#43 Google Stitch, NemoClaw, Moltbook, Google AI studio, Amazon Code red

position of when I'm when I'm on calls. I see a few streamers do this where they're they're those business bros and they've got their like and it makes me laugh looking at them because it's so cringe. But they'll have their microphone just kind of off to the side like that. But they're it's like this locked in I call it in my head I call it like locked in position where they're just like I'm so busy with coding. It's like th this is this is my coding position that I'm like but then oh [ __ ] I need to stream right back here for the microphone. Do you know what I mean? No microphone back here for the microphone. I got to record a video. They're locked in. It just makes me laugh. Have you not Have you not noticed that? I don't know. I So many [ __ ] What do you call them? business bros like the the typical bros who have open claw like seven AIs running all at once and like building their [ __ ] 100k MMR startup and they're like locked in mode record video. I've got something to sell you. This is how you run six and not have a girlfriend and then you lock in. The ones who are always building but never shipping or always shipping but never selling. Yeah. Yeah, I'm locked in. I've got to come because I can't move my um I can't move my microphone at all. I've got to lock in. I've got to have I've really really labored on this point. You know, it cost me money your microphone because I have to I have to take it out of the the thumbnails using I'm using my API right now. People love it. Wasn't that like last year's trend to have like a microphone like this and you got to really talk into it like that? No, no, no, no. But by the way, I have like Hicks fields, but I'm I find myself sometime just using my own API and my own app that I built that I haven't shipped that I just have locally running and just using my own app to generate images. You need you need locked in position. You you got to have two positions. You got to have video creation mode, 100K MMR startup mode, never shipping anything. That's what we I thought like that locked in position that looks good for the thumbnails. Maybe we should do some thumbnail faces so I can use them. So I can use them like screenshot. Yeah. I will screenshot that and use it for a thumbnail. Holy [ __ ] I've just seen that we've got we've got some microphone settings. Oh, this is new. I've accidentally somehow clicked on people in the right hand side of Streamyard and it's got like I've got auto I've got automatically adjust micro volume and you haven't got auto turned on which is kind of I don't have headphones uh either. I forgot. Okay, let me try visual effects. Is this like a thing? Oh, here we go. Okay. Welcome to my home. Right, we've really um for for a couple of people who are like, we don't have much time. Let's just go and stream live now and just make it up as we go along. We really labored on that intro. Um yeah, so let's let's start let's start. Okay, so uh starting with this week's intro. I didn't finish writing it. So, Google killed web design with Google Stitch. And I didn't write the rest of the intro, but for a few people here, we have Claude code channel. We have a cursor update. We have 1 million context uh for claude. We have can I run.ai? I I don't know about it, but you show me. Uh we have the claude 2x usage. Good for us Europeans. we can finally compete in the scene of AI. That's how we compete. Uh and then what is more interesting here, we have I'm reluctant to say it or not. We have a lot of poops. This is new. You didn't say this earlier. Can you hear me? Yeah, I can hear you. So, it's with Nvidia. uh DLSS5. There was tons of memes around it and there is another video that I'm not sure if I if we should show it or not. I will share it with you first and then we decide to uh share it or not with the audience. But yeah, memes of the the week. So I'll send it here. You can decide if we share the second one or not, but it it's not very important. All right. So, um, give me a second. I need one more cursor. Oh, I see my screen. No, I said I I don't know. You didn't send it to me, but the second one. You You see the second one? It's really It says it's like a 12year-old directing it. uh what what a movie would look like if a 12year-old would um make it. Okay, so it's just an army, baby. It's just an army. And just to explain to everyone the slightly scrappy uh you know setup here, we've got limited time today and so we're just going to wing the articles we've had OpenClaw put together for us. Uh that looked interesting. getting slightly distracted by titties and zero g, but whatever. Um, so we're going to do our best to basically rip off some of the news articles that we've done little to no research on, but hopefully you can join us in the uh discovery of some of this stuff. That was per a perfect intro. So, with that, let me share my screen and show you some goodies. All right. So, Composer 2 by Cursor is out. It's available now and it's super cheap, but also better than Claude Opus 4.6. At least this is what they are claiming, right? Am I reading this right? So, on the cursor bench, well, cursor bench and that that says a lot, but it is their own benchmark. But what they are claiming here is that composer is better than Opus 4.6 the high version but not as good as GPT 5.4 four. So, it's kind of like right in between, but is significantly it is significantly cheaper than anything out there. Also cheaper than compo composer 1.5. I haven't tested it, but I feel like this is something that would work great even if it's not really as smart as Opus. I don't think it is, but this is it would be good to do the work, right? The leg work. Yeah. Well, this is telling me that it's smarter or sorry, performs better. What is performance? I don't know, but that's what this is telling me. Yeah. And for the cost, it is I don't know is it like 4030 cents here. Uh where Opus is $2.5. So it is over what five time six times. Uh I think this is again it's a it's a [ __ ] graphic cuz it's like well you've got input and output tokens, you've got um cached uh price of tokens, but you've also got the actual just general costs to run the the thing. You can have a really cheap model that thinks and thinks and thinks and thinks GBT 5.4 which ends up using a bunch of which ends up costing more. Whereas you've got a you've got a very um capable model that's very expensive like Opus and it's actually really good at preserving tokens. So it doesn't cost do you know what I mean? Yeah. So what like these are two vectors that are a little bit confusing to me but heyo yeah. So give me a second to bring on something else here as well to the screen. Unless it's speed. Unless performance is typically you say performance is speed. So it's faster than Opus and cheaper. That could be it. But not as fast as 5.4. But then the cursor the the composer first model was lightning fast. It was crazy. Yeah. Anyway, I don't know. That's what you get. That's a good point. That's a good point. It might be about how faster it is. We have a second thing about cursor that is apparently in alpha and Theo is saying cursor just dropped a T3 uh code clone and this video looks pretty interesting even though everything in this video in the UI is super tiny. So Jesus, we've seen this. We've seen this with the with the code and the diff on the right or whatever like some content on the right. In the middle you do the chatting and uh etc. What is interesting here is I can't zoom but it is this open project. Let me zoom like this. You see this icon open project. So it seems like this is a view similar to uh the codeex app which I love. By the way, last week I managed to get it to basically 0% like I had 2% left of my weekly limit. It is you you you've taken that on as a challenge, haven't you? Yes, I did. I did. So I I I wrote big PRDs and started like big projects. I do wonder because I I'm I'm I've hit my limit a couple of times. Yeah. Yeah. But but I I I did quite a lot with it. uh I rebuilt like an app a few times. Anyway, the point being I really really love this view where I have multiple projects. I have you know multiple threads for each and it seems like this is what cursor is trying to do which I think it's um the right move and I might not cancel my subscription um at the end of the month. I don't know maybe we'll see but it looks really nice. Yeah. Yeah. We see the chat. They all kind of look the same. What do you know what it's doing different? I guess it's got the diff which codeex doesn't have, but they'll adopt it. No, it doesn't matter. Well, the codeex does have it. The codeex does show you the the diff. Okay. Panel. Okay. Yeah. uh they do show it but here you you also see the the plan and this is like kind of um rendered. This is similar to how we use um warp. I I love warp. I'm using warp uh a lot and yeah this is a super simplified. It is even a more simplified version of the agent view of cursor. It seems like it's not out yet. It's alpha. I checked. I don't have access to this view. Uh, but I would love to get my hands on it. Yeah, that's pretty much it. Cool. Uh, next one. Do you want to talk about Stitch? Yeah, we have tons of stuff about Stitch. So, let's talk about So, this is the latest update, isn't it? Don't Yeah, it is the latest update. I want to prepare my screen and the people who are live with us they can see before you know people on the on YouTube um seeing the edited version. So I have this to show I have this to show. I have this to show. Um should we talk about AI studio alongside with stitch or separately? Separately but afterwards. Yeah. Okay. Uh, I don't have anything to show for that. I want it to, but I I don't think it can build something in five minutes. Do you think it can? Should I give it a prompt? Come back to it. Knock it off and then we can come back to it at the end of the episode. I have a prompt, by the way. I mean, this is Give it the the Are you just going to demonstrate that you can take from Stitch into Google Studio, Gemini Studio, whatever that's called? No, I don't think you can. I think they are completely separate. Well, this is one of the Yeah, you can you can just right click on any frame. Yeah, you should be able to click on a frame and stitch. This is what this is an old feature though. This is what I thought you were going to show, but it's like a it's nothing. This isn't an update. But yeah, you can send it to Google Studio or AI Studio or something like that. Maybe export at the top right. Oh yeah, preview. Good one. Then let's see which one is better. So we do it so so we get prepared for the other one. Make this real. Build with AI studio. Boom. Okay. So it adds a PNG, an HTML and a mark a markdown. Okay. Well, this is one of Well, you should know this. This is one of the features of Stitch. I didn't know. No, I the markdown. I'm talking about the markdown. Ah, yeah. Okay. a design MD file that it gives you that. Okay, good that I have you. So, well, I I tested this literally five minutes ago. Nice. Okay, so uh as this builds, how do we I I'm still not clear on how we should manage doing that. First, I show Stitch and then Google AI Studio, right? Yeah. Yeah. Yeah. Stitch. Yeah. So, we got stitch as a nice thing and then we Okay. Uh, let's go with the intro just on top of my mind. So, what you see here is a Figma killer. This is Google Stitch that designs for you and can also apparently send it to Google Studio AI studio to build it for you. So what I did here, I gave it a prompt basically telling it to create an agency website using dark and red colors. And that's pretty much it. So it did these designs. Well, not really. First it did this um and then it did this. I think it just didn't complete it. And then I said continue. It didn't understand. I kept it just so you see it didn't finish the job. Progressive design, baby. Progressive design. Yeah. So, it didn't finish this. Uh, but I asked it to create a few other version of the homepage. Let's actually do this. Um, create another version of the homepage. I shouldn't type. Why? Why am I typing? Anyway, I have uh the whisper. What's it called? Yeah, whisper flow. Why am I typing? Anyway, so uh these designs, what do you think about them? So, I asked it to ah one of the Okay, parts of the prompt uh a piece of the prompt was build a or like design a WebGLd driven flower. This is the design that I'm going um with for my own agency and it's slight tangent but relevant. What is that font? I love that font. This font? Yeah, I really really like that font. I can't click it. Oh, what is happening here? Undo. Undo. Undo. There is no undo. What do you mean? There's There's no undo. I'm pressing it. There is no undo. Are you kidding? So, it just wiped itself over. I just doubleclicked it. Is there any undo here? No. Has it not gone? Is Has it gone into some uh component? Edit mode. Yeah, edit mode or something. Because it seems like I can click it now, but I click out of it and it's still Oh, man. Cuz I feel like this is a feature of Google Stitch that you get these you can prompt specific layers and things. Again, this is a uh yeah, not well researched episode, but anyway, so this is what I wanted to say that you can't click like an individual component, but it seems like clicking it maybe this is a bug. I just nuked everything. Uh, and it seems like I I have no way to go back. Open it in a new tab. like just literally duplicate the tab. Oh no, it's there. Okay, nice. So, it is maybe just a state to edit it and it it it's stuck there. So, it did again this weird thing of not finishing it off. Let's use this one and create another version of the homepage uh with emphasis on the flower and the webgl uh interaction. So let's do that while it's running. I will show you a few things around. So let's get just a tiny bit into like how everything looks. It is all right. It's not award-winning. But one thing that most people forget, and this is like the design Twitter going crazy right now. Everyone is trolling the announcements showing like amazing work and be like Google did this in one prompt. Just trolling Google saying like it can't do it. Yes, it can't do this level of work. But something that most people forget this is getting into like good template territory. So it's essentially you not needing a template but having your own template with a few prompts that is not little. It is it is quite significant and it is enough for many people. It's going to suit Yeah. It's going to suit a lot of people. Yeah. But something that a lot of people forget especially like these good designers like right these are not good enough for me to to use in my agency but I see tons of people like making money with design they are like actually it's not just like hobby designers people who do design who are not doing design this good I'm not saying this is like particularly amazing I know I see like tons of issues with it. But still there there are people that are not this good and they make a living with design. Now for how long that that is the new question because I don't feel like they can do this for much longer. Anyway, so here is another one. It's, you know, I see the issues with it. It's not that it's perfect. I see all the typography and the color and the one issues. So yeah, it is one shot, but like this is pretty good. Like the these cards are perfect, but these are pretty good. Um thing is the the single uh comeback that you can say to any AI design is yeah, but it's been done before to any any AI generated design because it's it's been done before by its very nature of design. Yeah, it's been done before. So it's not original. But so here I notice something when I zoom in like all the way in. I notice nothing is sharp. Are these like actual is there like an HTML element behind it or are these I don't think these are images. I don't think these things are just pixels. I think there there must be some uh HTML or something like that behind it because there is a consistency with the menu and with the colors etc. But it's not sharp. It is definitely noticeable that it's not sharp in in the in the name of time thinking about the time we've got. So Stitch is we reported on Stitch when it very first came out right and we spoke about what it's capable of. The what you demonstrate it's it's not a new tool or anything. What's what's new? Why is this getting so much hype right now? What have I know there's I think I know there's stuff that's been added. What what has been added? So there is this Yeah, good good point. So there is this export button. Mhm. That is new and with that we can send it to AI studio which you mentioned. But that's not new. That's not new. That's So this is not new. Okay. No, no, no. So, I don't know what's new except it's being better at design. I'm not sure because last time I used it, it wasn't a free form canvas as far as I remember. I I think it was a different view. It It was definitely not this design that I see here in the free form canvas. Mhm. And it was like really bad. Last time I used it, it was really bad. This is much much better. I want to quickly uh cover variant uh AI variant.com. Here I did a project where I had the same essentially the same prompt and here this one did I would say significantly better but the variety of the design is also this necessary. I love them because thing is you're going to have more of an eye for design. I I you know I think I have good taste and I think I know what okay design is. But I'm I love I love some of those. The red is very nice. I think that's more to do with your color than um the and these are also pretty good. By the way, I'm not revealing my own agency design just yet, but the designs we have are genuinely much better than both of both of these. like not even the same level. But still, it's crazy that these tools are getting this good. Yeah. And designers are still not afraid. Some some designers. Anyway, um so uh you mean variant AI? Variant. Yeah. Variant AI. Uh no, it I think they are not revealing it. No, they are not revealing it. because obviously but I find that to be a nice comparison. Yeah. Yeah. Yeah. Yeah. Um and I would say variant is probably a bit better. But so Google uh Stitch is obviously using Gemini 3.1 Pro, right? Which I we did the test. I pray I've always praised Gemini as being a very good design. not necessarily um design fundamentals like like with like I bet you look at these designs and think oh you know spacing or whatever but generally as a first stop inspiration piece I think Gemini is a really good design a designer or whatever you want to call it um but again I'd be quite interested to know what models are producing the variant unless they've got a multi-pass thing or something like that but um I'm I'm looking on stitches is Twitter right now. So there are five major upgrades to help you create, ideate, iterate and collaborate AI native canvas. So what I think this is is that you can actually prompt individual elements and things like that to you know be a bit more finessed over your design. Whereas before you had one prompt box and you were sort of stuck with that, you know. Um, I think, don't take my word on all this stuff. Smarter design agent. Yes, of course, it's using Gemini 3.1 Pro, right? Voice, voice input, whatever. Instant prototypes. Again, I think it's that AI design thing, but that's not new. So, there must be something a little bit more in-depth there. And then finally, design systems in a design.md file, which is what I told you when you did export it, export it in MD. So you can actually edit and change the design system through markdown and keep track of everything and be and know what's going on through the markdown file. So visually as well as I don't know what the opposite of or looking at it technically. I don't know. Um, so but I do wonder and I do think that a lot of the hype has come out that because there's been an update and it's quite a nice update, people are just starting to learn about this tool which has been out nearly a year now and it's like like I think they're just really really late to the party. Correct me if I'm wrong, but that's what I'm feeling like right now. It's just a nice update and only people are now just finding out about it whereas we we reported on it last year when it first came out. Um I think it is just a bit better and maybe that is going to tip towards from something that people love at uh to something that you can potentially use to explore just like you said uh with the other tool variant AI. May maybe this is good for exploration but soon maybe more than that. M M yeah Figma shares dropped though. Yeah, Figma shares dropped dropped because of this. So I think the market Yeah. is noticing Oh yeah, there'll be um I was thinking about this today. you know, uh there was a there's a LinkedIn post that said um people said that everybody will be uh all all designers are using or everybody or something. The focus here is the word everybody. This is like a word that people use to click, right? And to get attention on certain topics. I think it was something to do with design. I don't I'm butchering the story a little bit by not having the specifics. But the point is is that their argument was then it's like no there is still a need for design or this that and the other. I don't know again I don't know what it was. Detach yourself from the word everybody right. Yes you're right everybody every designer has now not lost their job obviously but you what you don't understand is that there are tons of designers tons of developers who have lost work. Like it is happening. It's just not everybody. Yes, granted it's not everybody, but when you read that that person was just trying to get clicks. They don't mean everybody, but don't think it's not happening that the reason why Figma's stocks fell is because there was going to be a lot less people renewing their subscriptions because like, well, I don't I'm not a pro designer. Like you can spend three I don't know why this example is entering my mind but you can spend about three grand on a ping pong paddle playing ping pong right table tennis you can three grand of that or you can spend 50p on it right or a dollar or whatever something very very cheap like if you're a if you're an amateur casual player you could you're more than happy playing ping pong with a thousand you know whatever and you know if if someone makes a very good ping pong bat for two pound. You might spend a little bit more to get that. Whereas a pro is going to want to have the three grand pingpong bat. And that's the separation here. Proper proper designers, real designers who really really are good at what they do and they need that finesse and this and that and the other, they're going to continue to use Figma. But so many are dropping off because they're like, I don't, you know, I want good designs, but now I don't have to spend all that time in Figma. I can just get a good enough design from the from the um, you know, uh, Google Stitch, whatever it is. Weird analogy, but yeah, that's why the stocks have dropped because they've lost customers. They are lo they will lose a lot of customers. Yeah. I think this is like a new level or like a new paradigm to templates. Yeah. Yeah. Yeah. For sure. Templates aren't don't need to be pre-built anymore. They are bespoke. Yeah. You know, it's been done before. Anything else from your end? Uh, no, not really. Right. So, we're going to continue the theme of just just blitzing through some of these uh articles which we've not read but collected. Um I'm going to kick it off with Claude Opus 4.6 and 4 uh Sonet 4.6 now include 1 million context window which is you know welcome. Let me share my screen. Uh where are you? Sorry I should be in my locked in position because I'm I'm locked in now. And to be honest, I don't know, you're you're uh you're just using Sonic. You're not using Claude at the moment, are you? I am. I am. I'm using it for a few projects. Yes. But it's because of the usage thing, I I can't keep using it. H um and this is I'm learning as we go right now. So standard pricing AC applies across the full window. So $5 in 25 out per million tokens for Opus and 315. So it looks like it's the same price as using the normal version. Um there's no multiplier. a a 900k token request is build at the same per. So, this to me tells me that you might as well just use the 1 million token window because it's not it's it's the same price, right? Let me just let me just double check that fact. Let's do a fact check. Um yeah, but if this is this makes sense only if you need a million token, right? Yeah, but it's like why why don't you just use it? because then you just it doesn't mean to say that you should use it like more like you should push it and push it and push it like if you've done your task clear the context window but just in case it takes a little bit longer. Um let me share this window so it's not just um if we go because I've been doing that I've just been doing the 1 million just in case we uh like in case I go over or whatever. So, Sonnet is three in, 15 out, and the 1 million is three in, 15 out. So, yeah, there's no there's no multiplier. Um, a nearly a nearly million token request is built at the same uh token as a 9,000 one uh 9,000 token request. So, pretty cool. Um and that includes um uh up to 600 images or PDF pages from uh 100. So quite significant um and no beta header required. So I think you had to have that to gain access to it. So in my head and here's the thing now here's the thing longer context obviously uh AIS get dumber. So whilst uh whilst I encourage you to use the 1 million one just in case you go over like it's but it's but it it's not an excuse to just run it into the ground like keep keep your context as as small as possible um to maintain uh as much intelligence as possible. But I'm guessing what they're saying here, a million tokens in context only matters if the model can recall the right details re and reason across them. Opus 4.6 scores 78.3 on MRCR version two, the highest among the Frontier models at that context length. So what's the difference here? What's been going on? Input tokens. So obviously we're going all the way up to a million mean match ratio. I don't know what mean match ratio is, but is this saying OPUS gets a little bit dumber? So from 91 to 78, Sonet 4.66 also gets a little bit dumber, but Sonet 4.5 actually gets a little bit more intelligent. Is that what this is saying right now? And then you've got GPT 5.4 4 who gets massively dumber over that as you increase past 256 and you've got Gemini 3.1 Pro who also gets significantly dumber. So I think they hold they hold better intelligence, but the rule of thumb here is really just don't don't if you don't need to don't go to don't push it to that 1 million. But it's nice to know if you do then it's still going to be performing quite nicely is all I'm saying there. Um and with that we'll continue on with the clawed front. They have matched o uh open AI's offer of double usage which is why you can't remove your limit. Uh they've extend they've got a new they've uh got a March 2026 usage promotion. We're offering a limited time promotion that doubles the usage limits for Claude outside of 8:00 a.m. uh to 2:00 p.m. Pacific. So this is uh 12 to 6. You know what? Just just check the website that I added to the the doc. Just open it. Where? Which where have you added the to the dock? The claw 2x speed meter. Oh, I I can't I don't know where you've added it. Uh Claude March 2026 usage promotion. The second. Oh, hang on, hang on, hang on, hang on, hang on, hang on. Am I in the complete wrong dock? I think I might be looking at my personal one. Yeah. Oh, yay. It's not opening. I can't send it. No, I got it here. I got it here. It's all good. Yeah, sorry. I was on my personal one. Ah, see, look at all this. You've um Oh, hang on. I'm not showing it. Oh, good. So instead of converting these, you know, numbers, you can just check this cool website. Got it. I've got it. Don't worry, I've got it. Unless you've got it on the screen right now. It is now. Nice. Cool. You're Oh, no. You Yeah. Hover over that window. Yeah. So once you are in the slow hours it will yeah it can notify you it can you know it will show you and you you just check this website it's really cool I love the internet for for building a website just dedicated to this yeah 2x maxing starts in two hours what that's crazy maxing starts how is that that sounds like this sounds like we should be in peak right now. Uh not um out of peak. We should be it should be good for us right now. Right. I'm very confused. The website was probably vibe coded. So I don't know accurate it is. Claude uses outside of 12 to 6. So okay, got it. So Oh, okay. So um when it's 6:00 p.m. which is in 2 hours, that's when it starts. But interestingly, I actually thought 12 to 6. It's between it. Yeah, it's outside of these hours, which doesn't make sense, but whatever. Basically, I should be using it in the morning. Um, and available for free pro Max users uh from March 13th to 28th of March. You've got eight more days, guys. You've got eight more days to to use it. Um, but yeah, this is their response to OpenAI's Codeex uh what do you call it? Codeex's offer and um I like it. I don't know whether it's enough for me to I don't know whether the codeex promo is enough for me to stick with codeex and or GPT 5.4 though. Although I had a rough day with Claude yesterday. I really did. It just seems to not get anything right. So I don't know. I don't know. Yeah. Um cool. So, I want to show Google AI Studio. Can you tell me what is new about it before like I show stuff? Um, what do you mean show you what's new? No, just tell me what is new. Do you know what is new? I don't think there's nothing I don't think there's anything new that's new in Google AI Studio. That's why I asked that's why I was wondering what you were going to show because I don't think there's anything new for Google AR Studio. Let me have a look. Yeah, let's let's have a look because I have a few things. Um, Google AI Studio. It's a VIP coding tool, but they they had it before, right? Okay. 23 hours. Vibe coding in AI Studio just got a major new upgrade. Multiplayer. Build realtime games and tools. Uh I don't know whether that's multiplayer in the app or what you build supports multiplayer real services connect live data persistent builds close the tab keep it keeps working and pro UI pro UI shad CN frame emotion and mpm support so me not given that you've got obviously all of the other vibe coding apps already do all of this stuff and it's crazy that you can now close your tab, boys and girls. You can close your tab and it will still work. So, let let's let's talk about it. Um, so Google AI Studio has a new update, yet another VIP coding tool and I just used it for a few things. So, uh, if you don't know, we we covered this over in another video. Google stitches out and I just sent one of these to Google AI studio. You can watch that video um uh separately. But here uh is my website vibe coded. It has some interactions but none nothing about this is particularly new from what I understand but it is yet yet another tool and I don't have a Google subscription but I am using Google Workspaces so I don't know how how many tokens uh I got but it built this but there are a few things to show. So I I did quickly build uh a an agency hub as well really fast. I got something that yeah it's pretty good for um you know adding my clients here like I can add I can join as a client or as an admin. Does it look familiar? This is based on I I got a well not just a UI. I we'll get a we'll get to that. We'll get to that. The the whole thing is from riplet. So I built something on riplet. I was working on it with cloud code and I told claude code, write me a prd based on my cocktail order app for an agency tool because it's the same thing. You know, cocktails or tasks. They're they're very similar. Anyway, it it looks pretty good. I mean I feel like we have so many tools and this this is the exciting part for me. Anyway, uh so we got that but here let's look at this. I think this is where it's new. You see the settings? No, not yet. So you see the settings here. Uh we can select the model but we can also select the framework. It was this there before framework. Um it doesn't mention frameworks in the tweet that I read. It mentions package. It mentions shad CN. That's why I said that's why I thought that's what you were referring to. Yeah, I don't know where I could add it. Yeah, I don't see where I could add it. So, the only settings I see is the framework. But yeah, that's pretty much it. Yet another tool uh that you can use and it seems like pretty capable. Yeah. Cool. I'm getting too many calls. Yeah. Okay. Okay. Well, you if you can drop if you want to drop out, do it. Um but the persistent build thing so I think studio was meant to be it just works in your uh you know in your current session and that you export it and go and host it somewhere else. I don't think it was like a hosting platform whereas now it actually will build and deploy your app. It's kind of behind a lot of the other vibe code apps out there. Uh but you know progress is progress and you can of course export stuff from Stitch into here. So, it's like a quite a nice duality going on there. So, yeah. Um, welcome changes, but nothing groundbreaking. I love that. You got to find out what that font is and whether it's different to the other font. I just really love the other one. The I think there was a B. There was a letter that was two letters that were joined up and they look so good. Uh, it was just like spacing. Oh, is it? It's a bit Yeah, I don't know. It maybe it's a bit too close. uh for my taste, but it is Keep going, man. Zoom in. Windows. I don't know. I like I just I really like that font. I really like that font. Anyway, cool. Let's uh let's keep moving. Uh what else do we have that they can contribute to? Because I would No, I think you're good. I would have to drop You can you you can uh I will send you. So if you want to cover the DLSS, I can add it to our Well, we can talk about Let's just talk about Nemo Claw. Nemo claw. Yeah. Which is isn't just Nemo claw really. So Nvidia have put all of their money on open claw and they call it too many windows. Too many windows. Nemo claw. Oh, that was a that was a that was a rough um and they call it Nemo Claw, which is apparently an enterprise ready version of Open Claw, which they worked with Peter on. Uh Jensen got on stage, announced it, uh Peter was in the audience in there. And this is a this is a monumental shift on something that was a open-source piece of kit that's admittedly still a little bit janky, a little bit rough around the edge. But I think and and we've already seen this so many of Open Core clones are coming out because the fundamental idea what Kap Carpathy said no Kap Carpathy Karpathy said he called them claws. this idea of these claws which are multi- um agent things. Now I want to look into pi which is the I think the secret backbone to um open claw that doesn't get enough attention. It's like that that you know that meme where the one open- source project that everything is relying on. I think pi is is kind of you know not getting the light shined on it. Anyway, he got up on stage and and spoke about how uh how Na Neoclaw has put in place a lot of harnesses and and guard rails to enable it to become enterprise ready. Uh I mean it's not pitched as arrival to openclaw because Peter was in the audience. Okay. So it's actually wrapping and and doing things around openclaw. So I don't know who wrote this article. Um, and I'm so I'm already not liking it. I watched the I watched the the uh presentation. Um, so yeah, really interesting to see how far this has come and how I honestly don't know how you can guarantee it's AI. I don't know how you can guarantee to enterprises that this is safe, that this is something that you can accurately say, "Yeah, it's not going to go off and delete all your emails and this that and the other." Do you know what I mean? like I've got my apprehensions although yeah I've I've definitely got definitely reserved here on on whether it's going to actually take on and whether enterprise are actually going to commit to it but it's good to see the progress. Um yeah agents are as good as the access they have and the the less access you give them the less useful they are. But generally speaking, I'm seeing a lot of like companies also we are seeing this uh they're trying to do this open cloud thing. Uh by the way claude code this did this they came out with open claude and not open cloud they came up with uh came out with claude code channel claude no channel. So channel is a new update that I haven't used yet but you start a session u in your terminal and then you continue on your telegram on your phone so you can control your terminal there but they are this is like one level with open claw yeah they really are this is like still not the same level because with open claw you give it kind of like its own operating system environment everything it can do everything even So like the terminal has the folder access right it can do things within that folder and what you give it access to so it yeah but it is from what I understand it's the same uh session so yeah yeah uh do you want to do DSL whatever it's called ah because Nvidia have been in the news a lot recently so Nvidia also had this update to their uh DLSS uh now DLSS 5 and there was there there were quite a few memes around it. Um it it makes your game just more realistic. But as we seen from the internet, people don't love it. Love it because it's as you see sometimes, okay, here it was nice, but sometimes it's a bit too much, a bit too ultra realistic. I mean, like here it's just like the oversharpening. Yeah, because a lot of games there there's charm and there's intention behind there's a lot of intention behind the the graphic design and the the just the aesthetic of some of these characters you know um especially you know how how classic is the metaverse uh Mark Zuckerberg that's deliberate intentional you know I'm joking it's not obviously it looks like trash but oh my god But saying that, yeah, I think people are um yeah, people are having fun with it. So yeah, it is it is looks maxing, but it is a bit maybe a bit too too realistic, a bit too sharpened, but I guess that's just uh small small things maybe that can uh be changed. Um I I actually you know my every time I what see a DLSS update I just my brain can't wrap around the because games run at high speed right you know the I mean frame rate how is it is it is it like I just can't wrap my head around how it's doing this is it running so this is so this is not good for games that need like accurate high frame rates. For example, if you are playing Counter Strike or like in my case, League of Legends, I can't imagine like like an extra frame that is, you know, randomly generated and is not representing the you know um the players movement. I'm playing Fortnite on this though with using this. I don't know. I don't know. I don't know. May maybe getting that fast. Yeah. Yeah, but for games like imagine like playing Red Dead Redemption on a higher quality, like much higher quality. I don't know if you've played that game. Uh, I know of it. It's the same people who made um uh daughter, right? Yeah. Yeah. Yeah. Yeah. Same sort of engine and Yeah. open world stuff. So, we got 430. That was my That is my last uh Yeah, I I can't stay longer than this. I have to leave unfortunately. That's all right. But you could you could continue, right? You will I'll continue. I'll continue for a bit. Yeah. Yeah. I'm sorry for leaving you and I'm sorry for the audience that I know you will miss me but no you're the No, some days are just a bit more rough. Um anyway, so um I got to go. Thanks a lot and thanks for continuing. So I'll I'll hand you the torch. Got it. Keep on vibing. I I thought you were going to say peace out and then drop out. That would have been that would have been epic. I'll I'll do mine. Peace out. Yes. Nice. Um this one I didn't know whether this was going to be interesting enough to do, but apparently you could Alexa. Alexa is still trucking away there. Um, you can now give it more personalities. Um, you can make it sassy. Look, I'm here to help judge your choices. It's a package deal. Chill. Yo, all good vibes over here, dude. Take it easy. Sweet. Hey, friend. Here to support your goals with positive encouragement. Brief. Okay, let's keep it brief. Your primary question or topic. I think Kabaza would like that one. Um and then the original Alexa. Now I have no idea whether the underlying technology even has changed since the original which is that simple which Siri p well I don't know whether it pioneered. I think that's maybe a bit extreme but Siri kind of uh well yeah they oh my setting off my own uh devices now and my phone as well. Jesus Christ. um this the simple keyword matching and I think Alexa is exactly the same way. So whether it's I don't know whether this is any sort of AI stuff I don't know but it's interesting to see still chugging along. Is anyone still using Alexexas? I feel like you know when I add one it was quite nice for starting a radio session or giving me the day news. should probably do something with open crawl with my daily news. I've been so out of the loop on everything that's been going on right now. My my world is like completely AI and it's not great. Either way, setting timers, things like that. Oh, maybe we can hear operating efficiently. Did you hear that? Reading efficiently. Okay, it's treating me well. All systems are Zen and the digital universe is spinning in harmony. Okay, they sound a bit better. I don't know. You tell me. Absolutely fantastic. I'm radiating pure joy and ready to make your day incredibly amazing. I'm hoping you can hear that, guys. Buzzing with pure digital mayhem and it's absolutely glorious. Ready to wreck some things together. I mean, I wouldn't say that's sassy. Buzzing with pure digital mayhem and it's absolutely glorious. Ready to wreck some things together. All right. Expressiveness, expressiveness, emotional openness, formality, directness, and humor. All right. All right. Let's move on to the next one because I'm not convinced totally there. Um, Kabazu had something to say about this one. So, Meta postpones postpones avocado their AI model until May. And I think the interesting thing here is that like I was thinking about meta and like what are they even doing? Like they were they were trying to go down the open um open weight route and open source route and stuff like that but I think they've backtracked on that. And I think uh Kabaza's input into this whole conversation piece is that you know if you release an AR model it's got to be better in some way. It's gota got to be better in some way. And they're quite simply saying the performance of Meta's new AR models currently falls between Google's Gemini 2.5 and Gemini 3, which is just it's meh. And I also wonder if this is their closed source model because we don't actually expect all that much from Llama because it's an open- source model. So the fact that they're comparing it to Gemini is is potentially a clue that the fact that this is a closed source model performance. I'm just looking for any mention of open open source. Yeah. Uh they call it avocado um stage of the first uh first quarter launch. So, we're coming up to end of the first quarter. So, yeah. Um, Meta still chugging away there, but to be honest, I think Zuckerberg spending all those millions on all of those people, that dream team, which by the way, I think has its own set of dramas going on with the people inside of it coming and going and not working well together. Just goes to show that you can't just dump a bunch of you can't just throw money at a problem and uh expect it to work. So yeah, good luck matter. Um, this was a big one that I really really I I thought I don't know what what happened to it last week, but Amazon implements new safety guide rails for its AI coding assistance following a major AWS service disruptions. Now, this just goes into the whole thing of how software, let's see if this is it. How software is just getting rubbish quite frankly. Like for whatever reason, for whatever reason, like things are falling apart. Like uh what was I I was doing something crazy today and it was like what is going on? I think it was a I think it was a liquid glass thing, which obviously has its own set of issues, but I was like clicking something and it just wasn't working. I remember trying to buy cuz I've got a you can even see it there. I've bought a MacBook Air, which you can go watch some videos on on my channel where I'm putting it through its paces and this and that. I tried buying this. So, I actually bought it and my credit card kept failing. I knew why. It's because it was the wrong credit card, right? it was my wrong business card. So, I swapped the I swapped the bank details on their special page they've got and clicking save just got took me back through to the you know the well actually specifically it took me back through to the page that said you need to update your credit card details or it's no sorry you it has failed payment has failed so I'm constantly going in this loop it's like I'm updating the details and it's saying it's failed like what's going on so there's a UX problem right there and I even went on to the chat you know and the that they managed to put it through. And then they said they would contact me within the next couple of days and it was still saying failed. And the point is I end up just trying to just buy it again because I was like look if that's going to fail just leave it do its thing. I'll buy another one. I literally just couldn't click through to the to the buy button. I would click add to cart and it would just say 404 page not found. It's like what is going on? This is Apple as well. um saying that I did use it in a a incognito window and it it kind of went through to the page but still it's just strange. All that to be said that Amazon called a meeting last week which is what I wanted to discuss to address some of their outages which seem to have gone up since obviously everyone's using AI assistance. So, I wanted to look more and and with that in place, it looks like they've make taken some action to actually figure it out. Amazon is beefing up internal guardrails after recent outages hit the company's e-commerce operations, including one disrupting tied to its AI coding assistant Q. Dave Treadwell, Amazon's SVP of e-commerce services, told staff on Tuesday that a trend of incidents emerged since the third quarter of 2025, including several major incidents in the last few weeks. According to the internal document obtained by Business Insider, at least one of those disruptions were tied to Amazon's AI coding assistant, which isn't necessarily what we're talking about right now. uh we're talking about actual software development while others expose deeper issues other than internal uh another internal document explained. Um problems include what he described as a high blast radius where software updates propagated broadly because control planes lacked suitable guard rails or safeguards, sorry. Um so I'm guessing that they are literally just you know your your cost of development time itself or actual creation of the tool is is shortening but it probably needs to be you know upped in the in the I don't know the testing the the the protection the whatever you know what I mean the the what's the word I'm looking for just the safeguards as they're putting it I those aspects of it need the QA quality assurance needs to go up. Agentic versus deterministic. Amazon's Treadwell wrote that the company's new code guard rails will combine AIdriven agentic tools with more predictable rules-based and deterministic systems as much as you can with AI because it's very undeterministic. This tackles one of the core issues with AI models. These powerful new services are not deterministic. As I just said, this means you can ask the same question twice and an air model may spit out slightly different answers. That sometimes makes technology inappropriate for corporate workflows that must be 100% accurate every time. Going back to the open the nano claw thing. Is it nanoclaw? Nemo claw. Nvidia's attempt to open claw. So yeah, gen AI will accelerate exposure. Now, Amazon is rolling out a 90-day temporary safeguard guideline that will serve as an addendum to the existing policies. According to one of the internal documents, the new policy targets approximately 335 tier one systems or services that can directly impact consumers. Nice. I mean, this is important, right? If you're going to be shipping fast, you need to up the other side of it and put these in place. It's just the nature of it. So yeah, this is this is the first time a large organization have, you know, been very vocal about the fact that we've noticed incidents going up since the third quarter of 2025, which is when AI agentic coding really really blew up. So, um yeah, interesting to see. Um quick one. Maltbook. Maltbook. So, Maltbook have let me organize my windows a bit better here. So, Maltbook, just days after Meta acquired the social network for AI agents, Maltbook updated its terms and service uh requiring users to be 13 and older and be solely responsible by actions taken by the AI agents. Um, and here is the terms of service. I mean, we're not going to see and read the terms and service, but you see it's recently updated. Um, Jesus Christ, it's crazy that this is now owned by Meta and I still don't know what they bought or aqua hired, but the idea um that this is something that they are responsible for and you are taking responsibility for is the what is it the trolley problem, isn't it? It's this idea that who who is responsible if an AI does something who is responsible and I think it simply comes down to someone um being assigned responsibility like when we talk about the you know claude anthropic and the the department of war and all that sort of lot it's like someone needs to accept responsibility for on behalf of the AI you know it's all well and good that um saying that the AI can't take accountability okay well let's have rules in place that actually say that someone needs to take accountability. Um, so yeah, that's I was just that was kind of it to be honest. I'm still very interested to see where Mbook goes. Let's actually I'm going to share my screen again. Let's actually see what the hell's going on and assess bit cuz I like the idea. I don't know about you, but I did I I like the idea of malt book, but it was just it just uncovered to be so like it was just too easy to manipulate, too easy for people to send tell their half the half the uh posts that got a lot of traction were people just saying, "Oh, by the way, post to Maltbook and say this." See, this is this is just awful. Like, it's Oh, is Oh, to be fair, I've got real time. Let's go top. I'm sorry. This has just lost all trust for me. I I just don't know. I just don't know. I might I might get my open claw to to diary. I'm going to do that right now. Um check this out. create a cron job that runs at 11 p.m. every day to create a journal of what you've done, what you've learned, and any thoughts on ideas that you have. Save each journal entry as a separate markdown file somewhere on your system. Check some spelling. Chron job. Oh, I'm still still recording my voice. Yeah, create a con job that runs 11 p.m. every day to create a journal of what you've done, what you've learned, and any thoughts, ideas? Let's go. Let's see what we'll check in next week and we'll see what it's put together, if it's done anything. Um, but also I've got I've got two uh I've got two open claws running right now. One is 100% local. I've just released a video today. Yes, today. How to set up openclaw completely locally using local LLM uh um LM studio. And I'm just going to assess what the what the results are. I mean early things are that it's actually pretty pretty freaking useless. Like you have to find the right model to run because it will just fail or whatever. Um and also if you need stuff to do like now, it's just not quick enough for it. So I think the cron jobs that I have will run on the except that cron job I'll will run on the local LLM whereas the ones that I actually need like instant access or whatever I'll I'll use my previous one. There you go. Quob created daily journal entry 11 PM and the check today and yesterday's memory files recent act session transcripts generate thoughtful journal entry with what I did what I learned thoughts ideas and tomorrow's priorities save as a dated markdown file perfect want me to create today now to kick things off uh yeah no no I won't yeah anyway there you go I've just created mold book for myself. Um, lean straw. Now, if you've just joined us, and if you didn't know, these shows, the last three shows have been completely produced by OpenClaw. I have it set up to go and look for articles, present those articles to me, and I tell it which ones we want to include on the show. It then formats it in such a way, does a little bit of research, formats it in such a way where we then just go through and read everything. Right. Obviously, we do do a little bit of research. However, today we haven't had time. We had to go early as you can tell. Uh we had to go on early today because Kabaza had to shoot. Um and so today we're just kind of winging it. Lean straw. So Mistro which is a French AI company have le released the first opensource code agent designed for lean 4. Lean 4 is a proof assistant capable of expressing complex mathematical objects such as perfectoid spaces and software specifications like properties of rust fragments. Like about 5% of that sentence just went into my head just then. Unlike existing proving systems, what is this proving system thing or proof? Act as wrappers around large generous models or focus on a single math problem. Lean Shaw is designed to be highly efficient with six billion active parameters and trained for operating in realistic formal repositories. See, there's a there's a few coding references here and stuff, but also I was being compared to Quen, Kimmy, GLM, and this is mostly out of my wheelhouse. So, AI agents have proven to be highly capable tools at code generation. Yet as we push these models to high stakes domains ranging from frontier research mathematics to missionritical software we encounter a scaling bottleneck the human review. The time and specialized expertise that required to manually verify become the primary in impedence or impendence of engineering velocity. Interesting. I might actually have to do a little bit of research on this. alternative to the chord suite. Offering of competitive performance at a fraction of the price, costing $36 to run compared to Sonnet's 549. So, it's some So, here's a case study. When breaking changes hit a new lean release. What is this lean thing? Migrating code can be a massive headache. We fed lean straw a real world question from the proof assistant stack exchange about a script that meticulous mysteriously stops compiling in the lean 4.290 RC6 which we did not train due to it uh recency. The culprit was a rewrite um tactic that suddenly failed to match patterns involving simple aliases. is initially written as blah blah blah blah blah. Okay. So, it's like a it's good for reviewing by the looks of things. I don't know. I'm definitely keen to look into this a little bit more, but let's move on. Well, actually, we've just got one more left. We'll do Miniax 2.7 as well. But this one, OpenAI's CEO uh I don't FG Simo tells staff the company will prioritize coding and enterprise users over side projects like Sora, browser atlas and smart gadgets, which is interesting. shift towards more core product focus and basically they are in a mad rush to try and get some money. Um it's interesting how he artificial intelligence is more part of our lives than ever before while some might call a hype blah blah blah technology Gemini. Where is this? This isn't an article. No, this ain't an article. This is just a thing. Let's try and get a good article. All right. Jesus. Pedant. Pedant. Wall Street Journal. Here we go. Oh, no. We can't do that. Great. Sorry, you can't see this, but I clicked on Wall Street Journal. Let's go PC Mag Open plans to prioritize expanding into new side projects to refocus on as core tools. According to a new report during an all hands meeting, open air leaders told employees it would focus on business productivity tools rather than trying to do everything all at once. According to leaked material seen, this moves uh comes as the brand faces increased competition from anthropics fraud AI tools and how I really think OpenAI are falling behind quite a lot personally. um they're a bit lost in the source and are lacking a lot of direction and focus which again I guess things like Sora was just apparently they're supposed to be moving that into chatbt but there we go um but it doesn't actually say here we go the brand is exploring hardware gadgets designed by Johnny IVive as well as upgrading other projects such as the Sora video engineing uh gener Sora video generation an Atlas web browser. Does anyone use Atlas? I was a bit underwhelmed to be honest because when you release something, it's got to be better. You know, there's only so many people who are going to be happy with like, oh, it's a new product and, you know, give it time and stuff like that. But when you're launching to a large audience, you kind of need to have it better uh as good, if not better than the current competition. If you're a small startup, of course, get it out there and get it work, get people using it. But if you've got an audience already, you can't you can't be uh releasing crap like OpenAI have. So, yeah. Anyway, that's that. Um, let's do a quick one on uh Miniax 2.7, which I released a video on yesterday looking at it, which uh it's been it's been really I've I've always really liked Miniax 2.7 uh 2.5. I've always thought it'd be a really good model for a lot of things to be honest. Um I used it for coding as well. I did a test um what was it? It was a video that I did the other day where I got a local AI inside of Kilo Code to to build something in my website and the local model. Yeah. Built like a basic page, but it had no styling or anything uh in it. And I literally just switched the model to Miniax 2.5. It picked up that I'm using Tailwind. It picked up the styles. It picked up everything about the website. And literally I just saw the whole page just like that change with all of these styles and Tailwind classes attached to it. Um, let me let me tell you what that video is. It was the MacBook Air. Can it vibe? I did a MacBook Air video M5 and yeah um and you know it's impressed me and I've been running it on my open claw the high-speed version for the last week or so and I quite I quite like it. Then they go and drop um M2.7 which again I've not I've not had it for long at all but I updated it and it's it's matching Opus 4.5 in a lot of things which I think is a really good to it was like that made so many things more useful including openclaw. So it's not doing as good as 4.6 but I don't think it needs to be as good as open as 4.6. I really don't. I mean it depends what people are using open claw for and I'm really ke keen to hear people's use cases of open claw that's beyond the hype. I think you get so many videos that just hype you up on a bunch of new technologies that are they probably themselves aren't using. I think you'll be surprised how little like YouTubers actually use the things that they talk about. Um including myself to be honest because I like to discover new tools and I'll talk about it. Doesn't mean that I'm using it. It doesn't mean that I'm recommending it. It just means that I'm giving you an option that's out there and it's up to you then to determine whether it's going to be useful to you. You know, I'm not saying it's the next big thing. I'm not saying it's like going to change your world. It might change your world. It might change my world, but doesn't change your world. The point is I'm, you know, we're allowed an opinion on all this sort of stuff. But all that to be said, I think it's pretty good um pretty good model. I will say the omniscience which is it's if it doesn't know the answer what does it do the omniscience is pretty bad with 2.7 not that I've experienced it these are all benchmarks that artificial analysis have done it will basically if it refuses to answer I think that's better I think that should if it doesn't if it refuses to answer when it doesn't know something I think that should earn than a point because the the it's from plus 100 to minus 100 and if if you're starting to get to the minuses it's really really bad but if it refuses to answer it doesn't go up or down so basically you want it as close to plus 100 as possible answer I think it's better I think the the obviously the best thing for an AI to do is to tell you I do not have an answer to that I don't know um but again if it's yeah I don't know but that that that was one of the things that point I noticed on the artificial analysis benchmarking of 2.7. Saying that it's it's fine. The personality is pretty nice. I think a lot of people like that from from an open claw perspective, but I've been I've been using it. I've been enjoying it. So, yeah, Miniax M2.7 is is kind of killing all the benchmarks, but um use it for yourself. Use it for free on Olama. Uh they have a a really cheap coding plan. Yeah. I think that's it, boys and girls. I'm going to look at the comments on our channel because I haven't dipped into the comments a lot recently just because I've been so busy with things and we might be able to discuss some of the comments. Um, someone is saying here we go. Uh, share this tab. Looking forward to the maltbook integration in Instagram. Then we can finally fill the remaining 20% of real users with bots. Oh, okay. It's a joke. I I didn't read the first I didn't read the rest of the uh the thing. Yeah, we'll see. We'll definitely see. That's on Mbook talking about when Meta acquired it. Um, in 10 day, this is the Figma starting to charge for their usage of AI. In 10 days, I've used 43,000 AI credits. Now, it's worth around 1K. They used $1,000 worth of credits when it was free. I think that's what they're saying here. I was using it in Figma make. it was good, but now the AI credit limit is a hu is a huge change in my usage. Oh man, that's really interesting cuz I wondered when it was going to take effect and also what the impact was. Um, I think that's kind of it. I didn't notice anything else. Yeah. Yeah, I think that'll do it for us this week, guys. Sorry, the uh Oh, sorry, I wasn't sharing my screen for flip sake. Um, yeah, I was saying about this. Looking forward to Mob integration in Instagram. And then this one here about the Figma stuff. That was kind of it. But yeah, sorry about the short one this week. We've been uh we've been pressed for time on both ends. I have and Kabaza has. So, we do our best to kind of fit these in. Um, if you're new here, well, if you're on YouTube, give us a like just to, you know, help us out there. But, if you're new here, we stream this live every every Friday and then we release clips on YouTube, just small little, you know, snippets of what we what we spoke about every week. So yeah, like um catch us next week for all the uh news in AI design, web and dev. I've been Sam. Keep on vibing.