a lot, bro. It's the most complex things we have even in the client, even on the server. It's It's so complex, bro, what we doing with MMT. It's crazy complex. It's not easy. It's really not easy. It's [ __ ] hard. We even have like Without technical skills, it's impossible. You cannot vibe code your way to something like this without shooting yourself in the foot. It's impossible. So, I wouldn't be too worried to answer your question. I would just let the model guide you and you will get somewhere. Because you're not do not need to be as technical as we are. Because probably your problem point is different. All right. So, I'm talking here more from a technical engineering perspective as you are an engineer maintaining something great. Well, you know what I mean? Like something substantial, right? But if you want to build something like a small business ### Segment 23 (110:00 - 115:00) [1:50:00] AI will do it fine. It's not going to be the best code and it's probably going to end up in problems, but that does not matter because you need to first get your validation, you know? need to get some users, some traction. Maybe it's a [ __ ] idea, I don't know. And just vibe coding your way to success is going to create a problem that I call How do you say it? Like um a problem that's actually not a problem. Luxury problem. It's going to be create a luxury problem that you can solve later by hiring some real technical boys. What do you mean by technical skill? With technical skill, I mean it's if they I this where I don't know where bro did something I can I cannot see anymore because of the changes made where I basically re-prompted this thing. It did something. I instantly saw it. I couldn't understand it and because I couldn't understand it, I was smart enough to ask it why and did some digging. This These little spots is a technical skill. Sometimes it will place a file somewhere that doesn't need to be. Sometimes it's using like uh for example like we are non ga MMT is written in a non garbage collected language. Sometimes it basically always make the mistake of skipping uh arena allocators. Arena allocators is where you allocate into an arena. Into a fixed block of memory that can be garbage that can be cleared freed at the same time. I'm not going to be too technical here. It's really super important to understand how that works especially for very high performance rendering computational stuff. Real-time applications like we are making. If you cannot spot when to use an arena allocator and when not to, that's a technical skill. And that's where people with no technical skill they never know. They don't know what it is. So, whatever the model is producing, you're going to take it for granted. While I'm not. I'm going to see like what the [ __ ] What This is something for an arena. And then it will implement the correct thing, but sometimes it misses this and that's where a great engineer will be separated from the slop. It's where you can spot these things by checking code, by understanding what you are doing, why you are doing it, what is it trying to solve and so on. So, knowing all your data structures very important. You know, like a dag, a b-tree, self-balanced b-tree, uh linked lists, a double linked lists, binary search, linear search, O N, O N + 1. You need to understand that. Because if you have a problem, you need to implement a feature, you can guide them all because you say like, "I would I think I would recommend using a b-tree here because A, B, and C. " And then the model is going to say, "Yes, that's a good idea. " You know? But if you say like if you say nothing, it might use a map which is not wrong, but it's not the best possible solution. You know what I mean? These things is what I call technical. What's the interview process for joining MMT? Uh that's a good question. Uh do you still use Go for the server? Yeah, yeah, we have big Go project. Everything on the server is Go. Everything. Except the server-side scripting. A bit of the server-side scripting is C++ V8 and engines which I don't understand because I didn't wrote it. How do you think about the And that's the last question actually. How do you think about the interviews for hiring process going to be for Defro? I don't know. I cannot imagine how it goes. How do you do this? Because this landscape is going to change forever, right? I mean, how is that if I would hire somebody, I really want to because getting an getting a junior on your team is hard, but getting a junior with AI on your team, that's even [ __ ] harder because they are going to [ __ ] mess up, bro. mess up so hard unless they understand, but they don't because they're junior, because they need to learn. So, how you going to fix this problem, bro? I think it's going to be a bubble. 100%. There must be something going on. There must be a bubble going on where companies are going to explode because of the slop they are producing. A tropic. Yes, you. Are you using Codex? Yes. Codex most of the time. How are you handling data privacy? We don't handle data privacy. Of course, for our users we do, but what I'm going to do like It can know my code, man. It can know It ### Segment 24 (115:00 - 120:00) [1:55:00] It can know my code, man. It can know all my prompts. They're going to ban me because I'm basically swearing and screaming at [ __ ] cloud all the time. Maybe I've $2 million company to create a billion dollar company. Um MMT still hiring? Asking for a friend. There is always place for great engineers. What the [ __ ] this? I'm still jacked. Yeah, I know. It's crazy, ain't it? Don't send insensitive data. What sensitive data are you going to send, man? What are you going to send? What sensitive data I'm not sending it any data. Only in my bald brain that data goes in. I don't even know what it does here, steps, but it's fine. I just seduce my Back off. I replied to the previous guy. Oh. Sorry, man. Hey. But this really crazy times, bro. I'm thinking about this every [ __ ] day like who has thought this, man? Who had I could never think about this 1 year from 1 year before like what the [ __ ] going on? How is this even possible? It's solving a lot of problems. It's going to create a lot of problems as well, guys, because it's not anymore like hiring is going to be way harder. It's Yeah, I always do voice inputs. Yeah, yeah. Yeah, of course. For long sessions. Yeah, always do that because I can express myself better. It's also faster. Uh why are you CLI tools and not the cloud code extension code? Um Because actually this is just I I actually never use cloud code. This is just for demonstration. I always use Codex and sometimes I use cloud code, yeah. Depends if I want to chat something or otherwise I do it in um It depends. Like you know me, right? Sometimes I open up a cursor to do it. the CLI. Depends. But I'm Codex is my buddy. Codex, that's where I do all my work. Because my Like I said, only for like big features where I really want to talk to something, I most likely use cloud. Otherwise, it's just Codex all the time. It's so much better. It's more pragmatic. The only thing that Codex does is over-engineering, man. Oh, this thing over-engineers like crazy. They need to tune that down. It uses terminology, man. Oh. Ah. But they all have their pros and cons and I think you need to juggle between these models. Sometimes I use Composer 2. I swear to God. Especially for UI stuff. I use Composer 2 for UI all the time. It's so fast. Composer 2 is the best thing for UI in my opinion. It just goes ham instantly. Um Do you ever use a word ultra think? I don't think it works anymore. I think ultra think just put it to max, which probably doesn't work if you use the 1 million context a model. Something is going on with the 1 million. You need to turn it off by the way. This the 1 million don't use that. It's not good. Always use the normal one. You need to do a special setting for that, I guess. How many subs do you have? Two. I have two subs. And I have three subs. I have good sort, which I forgot to uh remove. I have good sort. I have cloud max 200 and I have the codex 200. So, that's $600 something a month. $500 something, which is nothing, right? It's completely nothing. I mean, that's not It gives me basically a couple more engineers, right? I mean, if you think about it, if you if Let's say that I I hire ### Segment 25 (120:00 - 125:00) [2:00:00] if I hire a mid-term engineer, like mid, not the greatest, not the not the worst, mid-term. Mid-term engineer. What are you doing, bro? Mid-term engineer that already has an eye. The mid-term engineer. How does it work? I create a task. All right. There's a bug. I need to explain it. Yo, brother. There is a bug. Give it Mike. Timmy. Timmy. Let's say Timmy is my co-worker. I'm the boss, CEO of MT, the big CEO of MT. Wink wink. I hired Timmy. TIMMY! COME! TIMMY, LISTEN. A user reported a bug. Yes. So, basically, I'm going to show it because Timmy Remember that we have the TPO? Yes. Well, it has these extensions, right? Timmy. Do You don't know what it is? Okay, I will show you. Here, Timmy, listen. Average GG. Let me show you. It's in Chinese. English. Listen, Timmy. So, you open up the TPO here, right, Timmy? Where is this thing? Then, you basically set the value area, make it extended. Okay? Timmy, you see these things here? These labels? Apparently, I'm going to have sucked this out of my thumb. Apparently, that when you hide the indicators, these labels stay. You know, like we hide the indicators, but the labels stay. Timmy. All right. So, I think you need to check where the TPO code is. We have a function there called render axis labels. It might can do some interference there. Can you fix that for me, Timmy? All right, thank you, Timmy. How long do you think Timmy needs to fix this? Mid-year engineer. How long? How long Timmy needs to fix this? Like, tested, written, tested, and ready for review. How long? Chat, how long Timmy needs that? It's not It's not the biggest bug. It's not the craziest thing. Right? Come on, guys. How much How many hours does he need? Or minutes? Without cloud code. We're talking about Timmy. We're talking about the non-AI Timmy. Two to eight hours. Then Timmy will be fired. — And I'm going to get Timmy out of house. God damn it. Half a day. That's I No, not half a day, bro. It's like Is Timmy mad? Is Timmy? Give Timmy some [ __ ] credits. He's going to take the whole sprint. — No. I don't know because Timmy is probably busy with something, right? That's the first thing. Timmy is probably already working on something because otherwise, what the [ __ ] is he doing? You know? He's already has something open. He's working on a project. feature, a bug, whatever the [ __ ] Timmy's doing. I don't know. We need to check his his notion list. So, maybe Timmy can pick up the issue in the afternoon. I SAID, "TIMMY! THIS IS PRIOR! " Timmy has said it's prior, but still it's prior. He cannot drop his work. So, he knows it's prior, so it will He will pick it up in the afternoon, prior. So, let's say it takes 1 hour to do it. 1 hour. It's like That's [ __ ] 1 hour in the afternoon. It's still morning. So, it's like 3 hours. It's like in 4 hours from now this thing is shipped. At the best case scenario. Because maybe Timmy is going to get a bug problem, like a memory leak or something. Like, "Ah! " And he's [ __ ] Like, you know how that goes, right? So, Tim We don't know. It The best possible scenario is 4 hours. If I explained it the exact same way You can re- re- rewind the video. If I explained the exact same ### Segment 26 (125:00 - 130:00) [2:05:00] to cloud or to GP to codex, the exact same with my voice app, and I took a screenshot of the labels, this [ __ ] would do it in 5 minutes. In 5 minutes that would be done. So, I don't even know what we're talking about. the initial question was, to be honest. My ADHD brought me here. But, I just want to give you a point here that why would you hire a mid-term senior the mid-term engineer if it's going to take 4 hours whilst you can pay $200 to get your [ __ ] fixed in 5 minutes? I rest my case. Ask him if the other my opinion on the websocket Yeah, maybe [ __ ] Timmy is on a holiday. "Ah, I don't feel well. I going to go with my wife to IKEA, buying some furniture. " [ __ ] IKEA! GPT doesn't care about IKEA. GPT doesn't care that I spit on him all the time, that I scream and yell and tell him he's bad, and then I'm going to fire him, and I'm going to hurt his family. He doesn't care. He's never tired. He's never stressed. He's always Whatever I do, whatever I throw, GPT is always pro GG. So, why would I hire Timmy? But, wait. So, let's not to say we going to hire like the the intern. You know what I mean? Like the junior. Why would you need a junior? Like, is it Why would This is Hey, I don't like this, guys. I swear to god I have nothing against juniors. I was a junior. There will be You're probably a junior. There is nothing wrong with juniors. We all were junior once in our lives. But, it's going to be [ __ ] hard for junior to beat to beat our models. Unless copy I can copy myself or somebody that does the same approach as me, that knows his [ __ ] in his domain, knows how to use AI and how not to use AI. I'm still thinking about it because it's going to be so [ __ ] hard. It needs to be a beast of a monster. one incredible person. Because guys, listen. It's not so to say that I do not trust people using AI. There will be great people doing that. But, it's still I need to still explain how things work. Like, I still need to explain. So, I want to change how the heat maps work. That's something that I want to do. Which I still do not done it did I take what Getting out of here. Yeah. Which I still did not have done yet. I want to make the heat maps. Why is it [ __ ] slow? Because I was way too zoomed in or something. I don't know. Nicht gut. Why are we here? What's going on, man? Look how slow they are, bro. Anyway, I want to [ __ ] I want to do some stuff with them. The problem is that I cannot just prompt the AI with this. This is a problem that I need to take my time because I don't even know exactly myself what I want here. I know I have a brief direction. I need to discuss this with my boys. And I'm not going to use Claude because Claude is not going to solve this problem. ### Segment 27 (130:00 - 135:00) [2:10:00] problem. I already know that. Opus whatever [ __ ] Opus model Nicht gut. This is GPT work. This thing is hard stuff. So I need to go back and forth with GPT for a while. So if I can find the person what I'm looking for now. I need to find the person. So I do not need to explain the problem what we have with it. And what I think should be done with it. And which are the potential routes we need to go with it. Because explaining this to Timmy, Jimmy or Candy is going to take me the same amount of time explaining the problem to Claude or GPT. Are you seeing Are you understanding where I'm going right now? So if there is a person which I can just say like James check the [ __ ] heat map codes. F Tell me what the [ __ ] going on. How can we make this better? more memory efficient? How can we make it more this, more that, more thing? And I just leave it alone. That's what I want. That's what I need. Right? Ideally, James, the heat maps I think it's time for some little fuzzy wuzzy. And James said, "Okay, boss. " "Okay, sir. " And James goes away. And James comes back. Fixed. I used a special compute shader pre-computing the textures making sure we skip the visible range. We don't do the We don't basically upload every time we are using view transformation matrix which I tweaked. I tweaked this by not using the offsets of world position which we already were doing because we need to fix what problem? The float 64 problem which doesn't exist. Unless you use WebGPU. Well, it doesn't exist in WASM because it doesn't exist yet. Then I'm going to say, "Fucking James, you're hired. " That's the person that I need. That's what you need to become. Because everything else I can spend my time explaining to Claude or GPT because it's the same amount of time. And GPT is going to do it way better. Do you understand where I'm going? This is the person you need to be. There is a problem. Go [ __ ] figure it out. You have all the tools you need. Who is telling an overview of our history? And I think Dirk, to be honest, that's why I would instantly hire Dirk. Dirk us I already worked with him. I would always hire Dirk. I always told him that he's studying biomechanical crazy stuff. He's too smart. That's why I would hire Dirk instantly. Why? Because I can just say, "Dirk, we need this. " Or Dirk comes to me myself like, "Yeah, I think we need this. " And he already did the How Dirk works is like he all The guy doesn't get paid. Well, we pay them for a couple of work, but Dirk is somebody that comes to me out of nothing with some GitHub repos and some examples and some stuff he already experimented with. Problem is that I would not thought about features because it's just like way out of the current scope. He already fixed it. There is research, used AI, whatever. Systematically. That's the That's the people you want right now. want. I had a call with I always have a call with my co-founder CapTrack. We are basically married. No homo. And I told him like I think it was yesterday. I told him like, "Listen, listen, Cap. There are two people in this world. You have the suckers. ### Segment 28 (135:00 - 137:00) [2:15:00] And you have the blowers. Right? And the suckers they suck energy out of you. They suck. They drain. They drain energy. I cannot write it. I know. But I'm a blower. And the blowers they blow energy. They enrich your life. They always come up with something. They call you in the morning or they come up with a message at DM and they come with something to you. They just come with something. They help you instead of sucking energy out of you. And what I do in my life, guys, and I take this very seriously, literally. I do not want to deal with suckers at all. I don't want to get drained. sucked. But do not want to get drained. Technically. You know what I mean. I want the blowers. That's the guys and girls I want. So thank you so much for joining this live stream. I hope it was a fun one. And maybe I have a bit of a yap too much, but I had a lot of months that I didn't yap. So I yapped it all out right now. And remember, guys, I always love you. Don't be a sucker. Be a blower. I love you that [ __ ] I love you all. Guru shurus.