#39 Figma Code to Canvas, Gemin 3.1, Sonnet 4.5, Anthropic Crackdown

one and we're live. Nice. Hello there. Hello there. Welcome or what? Welcome. Um, where's your Red Bull? Oh, it's a monster this week. Oh, wow. Yeah, I haven't I I haven't had a monster in years. Literally years. And I got this one and I I don't know. It's It's okay. I bought it last week because I drunk my Red Bull and I was like in case of emergency. Crack can in case of emergency. That's That's how I do. That's still basically the drug that I take. Like I don't do more than this. This is like my Yeah. maximum amount of taking drugs. Yeah. Oh, you're a good Well, you know, the the Sundays Sundays with family uh beg to differ to be honest. That's alcohol is a drink. Well, those are uh Well, I make cocktails. I don't drink. Well, I I sip the ones that I want to try because I want to know what I made. Yeah. But I don't You don't drink? No. No. I I just like I like getting my friends drunk. That doesn't sound uh ominous. Yeah. Um yeah. Say no more. Say no more. Uh what's been going on this week? Any news? Any news? Uh no, not really. been doing a lot of design for my agency with my team and a lot of vibe coding like things that I want to share obviously in the next few moments. Uh yeah, but it's we got stuff to share with you. Yeah, always always vibing. Never be vibing. Well, I just did a launch, so um yeah, I should probably check that it's all working, right? That's not that's something you do, right? You you you check that the app works that you just launched. I'm pretty confident it works. You know, it works in dev. So, you know, work uh I'm just telling my AI to not make any mistakes. And since last uh stream, I learned to tell it to write tests, whatever that means. Okay. Oh, yeah. Because we went live on Wednesday, didn't we? Yeah. Yeah. Just write tests. Well, it's a weird one because obviously if it's just updating the tests all the time, then you you you want to only add tests for new features when they're complete. And you can you can physically test their complete and then when stuff like when test fails that your AI is not like you always just want to make sure you run all the tests at some point but the feature you only create tests for that one feature get them passing and then you can verify they're passing and then when you move on to a new feature and it runs them all and then it breaks something else you then need to be like okay we need to Fix the feature, not that broke. Ah, not not the test. Don't touch the test. Yeah, cuz the test works, you know. Um, see, this is this is giving me kind of like new perspective. Um, I I I completely get what you mean. Yeah. Yeah, it makes sense. But it's so easy to confuse the AI. Yeah. Yeah, for me to confuse the Yeah. Um, I'm testing my app, by the way. That's why I'm I'm down here like pressing buttons and stuff. But I've um I What are we on now? I've reached 10,000 subscribers. It's been a long time coming. We're currently at 1011. So 10,011 subs. Let me check. I I think I'm at 4K. Uh nice. You you should when you hit the the five. Yeah, I don't post which is really bad. But I have a video like right now working on the thumbnails for that Figma uh for that framer pricing. So what I did I I calculated basically you know if you have like three people in a in a team of with three projects how much it's going to cost and it's it's insane. It just does it stops making sense. Um and nothing like really too bad against framer but I don't understand why it should be that expensive. So that video is coming out. Uh I have another video um that got delayed about uh styles that you know are going to we're going to see in the like trendy styles that we are going to see this year. That video is on the way and I'm as I said like prior to the show starting and now we have a few people also watching. I'm not scared of AI anymore. Oh yeah. liberated. Liberated. You're free by AI. So that's good. I see this constantly. A lot of people are they they they they feel like it's just so much buzz around AI and it's so difficult to wrap your head around it and where do I start and it feels like everybody is building, everybody is shipping, nobody is building, you know, you're especially when you're not seeing the benefits. Yeah. A and I I see so many people saying like where do I start? And I was like that for a good chunk. Um but since I started using AI in my terminal, it it it shifted. It's like slowly but surely shifted. Now I can see okay where my shortcomings are. I see like okay how I could do better and just learning this new way of doing things. Mhm. Uh I'm veryware well aware of like how I could uh introduce a lot of like vulnerabilities and you know all of those things as well. But yeah uh it it's liberating to to see this is the state we are in and these are the tools that matter and they are not scary anymore. That maybe that's maybe that's it. Maybe that's the long story. You've taken command of the AI. Yeah. You now know how it all fits together. Yeah. It's not scary. I Yeah. No, it wasn't scary anymore. It was so it was so I I deployed my an app that I made like for tests and I deployed it on railway and it's just a new world in a way like databases and you know infrastructure and so much that coming from web flow you know you don't deal with any of these now you deal with it with AI but yeah It's It's not scary anymore. It's just I have AI and I'll figure it out. Yeah. Yeah. Well, that's good. That's good. That's feeling nice. Yeah. It So, I'm I'm working on a kind of kind of not core like I want I want to go into organizations and talk to them about how they can use AI to be better, work better and things like that. And I came up with a concept this morning, literally this morning. And what people tend to do, and I don't, you know, I don't blame them at all, is I think they tend to overwhelmingly try and figure out like, oh, what can I do? What can I do? What can I do? And they end up like just really confused. I even got a comment on one of my videos, some guy who's following my stuff. He was like, I'm still deciding what to build. I think we need to I I describe these things as there's two levels that I've introduced and I'm still working on this this concept this idea. The first level is well first of all you got to start with the problem that you're currently having problem that you're currently facing right forget you got to basically forget about the technology the technology comes absolutely last decide on a problem that you're having or more specifically a problem that you encounter where you you're doing the same thing over and over again. Right? automation is is basically where we're trying to get to level one. And and if you can get to level one and figuring out how you can automate things, how you can systematize things, that lays the foundation for AI tooling, right? How how AI can help you with your tools. And um again it kind of stems from this idea of another example uh some guy who's he's trying to get me to build something for him but I'm like you know he he thinks it's just some magic box that you can just chuck stuff into and loads of stuff comes out and um most of it most of the time it all it starts with a process that you're already doing something that you're already doing. Like I I had to say to him, where is the data coming from? Like what's the sources? Where is it putting it? And all this sort of stuff because it needs to the AI, if you're talking about an agent, the AI needs to follow and do things with real data. It's not just going to magically create stuff for it to work with and whatever. So yeah, I mean, you need to be specific. Need to be specific. And and it it's not it's not magic. It's all it's it just does things ba things that you're already doing like we'll talk about today the Figma thing. I mean maybe that's a slight maybe that gets a bit crazy that's that's probably level three when you talk about that sort of stuff but really all you want to do is you want to get stuff from that someone's developed into your Figma design like you were doing that anyway you just might have done it in a roundabout way. You might have downloaded the HTML or whatever it is, but you're just replacing things you're already doing. And I think that's if you can understand that without thinking about AI, you you're those are the foundational elements to understanding AI. It's like it's not doing anything magic. It's not doing anything that you weren't already doing. It's just doing what you already did, but quicker. And again get to level three and you start to understand that actually it's doing some technical stuff but it's not much more complicated than that. So uh you asked what I been doing and I forgot to mention something something really interesting. I had a conversation with my neighbor just earlier today about AI and he's not an AI guy. He does music and he plays dart on the side as a hobby with his friends and I think they go to a place and they play and I don't know how it happened but we started talking and he said they have they have an old software and he has tried to do something with chat GPT about it because they don't want to do it on paper and half of it as software half of it in WhatsApp groups and half of it on paper um and I was like this is super easy nowadays. But he tried Chach GPT which to us it it doesn't it make any sense like why would you use chat but I don't blame him he's not a tech guy he he he doesn't you know know lovable he doesn't know rilet he doesn't know any of these tools and he won't he will never probably install cursor right uh let alone clot code so I just told him I said send me a 10 minute long um WhatsApp voice message. I don't care. Make it as long as possible. Explain everything that you need and then I will put this I want to put it in a PRD like transcribe it and then I said I will make a list of the features and send it back to you and if it's fine I'll just give it to Vzero or Replet or whatever and I'll build it for you just like the drinks app and it's it's quite fun. That's that's the level where we are at. I feel comfortable like it's a small group of friends to make something like this. It's they they probably even don't need like Google authentication or what whatever. They will need just some password and something very simple. My my dad is back with uh with my son. Give me a second. So, how's it going? I haven't actually been watching this stream. So, hello everyone. We got 10 people. Um, give us a like wherever you are. Where are you? Got a couple people on the Command YouTube. Give us a like if you're on there. Um, what is he doing? I can see him down in the bottom dealing with his uh whatever it is he's dealing with. Um I what Kabaza was just talking about there is I've got a couple of videos on my personal channel um talking about how I take an idea and actually build it into an app you know from a discussion with someone and normally it starts with basically yeah as did just transcribing it turning the transcription into a PRD then feeding the PRD into an AI whether it's I like Replet. I really like Replet as a as a a product to start my um start my projects in, but then I sort of bring it down to claw code. Although Kabaza knows this, although I've been you I'm I love claw code. I really really like it, but I'm not opposed to using other tools. And I've been using codeex this week, the the model like um well codex the app, but also codeex the CLI ma mainly to try out the models like you know at the end of the day they're all harnesses. I think Claude code has better quality of life features than any of the other CLIs, hooks, MCPs, um, that don't take up context. You know, it's just it's a better CLI to be honest. It's a better app, but saying that it's not necessarily the best AI for all of the best for for the right um, you know, features and whatever. I was just saying that I've got a couple of videos on um that process that you just described of taking someone what they say put it into a PR put it into replet probably you you going to use replet for that or you going to try and brave I I will use bolt because I bro we we have the subscriptions and we are not I've never used Bolt never used it yeah I I want to use Bolt to you know give it a chance to see how it works Um, but yeah, did you see that comment? No, I was waiting for you. Yeah, cool. So, um, hey folks, do you do you know I think he's trying to do you know any folks or creators who make good money just by being the the creator, not an agency or doing design and stuff like that. So just a creator, a creator who makes good money. I don't know. I might know someone the I I come from my perspective, right? I don't talk about running an agency like I talk about how to use tools and build things, right, and write code. I'm not designer, but I make money. I'm not going to say I make good money, but I make money doing the creator stuff. Um, I'd be interested to know what your goal is. Do you want to become a content creator or something? Is that why what you're are you just trying to work out whether there is money to be made? Because here's the thing actually. Here's the thing. All right. You triggered something in me that I've noticed recently. H there are so many creators who we've talked about it the snake oil of here's how to make loads of money and here's how to become successful and they teach you how to do the thing and then you look at their own credentials and they are not doing that well themselves. So, here's how you make uh, you know, an easy example like become a blogger or vlogger or whatever. And then you look at their personal vlogging channel and it's like they're not doing that well. They're just good at teaching you how to do the things. And because there are so many people wanting to learn, they think they're really good content creators. No, no. You you're you're there's this I don't know what what would you call it? This kind of fallacy of Yeah. teaching people how to do stuff that in of itself is a successful niche. Um, but yeah. Yeah, there are creators who talk about creating. I don't really love that. Uh, it's like what do you like to do? Um, I I I like it. I like it when a creator they have some of something of their own and then kind of like on the side they talk about the the being a creator and not just talking about making content and their content is talking about making content. Yeah. Yeah. Yeah. And I I I see a bunch of dudes on Instagram and I watch the content and it's just talking about content and it's not what I like is like okay maybe a science communicator, maybe a tech enthusiast, maybe a storyteller, somebody who does something. Uh I came across this um woman on uh on uh Instagram a video of her on about some planets that are like like with amazing storytelling. And I went on her YouTube channel and she had like videos on the pyramids and I'm like this is good. This is like high quality content, high quality storytelling. I'm like, if she for example would talk about creating content and give advice, I'd be like amazing. People who are really good at it don't actually Yeah. teach about it, right? Yeah. So there there are few, right? But that's where I would uh look at learning like I would try to learn from those. But also to answer the question, you can make like really good money uh if you have reach. So these companies, they typically pay you for your reach. Um like and you get that reach by being a good communicator, but be by being good at storytelling. And if you are just starting, don't um don't don't get yourself busy with the technicality, with camera, with setup. I even forgot to turn on the lights here. Don't don't don't get yourself like uh don't go on that rabbit hole. Just try to unless you enjoy it. Of course, a lot of people enjoy that sort of stuff, but it's not the thing to learn, is it? It's not it's not about that's not going to get you views. That's not going to and it's not just about views. It's when I talk when I say views, I don't mean it in a shallow way. I mean it in a in a way that you connect with people, right? So do something that um that you know interests you and that you can enthusiastically communicate. Well, it looks like he doesn't he wants to get into content creation. Ron Wayne wants to get into content creation because not enjoying the agency stuff all too much. I mean, it's one thing. Just be aware that it's I'm just hitting 10,000 right now. I've been doing this since 2020. That's six just under six years because I started around March to March and June. I had been doing a few little bits and bobs here and there, but you know, six years, I would say my journey was pretty slow in comparison. If you can hit on something, then you know you can grow very very you can grow very very quickly. But you need to understand the amount of content that's being uploaded to YouTube. The amount of stuff that just goes unseen, the ones that pop off, um a few a fraction of the people that actually get to a point where they're making money. Um, I don't know whether my timeline is good, whether it's bad, whether I don't think it's good, but I don't know whether it's bad and I or I don't know whether it's average. Some people might take a lot longer. I don't know. I don't know what the averages are, but the point is it takes a while. So, if you want to do it, start sooner than later because it takes a while. How long have you been doing yours? That's what I want to also mention. It's hard to put a number on because I've been doing it on and off. Uh but when did you decide when was that first upload? Probably 2020 uh or 2019. Five years ago like is my la my first uh YouTube video, but I was doing like I w that's what I want to say. I was putting myself in front of the camera way before that. So, it's been almost 10 years of me trying to do it. And believe it or not, it So, this is like what most people get wrong. People who have not done content creation yet, they don't think about how freaking difficult it is to get in front of the camera and just deliver something. We just forget that it it's it's not easy at all. It's not for everyone. You make people make it look easy. If you look at any job, like there's nothing more exposed than this YouTube creating YouTube videos. But if you were to watch anyone in action, they'd make it look easy and you actually sit down and actually do the thing that they're doing, you find it's a lot harder. You should I think a lot of people in my dayto-day, they always undermine, they think their job is like the hard one or their job is the most important one. It's because we fundamentally cannot empathize or or appreciate how difficult another person's job is because everyone just makes it look so easy, you know. Um, saying that um what was I going to say? Saying that, what was I going to say? Um, I don't know what I was going to say, but the to your question, mate, I'm I'm a I'm a developer. Like, I've been I've got my I have got my own agency. I'm a developer. I've been a developer for 20 years. Um, probably over 20 years now. and I talk about code talk about and like I went heavy on the AI stuff in the last year. Um, so I've moved over to transfer because as a software engineer my skills are dead now. Do you know what I mean? So I needed to shift my my content needed to shift into AI coding because coding just doesn't float anymore. No one wants to learn how to code. I still think it's important mind. I do think it's still important, but um but yeah, what was I think I was gonna say annoying. Never mind. Um we should we should move on. Yeah, we should. Uh I'm sure it was something about on YouTube or something like that, but I don't know. I I want to go with the intro, but first I turn on the light before it gets really dark here. still living in Germany. I think I think that it's debatable whether no code is too powerful. Like no code tools I find them really slow and aborous. Although we'll show you something today that makes it a lot easier, but it depends what you're building. If you're if you're if you want to design interesting websites then yeah no code is pretty much design if you want to design interesting marketing sites and e-commerce sites and things like that then no code is a pretty good tool if design is the way to go whereas apps and productivity tools design is becoming much less important and learning how to code with AI is becoming more about uh the way to go. However, you'll be a lot better at AI coding if you know how to code. You know, you just how how everything fits together. Like, you can you can vibe it, but you're not going to get anywhere by vibing it. You you're you're not going to get anywhere professionally by vibing it. You need to you need to um be at least curious in time. Have some fun like Kabaza is doing. Like, have fun creating loads of wacky [ __ ] But then I hopefully your curiosity will start being like, well actually what's going on here? Because I just remember back to when I first started coding and I was just copying and pasting stuff from Stack Overflow, from Google, whatever I found. I was plugging it together and it was a couple of years of that. And then I was like, well, what is going on here? Because it's it's annoying me that I just don't this is all magic to me. It frustrated me. So then I started digging deeper. So find find your purpose, find your fun where the fun lies and then uh dig deeper. But coding itself um that's that's not the first thing to go for anymore. It's I think it probably should be have fun with AI first and then then learn the code side. I agree. Uh what you what were you saying about the intro? I just said I'm going to turn on the lights and Oh. Uh, and then we'll go with the intro. We have quite a few things to show. So, yes, uh, let's get it started. So, this week we got Google Pomelly. We'll plly We'll see what that is. I don't Yeah, it's strange name. The creator of Open Claw has joined Open AI. It's a big news. Google and Anthropic came out with their new models. Uh we'll talk about them. And the thing that I'm most excited to show you and talk about is code to Figma, bringing your vibe coded stuff to Figma. Uh which is very exciting. I've done it and I'll show you and all of this and a bit more uh in today's episode. This is command AI. I'm Kabarza a little bit more and I'm Samuel and this is a week in AI design and dev. Cool. So, let me share my screen. About to share screen. This is Oh my god, this was so funny. This is guys, this is the meme of the week. This is a new thing. We have a new section. No, I'm kidding. That was so cringe. Yeah. And this is by Chad GPT. They their official. Yeah, this was painful. So, uh who is this guy? He is the prime minister or like the the president of India bringing everybody to India and like having this AI summit, AI impact summit showing that how important it is to have everyone you know united together and Sam Altman and Darius the the two biggest like AI CEOs they they are not holding hands actually and it's So it's so bad. They refuse to hold hands. Yeah, they they refused to hold hands. Uh so they are not actually. So they posted this and this is quite funny because they bought they bought well they didn't buy uh Open Claw but they hired the the the developer uh of Open Claw. And this one uh they uh they they didn't sue him but they asked him to to change the name which is which shows like two extremely different approaches like one sees the opportunity and like I need that guy and the other is like too close to me change your yeah whatever. So with that, let's uh probably let let's talk about the the first thing here on our list. Um I'll copy the link and I'll get that ready and we'll talk about it. But we can't hear it. That's fine. We don't need to hear. Should I play it so we can hear it? Uh it doesn't matter. It's like just the music. So, uh, but I need So, this is taking the internet by storm. Everybody's talking about it. It has nearly 17 million views in less than 24 hours. It's insane. This is a product, a new product by Google that essentially kills all product photography. You just upload like a shitty picture of your product that you take with your iPhone. You select a template. It seems like uh the things that you want and then it boom. It it it puts your like as we see it puts your product inside of all these like like amazing mockups and pictures all with AI. Yeah, it it is quite like like a bad day for people who do this type of like product photography. So, if it wasn't dead already, this is Yeah, this this is quite quite good. Uh, but a note on that, it to me it seems like this is just Nano Banana Pro but wrapped around like templates and probably like nice prompts and uh example images and maybe some refinements uh of the AI. They don't talk about um exactly what the model is. Like you could vibe this, shouldn't you? You could you could absolutely vibe this yourself. And I I feel like a lot of these are really good stationary and like um like sets of images for the AI to take as um like a source of inspiration and then add the product instead of whatever product that is there. It's just replacing it. And as we've seen um Nano Banana Pro is extremely good at that. Like I I've been I cancelled my I I don't I I don't know if I should name it uh ls.graphics. I canled my subscription after many years. I I I was subscribed I think for four years now. Their mockups are really good, really beautiful. But now now I'm like, man, do I need to pay a hundred bucks per year for this? Uh, when I can just ask um any AI to do this? Well, and presumably you need to edit these, right? You need to put your you need to skew whatever it is that you're putting into. No, you don't have to skew it. So that's also really nice about it. So I still have my subscription remaining, you know, half of the year. Pay for the year now. Yeah. Yeah. Uh I feel bad because I I I really want it's one of like these strange feelings like I want to support them as well. Um because look, these are so high quality. These are so good. Um but yeah, uh AI can now do this. And to answer your question, um, you just upload your image. So they have the Figma plugin. They have their own tool. You can install it in Photoshop depending on like the mockup. They have like different ones. Um, you you just upload the image or the design and it just puts it there. It's it works beautifully. It works just so so nicely even with let me show some really complex ones. Uh they I know they are like below. They were older. If I find them. Yeah, look at these. Like even with these you just add your design and it basically works. Uh things like this I could imagine being more difficult with uh for AI. But yeah, I don't think I'm going to I don't think use this anymore. What was that tool that you showcased recently where it was a nodebased creation and you could granted it was video 3D? Yeah, it was free pick thing. It reminds me of this this this nano banana rapper thing. It's not the same. However, it's the the output could be used the same way. you know, I've got an image of something and now I want to see it in loads of different product shoot mockup things. But either way, yeah, I wonder why this got 17 million. I'm not I mean, it's good. It's fine, but I'm not blown away by it. Maybe I'm too blindsided by the the technicalities or the lack thereof of it because you could vibe code this. What do you What are people saying in the comments? What are they so blown away? I fully agree. I fully agree. I don't know why it has 17 million. Maybe it's just the video, the showcase is just so compelling. It's selling it so good. The the examples are just so perfect. Um maybe Nano Banana Pro is not an amazing name and it doesn't, you know, I mean, even this is not an amazing name. I don't know. Uh people are like, "This is good. Well, this is didn't work all that well with my product." didn't it? I don't know. So maybe it Well, not maybe we know it doesn't work perfectly fine with small text. It's just it tries to replicate it, but it uh it might get it a bit weird with small text, but I'm pretty sure if you use um like a JSON based prompt, you'll get better results. And this is what I do. Uh I you crazy [ __ ] people create um I I prompt like you know um based on with with JSON and with that everything is just more exact. So yeah. Anyway, very good. Well, you know, it's going to be used by people. You know, it is what it is and we'll see. Uh, I'm going to recreate like we're in the process of recreating like redesigning my agency website and we will need mockups for uh our case studies. Our main focus is on our case studies. I still have my subscription for this like the remaining and we'll see we'll see I'll talk about it if I'm going to continue using a tool like LS graphics or AI is going to be enough. Cool. Yeah. Peter Stein Steinberg. Steinberg. Yeah. Steinberger. Is it Steinberg or Steinberger? Um, so Sam Olman has announced that Peter Steinberger will be, who's the creator of OpenClaw, will be joining OpenAI in potentially a way to incorporate Open Claw in TGBT. Um, which is great news for Peter Steinberger. Show my screen here. Yeah. uh you know that chat GPT was the the fastest growing product ever and OpenClaw is the fastest um growing GitHub project with like 200 plus K stars in a really short amount of time. It's interesting that this company is acquiring you know the fastest growing GitHub uh project. M well not a project they are acquire they are hiring the the guy who made it. So Peter Steinberger joining open to drive the next generation of personal agents. So we don't really know what this entails just yet to be fair. Open claw will live in a foundation as an open-source project that open AAI will continue to support. So they were already um supporting it presumably from this uh from this announcement here. So great news for Pier Steinberger. Um, I checked out this interview with uh Lex Freriedman where he talks quite a lot about the decision because, you know, he he goes on to describe that he had so many offers from all the major labs when it comes to uh hiring him. and it was a tough choice. He doesn't announce it in this video, but he said that um yeah, that Meta and Open AAI were the two people that he was considering, which is interesting. Maybe from a personal morale perspective, if you didn't know, the the journey of um Claudebot, which is what it was originally called, got a cease and desist letter from Anthropic because, you know, there's a part of me that agrees, part of me that kind of disagrees completely because you've got open code, you've got now open claw. These, you know, are people getting confused with these because they're open AI? I don't know. However, I did feel like the video that I made about Open Claw or Clawbot as it was at that point, I did have to say and it did confuse me. What has this got to do with anthropic? Where was the ties? Like I was looking for these ties that just did not exist. And uh again from this interview with Lex Freriedman, he talks about how his V1 of this was based on the fir they he bought like a $100 subscription plan to Anthropic and gave it to his friend so that he could use it. And that was the first time he saw someone else and was like, "Holy [ __ ] this is a big deal." So it the name did come from the fact that it was anthropic related. It was part of the it was the the intentional ecosystem that that that it gave birth to basically doesn't really matter. That all being said he went with open AI and uh one of he says one of the things that he wanted to maintain or ensure is that openclaw remains open source. Does this mean that Meta wanted to take not make it closed source but own open claw? I don't know. However, that was one that was the main thing he wanted to keep that that uh open claw is open source and it is it's part of a foundation. It will be continued to support. A lot of people are saying now that it's downhill. We we don't know what the future is and I'm not necessarily going to speak to those things but I I have faith that actually it will remain an open source project and you said that he's actually looking to not maybe not hire maybe you can tell us about that. So here as you see uh I'm looking for open claw maintainers. If you love open source, have experience with running larger projects, are security minded and want to help, drop me an email. So he's looking for somebody. Um he said also in that interview a lot of people are trying to help, you know, with uh pull requests and he he says like these are prompt requests. uh for a lot of people it's just like people who don't know coding and but he finds it valuable still for people like us like me you know haven't done pull requests yet uh to do it sometime just do it with AI still like you know starting to do it which is interesting and now he's looking for somebody more serious somebody who can help and apparently a lot of people are DMing and replying here he's saying like send me the email. But yeah, the project um I also saw that is getting uh updates now. You can like trigger it from your Apple Watch. So, it's moving really fast. Yeah. And and more importantly, I was going to mention as well these updates like they're working with a lot of security is obviously the biggest thing that's the worry. And what I noticed that if if you go to Claude Hub, you'll see that there is actually security scans that are happening on everything, vulnerability checks. With every update, there are vulnerability patches and fixes and and the system now is working with another um claude.com. Uh claudehub.com. Just type in the URL. You're so like Googlerained. Yeah, I wanted to to see it to No, it's it's not Oh, see there you go. It's not a claude.com. Yeah, it No, it is claudehub.com. It is claude. It's with a W. A claude with Claude. Yes, exactly. Claude hub. Claude like this. I can't read it. That's my days. Claude because it's claw. Okay. Ah, okay. Claw, not ed. Claw clawhub. There we go. Clawhub.ai. Yeah. Okay. Okay. Yeah. Um, and if you go to browse, I mean, you might be able to see it elsewhere, but I want you to go to browse. Browse skills. Oh, actually, do you know what? It's a lot clearer on your on the mobile, but on the mobile it shows you like um what's in that red button there? Oh, yeah. Hide hide suspicious even. I mean, that's a minor feature, but the point is it's a lot of security updates that going into this this thing. So, I'm excited about it. And and I wonder as well on a lot of my videos uh people and a lot of my Tik Toks like people are laughing being like haha uh until it gets hacked and this and that and it's like oh those people are just going to drift into like the the background and that they everyone's like rejecting this tool because they assumed it's going to hack you or they hear that it's going to hack you. Nothing has happened yet. Don't get me wrong. My mic, huh? Your mic, Mike. Oh, you're Yeah. Yeah. Yeah. Yeah. My agent. It's not It's It's not happened yet, you know. And it's not to say it is still an opensource single person vibe code. It's not to say that we we freaking don't know that there's risks involved, but it's good to hear that nothing has happened thus far. So, yeah. Yeah, my my next uh clawbot agent name will be Jinyang. Jinyang. You don't know this because this is from Silicon Valley show that you haven't seen yet, but he's a legit. He's really he's really funny. Anyway, so uh about uh uh Peter, another thing that I found really cool uh from his interview was that he managed to move really fast. Even though at the start he was like just playing around with it and just like feeling how it is and maybe sharing it with friends or like he said he he mentioned he used it while traveling. But once it gained traction, he managed to make updates uh to add like Discord, to add like Telegram, to add so much more. He he managed to keep up and like basically help with the project growing. And he also mentioned how hard it was to basically didn't sleep much because it exploded. They have like like 300k plus uh followers on Twitter. It's crazy. Like the amount of interest that he's getting is insane. And that's where you start to panic about about the vulnerabilities or the application because you can you know I won't I won't lie and say like when I release something there's there's always that thing in the back of your mind. And it's like, have I, you know, it's it's nerve-wracking like releasing something out there because there are people a lot smarter than you and a lot more malicious than you who will try and take down. And what if you've accidentally left something in there? You know, no one's perfect. And when you get successful very very quickly in any in any walk of life, when you get that many eyeballs on you, you suddenly you feel exposed, you feel naked, you feel vulnerable. So, I can only imagine the amount of um emotional trauma he probably went through. And he's held up pretty well. He seems like a really chilled out, nice me, well-meaning guy, you know. Yeah. Yeah. Um, so there was also a conversation about him leaving uh EU because as you know, he is Austrian and he had this interview. I I I think this is like uh I don't know this is an Austrian um TV media whatever and it was painful to watch. I watched it. I I I don't want to uh play this because of copyright. I don't know how it is with nonyou stuff. Um but yeah this um interview was extremely painful for me to watch because the entire time there is no like even a single like interest in what can this technology do. Everything is what can it destroy? the whole focus is on how bad this could go uh and not on what could it also get could it uh do right and he's holding up really well he's very chill and he's just explaining the the nature of things like how AI can hallucinate how AI can be you know everything that we know about AI that can go wrong we we know it it's it's nothing new but it seems like in the do they they cannot shift their focus away from that and it's stuck like they is stuck in that point. Um so here he answered uh let me just show the translation it's easier. Uh well starting with this one it says like Peter had this um interview and he's going to open AI why couldn't EU not hold him this talent why it it wasn't about money for him obviously right that's the tweet and he answers he says in the USA most people are enthusiastic in Europe I get insulted people scream regulation and responsibility and if If I really build a company here, then I get to struggle with things like investment protection laws, employee co-determination, and uh paralyzing labor regulations. At OpenAI, most people work six, seven days uh six, seven, baby, six, seven days a week and get paid and get paid. Yes, I know. I know. Flame him in the comments. My younger brother, he's 18 and he shakes his head his head when I do this. Uh six, seven days a week and get paid accordingly. With us that's illegal, which is true and it's very interesting. I I understand. Okay, we need to be responsible on regulations. I I'm not necessarily against them, but we I we don't need to be baby. We have too many of them. Like you cannot do and I've been I've seen this with my own eyes times and times again. I've been living in Germany for 10 years. I see many people starting like trying to start a business. They get stuck at uh like even the the footer notes that they have to make for their websites and all the cookies and all the protection laws and they they can't even start exhausting focus on their business and service and like learning because the law is just insane. Anyway, so there is a conversation around that and it's you know it is what it is. Europe obviously is not able to have somebody like him which is sad in a way. I mean yeah uh as most of you know this is another tweet that I found interesting. As most of you know I'm from Germany so I was able to watch the entire interview with Peter Steinberger on site in build. It was incredibly painful. Yes. uh not because of Peter's answer but because the journalist questions uh typically German Austrian almost exclusively revolved around whether we should be afraid of AI whether data privacy is being protected uh being respected uh what dangers open claw poses and so on. Yeah, that was that was basically the whole interview. The hottest topic in the world in the world was talked down. Instead of sparkling curiosity and enthusiasm among viewers, the program ultimately only stirred up more anxiety and resentment at damining uh in decment of Europe which is yeah he left. There is a balance. Yeah. Yeah. like we we do need to take it seriously but there's no that you can do it in such a way surely where it's like this is fantastic like what like yeah just having a balance of the the good and the bad you know what I mean but I haven't seen the interview I'd like to see the interview but um yeah it's in German and I mean I'm not native speaker and he has an accent as I don't know if it's an Austrian accent but it's the the interview is pretty it's yeah it you need to budget, especially if you live in Europe or you care about uh about the future of Europe. Uh yeah, the next Open AI won't happen in Europe. Just like this, this interview just kind of shows the the vibes around here. And like a few minutes ago, uh I mentioned uh in a fun way, I mentioned this Jin Jin Yang um a character in this show Silicon Valley. And I think that show is still extremely relevant and it's also very relevant to this topic that we are we are having here about Europe and the vibe around creating companies like this show pictures Silicon Valley so good so so finely done. Um there there is this guy not not Jane Yang, there is another guy who has uh a place and he allows other people to to live with him as roommates in and they don't have to pay they don't have to pay rent. They just pay him a share of the company that they are building. So, so he lets people live with him, but he he interviews them as a tech co-founder or something like this as like investor. So, he invests like 10% in them by just allowing uh the companies to exist in his um yeah in in his house. So that is my daddy pays for it probably. I I think he has the house from I don't know like previous um I don't I don't remember where he has the house from but there is this enthusiasm about creating and there is it's it's not about work life balance. It's about you want to build something so badly that you let it consume you like what Peter did here. Yes. He he uh he was not like programming for three years. By the way, that that's also very interesting. He he built something for 13 years. He left. He sold it. He left for three years, vanished. He came back and he found love again for programming and he let it consume him like he he found this idea. It was interesting to him and he worked on it. And the issue with Europe is you can't build a company around that idea. You can't have a company a bunch of people coming together working like that. It is illegal to do that. It is um there are so many issues with regulation and yeah whatnot. So watch that show. Watch watch watch that show and you'll see how Europe is not that. Yeah. Cool. Cool. Well, from one claw to the next, Anthropic just dropped Sonet 4.6, which is absolutely killing it on the uh wow, it's beating Opus 4.6 in some tasks, specifically financial analysis and office tasks. It's kind of killing it. But um I thought this intro was quite weird. Claude Sonic is our most capable Oh, sonet model. I missed that. I missed that. Sonet model. Yeah. Also features a 1 million context size window in beta. Now we've had this in Opus in the past. Now they've got it in uh in in Sonet. And I think you can even if you in claw code if you slashmodel you actually see it in the in the um what do you call it in the models list. I'm hoping I didn't set some sort of environment variable in there to to see it. Hopefully you will see it if you see that /model which is great. Um hopefully they they're not just like Google just give you all of that 1 million context. Hopefully, it's been tested and you're actually able to maintain some level of of alignment in the not AI alignment as in task alignment in that in that flow in that context. But anyway, pricing remains the same. You know the drill. Sonic 4 is not cheaper. They just keep it the same price, which is kind of annoying. But this is performing marginally better than um Opus 4.5 which was released in November which is super interesting because Opus 4.5 changed the game. And I know we say that a lot, but the amount of people who were AI enthusiasts still didn't accept AI coding until Opus 4.5 came along is is mind-blowing. There was a huge adoption of AI coding when Opus 4.5 came along. And the fact that it's now cheaper, you know, and it's the default uh they say in here somewhere, it is the default model in um under claw. AI, the free version is is pretty cool, pretty remarkable. So they mentioned about computer use, which is becoming very important, especially as we just spoke about open claw. Kind of interesting how it's um you know, so high up in this article here. But yeah, they're remarking on the computer use and you can kind of see it trajectory here. Um, and here's some of the benchmarks. And like I say, financial analysis and office tasks coming out on top. You know, maybe there's a speed thing involved in that. I don't know. But still, Opus is the best. Obviously, funny how no one's beat Opus 4.5 with a gentic coding yet. I mean, we're talking about one we're talking about one point uh.1 of a difference. So, you know, but it's still there. And scale tool use, you know, that's quite a big difference. That's why I say big difference three you know percent or whatever it is but yeah um and this is also another interesting graph this is the vend bench and we got a video on the vend bench uh we talk spoke about the vend bench which is actually a vending machine that it's told to uh run to maximize profit. Now you can see it had a bit of a head you know a bit of a sketchy start and it's wiggling around and then you see it pull away pull away and then roof flicks off to the end there. So there's definitely some uh I think they even say in here there's some early investment that it did which you can see it dropped there because it dropped the balance. It invested early to make the gains you know um in the later in the latter stages of the of the experiment which kind of goes into the agentic financial analysis there. So yeah I mean nothing more to really say on it. I'm not gonna I'm not going to boil an ocean here. It's just nice to know a new new um player um new update has has been released. Uh yeah. Anything to add on that? No, that that's cool. I'm excited to to use it and give it a try. Yeah, because you're a claw code fiend now, aren't you? Yeah, I I used my current section session 99% uh and it's going to reset in about one minute. waiting by the clock. Are you an Opus person? You seem like you would be an Opus person. Yeah. Yeah. Yeah. I I just ultra max everything. Yeah, you do. You are a maximalist. So, in the same breath, literally last night, Gemini 3.1 Pro gets released. Um, which Google seem to do this. They wait and they wait and they wait and then they come out with something that this is actually kind of killing on all the benchmarks. Humanity's last exam, ArcGI 2, it's winning. It's beating um not only is it beating three obviously, but it's beating Opus 4.6 on a lot of those things. Sweetbench verified again, they're actually reporting 80.8 whereas Anthropic say 80.9 or was it? I don't know. I can't remember. Uh anyway, I've never been a fan of Gemini as a coding tool at all, but saying that I do use it on all of my nan tasks. I think it's a great just generic generic AI, you know. Um it's great at writing. I think it's good at writing. I think it's good at analysis. I think it's good for scientific purposes. Um, which is kind of yeah, why I use it for sort of administration tasks. So, um, we we'll talk about the animation stuff. Yeah. Do you want to talk about the animation stuff as well? Two seconds. Yeah. Um, we have So, I've got this. I've got it here. Um, so we have this. Oh, I'm not sharing. There we go. So, so yeah, here they shared. Hang on. Was I sharing my screen going through that Gemini 3 post? I don't think so. But you were sharing. Yeah. What? What? Hang on. I'm gonna go back on the live. I'm going to go back. Oh, man. Come on. You got to do better than that. I wasn't sharing my screen. You I thought Yeah. Uh I was preparing the next thing. Okay. Yeah. Anyway, go for it. So they are also sharing these animations done uh with Gemini 3.1 Pro versus the old one and it's getting somewhere usable I would say. So the past animations you see it's clunky. It doesn't really make much sense. Uh it's not beautiful and you would say okay it looks cool but I'm not going to use it. the new version it's like yeah it is getting usable maybe not quite yet but it is getting there I mean you would pay an animator if to do this in Lahi to embed it in your website I don't know like 500 bucks I depending on you know the market and the quality and now the AI can do this this is to me it is impressive and some of these I could imagine maybe a bit more minimal version of them being usable straight out of, you know, the AI. I don't know how much of like I don't know what exactly the prompts were, but I could even I would be fine with like giving some design direction to make it better. Uh, and not just prompt it away. Just design direction, make it better, and animate it. That animation part is very impressive. It is good. I mean like I would Yeah. I mean I I said to you earlier like a normie would look at this and be like what? It looks absolutely [ __ ] I mean but we have to remind ourselves on what's being done here. Um and yeah, it's a free animation so why not take it, right? Yeah. And maybe the designs are not good. That that's why I'm a bit hesitant. the designs are not amazing. Maybe it's just a bit too childish, maybe too cartoonish. Maybe it is a bit outdated the style. Um, but maybe if you could get like a more nicer modern u style and animated the animations like the physics of the animations look really spot-on. It it has details. It has like parallax. it has like some sort of like understandsive understanding of the physics. So that that part to me is very uh very interesting and yeah I see um I'm trying to think what model it was. I think it is Gemini that uh Replet uses for its designs is UI designs. I think it's re I do think this on in hindsight now I'm thinking I think Gemini is really good at UI design like a bit of a underdog cuz I mean look at this now you know it's and and I remember like doing a doing a test inside of Replet and it being quite creative when it comes to UI design. Don't get me wrong it's not going to be insane. What was that one? Claude be like um you know it's not going to be the same as a designer coming up doing loads of research figuring out it's not you're not going to beat that of course but for someone who just wants something a bit less AI generated and a bit more creative I think Gemini is your your your go-to to be honest so yeah and replet streaming now look so you can vibe code your video oh because they won't so now they've got an animation platform so you can vibe code your animation which is interesting. That's That's cool. Yeah. Is it using Gemini? Is it like motion? Well, yeah. I mean, it's same output, right? Same output. I don't know how it does it, but um maybe it's using Gemini under the hood. Why is there not something like this for Gap? For Gap. Well, rem um Remotion like Remotion for Gap. Is it using Gap? Probably. Probably. But why why does it matter whether it's because the output's the same, right? Yeah. If the output Gap has like tools for like I mean for scrollbased animations for like SVG um more things. So there are like these GAP plugins that are like specific for these things or I'm sure you can install GSAP on remotion, you know. I'm sure. Um, all it's doing is Well, I don't know. I mean, these are react components, right? They're react components. Yeah. But like, yeah, I don't know. You'd have to just have the play, but I'm pretty sure that it can. Yeah. But I mean, how difficult is it to record? I I don't know how that part, the technicality of that works, but how do you record the browser screen and uh export a video? Well, that's Raspberry Motion's job. Yeah, that so-c because like the rest is just vibing with Gap, you know, you could vibe anything with Gap and the AI could also handle creating the HTML elements and the styling. Um, and then there is a layer on top that captures that essentially. I'm guessing it's um leveraging the get next video frame API in JavaScript. So it kind of freezes it, saves that snapshot, plays the next frame, takes a snapshot, plays the next frame, and at the end of it, we'll just put them all together in ampp4 or something like that. So all rendered on some sort of canvas or something, you know. Now mix that with um theaterJS. Is it theaterJS? Yeah, I'm not too sure. That would be so interesting. TheJS. Yeah, it is uh theJS. So just as an idea, I mean theJS is uh used for 3JS. Yeah. Um, so it has like this uh I know it's like [ __ ] quality, but doesn't matter. Um, it is like a timeline similar to uh what you have in After Effects and you can play with the speed ramp and the easing and you can drag and imagine all of these could be your different components that you animate. This is a plugin like this with 3JS with Gap with Votion on top that that like you could create like any sort of and video and animations it it could get really interesting. So yeah uh it is a bit busy the UI if you don't know video editing but if you have used After Effects you will feel right at home. Yeah it makes sense. Yeah. Yeah. Anyway, cool. Cool. Right, quick one. Um, and I just wanted to get your thoughts on it. So, I can't confirm this. It's it's a LinkedIn post from a Twitter post. 4% of GitHub commits are now written by AI specifically clawed clawed code which what was the statistic we read out like a few months ago which basically saw a quarter of GitHub's entire um user base was um you know uh joined in the first quarter of Janu your age. Do you remember that? No, not really. I think there's something like 20 30%. Yeah, something like that. Like, you know, 25 I think I think GitHub I'm going to throw numbers out there that might sound crazy, but I think GitHub has like 25 million users and like 8 million of them were were discovered in January or something crazy like that. Like it was a large number. and then thinking that 4% now is you know uh and and projected north of 20%. So they've he's he's you know extrapolated the the trajectory and apparently 20% will be end of 2026. So this you know thinking about what anthropics said that most code will be written by AI you know should you learn code this and that like it's writing code lines of code is not the is not the output anymore is it really it's the system it's the product it's the it's it's the output um yeah you're probably going to start now like previously you would start at the very bottom you would be like hello world right printed or like console log depending on the language. That's where you would start and then you would build like on top of that. Was I not sharing my screen again? No. Yeah. I thought like it is a short one. You just explained it. It was fine. Okay. Fine. So, this is the Twitter post. This is the Twitter post and it's a LinkedIn thing and this is the this is the stat here. So, okay. uh north of 20% by the end of 2026. Okay, cool. So, um previously you would start at the bottom and then build up your knowledge and then like later on you would build bigger apps. Now you're kind of like starting at top like you're building some app and it's getting really easy to build an app and as you get better you are probably just going to dive deeper and understand more like what is going on like maybe next thing is the structure and the technologies that your AI is using and you become aware of that and then you become maybe a aware of your components that you are building and then maybe a bit more about like the classes and the scopes and maybe syntax sometime and I I feel like this is the route that a professional like the next generation professionals uh will go. Yeah, I think like again like understanding the understanding the code is uh still still really crucial like but it's more so about the systems that that pull it all together and and whatever like you know and then the best of the best the Netflix the Google the these massive companies who where every single kilobyte matters um yeah they're going to need those people top tier who really understand the code because you can get away with you can't get away with failing. I mean, Twitter is [ __ ] doing it, is it? Getting away with sloppy software. But yeah, you you even that is not, you know, as bad as my vibe coded app, right? It it's still like people there who understand um the code. It is getting sloppy and it is like lower quality. But like your point there completely holds on the level you have where you have like YouTube where you have like billions of videos being played. Yeah. you that that's a that's a different problem to solve and it's also not something that AI is really like has any experience on or has like enough data on like these things are so unique and you need humans who deeply understand the system AI comes to play from my understanding in like searching the database and maybe come out come up with like new perspective is like an assistant to the the human who can listen to the AI, read it and like have a clear understanding of is this actually true what the AI is saying or is it like just nonsense. So you you you still need that. Um and we kind of know this from also this thing that Anthropic did with um what they said. They said they let uh cloud code to write a C compi compiler. Oh yeah. They run it for like a week or two weeks straight. Two weeks. Yeah. and they they spent like $20,000 of like API uh tokens worth on it and it they said it kind of works. Uh and there was this video from Primagen basically saying like no like this is just like [ __ ] marketing. It is deceiving because you're doing it on something like completely open source and you're not like completely walking away and it's not actually like working all that much and it's not let's not even talk about the performance and things like this. So there are a lot of like asterisks around it working and even that is around an open-source uh project that already exists and the AI you know has been trained on it heavily. So yeah for that level humans still uh yeah beat AI. Yeah, the big one. Code canvas. Yeah. Uh you we didn't agree on an uh uh on an intro. You have your intro and I have one of mine. Let me also I'm grabbing my screen and like putting things together, things that I will need. Yeah, because we spoke about this at the beginning of the show, this idea that this is a re very very clever way to use AI. In my opinion, it might be more simple than I'm anticipating. Okay. So, uh let me get a few things ready. Um I don't know if I should show the announcement as well. Just just a second. Um Claude, maybe if the announcement is nice, I will show that as well. Yeah, I I'll All right, I'm ready. So this is probably the biggest So this is probably the biggest news for product designers and developers. Now you can vibe code or like program your front end whatever app that you are building and then after building it bring your code and visualize it inside of your Figma and then a designer can work on that already existing you know functional app uh change the designs and then send it back to uh Claude via the Figma MCP and then continue uh working from there. So, uh I can let me actually just make sure that I got this right because I think this is big. Maybe I'll try it again. I feel like I rushed it too fast. Uh you're muted, but it's fine. I know. I was I was texting. Yeah, it's okay. It's okay. I'm texting my design friend Susie because I told her about code to canvas the beginning of the week or whenever it launched because we've been talking about a workflow that for a project that I think will this will benefit from. So I want to make sure she's checked it out. Okay. Am I sending code? Yeah. Sorry. You go. All good. I'll I'll just do it do it again. clot code to Figma. This is probably the biggest news both for product designers and also for developers. The bridge between designing and developing has shortened again and this time it is becoming very interesting how you can collaborate as a designer as a developer with each other from your codebase all the way to your Figma and then back to your codebase. So with this new update, you can code something and then send it to Figma. And then the designer can change the design inside of Figma and then send it back to Claude. And I have a demo to show you. Okay, let me which you know just jump in here as well like it does solve a big problem that a lot of people are starting with the prototype starting with the vibecoded prototype which doesn't care about design which just gets the prototype up and running and then it's a way of how do we now synchronize that process of applying the design and then getting it back onto the the the vibecoded prototype and this Yeah. And I Yeah, I have something to show on that as well. So, just to just to show you what it can do, what you see on my screen is an app that I'm creating uh for Web Flow users, potentially for more people uh to manage your website's images uh from this app. Now, I didn't design this inside of Figma. I just vibecoded this. Uh there are many design issues with it though. I there are many small things that I want to change and prompting it is just like painful. Uh with this new update I just asked clot code to um send all of my screens like when I say all of my screens I mean like when I click edit this is a different you know component. all of these and all of the different states like the hover states, the select states, everything send it to Figma. And here's the the result. It just sends everything to to Figma. You click in there, you can you can change things. Is identical? Yeah, pretty much. It's not perfectly identical. Um because it creates So, here's how it works. It creates this HTML structure and then from this h uh HTML structure it like copies it to Figma and for that you need the plugins and I will show that as well how how you can get it set up and running. So here I have um my claude and I just want to say do it again. Do it. Is is this Claude code? Presumably it uses Claude code, Claude desktop, Claude probably called co-work to be honest. But it's just an MCP, right? Yes, this is an MCP and a plugin. And that's the thing. There is this new plug-in thing. And by the way, this is how it works. And now it it opens my project in a new tab. And you see this select element to capture. And now I can like hover over these elements. I can tell it to directly send it to my Figma or I can do it like this. I can let's say if I if I'm interested in by the way look like these views. If I'm interested in one of these or this section, I can just click it. And now it sends it to Figma. It takes a little bit. It takes a little bit to send it, but the whole point here is if I wanted to tell my designer to fix my design, how would I do it? I would have to take a screenshot, send it to him, and he would have to like recreate everything. He doesn't have to do this anymore. He can, you know, do it like from here. In fact, we did this for a client here uh is a client we did we did this for two months ago. It's painful to think that these are screenshots. They sent us screenshots of their app because they were building it. Developers were building the functionalities and they were not caring too much about the design first. They wanted to get it up and running. They sent us the screenshots and based on those screenshots we came up we came here and we made designs like everything here. we had to like recreate the frames which is you know now the old way of doing it it's it's so time consuming and I messaged my client and I said like check this this new tool like if they want now if they want to redesign something or if they implement a new feature they don't have to send us a screenshot they can send us directly um the component inside of Figma we design it and once Once we design it, here's something cool. So, this is something very simple, but I haven't seen anybody talking about it. When you're in Figma, you can go to Figma, the the dev mode, and here under the MCP, you have this copy example prompt. It's nothing crazy. It just says uh implement this. And it it it has this link that is like a specific link to whatever you click on. And I found this to be very helpful. U actually in my example, I I went in and let me if I can find it. Like for this slider, I used a component library. I tag I I selected slider. I changed it to the style I want. I copied the prompt, gave it to Claude, and it did it perfectly. Nice. So that was really nice for me to to say h the the the output from the AI I don't like it. I go to the one that I like send it and it executes it perfectly fine. So I think what will really help if you can you break down the different components that make up this process like what do you need to download? What's how can someone get up and running with this? Yeah. Uh, I will get that in a second because I think it's pretty clear what problem this solves. And um, yeah, it's really cool, really handy because it used to be that you would start with the design and with you. Again, you need to pick your battles because some apps need you to think about the UX. Maybe there's something really complicated about it. Maybe there's very specific needs. Maybe it needs to follow something that has already been created for whatever reason. Yeah. Whereas there are tons of projects that are going to benefit from just starting out with a shitty AI coded well not shitty sorry just a AI coded or vibecoded um prototype that the the client can actually use. It makes m so much more sense for so many projects just to do that and then think about the design later on. Case in point of the project you and I have been exposed to recently, you know, where where we we'll start with the vibe coded prototype. They can see that it works, they can use it. It's like, okay, now while you're using it, we're going to go and take it and we're going to design the heck out of it, you know. Yeah. So, in order to get this working, you have to do two things. One is the claw, the Figma MCP, the Figma MCP install. Let me actually bring this up as well. Um, I mean, it's so simple. I just asked my claw to install it. I haven't I haven't even used this. Um, I just asked my We'll leave links down below so you can you install the MCP and technically this could this could be installed in cursor. This could be installed in codeex. It doesn't need to be clawed, right? It's just an MCP. Yeah, I can show this. So to to do all of that, you come to your cloud or codeex or whatever and you ask it to install your MCP. Now, if I list my MCP, uh you see that I have here I have the Figma uh the Figma MCP already connected and installed. Just ask your AI to install it. And then we have this Figma plug-in thing that is the the code for it is here. You just paste in paste this inside your Oh, yes. editor or yeah you can ask a cursor to install it. Just paste this in and now um you have to also authenticate it. That is also important. You authenticate your uh Figma workspace and you're good to go. You you connect these two and you can now send things between these back and forth and you can even like design potentially with this because it is the MCP you can generally use it. So what does the plug-in do then because that's where I'm lost and this is where it becomes claude specific because claude other things don't have the concept of plugins. Yeah. What what is this responsible for? It it is it is I I'm not sure because it says the MCP see it is like all about the MCP. So here for Figma claude MCP ad you know this is the MCP. I'm not sure but I I did it I did it as it was said in the post. I I just added this to my cloud and it installed it and it maybe maybe this is the thing that runs the server. So you can go in and click and then then this communicates because the MCP is there to communicate from claude code to Figma, right? That's the back and forward. It's the push and pull between claw code and Figma, which by the way, you need a paid Figma account to access dev mode to access the MCP. So there's the connection between the two and I think the plugin is the thing that springs up a server of your app and allows you to click around. That's what I reckon. Yeah. Well, from here it says Figma cloud plug-in directory. So I I think this is probably just coming here and installing the MCP and maybe more. Maybe maybe uh at the end of the day it is very simple. You can do none of that and just send the links to your AI and say install that for me. You could do that as well. Yeah, it looks like from here I'm just I'm just seeing on the tweet to get started install the Figma MCP server and then you install the plugin. So actually the plug-in contains the MCP. So as you see here, it worked. It brought the the UI part that we clicked on. Nice. And that's pretty much it. it. Um I already asked my AI and now it says it timed out but it worked to put every state in like one view and from here I'm like entering this phase of like refining my product because it already works. I already, you know, I'm already pulling data from the a web flow project, uh, CMS and non CMS, and I can, you know, do a bunch of things that I want to do. It's kind of crazy that I vived this into existence. I can't like edit something. Well, it didn't work. I'm in the middle of building it, but it works. I can upload the things back to Web Flow, uh, and all of that. So I want to also talk about something that a lot of people might like they they might not have a clear understanding for. So why is that you why is this useful? A lot of people might think that the process of designing it is going or like building an app it is going from design to dev just like that. that that's probably what people think like you design it in Figma and then you build it. But that's like at least in my experience it's pretty much never the case. Uh what you do is you go from dev to design. You build something and then you fix the designs and then you go back to dev and then you go back to design and then you do the same thing over and over again. And this is what I'm calling like this zigzag approach that you're constantly like passing between these two states. And that's why like the the best people in space are in this space are like the the devs who understand design or like the designers who understand de like connect this or like shorten this bridge. Um but yeah and this tool is making this much simpler like it to to me it seems like with this developers and the de designers don't have to do much of like manual work anymore this is more like the AI is taking care of like the leg work like the the layouts the layers like the simple stuff and the designer can come in with like strategic IC thinking for the flows. Maybe the components, maybe there is like something about accessibility, maybe there's something about creativity. So the designers can focus on that instead of the layers or naming the layers or you know creating the outer layout. It's already all being created. Same for the developer. That's why I'm so excited about it because it's to me it's not replacing my designer like my actual designer in my team. This is like enabling him. Now I can give him work and I feel like I I feel like it's so much easier for me to give him something and he can just take it and start working on it instead of me thinking, "Oh, if I give him this task, it's going to take so much time to just set it up or just to understand it or like just to list everything." Now it happens all all in like minutes. So yeah, very cool. So to sidestep from this, right, what does this mean for charging for a website? Like if someone comes to Oh, sorry, not a website, an app, right? Could be a website, but it's more than likely going to be an app. What do you reckon? How do you reckon this this changes things there? Because like you know the development traditionally when when someone comes to me and says they want an app the actual development costs become very vague because it's like well we don't know what we're design we don't know what the design is going to be we don't know what the features are going to be and this and that. So you sort of what I've tended to do is give a rough sort of figure on based on what we think is going to be there. we let the design and the features flesh itself out after some meetings and whatever and then you can start to solidify the the development costs or you take a punt and you're just like, "Yeah, it's going to cost this much." Um, does this mean then the design itself becomes a bit more of a vagger uh scoping or or price or whatever for the client? What do you reckon? I don't feel like it's getting vagger, but I feel like it will get more important to do the branding and the positioning and strategy strategic thinking and things like that. Uh, and I don't think at the end of the day it's going to make it much cheaper for the client. And it's not in a bad way. It's just that the client won't be paying for me and my designer to to basically draw rectangles, right? they will be paying us uh maybe the same amount but maybe even less but to do more important work and to do it faster like so much of the work would just go into this like early steps of creating that like these layouts. Um, and I think with all of that, you know, um, being eliminated, the work can get faster, it can get cheaper for the client. And I think it is, it is a bit more clear because now I can ask my client in order to get a quote from me. Just send me every screen and every state that you have. And they can just do it like that. What if they don't have anything? That's not what I'm I'm asking. They don't have anything. Like they're literally just like, "Hey, we want to build something." Yeah. But it's still getting cheaper, right? It is getting still cheaper because we can like come up with something like an MVP much faster. But what I'm asking for is not how cheap. It could be million pounds or I understand. Yeah. What's the what's how do you charge that person in terms of do you say you know um does the does the price on the design become vagger because you don't know what you'll be designing or does I don't think it's different from before because if it's if if it's completely green field and I don't know what the app is that we're going to be building like and then but client and clients want the price for the app it's like well I We we don't know what we especially if you're part of the strategic side, right? It's best to and and here's the thing, right? I understand. I mean, this is a debate that will happen forever and ever and ever, but like I understand the difficulty in someone saying, "Look, if you don't know what you're going to build, then we need to run the strategy." It makes sense to run that completely priced completely separate to the website. like don't even think about how much the website is going to cost to design and develop because we don't know what we're going to build. So we do the we do the strategy first and and run that at a fixed price or whatever. Yeah. And then after that Yeah. Yeah. And then after that you say okay we're going to design it but well I suppose you know a bit more but like the prototype the design it might come up with certain let's say animations as a stupid simple example. Obviously in apps, animations don't really come into play because you're really talking about integrations and things like that. Maybe some sort of complex integration comes up through the design process. You don't want to you don't want to quote for the development because the design or the UX could turn over some really intricate features that you haven't yet figured out, you know. So then, but point is the development is such a vague thing. It's so difficult to price development. Yet, I understand that when someone comes to you and says they want to build an app, they're just like, "But I want to I need to know how much is it going to cost you?" It's like, well, I don't know how much is do you know what I mean? Yeah, I I know. But I don't feel like it is going to be different from the past. It is always the question of like what do you want to build? And what is changing now with AI is just some of the steps are getting faster. But if you look at the entire thing, it's not like 10 times faster. You you go for more complex stuff and you go for like more ambition goals. Uh and I think that is going to make the price at the end of the day similar to how it was before. M uh and regarding the the quote thing, um it is like custom in many other industries for you to charge the client for the proposal for the for the quote that you make. It's not like free everywhere like they might need like in many industry they might come to your house and like inspect things and go and like plan things and come up with a quote. They are going to charge you. they will tell you okay we will we'll charge you this much just for the quote uh and then you decide if you want to do it or or or not. I guess in this case we with experience we can say like about part like it's going to be between 20 and 100k uh so the the client knows it's not going to be over 200k right something like that so with all that being said I think it's not too different from the past yeah well yeah it's a cool feature I'm looking forward to try it's gonna it's it's gonna be a it's guarantee it's a it's a I'm going to use yes. It's a yes from me. Yeah. All right. Well, let's wrap this bad boy up because there's been some major clawed code policy updates from Anthropic, which has affected our old boy Open Claw. Mostly affected our old boy Open Claw. um to be honest and I think it this has been brought about by our old friend Open Claw. Um oh that was embarrassing. Basically, uh this is a kind of summary of all of the tweets using oorthth tokens obtained through clawed free pro and max accounts in any product, tool or service, including the agent SDK is not permitted is not permitted. Now, Tar here is is is confirming that nothing is changing about how you can use the agent SDK and max subscriptions. It's it's getting more and more confusing. And uh you go in here, we want to encourage local development experimentation with the agent SDK and Claude P, which is a way of prompting Claude from the command line outside of go outside of claw code. The point is if we come to this thing here, free pro and max plans is intended exclusively for clawed code and claude AI. So they're subscriptions because they're a huge discount on you know um the API usage. So they are only intended for claw code and clawed AI and presumably claude co-work but they've they need to clear that up. It is not permitted and constitutes uh sorry it sorry I clicked on one of the links there. um in any other product or service including the edge SDK is not permitted and constitutes as consumer uh is against their TOC's but this basically means that what what a route to using open claw is you could use it through your claw code max plan which we spoke about at the beginning with with that was the whole Peter's naming of claudebot and all the rest of it you just can't use it and I actually started noticing this I was getting uh a API blocking and things like that. And I didn't really know what it was, but just so happens that I was interested in moving on to Miniax, which I've got a which I literally just released a video during this stream uh talking you through how you can install Miniax subscription with OpenClaw. And I'm getting really good results with Miniax. To be honest, um I'm not into the personality. You're into the personality, right, of Open Claw? No, I'm I'm I'm fine with it being like short and just, you know, just do it. Yeah, you can you can prompt all this in. You can prompt all this into your soul.md and this and that, but the out the box personality is something people do enjoy from Claude. I don't care. I just wanted to get the [ __ ] job done, you know? Yeah. Um, so yeah, I've set up Miniax. I paid 10 bucks a month for my Miniax plan. I'm going to see how that scales up. But um if you want to learn how to set that up, I've got a video on my channel which my link is there. Um but this is this is yeah but the bigger question here the bigger conversation that's happening here is anthropic are really creating a lot of enemies by creating this wall garden approach this applelike approach to all their tools their services and again go to that conversation with Lex Freriedman Peter Steinberger says that in this in this world I mean I understand both sides of the conversation as I normally tend to always do. But in this in this time of uncertainty and discovery and curiosity when you've got anthropic really locking things down and not open sourcing their software, dmcaing open code and and this all these companies and and Claude code obviously Claude Claudebot. Yeah. um they're creating enemies for themselves and and creating an ecosystem of of secrecy and privacy and I think that's rubbing people up the wrong way. I hate capitalism, right? But I also see that this is this is the way it needs to be done. They are on track for becoming a profitable company before open AI. Um, so there's obviously method to the madness, but I can't you can't deny that this is robbing people up the wrong way. And this is just another example of them being like, "No, because all of the other accounts, Open AI, you can use OpenAI's codeex plan with uh Open Claw with Open Code. You can you can use Miniaax obviously GLM all of the they they don't care about how they're they're getting money at the end of the day but yeah it's just such a huge discount from the API usage because all of these other coding plans are not uh sorry the API pricing is not as expensive as open co anthropic so I who knows yeah um they are destroying their relationship with developers which is I don't know it sounds very stupid. It is very stupid because who who loves uh open code? Uh not open code. Uh it's so confusing, isn't it? Yeah. Yeah. After two hours of streaming like developers love uh cloud code and anthropic is destroying this relationship. It's it's really stupid uh what they are doing and I don't understand why are they being this harsh on people for doing this. Um are are people like using way too much of the you know the subscription is because the subscription the 200 bucks gets you roughly around like $3,000 uh worth of API. So, you pay 200, but the actual like if you use it to its limit, um it's like like 3K, which is crazy. That's why like it's so much more worth to use like the 200 bucks subscription compared to using something like cursor. And it's quite unfair to cursor because cursor has to pay the API cost. But you when you don't use um anthropic models through cursor when you use them directly you are paying them like much less and they are trying to lock this down as much as they can and I I'm not sure because people are abusing it or it's just that they are they're trying to be Apple and it doesn't make sense to be Apple in this you know in this industry. like AI with you see like you can just change whatever tool that you use uh just just like that it's so easy to switch they can't build walls that's that's my point like you cannot build walls with AI with because the nature of the product doesn't allow you but this is something actually I'm going to mention web flow right because this is the thing that I really dislike or not dislike that's the wrong word because there benefits to web flow encapsulating you in their wall garden. However, the internet and web technologies are based on open open standards and cross-pollination and interoperability which these wall garden approaches just don't don't work with. Right. I I understand that they need to start somewhere. Um and maybe well I'll finish what I was going to say. It could be a it could be a question of um subsidization because a lot of people pay those 200 those $100 plans and don't actually use the tokens that um they offer you. Yeah. Clawed Code makes uh sorry, Open Code makes damn sure you use those tokens because it's using them without your prompt. You're not do you know it's just churning away. It's just doing stuff and it's quite token hungry on top of that. So they're they're making less profit because people are actually using up their $200 $100 a month plan. So it could also be that. But yeah, back to the back to the web flow thing like it helps when you and Replet is also another example. It helps especially non- techy people to come in and just know that their database being taken care of, their hosting's being taken care of, domain names, all this sort of stuff, the tool itself, they know they can just log into one platform and use it. That's great. But when you start to need more from it, it doesn't scale that well. It doesn't scale well when you can't use a different or system or connect to a different database because they're just so locked down, you know. And I know we're talking about Anthropic and their policies here, but it may very well extend there. I'm not smart enough to know otherwise. Well, we'll see. We'll see how it goes. Um, they might be locking it down just to learn, you know. They might just be like, "Let's just let's just keep keep a lid on things. Let's see how things play out." And and you never know. Their their master plan might be to open up everything. But they're just they're very cautious company. So, it could we we could see a complete open openness when when things start to play out, when they've reached profitability, when they've when we figured all this [ __ ] out, but I don't know. I don't know. For now, I'm going to be using both. Uh, I wanted to cancel my Chat GPT subscription, the 20 bucks. I'm not doing it yet. I mean, the codeex app allows me to build so much. So, I'm just going to keep um having both for now. Yeah. Yeah. I don't know cuz you can't argue with the price of of codecs. 20 bucks and I'm getting okay usage, you know. Um I I had an intense project that I needed to get done over the course of a couple of days and I was really excited to learn a bunch of stuff. So, I really [ __ ] hammered it. I had three agents going off on and and a fourth agent overlooking the three agents and this and that. I used it very heavily. I couldn't do that every and I dreamt of code and agents last night because I was so heavy on it yesterday. Um I used up I've used up 50% of my week's quotota already. That all being said, yeah, you know, the next the next price jump is 200. So maybe $100 for claw code is the way to go. Yeah. Yeah. Who knows? Um I could says this by the way like it is so context um heavy when like for your brain to have like four agents running and actually keep them busy. Read what they are doing. Read the outputs. prepare the next because like at the end of the day you are building something that would take two month within two days or like you're building like a prototype of it or something close to it. So it just makes sense that your brain has a difficult time trying to keep up with everything. Yeah. Well, anyway, we'll wrap it up there. But in case you didn't know, this show was brought to you by OpenClaw. Our producer, my OpenClaw that I've created, pulled together these articles. Every day I would get messages saying, "Which articles do you want to bring in?" And yeah, it kind of worked. I think we need to we need to work on uh it giving us an intro. We need to give us talking points, but this is open code, right? open uh Claude, you just work on it, tweak it, and have it do it. But yeah, so we've we've got a we've got a AI producer now, which is nice. Does it have a name? Well, I mean, my mine's Claudius Maximus. Cool. So, anyway, um that's all we have time for. Make sure to follow us on in Well, we're not on Instagram anymore. just just YouTube. We've been streaming on on uh Twitch. If you are on Twitch, we don't really get many people on Twitch, but we are on there. So, if you prefer that platform, go give us a go give us a follow on there at commandai.show. Join us next week for all the news and AI design and dev. I've been Sam and I've been Kabar. Keep on vibing, baby.