#33 Tailwind Paradox, Andrei Kaparthy, MKBHD, Claude Code

Happy new year. That sound. Happy [laughter] new year. Happy new year. How is everyone? Sorry, I'm looking a little unprofessional now with my dressing gown on and not uh not professional at all, is it? We have similar like colors on. So like Oh yeah, we do. [laughter] We're merging merging into one. Get this set up. So, um, how do we start? It's been a while. Like, I I feel like we're we're rusty. We haven't It's been two weeks we had off where we haven't done anything and now we've got two weeks worth of news to condense into. Well, we managed to shrink it much much more down because we had loads of nonsense in there. But yeah, we got to got to loosen up, try and get back into the swing of things. But yeah, pretty cool to be back. I'm really happy to be back and a lot to talk about like and just news and also what we've been up to. Well, yeah. What have what's been going on the last two weeks? Well, a lot. We had the pre Christmas time. [laughter] We we had the pre Christmas time and then we had the post Christmas time. So pre Christmas time I spent a lot of time with um I cut and payload. Oh yeah. So a CM an a headless proper CMS and that is crazy to call say proper CMS that is opposed to you know years and years of using web flow prior to that WordPress and now payload and it just what was the what was the biggest kind of like holy crap what I've been what have I been missing out on? So with payload you can define like super custom properties. You know how in web flow you have image property, text property like select what whatever with payload we just have like super custom properties. I don't want to reveal too much. I will but not today but it's like very custom like any property type that you want. And then within the same CMS item, you are not limited to one of them or five of them. You can just add as many as you want. Basically, yeah, nested. Nested, right? Is that what you're talking about? You can nest stuff. So, so I'm nesting code components within articles with properties going next to them and just like in a free form and just like ah it's nice. it it and it just works and it's super like customizable like I can make essentially an article can be like super custom there is no limitation uh and we are rendering it by the way still within the web flow CMS so we are adding everything mixing it with web flow CMS I was going to say are you how what's rendering this is it an nextjs website or I believe so this is done by iicot he he wrote the whole. But then how are you getting it into web flow? I should check it. I don't know. Be be careful with your Rubik's cube because it's right next to the microphone. Oh, yeah. Yeah, yours as well, but mine is probably much more sensitive. Yeah. Um, yeah, I should ask him more. I don't know the specifics. I don't want to give a dumb answer. Welcome to the f. Well, welcome to code. The flexibility of writing your own code and owning your own, you know, whatever. Welcome to open source. Yeah. Yeah. Yeah. Although I wonder how long because it's payload they got bought by framer. I wonder how long Figma. Yeah. Figma. Figma. I wonder if they're going to change. Well, it's open source and what we have is more than enough for us. And also like changing CMS. I mean I know it won't be like super super easy but it's not going to be too difficult. It's we we are now in the different realm. We are operating with like you know in a different level. We are dealing with code and I asked this I I asked I could like how difficult would it be to move away from payload because they are bought by Figma. I had the same exact question and he was like the most annoying part would be the content. I was like, "Okay, that doesn't sound too bad because you won't have like a lot." Yeah, because like my my personal website, samuel gregory.co.uk is uh built on Nex.js and the whole website is built around payload like their starter template and it's very much embedded in the website. So either you're using a headless version or something like this that um yeah it would be it would be a pain to redo my website in any other CMS. I think I think I mean I just chuck you just you just have like 20 AI agents going on uh do doing the work piece by piece. This this is what I'm seeing more and more like people coming with bigger and bigger screens for more and more AI agents uh to run simultaneously and talk to simultaneously and review their work which is getting really fun. Like I could was here and he had like three screens with like multiple agents and he was just just talking all the time. uh and very often like very technical like saying like doing something in a spec very specific way. Um but yeah that that's I guess the new coding era which we will be talking about. Yeah for sure. Um any other news? Well there there there's a lot uh I want to show what I did for Christmas your clients. Oh your Christmas cards. Yeah. Yeah. Yeah. So every year I send a Christmas card to my clients and this is the one from this year. So it's not perfect. We had like timing issues and you know it's Christmas time. I have to ask what is it? It says 1226. What's that what's that number relevant to? So it's instead of the year 2026, it's 12,026. It's a calendar mostly popularized by Kurtskazak. It's a YouTube channel. Okay. And the whole thing is instead of adding creating a whole new calendar, let's add the number one which is 10,000 years to the calendar to have a different starting point which is roughly around the time the civilization we know it as today with agriculture has started. It's about 10 to 12,000 years ago. So when you say this year, it's like you're not celebrating the birth of Christ. You're celebrating the start of the civilization but using the same calendar without messing things up. Uh I mean it's still merry Christmas. Um it's still like you know with the Christmas tree by the way. Do you see on the Christmas tree we have web flow logo? We have blender. This is what you we used for creating this tree with. We have Figma. We have uh my agency, we have flowing, we have unicorn studio. So we have we added some no command AI logo though. [laughter] That's true. Shame on me. Shame on me. Next, by the way, like command AI logo would be so perfect. It would just work so perfectly, you know, to hang it. But ah yes. Uh, and this is a this is a alpha channel video made in Blender. And then oh, by the way, the the main the the main showstopper is this. Like you you pull it down and then shake the snow. The Christmas song plays and the the snow with WebGL snows down. And I I sent this to my clients and every client got their own like specific messaging with their own like screenshot of the thing. Like I did it for every single one of them. It took me hours and hours to do it for all the clients. But yeah, worth it. Worth it. Yeah. Watch the money just come rolling in now. They'll be like, you know what? snowing like the coming down like the snow. Exactly. Exactly. Yeah. Yeah. It's very good. Yeah. I didn't do one. I was too I was too busy this year. But um Yeah. Yeah. We did two actually. We did this and we did also another one if I can find it quickly here on our our half launched uh platform. Well, we are we have launched but we have we are still in beta. Uh let me show you the other one we have. So we did also this one uh this card. Where is the link to it? Uh I don't have the link and I Yeah. Anyway, actually I have Yeah. So you see this is some WebGL effect. That was brand new was it? That one. This one you made this year. But you you you've used that in the previous year though. This is like an old No, this is not old. This is new. Yeah. And you see like we have some gap here. We have this WebGL effect. And then when you press T on your keyboard, this thing comes up where you can like actually change the properties of the tree. Um, just I mean it's kind of fun and you could like do whatever you want. Whoa. Very cool. Yeah. Like background articles. It's a lot of fun creating these. I love it. It's a chance to just do something fun and creative, right? Yeah. Yeah. I love it. I love I love doing that. So, that's been what what has been going on before Christmas. Mhm. Obviously, we have stuff to talk about that happened after Christmas. Like what? Well, we'll we'll get into it. Should we start? Cuz I'm just mindful that you've you've got we've got an hour and 15 minutes before you need to shoot. So, we do we were going to do a race with the Rubik's Cube. I feel like we won't have time. Oh, well, we have two minutes. We have literally two minutes. Should we Should we um We should time it. We should time it. Don't start. Oh, no. I I'm not starting. I mean, we see the live. It's at 11. Ah, but yeah, use the timer. No, well, no, no, because it's just whoever finishes first, right? Basically, I I got a Rubik's cube for Christmas and I learned I was like, you know what? I'm going to learn how to do a Rubik's cube because it's always escaped me. And I will say the magic kind of goes after you learn how to do it. But then you just work on your speed and then yeah, you get the chicks. Do you know what I mean? Then they then [snorts] the then the girls come rolling in once you know how to do a Rubik's cube. So, should we should we give it a go? Like the snowflakes down. Should we give it a go? Right. I've messed this up. Yeah. Uh I'm very rusty on it. Very very rusty. I'll give you a minute to look at what you've got going on and then we'll um Yeah. No, it's okay. We can go on. We can All right. We can do it. I have mixed it. Okay. Three. Yeah. Two, one, go. Yeah. Which color do you start with, by the way? I start with a daisy. I start with the daisy. What is daisy? Well, there we go. See, you do a different thing to me. Uh different technique to me. Uh okay. Should try and get this uh so you can see what I'm doing. But uh yours seems to be a lot louder than mine. H. See, now I've already forgotten which one. Okay. So, I'm doing yellow. Uh, this is where I started with and um I I mentioned I'm I'm really rusty on this. So, you don't know the daisy trick then? No, I don't know. I I learned it b from a PDF a long long time ago. Uh this is like prior to YouTube. Uh well, not prior, but I didn't have YouTube uh access. So, I did it with a PDF and it was okay. Um, but I last time I did it, I messed up. Yeah. So, I'm not really I'm not really fast. And I'm not trying to do it super fast now cuz I'm scared of messing it up. Yeah. Yeah. I'm trying to do it super fast. I've kind of I've already forgotten kind of how to to to do it. Uh, where you at? I'm I'm not I'm not looking up yet cuz I'm afraid to look up. But where are you at? Well, I think I messed it up the first part. Yeah. So, I did mess it up. Uh because I got a bit unlucky. Okay, that's the thing. So, you get also unlucky with it. That's what most people know. That's how I got my best speed. I got lucky with it and uh and I got Is this going to be a fail? Yeah, I'm I should have done it before the show just to check that I remember. See, I I've already got lucky. I can see I've already got lucky. So, and that. Okay. Right now, I've forgotten how to how to do this. Um, is it this one? Oh my god, yours is so loud. [laughter] Yeah. See, now now I've genuinely forgotten. So, I've got my second layer. I've got my second layer and my white, but I'm not And I've got my yellow top. But, uh, how the hell I've forgotten. See, my mind's just gone blank, I think. Oh, I know. Okay, I think that's that. Yes, it is. No, it isn't. [ __ ] the bed. Okay, so see, I totally messed it up. Yeah, the first time I got unlucky and I and I messed it up. So, I basically did it twice, but now it's done. So, look. Yeah. Now, ah, pause. [snorts] Very good. But it took me quite I messed it up in the middle. Yeah. See, I I just totally drew a blank on the next step because I haven't done it. I haven't got this far in a while. But it is what it is. As you do it, I'm going to show the audience if somebody is watching something cool. So, what you can do with it is creating a flower. So, what I'm going to do is That's the daisy. That's what I said about the daisy. this, this, and this. [snorts] So, I'm doing the same move four times. And now you have kind of like this flower thing. And this is what you present. You You give some somebody a flower, right? Mhm. Well, that's a very geeky thing. They they should uh I I mean, if they don't appreciate it, they don't. Uh, another type that you can create is this type of flower. I I don't know. I call it flower. It's just kind of like mix of colors and it's really nice. Why am I not getting this? Maybe I've got the wrong I'm doing the wrong algorithm cuz they're called algorithms, aren't they? So, yes. And these are essentially um kind of like precomputed moves, right? So, what one thing is cool about Rubik's cube, it has like a lot of different combinations. So, this is a combination. The solved one is one of the combinations. So, how many combinations? I believe it's like 43 to the power of 80. Let me actually check double check that because I'm rusty on it. Uh number of rubix cube combinations is it? Yeah, it's 43 to the power of How many is it? Uh, they're not showing it. Oh, [ __ ] Have I just completely messed that up? No, I haven't. P. Okay, I've got it. We uh really went off piece here. I wonder how many viewers are still stuck watching. Yeah. Uh we we have some viewers. So nice, nice. Good job. So this is okay. Let me give you a few fun facts. This is one of the combinations, right? This is what we call consider the solved combination. [snorts] This is 43 to the power of 18. So this is 43 with 18 digits in front of it. That's a huge number that that number is so big that there is a likelihood of this combination. I'm just randomly doing something. This combination being completely unique in the history of Rubik's cube and in the history of the future of humankind. meaning that nobody will ever have this ever again. That that's so hard to believe. Yes. Yes. It's just that, you know. Well, it's it's even crazier with uh playing card number with playing cards because that's 52 factorial. That's a bigger number. But even with this, the number is really big. So now when you say we have a big pool of like big number, right? uh and the the solved version is one of these numbers. What is the maximum amount of moves that you need to go from any combination to the solved one? So what is the maximum distance from any of these two combinations? Yeah. And they they computed that. They computed that and it it's called the god number. That's Rubik's cubes god number which is 20 which means you need on theory 20 moves from any combination to get it solved. So if you were a superco computer in your head you could look at this and say ah if I do these 20 moves I'll get to the salt. And this is what these robotic supercomputers do. They look at it they premputee it. They know the exact routes to take which is under 20. The maximum is 20 under 20 to get it to solve. That's how they you get subsecond. Yes, these robots do it under a second to solve it. Uh so that's really cool. And just one last fact, we have the solved version. There is an algorithm that you can learn. I I have forgot it. I I knew it but I have forgot it which is the opposite of the solved number the solved version. So it is the chaos mode. So it is everything twisted to be in the furthest distance from solved like being at the number 20 which is the the corners are all twisted and these are also all twisted. So everything is in the position of solved but twisted. Got it. It looks also really amusing because it's kind of like super colorful. So you you should check it out. Learn it. It's just like one algorithm. You you you solve it and then you go from the solved version to the complete opposite which is I don't know. I'm geeking out on I'm going to I want to redeem myself with uh getting that was an atrocious performance. I was I was getting it quite consistently about three three and a half minutes when I was doing it a lot and I've done for a little while. But I still think you might have been faster than me even though you've got a not got a a speed cube. Uh yeah, my cube is very very loud. Yeah. And and again, sometimes you just get lucky, don't you? You know, sometimes it's just, you know, you might not We should We should do the We should do the news. Okay. So, we are um we are canceling the show and we are going [laughter] into a Rubik's cube right now. See, look how lucky I am here. Look how like I've been like that second row is almost all done. Yeah. So, but you which color do you start from? White, right? Uh I I complete white first. Yeah. Yeah. Yeah. That that's that's what everybody does. So, let me tell you I'm this is this is me confessing, aren't you? I'm confessing right now on on a live show that I start with, as you see, yellow. So I do white last. I mean it doesn't matter like it's it doesn't matter but it's one of those things if you can be if you can be color agnostic then that's better. See already I've got a cross there like and this is where I can now finish it quite easily. So so where am I? Yeah. Uh blue. There we go. See Okay. So, I got it for the second time now. I [ __ ] it up already, man. Yeah, I [ __ ] it up already. God damn it. So, done for the second time. Yeah, I [ __ ] it up. Yeah, it happens like especially right. We are We need to We need to crack through these now. Yes. Let's go, bro. My record was 47 seconds or 46 seconds. I've never got sub minute. I've never got sub minute. Yeah, but I got lucky. Anyway, so we got some AI show to run and [laughter] to start you're flustered now. You're like, ah. Yeah. Uh like the aden Yeah. It's it's crazy. Uh do I will I'm preparing for the intro. Okay. So all right you ready? Let's go with the intro and we'll get into the videos. So this week uh we have a this week we have a big tailwind drama. Uh Andre Kaparthi the top G and the AI god is feeling behind uh dynamic context uh for cursor is a story and uh how creator of clot code is using clot code. Now this in today's episode, this is Command AI. I'm Kabarzo and I'm Samuel and this is a week in AI design and dev. Awesome. Cool. Let's start with site of the week. Um, but before I before I start, let me get my screen prepared. Also, before we start, if you're on YouTube, give us a like. I'm coming on to YouTube now to spot you to see if you're gonna how many we got on YouTube. [snorts] Yeah. Wait, give me a like. What is this? All right. Yay. Cool. All right. Uh join us on YouTube and please like like write something in a comment. Well, not something like what you tell us where you're at. Tell us where you're at in the world. Yeah, exactly. All right. So, for this week's for this week's website, uh well, how what was in my notes? I I messed it up. Anyway, all right. So, I have a super super cool website to share with you this week. And this one is a bit special. It's not completely new, but it's in solidarity with my people. I'm originally Iranian living in Germany. in solidarity with them, being on the streets and um protesting for their rights. I'm going to show a website that is about history, like some old Iranian history, and it's super cool. Like this website, I cannot I cannot wait to show you. It's amazing. It's insane. So, this website is by Getty, and this is what we see at the start. And then we basically continue to scroll. There are some texts about this old civilization and this old like building whatever you call it uh calling we call it pers police p police uh it's pers police and then we press enter uh and then we continue scrolling we fly and at the start this is just a video being played so this is just like oh static frames mhm and then this is where it gets started it. This is all 3D with insane level of details. Uh going through how all of this like uh purse police was. Um it it forced it is called TA jam sheet. Uh and here you can interact with it. You can see how it is today. You can read about it and you can also just scroll past and see the whole grow glory. you see uh the color red being used a lot. It's also like showing power. It's one of my favorite colors by the way. Uh and it's just the website is really beautiful and the level of like details in the 3D is genuinely insane. By the way, you see how huge these doors are. This is just like unbelievable. What's the uh I'm looking at the photos of what how it looks now and it's obviously ruins now. What's the what's the importance of this building this this site? Um so this is in in a very old ancient city called Shiraas which was the capital of the old Iran. Yeah. So this is essentially the capital palace in a way and this is where yeah the the whole political power uh was and here we see a lot about that. So you see like these models and these like scroll interactions are absolutely amazing and then here we are. This is probably like one of the biggest uh areas here. It's like huge. It's insane. Unfortunately, I've never been there like in person, but at least I guess in the middle of in the middle of the desert basically, isn't it? Yes, it is. It is kind of like in the like southwest of Iran. Uh Iran has like a very uh like like we have like from green all the way to desert desert. Uh we have it all. Like I'm from the the place with like a lot of mountains and snow and we have like minus 20 degrees in winter which is like what we have right now in Germany. So yeah and here we see like the the old dress and whatnot. So yeah the it's beautiful. And now the menu look at this menu. You click it and then you are presented with this like beautiful interactive 3D map. Like when I say map like actual map this is the point like this the menu is an actual map of the whole palace and everything around it and you can read about it. Uh the website also is available in multiple languages so you can change the language. Yeah, I was gonna say hit the settings because I noticed like you can reduce the motion, high contrast as well. So yeah, large text and this is what people like typically when you show a website like this, people are like, "Fuck this. This is like so difficult to use. I don't like websites that take the control everything. Here you have a good amount of like accessibility settings. If you don't like to experience it this way, you can do it in other ways. Um, and yeah, the the menu. I really like the menu. And also the art index, which is kind of like a slider. Yeah, it's really cool. And this is, by the way, let me talk about sliders for a second. Like half of the sliders I see are not half like 99% of the sliders I see on the web are incomplete in my like from what I expect them to be. A slider to me should work with my mouse uh with my touchpad. Not just the mouse but like with my trackpad. When I scroll with my fingers it should scroll on to the sides. This is what I mean. Like right now I'm scrolling to the sides and this is working. But I can also grab and drag it. So there are like multiple ways of doing it. I can also use the the top navigation to navigate it. So it has it all. And then on top of all of that I can scroll vertically. So right now I'm scrolling up and down like up and down or right or left does the same thing. And it's to be expected because this is just a section. And it's not a page to be scrolled. And I wish to see more sliders with this like level of control. Well, are they using um are they losing a library? Can you infer what they're using? Uh I haven't checked actually uh for the slider, but Swiper is one of the most popular. So from the classes, I don't see see look at all the um you know data attributes and things like that. like yeah if you scroll oh have you have you got your plugin that tells you about the libraries and things they use? Yes. Yes, I will check that. But it's it doesn't say uh like it doesn't show these like smaller libraries that are like a slider library most of the time. It might I don't see much here. So I don't I don't think it is a library even the navigation here. We see an um like a list and then we we see like touch action pan Y uh cursor grab. It's like interesting like the classes they have and if we look at the text stack that they are using uh wait which one was it? Uh because the icon changes. Oh um they have okay let me actually reload because that one sometimes takes a second to to show the actual libraries. Ah you see more loaded. No we don't see a lot. So they are using GAP the library the the slider might be a GSA a custom GAP slider. Yeah. And they also have this uh this video or like sequence of images to be played which is an interesting choice by the way. They don't bring you into the actual 3D right away. They want they probably wanted to fly you over the whole map but they but they didn't want to reduce the the detail here. So they are using pre-rendered images. I'm just playing around now. Where which um where is the slider? Which little area? So if you go to the menu and then art index that's where you see the slider. Oh yeah. Okay. Cool. Got it. Yeah, just going to see. And this is done by Getty. Gety um like you can go and read about them. They do this type of stuff. They they have like really cool uh like projects they've done. So yeah, and this is a part of like the old Persia ancient Iran and the classical world. So they do these kind of like studies and exhibitions and then they do also like 3D and like digital versions of them. And this one was about pullies. Really cool. Really really cool. Yeah. Yeah. And yeah, obviously you can read more here about every part of it. H it's it's so nice. Also I love this interaction. You see when I get to the end of like a horizontal scrolling section, there is this like elastic Mhm. interaction. So it does tell me this is the end of it. Like these little details like a lot of websites lack this level of details and this is what Yeah. Um is missing so so often. Yeah. Very cool. I'll link it below and go check it out. Uh it will probably make your laptop's fan turn on uh [laughter] or like your PC, whatever, because it's quite heavy, but it's it's worth it. Yeah, cool. Cool. Are we moving on to Tind? Oh, yeah. [snorts] Are we? Yeah, we should. Okay, cool. Um well, who you want to wait? Uh uh I have my I had my screen. Okay, let me just do that and then remove. Okay, cool. Well, who knew that having a sign well who knew that having significantly more users would mean you would earn much less money. This has caused Tailwind who know Tailwind to lay off 75% of their engineers. Um but let's see how they made money in the first place. So you don't know what Tailwind is. Um this is their website. It's basically a utility class library. So instead of writing CSS, they have a bunch of classes that determine this is display flex flex direction column um you know items um uh align item center various things like that. It just means you don't write a lot of CSS and traditionally um you know as a developer back in the old days. Sorry, hang on. Uh back in the old days, I would if I'm using Tailwind here, I would come to these docs and I would look at I would press command K because they got a lovely um search thing here. If I want to do some box shadow, I'd search box shadow uh box shadow and I'd learn a little bit, you know, how to how to customize the box shadow and what it all does and it's a great place to to learn how to use it. thing is if you've been using AI to do any sort of coding recently, whether you know or not, AI is probably choosing to use uh Tailwind to do all of your CSS or all of your all of your styling and stuff like that. And what that means is is that I'm not now long I'm no longer coming to this because instead of learning how to you know how to make the box shadow bigger by going box shadow MD or make it inset um I'm just asking AI to do it for me and it's just figuring out all of the classes. And what you might have noticed or you know they don't obviously don't want to push it too hard but what you might have noticed is that there's opportunity here to click plus and because this is an open-source framework this is all free but how they make their money is things like this um Tailwind plus which is a subscriptionbased sort of advanced component system which uses Tailwind and people can buy these and and this is this can boot companies can bootstrap uh with with this library to help kickstart their product, but they don't need to do that anymore because AI is taking care of it for them. Plus, you're not coming onto this website to even know that it existed in the first place. Similarly, if I go let's try a different um thing. This is one of their this is their book, which is apparently a very very good book. I've not written it, but this is another way that they can make money. and you're finding out about about all of this when you're on the website, when you're going to the website, but alas, AI is preventing you from going to their website and and therefore converting as a paid customer. So the paradox here is that whilst if you were to look at the number of users or number of downloads of of Tailwind, you'll see that massively increase over the last, you know, couple of years. But what you're also seeing is conversely is the income and the the well yeah the income from the company is going down and that what that's meant is their co-founder Adam has come to this realization through maybe I can get the um GitHub issue. It all stems from this gib GitHub issue which was closed which is just a small text file that allows the documentation to be more easily understood by LLMs. And there's been a lot of uh frustration from the community as to why this hasn't been merged in. And it all stems from Adam coming to realization that basically says it's something he really wants to add, but he's just had to lay off 75% of their engineers because they've he's witnessed that they've only got six months worth of runway left in their of of revenue in in the company. to to protect the company to give the engineers a a severance package, he's had to make the harsh decision to fire. I mean, they only had four engineers anyway, but they've had to fire three to save that and to give them severance. Um, and also just the fact that this is this is supporting the very thing that is killing the company. So whilst he's not completely ruled it off the he's not removed it from the table, it's like I we have more important things to to figure out right now and that is how are we going to make money given that we are used so widely on the web right now because of AI. So it's a real tricky situation and I think more and more companies are going to see this trend uh this this difficult situation where the traditional model for earning revenue which is open source something give something away for free Nex.js being a perfect example of that you use Nex.js here have Nex.js JS totally free but where do you end up using that you end up using on Versel and you end up paying for so there's payment on that back end which is you know fine this is this is business this is how it works you you feed the free thing you get the money on the back end but now this is no longer working certainly not working for Tailwind so it's it's a sad state of affairs but this is this is where we're at in the world of AI and open source yeah this is this is quite interesting and uh just One uh thing to correct they are not a subscription based uh tool the plus you buy it. Yeah, the the plus you just buy it one time and maybe this is something that they could add, you know, subscription. Well, people hate subscri uh subscriptions. [sighs and gasps] Um but yeah, also like it's so the issue is two things. One is that people are not going to the website in the first place because AI is doing the job and people are not many many are not even aware of tailwind being used like the AI is completely masking that layer. Yeah, exactly. They just AI just chooses that for you and you just ignore what the AI chat is saying probably. So that's the first issue. people not buying it because they never see it in the first place even though there are significantly more users. But the second issue is what they are selling is kind of like simple templates, simple UI components that AI can't build perfectly fine but good enough in most cases. And that's probably what people are doing. instead of buying a really well-gineered template, they're just prompting AI over and over and over again. Uh so this is also interesting because the money is now going towards the AI companies even though they are also not making a lot of money. That's also I find it to be quite funny. So Tailwind is not making money. Open AAI not is not making a lot of money and at the end like the UI is broken uh and the the good stuff that Tailwind has uh made are not being used. Instead you just duct tape some some stuff randomly AI put together. That's where we are at. M and you know this this has caused me because um I think we all need to be thinking about this because this isn't a tailwind issue. This isn't a tailwind problem. This is a an industrywide problem where you know the information we give out the the the sort of the models that have worked for so long just don't cut it anymore, right? So we all need to re like especially you and I kabaza as content creators we're giving out free information uh on YouTube where does where where in this kind of pipeline does it matter that people don't even go to YouTube anymore because they can just search based on the free information we're giving out they can just search it on AI yeah using AI our our contribution to the to the world knowledge is being sucked up and we're we're not able to then convert people to whatever whatever it is and in two ways by the way like in one way it's ending up as the AI uh training data which will be then outputed in a very generic way it might not be as like exact as we put it out there and the second one is uh AI is just reading the transcript of whatever we put out there and the person will never even know from see yeah um and it it begs the question like who's next because this isn't like I say this isn't a tailwind only problem. This is this is a industrywide problem. Um, and I think the categories that we thought about that were that, you know, the the yeah, the the the criteria that this falls into is really those companies that have a free um lead magnet that they they that's information or information based or that is like again open- source based that they rely on people coming through to learn about it to then sell it on the back end. And luckily, you know, we spent a bit of time thinking about who's next. Luckily, a lot of these companies, a lot of these free open- source software have already kind of fixed, not fixed the problem, but they uh Nex.js has Versel. Um Shad CN was just recently bought by We just looked at it before the call. I've already forgotten. Maybe it's Versel again. Maybe they've already It's Versel, I believe. Yeah. Versel. Yeah. Yeah. Um so, also by Versel. There are there are some and that that did get me thinking like it did get me thinking of a silver lining that you know we're always facing change through history through um you know we're always confronted with a new paradigm that shakes things up a bit and I think all this has done is really just prove that this this model of offering something for free hoping to sell it on the back end just does not work in this day and age and while I don't have necessarily the answer to to it all. I'm optimistic that we'll just find another way to support all this. Maybe it is that you these companies are just less inclined to give away stuff for free because they just know there's no um route to monetization because of AI. I don't know how long that might take. Um but it it does again it's cyclical, right? There there's bound to be a solution. It's not sustainable. It was never sustainable to just give things away for free, right? So, I don't know. There's one aspect of me that's just like I wonder this can't this can't end this way, but it's it is sad all the all the same. You know, people lost their jobs. Um, and it's also just like I say, it's very paradoxal that something could be so so uh popular, everyone's using it. Maybe we're not not even knowing it yet. They are more popular than ever yet, you know, not a lot of actually. Yeah, they they are not being able to sustain themsel for more than six months, which is crazy. This also like makes me think about like the like the growth and distribution channels that you have. So th this tool the this tool's distribution has been powered by AI in a way like AI is choosing the tool like the library and just just choosing it and using it in opposed to the users like actual people's um ability to choose it like um consciously and make a decision maybe help the company or pay uh their premium products. So that's a question. But also on the template selling side, so whether it is a full template or just a component, this also makes me think of I think this is what Sam Altman said. Um for all products to think about what AI does to their product. So I'm just paraphrasing here, but will the next AI which will be more powerful and better whatever like the next updates that are getting better and better will that make your product better or worse and in case of uh shuts and tailwind and shutsen as well and others in many ways as AI is getting better the value you get from these products is kind of decreasing because AI can do it. Even though I understand like AI is literally stealing from them, but unfortunately this is what we are facing. AI is stealing and is getting better and better at stealing and getting better and better at recreating what has been already created. So the value of these products is diminishing as the AI is getting better. And this is true about anything that you are trying to do. So whether you are a web designer or a developer or have a tool or have a company or have a startup idea, you need to think about this like is the next uh update of AI going to break you or make you a super like powerful tool and you know it's in your best interest to align yourself with AI not against AI even even if it's a bubble even it Yes, it is. And even when it pops, it's still the value will be there which is like these level of automation and this uh change in the like the the detail that we will work with. So instead of code, we will work in a different level. So yeah, it's really hard. I mean I'm building a UI library, right? While we see a UI library is go going out of business. So, it is it is [clears throat] kind of crazy, but I'm trying to align myself in a way that we we build more and better stuff as the AI is getting more powerful. That's what I'm trying to lean to instead of like a small system. I'm thinking about how can I build stuff that are better and better as the AI gets better and better. How can we scale with that? So, but it's quite difficult. Yeah. Yeah. So, we'll leave links to everything down below in the description when it when the clip comes out. But be interested to know what you guys think, whether you are thinking already thinking about this, whether you have some ideas on how to again change this pricing, this mental model on how we how we earn revenue and whatever because AI is here to say even if even if AI bubble burst, it just means that it's it's what do they call it when the stock market balances out. You know, it's just a a correction. It's just a correction. Well, after a I don't know, after a bubble, I don't know if it's a correction that happens because we're not just going to get rid of we're not just going to get rid of AI just because it is. That's what I'm saying. It's going to just find it place. So, yeah. Um, yeah, let's see. Good luck to Tower and let's see how it goes on. There was a third way that you can uh give them money and that's just by physically sponsoring them, you know, if you have money. Oh yes. To just give them. Good point. Yes. You know, but um yeah, let's move on to Andrea Kapathy. Yeah, let's Do you want to take this one? Um sure. Let me actually You should go for the tweet. I'll just go with the intro. Yeah. Yeah. Yeah. Um [snorts] Uh let me Okay. So, Andre Kaparti, you Andre Kapi. Andre Kap Andre Kapi. Yeah. Andre Kaparte. This is I don't know how to pronounce his name. So, okay. I'll Andre Kaparti. Uh if this guy feels behind, we are cooked. So Andre Kaparthi, he's the he was the head of Tesla uh AI at Tesla and he's one of the co-founders or like founders uh early founders of open AI and he just tweeted that he feels extremely behind when it comes to development in the age of AI. And by the way, this is the person who called vibe coding vibe coding. Vibe coding and you know more about the tweet. So yeah, let me know how is that that he's feeling so behind. Yeah. So basically feel Yeah, this has got 16.2 million views uh this tweet. So it's it's a pretty big deal. Insane. I've never felt this much behind as a programmer. this the profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between I have a sense that I could be 10x more powerful if I just properly string together what has become available over the last year roughly and a failure to the boost feels decidedly like skill issue so I won't read it all but he's and he mentions a few things here that maybe you should actually it's very interesting like why not like let Let's go. I I love this. There's a new programmable layer of abstraction to master in addition to the usual layers below. Now, this is very interesting. involving agents, sub aents, their prompts, context, memories, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, SLC commands, workflows, ID integrations, and a need to build an all-encompassing mental model for the strengths and pitfalls of fundamentally stoic uh stochcastic, fallible, unintelligible, and changing entities suddenly emerge. This is a too long par sentence by the way Andre intermingled with what used to be good old-fashioned engineering. So what he's saying here just in this little bit is that there's all of these things and a lot of these things are introduced and are used by claude code hooks, skills, tools, plugins, whatever. So he he's certainly alluding to Claude code being the thing uh you know the de facto tool for all this but there's so much that you need to learn not only from like a a technical perspective of just understanding what all of these tools do what all of these things do but also the mental model around how to use them because it's all well and good knowing uh how to use a hammer. You know what a hammer does, how to hold it, you know how to aim it, all the this, it's not, you've got to rethink, you've got to re understand, you got to throw everything you've understood out the window because it's not relevant in this new age of, you know, prompt engineering or AI assisted engineering. Clearly, some powerful alien tool was handed around uh except it comes with no manual and everyone has to figure out how to hold it and operate it. There you go. while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind. And I'm, you know, I'll I'll throw myself in the I'll pat myself on the back here. This is something I've been really immersing myself in. Like we're dealing with so much right now. the tools, the mental model around it, the industry, the um not just the industry, but the the planet in general, the economy, everything is just moving at the moment. And we've there's a mad bum rush to figure out figure it all out and be the person or be the um leaders to figure it all out. But no one knows. No one no one's figured it all out at the moment. But his message to everyone is like get stuck in because something big is already happening over the last year. So quite quite a profound and um interesting tweet that came out just after Christmas and a lot of people were talking about it and this is very interesting um especially like that ending while the resulting magnitude 9 earthquake is rocking the profession. So especially after what we just talked about with with tower. Exactly. Yeah. Exactly. So the good thing is or like the good news not the good thing it's the good news is you are not alone in this. We are not alone in this. Nobody has figured it out. If he hasn't figured it out, if Andrew Kaparthi doesn't know how to, you know, move forward with all of that, that's kind of like okay reassuring because I feel I feel very similar like I see you use clot code. I see I could my you know uh friend and developer he uses like 10 agents well more than 10 agents at time and I'm sitting here with cursor I feel like I'm so behind and then I show what I made to my students or like to other people who do web flow and they're like wait you're already there using that they feel behind I feel behind I could is like these agents I cannot do what I want them to do. So he feels like it's not going where it's going and Andrew as well. So this is very interesting but also potentially something to be figured out. And also I want to mention what the the primogen has been talking about the fact that all of this is constantly changing. So we can't go in with the mental model thinking that okay I'm going to learn agents and these toolings around AI and these essentially workflows. These are all workflows, right? That I'm going in, I'm learning these workflows and I will be the god here. I will create whatever. The the problem is it's changing like the next month there is something new and the the change changes are rapid and big. It's not like a small change. like the whole new paradigm shifts every 3 months or so is introduced to the industry like as a programmer it feels like yeah crazy. It feels like you don't know where to go and it's yeah it's not clear and because of the horrendous capitalist society that we live in. You've also got grifters and chances trying to sell you on their system. There's the system that works. they've done the research and it's you we spoke about fatigue before the before the show like you know it's it's it's real the the feeling of of being left out and some people are capitalizing on and this is potentially a different topic altogether but it's just it's something I've been tuned into recently of just like oh my god like what there's so many people making you feel left out to buy their product or to comment give me please in the comment section below to like get their system that they've you and no one knows. No one knows and I think you it just comes down to you getting stuck in and and trying to trying to be optimistic and and curious about it and trying to get ahead you know but it will settle you know we will still figure it out. I think one of the biggest problems with AI and I had a call earlier with someone because you can approach it you can approach it at such a low level or a very very high level. You can ask AI to implement your whole feature or you can get it to change a line of code and that what that means is people's and and that's just in code, right? Let alone video production, let alone script writing or any kind of other creative thing. It means that everyone everyone's got their own interpretations of whether it's good, whether it's bad or how you should use it, how you shouldn't use it. Like no one no one can tell you how you should shouldn't use it. It's just by its very nature so fallible, so indeterministic, so whatever. So again, just that encouragement of of just try and stay curious, you know. Plus, not everyone's here to change the world, you know. Maybe you can sit back and chill and then when all the big wigs like and Andre McCarthy has figured out how we should use all some sort of stuff, then scrub up on it, you know. Thanks. Well, it's hard to sit and not do anything because No, of course. Of course. We we we are still like working right with with the old tools and trying to use these new tools. Like I build a website recently and I I I thought I would do it all in cursor. But then I had to go back in web flow because of like more reliable things around, you know, having a platform that I'm familiar with. Uh while I could just potentially do everything with AI and then ask it to host it, you know, host it somewhere and all of that. It's just a lot to figure out. A lot to figure out, right? And and then also there is the joy about it like the things that Andrew mentioned there are uh a lot of it is is around toolings around AI and not building the thing itself and that is also a little bit about like you know having fun with it or like you know the lack of uh fun existing in it. I I did enjoy going in Webflow in the past few days to to build a few layouts even though I could absolutely ask AI to in cursor to do it, but I knew it wouldn't like one-shot [snorts] it and I would have to go into like the mode of fixing it and prompting to fix bugs and I I I just don't enjoy that as much. Yeah. And we all like familiarity, right? And it's nice to go back into that. And just I don't want to dwell on this too much, but just that idea of knowing what to build. Now, my gut instinct was to respond to that and say, we already know what we want to build. We want to build websites. We want to build apps. But then something in the back of my head said, and it goes back to this magnitude 9 earthquake that he mentions in that I think we're and what I mentioned earlier as well about the economic climate that we're in is being shaken up. I've said it before, I've said SAS is dead and we're out we're all out here trying to build tools that we can sell to people um because that's all we know. That's the only way we know how to make money, right? With this whole idea that people can just build their own tools, we we probably don't know what we should be building. We're probably going to be building very different things in the next few years than what we have been building. We've been building products. We've been building tools. Maybe we don't maybe we're not going to be coding that. I don't Nobody knows. And I think this just all goes back to this this tweet here um that's specifically talking about development. Um we're in a very interesting time at the moment and it's all so much stuff is all happening at once that it's just yeah it's hard to it's hard to not feel tired and it's hard to um feel uh steady like we should all be feeling well we are all feeling the the craziness that is AI but I honestly think we will be building we don't know what we should be building that will figure will be check it will sorry I honestly think we we don't know what we should be building because the future is we're figuring yeah we're fig we'll be all right we'll be all right everything's fine set everyone like before you know catastrophe hitting we'll be all right yeah um saying that um Boris for Borg Churnney um the creator of cord called Claude Code has and I actually haven't gone through this. I I wanted to try and go through it before the show, but [snorts] it's something I' through it together. Yeah, let's go through it together as a family. So, but he invented core code and I I just want to say that although he invented Claude code, he claw code would not be what it is if it wasn't for the team around him. So this is not to necessarily say that they are the you know what they say goes but there's absolutely something you will probably learn from this and you might learn a thing or two as will I just kind of going through this. So he runs five clauds in parallel in his terminal. I number my tabs one to five and use system notifications to know when cla code is input. So I'm guessing work trees play a big part in in his to to be able to do this unless he's using different git repos for you know they're but I guess this is building clawed code so I'm get I reckon this is work tree. So I have I have a video on work trees on my own channel if you want to go check that out but it allows you to run different clawed code um instances all working on different features. So yeah, my experience in this it can get very confusing because you think [ __ ] where was I? What was I doing in that one tab? What what you know where was five is quite a lot. I struggled with three put it that way. I also run five to 10 clauds uh on claude.co. So that is claude for web. You can also achieve this with the desktop app. It looks and behaves exactly the same. This is a really nice way to like just send off something and it it creates a pull request on on your GitHub repo. Um, and you you don't need um you don't need to run it locally. Um, as I turn, I'll often hand off local sessions to web using at which is something you literally type at and then it will send it off to web, which is quite cool, or manual kickoff sessions in Chrome. Sometimes it will teleport back and forth. So basically he's got up to 15 tasks running in parallel which is quite mental. Mhm. I use Claude uh I use Opus 4.5 with thinking for everything. It's the best coding model I've ever used and even though it's bigger and slower than Sonnet since you have the steer it less uh it less and it's better tool it is almost always faster than using a smaller model in the end. Now he obviously works for Anthropic. He probably has free use, infinite use of Opus 4.5. This is not practical. Really, this is not practical. The price of Opus. But what I do find quite interesting, he's got thinking on for everything. Now, you can enable thinking with um uh option or alt uh P and you can turn enable [clears throat] thinking on. I'm in an open question at the moment. What is the difference between thinking and planning? because you're not doing any both are not doing any code. thinking tends to think before it implements all in the one like uh shot whereas planning gives you the opportunity to you know thumb up or thumb down something I'm discovering but it is just interesting that he's just got thinking on for everything again he works for anthropic this is probably very uh unachievable by normal people our team share a single claude MD file for the claw code repo we check into git and the whole team contributes multiple times a week anytime we see claude do something incorrectly we add So they're actively contributing to this cruel MD file. Um, which is, you know, I think we all should all be doing this like you're it's not going to work because it will be unique to your way of working. It'll be unique to your team's way of working, but it could also be unique to your um the way that you use cord code. You know, you might you might prompt it in such a way where it always does uh what have we got here? um you know it always uses npm because of the way you write code. It's a stupid example but you know what I mean like your experience with claw code is going to be very different to another person. So your rules are very unique. If I can have number five during is if this is not interesting let us know but we can we can no this is it speed through it. [snorts] Uh during code this is interesting. During code review, I will often well I'm hoping I'm hoping I'm providing a a yeah contributing something to what he's saying as well. Uh based on my understanding during a code review, I'll often tag at Claude on my co-workers PRs to add something to the Claude MD as part of the PR. We use the Claude code GitHub action which you can install using slash whatever um for this. It's our version of Dan Shipper's compounding engineering. Now this is something I don't quite understand. So you you can claude. It says doclaude. I guess he's trying to avoid the um Twitter's tagging here. So it's Claude on my coworker's PR to to add something add to Claude MD to never use enums. So basically he's just encouraging I think he's basically just encouraging number four here. I think we'll use core code GitHub action for this. It's our version. Yeah. Yeah. Most sessions start in plan mode. There you go. If my goal is to write a pull request, I will use plan mode and go back and forth according to it's I like its plan. From there, I switch into auto accept mode and cause can usually oneshot it. Again, he's probably using opus, but I will go into op. I use opus plan, which is a hidden plan feature, which is a hidden model. You type slashmodel and then type opus plan, and it will unlock. This is literally like secret stuff here. It will unlock whenever you're in plan mode. It will go to opus and then it will switch to implementation mode and automatically switch to sonet to actually implement that stuff. He's rich. He can do opus all the way, but that might be a more cost-effective way to to do this um to get the best out of it. Um I use slash commands for every inner loop workflow that I end up doing many times a day. This saves me from repeating prompting and makes it that Claude can use these workflows too. Commands are checked into git and live in claude commands. So in a loop, I don't know what that means. For example, I use I use commit push PR/comand dozens of times a day. The command uses inline bash to premputee. So I'm I'm struggling with this one. And a few other pieces of info to make the command run quickly and avoid back and forth with the model. Not quite sure on that one. Let me know if you if you understand that. But what this what this kind of does is what he's going into here potentially is using sub aents for every task in the task because so cruel code will create tasks. I had an idea today to say for every task create a sub aent so it creates a fresh context. That's the thing about sub aents. They have completely fresh context. I don't even know if it's possible. I just had the idea earlier, but um it seems like he might. Yeah, I don't know. I use a few sub aents regularly. Code simplifier simplifies the code after Claude is done working. Verify app has detailed instructions for testing Claude code end to end and so on. Similar to slash commands, I think of sub aents as automating the most common workflows that I do. So he's a big sub aent user on call guide. We use post tool hook to I really want to get into hooks. Post tool hook to format Claude codes tool. Claude usually generates well formatted code out the box and the hook handles the last 10% to avoid formatting errors. So okay so it just runs format basically um based on some rules that they've got set up. Um I don't use dangerously skip permissions. Instead, I use permissions to pre-allow common bash commands that I know are safe in my environment to avoid unnecessary permission prompts. Most of these are checked into. So, okay. So, yeah. Um, you can basically set a bunch of uh permissions that it's allowed to run. Instead of giving it basically you can just do anything, he's being a bit more controlled about what it can run while it's doing its thing. Um, [snorts] claw code uses all my tools for me. It often searches and posts to Slack via the MCP server. Runs BigQuery commands to answer analytic questions using B big BQ CLI. Don't know what that is. Grabs error logs from Sentry. Uh the Slack MCP configuration is checked into our MCP JSON and shared with the team. So what is he saying there? Use all my tools for me. I think he's basically saying that he just does everything from SL uh claw code. I think via MCP. Uh nearly the end. 12. For very [snorts] long running task, I will either a prompt claw to verify its work with a background task when it's done. This is something I want to talk about. Use an agent stop hook to do that more deterministically or use the Ralph Wigum plugin originally dream up by Jeff Jeffrey Huntley. I will use either permission mode, don't ask, or well, he just said he doesn't use that in a sandbox to avoid permission prompts in a session so Claude can cook without being blocked by me. So, um, you kind of mentioned it a little bit earlier, this idea of I think what will be more common is kind of what, um, what's his name? the guy who built his own agent, uh, PewDiePie built where you've got a council of agents all verifying each other's work. I think this might be what he's kind of doing here, like creating a sub agent to verify others work, but not to 100% sure. I ask my AI like I always um give it like I ask it to rate the code and sometimes I do it like manually with other AI and what I've recently started doing is I find like either open source or like yeah I find code that I think uh would make sense to kind of like um compare it to. So I say like compare this what you created with this and then rate it. Yeah. And then uh I also started doing like doing that and putting the end results in an MD file. Mhm. for either for AI to learn or for myself may maybe both of us like so they learn from the the code that they did or sometimes you know when you get an error or like you get a bug or something or something doesn't work out and then you continue prompting until you get it fixed uh I say everything that you learned through like this back and forth documented in an MD file why was it so and how is that that it's it worked so I do try to gather these either for myself to learn from or you know I could feed it back [clears throat] to AI in another session if I have something similar. Yeah. Nice. Um I don't know whether you've heard of Ralph Wickham. It's doing a lot of rounds at the moment but it was um it's basically a task that runs infinitely until the job is done. So and it feeds the result back in. It's just an infinite loop basically. You can set up max iterations but iterations but if you look at cold code look at plugins the plug-in directory you'll see Ralph Wigum um I made a short about it actually but um yeah it's becoming more popular like in the last week. So final tip probably the most important thing to get great results out of Claude code give Claude a way to verify its work. As you were just saying if Claude has that feedback loop it'll be two to three uh times the quality of the final result. So there we go. Um, again, everyone's going to have their own ways of doing things. Everyone's going to enjoy different ways of working. That might not work for you, but certainly works for them. Can't argue with it. Um, still a good read. We should do this more often, like go through an article or something and just read it together. Yeah. I just don't know. Let us know viewers if you think that's boring because like I was getting bored of my hearing my own voice reading all that. So no like to to me it was like still like there are parts that I don't understand uh because I'm not that technical yet but it's still definitely valuable and this is where I might see an article like this. I might start reading it and then feel like okay I can't uh continue because it's too technical or whatever and I stop but I sometimes just watch videos on this. So I instead of um you know reading it myself I I watch the video of somebody reading it which is often just primogen. I wish I could, you know, obviously literally just show you. And I mean, I probably could, but it would just make this section like 10 times more Yeah. longer if I'm there showing you exactly how to do it. But I hope that's interesting. That that's the idea for the comment section. Like just let us know if you would be interested in that. Yeah, for sure. Um, so another one I haven't necessarily done much research on is the context discovery. Where am I? Am I no longer sharing my screen? Uh, context discovery in cursor. Uh, where are we? Here we go. So I'm not I'm not a massive user of cursor anymore to be honest, but you so you might be interested in this, but I am interested in this idea that particularly around MCPs that apparently they've figured out a way what was three days ago they figured out a way to only use tools um and certain aspects that are filling up the context only when they're needed. We've spoken about this with cord um releasing the tool discovery tool, but that's only through that AI. It's not actually something that you and I can use while we're coding and developing these apps. It's something we can use in our API calls um you know in the apps that we are building rather than the way that we're building them if that makes sense. Point is um efficiently loading the only MCP tools needed is something I wanted to scroll. Well, actually, let's have a little look. Turning long tool responses into files. So, I guess this is the way of doing it. Referencing chat history during summarization, supporting the agent skills open standard and treating all integrated terminal sessions of files. So, clearly I don't understand those, but clearly that's one of the ways that they've used to reduce the context. Let's skip forward to number four and see what they say about the MCP thing. MCP is helpful for accessing. Do I need to explain anything? right here now about MCPs. I don't think so. I think it's pretty clear. Yeah. Um MCPs are just a way of accessing an API with natural language, right? MCPs accessing secure resources behind OHO that could oorth could be production logs, external design files or internal context blah blah blah. Some MCP servers include many tools, often with long descriptions, which can significantly bloat the context window. And it's terrible when you just don't even need an MCP. Like, it's useful to have the GitHub MCP as an example, but you don't need the GitHub MCP when you're just coding away. You need the GitHub MCP when you're trying to push stuff to GitHub, right? So, why it sits there in context is is really annoying. Most of these tools go unused. There we go. Even though they are uh always included in the prompt. This compounds if you use multiple MCP servers, it's not feasible to expect MC every MCP server to optimize for this. We believe it's the responsibility of the coding agents to reduce the context usage. In cursor, we support dynamic context discovery for MCP by syncing the tool descriptions to a folder. So we've spoken about this where we one of the options is to actually have the documentation or the code um you know in a in a in the in the actual code that you're writing in the in the folder that you're actually working in and then the AI can look at that and understand it and then use it based on that. So the agent now only receives a small bit of static context including the name of the tools prompted to look up tools when the task calls for it. In an AB test, we found that it runs that called an M uh that it that in runs that called an MCP tool. Hang on. We found that in runs that MCP called [clears throat] Oh my god, I can't [ __ ] read. You can tell I'm getting tired. We found that in runs that called an MCP tool, this strategy reduced agents by total agent tokens by 46.9%. Where I'm lost here is yeah, you can stick the tool tool names in a in a file, but how does it switch on? Can it dynamically switch on MCPS on and off? You know, um yeah, I that's something I don't understand, but apparently they can do that. And I wonder if this is already accessible inside of cursor right now as well. So I'm not too sure, but it's clearly something everyone's working on. Um, and I I definitely welcome it. [sighs and gasps] You got to go in four minutes, right? Yeah. Yeah. So, should we just round off with MKBHD? Yeah. Dark do that uh MKBHD? I think we can do MKBHD now. Uh because dark matter I feel like it will go deeper. But that video from MK Yeah. Yeah. So, this I pretty much everyone's watched this, but this is probably one of the best videos MKBHD has done in years. I mean, come at me in the comments, but it is fantastic. It's in collaboration with this guy called Epic Spaceman, and he's a British content creator who um I mean, what he basically does is make himself in 3D. And you see this is a 3D scene. This is all 3D here. And MKBHD demonstrates that by going like this. He turns himself into a spaceman and uh rotates around himself here and he basically shrinks himself down banana for scale uh to the size of the transis transistors that make up the M5 MacBook Pro and the size that we're talking about because we're talking about billions of transistors on a single diode, right? the size that he needs to shrink him down to is just unfathomable truly. And I think that we saw it just there where he compares the original transistor, the um the one that started it all, I guess. And even at this scale, there it is. Even at this scale, there's the M5 chip. This is one transistor. This is what they used to look like in the old supercomputers that used to take up the size of a room. And this is like him, the size of a rice grain or something like that. But it goes all the way down. There's a white screen. All the way down, smaller and smaller and smaller. And it's just such a good video. Such a good video. You should also show the Epic Spaceman uh channel because yeah, I've been following him for quite some some time now. Epic Spaceman and his videos are really good. Uh, one of like my favorite videos is putting all the stars and galaxies in a in a pool and visualizing how big everything is. And here he's visualizing how small something is, which is like transistors. So this is the complimentary video to AKBHD. So let's ignore that one just for now. Let's go popular. I shrunk every 10 times every 21 seconds until I'm an atom. So I guess he's just talking about [clears throat] he's talking about what he can see, think, and feel because I think physics starts to break down. And in fact, MKBD says this in his video that physics begins to break down at the level that the transistors at the size the transistors currently are around like a few nanometers. um we will start to deal with the world of quantum and this is where things like electrons just randomly move across um without you know you can't like gate them which you know that's what transistors are mini gates for electrons but electrons just teleport from a point to another point but yeah uh it's really cool to see it in action uh in that video we'll leave links to everything down below. We'll call it a day there just cuz Faza needs to shoot. That's all we have time for. If you're on in if you're on YouTube, give us a like, give us a comment right now because we need it. Um, follow us on Command Show. Like, subscribe, and all that good stuff. And join us next week for all the news and AI design and dev. I've been Sam and I've been cover. Keep on vibing. Peace out. Very good.