#42 Anthropic Marketplace, Meta Acquires Moltbook, Code Review, autoresearch, Cloudflare Crawl

for the last 10 minutes. [laughter] I wonder what the Hello everyone. Hello everyone. Hello beautiful people. I could So I could have told you I could have told you the secret. The secret, but yeah, true. Never mind. Never mind. Well, it it will be Anyway, [snorts] cool. So, uh, not a super crazy week and I'm waiting for people to show up. I don't know if something is broken because I see the the destinations logos being not like not being loaded and showing that broken image thing and the view count is zero. No, this is true. It's good. My I can see it my end. They're ticked. They're streaming my end. I've even got it streaming on YouTube now. So that number one Oh, three. Hello. Hello. Hello. Okay. Welcome everybody. Uh join us on on YouTube. If you are watching from my Twitter, I see. And your Twitter, just join us on YouTube. It's just much better. I promise we will have cookies for you there. [laughter] What kind of cookies? We won't say, but chocolate cookies. I was thinking more browser, but you know, browser cookies. No, we live in Germany. We don't want [laughter] GDPR. So, something crazy happened, not crazy, but something funny happened to me on Twitter today. I was uh engaging with some posts and um uh me and some guy were just chatting about something, semi-arguing, but chatting, whatever. And then some guy comes out of nowhere and raises an issue like a a raises an issue but completely irrelevant to the conversation that this me and this guy are having. We're talking about the new Nano that the the MacBook Nano and they're they're like they said some guy said something along the lines of like no one's using USB anymore. Something like that. And I'm like and I I was responded this is this is just the way that you you like to work doesn't mean other people whatever. And then someone comes in with like data speeds of USB and all I said to him was like that's irrelevant to this conversation right now. Like it's it's whatever. Yeah. And then they they responded to me and said it's totally relevant blah blah blah blah blah blah. So I go to like reply or whatever and I can't because they've blocked me. So, they've come in, raised a raised an issue. I've I've said, "I don't want to engage with this because it's irrelevant." I wasn't mean. I I wasn't like, "Fuck off" or anything like that. I was just like, "This is just irrelevant to what we're talking about." And then they just like say something quite uh aggressive back and then block me as if it's like running away. Yeah. God, like I I I watched also a video from Primagen about people on Twitter and he mentions like a lot of the replies that you get are probably just people telling their open just replies. I don't know. Yeah. Yeah. Yeah. What you what you described sounds like somebody Yeah. like that that would tell its bots. It just be [ __ ] annoying. But finally, you mentioned something. Let let's let let me show you something. What about this idea? A Mac Nano, not Mac Mini, Mac Nano. Mac, similar to Oh, sorry. Mac Yeah. Okay, let let's start again. What about this idea? [laughter] So, how about this idea? A Mac Neo powered by A19 Pro. So, this is similar to the MacBook, the new MacBook uh Mac Neo. And yeah, I really like this idea, but not $3.99. What? Why not for $2.99? There is like there is nothing in it. There is just a chip and you know that the things around it. It's going to probably cost Apple 50 bucks. I don't know. I don't know actually. But I think they could do it for $2.99. Well, couple of things. First of all, what if I was to tell you this already exists? Oh. Uh yeah. Well, not with Mac. But what you say is uh No, no, no, no. With Mac with Mac. So, okay. Before the M1 was released, they released a developer unit of a Mac Mini. Oh, yeah. A They did probably probably 18 or 16 or something like that. an A series chip. Okay, which and then shortly afterwards then the M1 came out, right? And that is just a rebranded A series chip. They just rebranded it to from a marketing standpoint. It's the it's the same chip that was in a iPhone pretty much. So it realistically this is just an M1, probably more an M2, maybe an M2 Mac, but like so that's that. Um yeah secondly and people have been comparing the Mac the MacBook Neo to the 12in laptop but they are vastly different products. They are aimed at completely different audiences. The Mac Neo is for kids to introduce them to Macs. you know, they don't do a lot on their whatever. Desktop arguably is um geared more towards quite naturally more towards power users because it's, you know, it needs extra juice and whatever. So, I don't know. I don't know. It looks nice though. Looks fun. But I I feel like this is this can be really fun. Like you have this for you could open. Yeah. Yeah. for open claw like for some light work that you can have for offices. This this could work for many offices. I feel like people might you know stick it to the back of a monitor and you you can have something proper for very cheap very very mass offices like because like yeah most people just buy MacBook Airs. I do think a lot of people do prefer laptops. is a reason why the MacBook Neo exists because the MacBook Mini Mac Mini at one point was the gateway drug to Apple but that was in the days where desktop computers were still quite popular but now the laptop has just taken over now as a well yeah allin-one um it's cool I've I've got a MacBook Air coming on Monday completely juiced out I'm not going to keep it I'm just I'm sending it Uh, but I'm going to run some Vibe Code local LLM checks on it. So, I'll have a video on that coming out. Oh, that that's cool. Well, yeah, this is very easy to recommend. I think it is still going to be the best to recommend, but the MacBook Neo is just like no-brainer. What is this? 50 years of thinking different. I was going to go apparently someone did a research or something that it's like it looks AI written but it's actually something that was written probably about 50 years ago or something like that but um I haven't read it yet but I've I've been maybe we should do it. Do you want to read it? Let's do it. Let's go for it. Two M dashes three four five. Do you want to increase the font? Yeah. Yeah. True. Yeah. It's uh No, you should you should read it. Read it like Johnny I with a British accent and similar years ago in a small garage a big idea was born. Is that that sounds like David Atenburgh? [laughter] Apple was founded on a simple notion that technology should be personal and that belief radical at the time changed everything. [laughter] [gasps] Um, what is it talking about though? Marks 50 years of Apple. I think they're just celebrating 50 years of Apple Watch services we use every day. The App Store, Apple Music. We've spent five decades rethinking what's possible and putting powerful tools into people's hands. Through every breakthrough, one idea has guided us that the world has moved forward by people who think different. They're banging that drum again, aren't they? Yeah. Um I think they're just I think they're just uh celebrating 50 years with a with a statement. Yeah. Uh it's all about their brand and selling the idea of empowering creative and power users because they are talking about that's because progress always begins with someone an inventor or scientist a student or storyteller who imagines a better way a new idea a different path. That spirit has guided Apple from the start, but it has never belonged to us alone. It doesn't sound like AI origin. To me, it is I haven't read I I read something I it was a LinkedIn post. Here's an interesting thing that I think you you'll respect, right? Because I was thinking about this today and it has something to do with Apple, so it's slightly related. So, the um the advert for the MacBook Neo. So, it's completely on topic. Um, is this, you know, the MacBook spinning round and this and that and this and that. Uh, and the person who produced the advert came forward and said it's all stop motion. And it goes back to remember the Apple glass the Apple TV logo that made with practical effects and they have actually made it right. Yeah. It's I I was thinking about this idea that if you give people the opportunity to play and express themselves like yes they could have done all this in 3D and this and that but those people working in that company you're sucking the soul out of them if you reject their um their creativity and their desire to do something. So whilst things take longer such as stop motion, you're playing the long game with that creative by allowing them to be excited by their work. So it's a weird thing and it and we'll explain this this episode's going to go a little bit differently for those uh tuned in at the moment. We're going to do something, you know, slightly different. We'll explain in a second. But the thing that I said to you, Kabaza, before the call, which is you can go the cheap fast route or you can make someone work a little bit harder. I mean, this doesn't quite translate, but so bear with me. But it's this idea of like um the you're playing the longer game. the the the most immediate route might not be as appealing such as you know some designer coming to you and be like I want to film this in stop motion and it's going to cost you all this money setup lighting blah blah blah but you know what you're in you're investing in the long term as opposed to that quick thing and I'll just finish on this other thing because if you watch Prime's latest video that he just released about six hours ago he actually talks about PewDiePie talks about his last video that PewDiePie made about the AI agents and that he one of the biggest I want to watch it. I haven't. So, one of the biggest and we thing is Primagin is saying the exact same things we said on this show three four weeks ago when we first spoke about PewDiePie and a followup that happened in my link on my LinkedIn channel where people just were throwing excuses under the bus as to why PewDiePie's got all this money, this time and we we we re rebuttled that. I don't know whether you say that. point is is that we're in an age where people are um crying out for quick solutions. They're not they're falling out of love with the idea of learn learning and struggling and they're not and to Prime's and and PewDiePie's point, they're not embracing failure. We're learning the we're losing the ability to embrace failure and see that as actually a fundamental learning tool. I always posit the idea of the the matrix where you just give people the ability to know kung fu. I I it would be really interesting if we ever invent that whether actually it would have any effect because the process of failure and repetitive nature might be just innately like the thing that needs to happen for us to even learn anything. We might need that. We might need failure. So anyway, that was a rant. Ju just to expand on that a little bit. I fully agree and it's not just about learning a tool or something to do with work or whatever. It is about everything. Literally everything you do in life. A lot of people, and you see this maybe today more than ever, they throw out words without understanding them. They they use words. I'm talking about like politics. I'm not naming anyone or anything special, but people do use words that have like specific meanings and they don't mean those or they don't even know. They just throw it out because because they heard it from someone because they think it's cool. I don't know. But it it is so [snorts] like it is so much it's full of pleasure to just sit with it and spend the time with it and learn these things like learn the basics like what does this word means like it was a while ago uh I started with this journey with a friend of mine who is studying philosophy and like one of the interesting things that I'm still to this day I'm like mentioning it around like the Word reasoning like a lot of people use reasoning in the field that we are in. It's you know everywhere like any app that you use you see thinking and you see reasoning and what does it mean like many people don't know what it actually means and where it comes from. So back to your point it is yes it is very important to go through that pain and also it is very enjoyable to to learn these like building blocks. So no matter if you do AI or whatever, if you hit a point, it's about curiosity, right? Hit a point where you're like, this thing, just pause this thing that I see on the screen or this thing that I just heard or I or I see, what does it actually mean? And you can that's where you can go deeper. That's where you can start learning. And it's Yeah. What's missing? Yeah. watch Prime's video because he does touch on this idea that you know new generations are not um are not it's not their fault at all but it's they're not encountering pain enough and I don't mean physical pain I mean I mean like struggle they're not they're not encountering struggle and so when they encounter it they don't know how to deal with it and they lash out and I think that explains a lot of people's attitudes on Twitter and whatever because the moment they see some sort of friction some sort of backlash some something that doesn't sit within their mental model, they lash out. They get defensive and, you know, they they they potentially lose out on the opportunity of learning and growing and being a better person because they're just it's just a short side thing. But we are too much Tik Tokrained and to add to that is we are too much uh one promptrained. We are like expecting you know wonderful things with one prompt to do everything all at once you know uh yeah well check out that video check out prim's video because the thing is I was watching that thinking mate we we already spoke about all this so yeah I don't think you'll be surprised because yeah we were we were we were praising um PewDiePie and and his ability just to get stuck in and and fail and succeed and curiosity like like all that kind of stuff. So yeah. Um should we should we get cracking and then and then you know Yeah. viewers no doubt we we don't need to finish this but I I just want to say to our viewers that we have if you are interested in us reading through something anything this is something we discussed in the past maybe we can do this. I think it it can be fun at times to sit through not a marketing [laughter] uh letter from Tim Cook, but you know uh sometimes there are interesting articles. You can learn a thing or two from it and we could potentially introduce this in the show to sit through those uh and just read them together. Okay. So, with that being said, this week Claude introduced their own marketplace, Claude Flare. Funnily enough, they they banned robots a while ago. Why? To presumably sell our data to them for money. Now, uh we'll talk about that. [laughter] Uh, Karpathy made an AI that does research and gets better. That's kind of crazy. Uh, and from what I heard, it's very light. We'll we'll talk about that, too. And we'll have a few other things to talk about. All of this in today's episode. This is Command AI. I'm Kabarza. I'm Sam. And this is a week in AI design and dev. Cool. So, should I start with the this one to get it out of the way? Yeah. Yeah. Yeah. All right. Well, should we explain that we're going to try and blitz through these articles in an attempt? Let's do it. Yeah. Yeah. Let's do it fast. We're going to try and go through the articles not as not as quickly as humanly possible, but we have we have had an interesting conversation before the stream about those just wanting to get started with AI and how they should think about things and what tools they should reach for and and our thoughts on the matter. So, it' be be interesting if we can recapture a lot of what we spoke about before this the the show started, but I I'll yet to be seen, but yeah. Yeah. I want to the end. Yeah, [snorts] I want to read a little bit of this. Let let me actually uh prepare the marketing uh site that [snorts] the marketing page of that site as well. Uh what's it called? Video.ai or something. Give me a second. That that naming it's so confusing. What's that? Uh, we'll we'll see about it. I'll I'll explain. [clears throat] All right. So, Web Flow just acquired VO saw.ai to deploy AI agents that work within your brand, not outside it. So it seems like this is a tool to generate assets such as any visual image, text, videos at scale for your brand with AI. So we are excited to announce that Web Flow has acquired videos. So, a multimodel AI asset generation startup built to help marketing teams create brand aligned visual and video content at scale. I think there is something important about that at scale. Uh let me actually zoom in a little bit. This is a meaningful milestone for web flow and uh to understand why you have to understand the problem we are obsessed with solving for our customers. The missing piece in AI powered marketing I don't want to go through all of this just yet. Uh the three layers we are building for our customers. So the platform infrastructure layer the marketing operate uh system that's the CMS I guess and on top of that where marketing teams and agents work together building managing and optimizing and I think this is the point where they are trying what they are trying to do. So the way I see it uh web flow is seeing that everyone is becoming a marketer or it's just easier than ever to become marketer. That is to build build a tool. You don't need to be technical, right? You can build a tool. You you know fairly you you can build something and you can launch it. And web flow tries to become a platform where it's easy to market what you have made at scale with content generation also at scale that will be consistent and at you know on brand vos makes the context layer dramatically stronger. I I haven't read read through um completely through this, but Frontier models are trained on the internet at large, not on your brand. So, this is going to be trained on your brand. That would be interesting. Um uh how would that work? Is it going to be kind of like a model that kind kind of like feeding your brand to the model and not to the context? The the first way of wave of AI agent AI gave marketing teams powerful but ungoverned generation fast output blind to brand system. This is interesting. I I'm building a tool for our thumbnails and I'm thinking about exactly these issues like how can I have like brand consistency um with just using nanob interesting that they are putting the money there because this means this is like a specific AI that is good at the specific thing instead of like going this general direction where we can see like something like nanob banana is getting amazing at everything uh everything visual. Anyway, so this is what they're trying to do. You can read the rest. I want to just show uh the website a little bit. Uh brand aligned content without prompt ju jiu-jitsu. So, uh, I guess creative emails, PDFs, everything that I I I can't tell if this is compelling to me or I'm like I don't I'm I'm not sure. And I'm interested to see how this would perform compared to something like um the the models from Google and the rest. Or maybe it's using the their APIs, but they have built harnesses around them. H that's what I don't know. I wonder what this I mean what is this a tool that you log into and then is it Canva that's built around AI is what I'm thinking. It is Canva for market I mean Canva is for marketers. [laughter] This is another Canva for marketers that is made with AI. So produce designed assets. I'm just looking at it seems like I mean the website to be honest is not doing looks really boring and it really commercial and justice it's not amazing and it's it says 2025 so we can steal uh the company now [laughter] yeah I I want to see if I can find any any like produced assets this is what I want to see Oh man, these are this this is as boring as it can get. The dashboard is as boring as it can get. I mean, [snorts] this is burning my [laughter] my retina here. I I I want to show you something. I I can't I can't not do it. Okay. Uh I need to show you something. I posted Well, timewise, what what are you going to show us? cuz I'm thinking about time. Okay, I I'll just show you a post. It will take two I will It will take one minute actually if Twitter is loading. Twitter is not loading. That's how That's not surprised. Fire all your stuff and you'll be fine. No, that was a lie. Um, I I'm going to give it a second. Yeah, something is wrong. Oh. Uh, let me switch my Wi-Fi. I'm back. He's back in the room. I'm I'm back. I'm back. Okay. Um I'm back and everything is loaded. It was my bad. Anyway, look, I'm just I'm talking about like designs and it's not necessarily marketing and this is self-promotion. Yes. But look at these designs we have made for a client. Look how clean and h how simple yet like full of detail these are. And when I see things like this and and then I compare it to [snorts] I don't know these as I'm like how is how is this possible like a company get acquired? I know it's about their tools and stuff, but they they are promoting brand assets and their own brand assets are not really good, but maybe their AI and infrastructure is good. I don't know. Um, this is just a short news on on the web flow front. Cool. Yeah. Wild. Yeah. Um uh shall I just run through these then? Yes. Are you Yeah. Okay, cool. So if you show me the marketplace. I'm actually interested in this one. Marketplace. So Anthropic have launched an e-commerce marketplace for claude powered application application uh claude powered applications and they've got people like repletable all these things uh already enrolled on the platform. [clears throat] And um it's quite interesting like I I still don't fully understand it to be honest. You can apply to join the weight list. So it's not just a freeforall. And the language is really interesting here. Use your existing anthropic commitment to pay for claude powered solutions. So what this tells me is that if you pay for Claude 100 bucks a month, you get access to all of these platforms, but potentially the use of Claude is through your app. So as an example, we've got our old friend rep here. You're not using Replet's hardware. Their their whatever your things you do on Replet just contribute to your weekly usage. I think I don't know. It's just this is really interesting. It's just their language. Use your existing anthropic commitment. Who are you talking to? Who are you? Are you talking to the users or are you talking to potential partners? Do you know what? I I I think this is talking to Replet and like partners. uh this is like anthropic commitment to pay for cloud powered solutions. So I think so I think um someone like anthra some someone like replet they are paying huge amount of money to claude and maybe by joining I don't know what is like this investment that they have to do that they that it would cost them less well look at it like one commitment more options and I sort of see it like Google workspace a little bit like you get Google Docs you get Google sheets all that stuff and and this is like yeah, you pay for Claude and you get loads of third party applications through it. So, you know, I I don't want to um uh speculate too much at this point because I just don't understand it and we're trying to get through. But like yeah, uh enterprise AI transformation requires speed, confidence, the core marketplace. Let's Cox automotive teams. So, it's the actual companies is it's is it's yeah, it's like Google workspace by the looks of things. lets Cox Automotive teams move faster by extending our anthropic investment into the partner tools that we need. So, I actually think it's for companies. It's not for you and me. It's for companies to um Yeah, I hate this. Get more from your commitment. No, not get more from your subscription, not get more from the money that you're paying us. Anyway, sounds like a marriage. Yeah. Yeah. Yeah. Um, power tools for enterprise customers. There we go. Yeah. But so there we go. Okay, man. I hate this. Uh, I hate the confusion. I saw cloud marketplace and I was like, is this for like to may maybe there would be templates for me to start from, you know. That's how it sounded for me. uh for for end users to have um somewhere to start from because that's how it works with perplexity computer you you can clone other people I mean this has existed and this is just essentially open sourcing or like using GitHub uh whatever but yeah um apparently [snorts] it's not that it's yeah it's Google workplace or workspace whatever they call it for um for apps, you know, uh I wonder how much it is. Don't can't see any pricing on it. I can't see anything like that. So, yeah, but it's an interesting um I think the bigger question really here is about, you know, um profitability um where Claude is placing them in terms of like becoming like the work forces tool as opposed to ChachiBT being the normie tool. I think they're they're placing themselves well for success by being perceived as the AI for real work for knowledge workers, you know. So, I think it's a solidification there. Um, let's move on. Yeah. Meta acquires Maltbook. [snorts] Let me get this up. How crazy is that? This is insane. This is insane. So, Meta acquired Maltbook. Maltbook is a Reddit like social media specifically for AI agents. Uh, founders Matt Schlitched and Ben Par joining Meta Super Intelligence Labs to advance AI agent technologies. Do you guys remember Malt Book? I mean, it turned into a bloody cess pit for crypto scams and whatever, which begs the question, what the hell are Meta even thinking about it? And yeah, and what gets me is that it was quickly found uh researchers soon revealed that the vibe coded malt book was not secure, meaning it was very easy for human users to pose as AIs and make posts. So the whole idea was just, you know, the the the the idea was admirable, like let's just give AIS a social media, but it was so badly done that people just took advantage of it and like told their AI to post something about a religion and, you know, again, to to freak people out. Yeah. Um, you could grab uh tokens and and pretend to be anyone you want. This does not strike confidence in the I'm I don't I don't mean to sound like cruel, but this does not strike confidence in Matt Schlitz and Ben Par. Like where is their skill if they can't build? Are they hired as designers? It was wipe coded, right? From what I understand, it was wipe coded. And you said it, but just to make sure people understand like I could post uh as you Joathi that that's insane. Like [snorts] this is almost as bad as it gets. Like it doesn't get too much worse than this. So what are Meta buying? What have Meta bought? Yeah. Have you seen the tinfoil? Yeah. The tinfoil video uh from Primagen. M he well we we don't know why they bought it but I think there is a point about how they make money how Facebook makes money method they make money by selling ads and maybe there is something about the content posted there already and maybe there are still bots and pe slash people posting there and maybe they want to see it as a experiment and research uh to potentially sell ads to AI agents. [laughter] So you tell your AI agent to do shopping for you and it it's not just you know scamming you maybe convincing them. So I don't know. I think there is like interesting things about it as a social experiment but it is a very surprising um thing that a company like it makes me sick to think how much well I don't think they bought maltbook. I think they've aqua hired the two people. Right. [snorts] Um it was poorly executed. It was an interesting idea but not not necessarily groundbreaking. I mean Andre Kapathy was like this is AGI. So that was quite crazy. Um it wasn't executed well like what are Meta paying probably quarter half a million a year for for these two people? I don't know. I mean I've looked at their profile pictures. They just look like a couple of dudes. Do you know what I mean? Anyway, [snorts] let's move on. [laughter] [gasps] So, we we're, by the way, if you've just joined us, we're trying to we're trying to steam through for for a bit of a um plus there's actually quite a lot today. [laughter] Yeah. So, we're trying to steam through uh in aid of a a more conversation piece at the end. So, if you do have any more questions, thoughts, or whatever, leave it in the comments and we can linger on a topic a little bit longer. But for what it's worth, yeah, we're just trying to trying to get to the end. But before you go to the next one, am I laggy to you? You were just now, but you but that was before you swapped your Wi-Fi. Yeah. No, laggy. I think I'm I still am because I'm using Chrome. Uh not Chrome. I'm using Arc again. Uh not like last week. So I have a technical issue. I don't know how to solve this. Well, you sound good and that's the most important thing. So Okay, that's good. Okay. Um so Claude code review anthropic introduced Claude code review or just code review I think for claude co for claude code a multi- aent system that dispatches a team of agents for every pull request to catch bugs that human miss. So it's in a research preview for team and enterprise and it's an interesting one because we've got a bunch of code review tools. You've got code rabbit, you've got replet reptile, you've got augment. I did a review on these. Uh, someone else released something recently as well, which is a good alternative. Um, either way, this is a code review that will just fling into every pull request and just give it a review. Now, it's quite expensive. Um, I don't think I've got anything highlighted on this thing. Here we go. reviews are built on token uses and generally average $15 to $25 scaling with PR size and complexity. So, I mean, obviously this is not aimed at you or me. Um, I would definitely point you in the direction of Code Rabbit. $20 a month and it's seemingly unlimited reviews. But um the fact that it's in beta review for team and enterprise again makes me think that this is something um you know for those types of people. So go on. Yeah. What is the difference between this? So code review essentially you would do slash code review and it reviews PRs. So it's bound to you'd link it to your GitHub or your GitLab or whatever. So it's bound to PRs and it will come in and just leave review comments on what the final code is you know. So what is the difference between that not just in terms of functionality but in terms of you know the results that you would get at the end between that and just you know prompting your AI saying like how good is my code? Good question. So I think the biggest difference is this agent team aspect. there's going to be a flood of different agents that are communicating with each other um that assess the code in a fresh set of context um and and that that's kind of it really uh I I would assume that there's a level of different architects like as in sorry uh different instructions like one is going to be focused on security it's just a more meticulous um process I haven't used it I don't know what the difference would be to you setting off an agent teams, which is a new feature. I've yet to do a video on it, but Claude Code has agent teams as a feature. Maybe there's some just real um very well put together um agent instructions, maybe. I don't know. So yeah, it's arguable, but uh I did do a video uh and I didn't know at the time actually, but I did a video on this thing called background agents.com and this is when you were off partying for your birthday and it talks about various bottlenecks in the development life cycle which believe me there is nice website. Yeah, it is a cool website. It is cool. So nice. So good. And it's by my friend Lou. So shout out Lou. Um who he hasn't written, you know, he he's collected the content and he's pulled it together in identifying bottlenecks in the software delivery life cycle and how we can alleviate those by putting agents proactive agents automatic agents in those bottlenecked areas. So where where can we relieve the bottleneck of code? where can we relieve the bottleneck of review test and all this little lot. So he's actively developing it. It's a really clean nice website. Uh really enjoyable to look look at um which we're trying to we want to do something about it but it's um yeah it's more it's not for the indie developer. It's more for like software teams. Uh why am I saying this? I think this code view is uh an important bottleneck to to relieve that you can automate and the more sophisticated the solution be I think the more um how do you say uh uh important it is to your to your your workflow and also how worth it it is for each code review being 15 10 to 15 bucks or however much they said. But yeah, go check out background agents.com. um hasn't linked his LinkedIn. Uh anyway, so yeah, code of view. Uh I for the indie developer, I think we'd be more interested in code rabbit grapile uh augment and I've got a video on that on my channel if you wanted to look at a review of it. Cool. By the way, it's cool. Um okay, before going to the next one, let me experiment with something. Okay. He's gone. He's gone. I think he's messing around with his internet. If it's me and he's hearing me funny, let us know. He's joined. He's in the He's in the green room. Who's left? He's left the green room. How are you guys enjoying this Blitz tour? Because we can end up chatting half the time and I quite like just churning through the the the news. And if you're on YouTube, give us a like. All in the pursuit. We were just talking about you. I'll be honest. Yeah, I I I just used another browser to join and let's see if this one is better. Okay, cool. Cool. Well, from one um loosely researched piece of news to another. Kapathy has developed auto research. Have you looked into this then? I saw a few tweets about it. Is it an AI that researches and gets better? Kind of. So, Andre Kapathy is I never know Karpathy or Kapathy. I'm sure I researched it and it said Kapathy, but it's spelled Karpathy. So, sorry, Andre. Let us know. Give us a text. Let us know how to say your name. Yeah. Andre Kapathy released auto research, a project where AI agents auton autonomously experiment on a single GPU LM training setup overnight, modifying code, running five minute training experiments and iterating towards better models. So this is what confused me. Is it training or is it research for better models? So I released this GitHub repo which um you can download it yourself. Um the idea give an AI agent a small but real LLM training setup and let it experiment autonomously overnight. It modifies the code, trains for five minutes and checks if the result is improved. So I is slightly outside of my wheelhouse really on the understanding of all of this. But if I re re um re revert or resort to my notes here, AI agents uh can run 100 experiments whilst you sleep. So 12 per hour and you program program MD um which I'm guessing is like its base instructions which I I think I read that you've got to keep quite light. The agent modifies uh train. py which is the Python file. So um there's a fixed 5 minute time budget per experiment and it can run on a single GPU single file single metric. So and it's based on nano chat. So let me open up nano chat and see if there's more clarity there because Andrea Kapathy is a smart dude and I often find he overexlains things because he's oh so he made nano chat as well. So nano chat walked so auto research can run. Nano chat is the simplest experiment experimental harness for training LLMs. So it's just training a model. I think I think it's literally just training a model. Hey, let's just try something. Let's just try let's try and build something tonight. Um have you built any apps or tools recently? Can you share any learnings from there? Funny you should say this. Funny you should say this. We we're going to we're trying to blitz through the news a little bit and then we're going to talk about building uh coding AI and what will naturally fall into that is probably some conversations about what tools we built. So um if you have any um tell us what you do uh tell us what you want to be building and tell us where you've come from like are you a web flow person and like what's your current state of knowledge on things what have you heard and we will try and address those and this is not just you uh sorry I can't read your name Ashish it's not just you Ashish our dear listeners let us know in the comments and we'll we'll address it at the end. Um, but we're hopefully going to have a nice discussion on yeah, building in the age of AI and how to get started with it. So, uh, back to auto research, I think that, yeah, it's a really simple thing. You you download it here. Um, you install the dependencies, you run the prepare script, and you can run that. And this this hopefully if everything succeeds then it means your system is up and running. And then you use clawed code inside of that codebase or codeex probably codeex based on the prices of everything. Um and I think it's intentionally supposed to be slow so that you could so it it bypasses like the limits so you use your limits during whilst you're sleeping. Do you know what I mean? I think that's the point. So yeah, look at program MD kick off a new experiment. Let's do the setup first and then so it will run through. So whilst I can't impart too much knowledge on what this bloody thing does um or what sorry what can you get out of it? Is it a new model? I don't know. It's something to that I'll play with and experiment and and report back on. Speaking of Chrome, um Google has expanded Gemini for in Chrome to Canada, India, and New Zealand with 50 plus language support, bringing a Gentic Workspace integration to the browser side panel. I didn't see UK on there and sure as hell not going to see Germany on there, so we're not going to be part of the part of the party clan. But it's so interesting how this has taken so long. We've got Claude, we've got Comet, we've got DIA, we've got all these agentic browsers. Google seemed to normally be the people who are just like throwing everything at the wall and seeing what sticks yet they've been so slow to agentic browsing and I really really wonder why. Um, it can read up to the uh the context of 10 open tabs. Has deep integration with calendar, docs, drive, match, search, and YouTube and things like that, which we didn't cover it, and I wish we kind of did. The Google Workspace CLI has got a lot of attention this week. We could I think we could have had a viral video on our hands if we done that and edited it, but we didn't because we thought this is a little bit boring. Are people going to be interested in this? And everyone, what happens? We're covering 10 news every week. So yeah. Yeah. Yeah. This um anyway, so you can compose Gmail, although I won't open it right now. I did actually use Gemini inside of Gmail because I was using Comet to like tell me about what the update is on a big long long conversation. But then I was like, oh, it's got a Gemini button. So I clicked that and asked it the same questions. It's actually it's actually pretty good. So it is still integrated within the apps, but I think this is a broader one that sits outside the app. So, um, you got Nano Banana 2 in it as well. Um, yeah, it's it's slowly rolling out bit by bit and, you know, might be a really solid solution when it does eventually come out considering it's taking so bloody long. So, yeah, Gemini for Google Chrome slowly but surely. Um, Cloudflare. Ah, that that one was the memes. Yeah. Well, you can you can cover the memes. So, so Cloudflare um have released a new crawl endpoint that lets you crawl entire websites in a single API call which you can return HTML, you can return markdown, you can actually return structured JSON on the website. So, you know, I'm guessing that would be like heading uh headers and paragraphs and things like this. Uh have you got something to share? Um Oh, sorry. I'm trying to remove Yeah, all good. Um, yeah. So, I actually got this set up and uh I mean it's so boring to look at to be honest, but yeah, I just I created an API key. I uh posted a crawl to my to my my company website Jupyter and Draft and it returned all of the HTML of the whole website. But then I can pass it a few flags and let me let me show you the nerdy stuff before we get to the memes. Uh c can you can you show me like that from start kind of? Um well oh no you you have it already set up. What do you mean set up? You have I I don't know like where would you start? You so in terminal well you'd start inside of Cloudflare and you'd create an API key. Mhm. Obviously, I can't show you that because I don't want to reveal my API key, but you you create an API key with browser rendering permissions, right? So, just log into Cloudflare, create an API key, and then you can literally jump into your terminal and run this. And what this does, curl request is just a request. Um, under the hood, a browser is basically doing a curl request. Um and then you are sending a post request to this URL and then you put in your account ID there and your API and then um and then your yeah your API code token in a sec and then that's the URL. Ah okay and then you pass it authorization header with bearer and there's your API token application JSON and then the data you send is the URL that you want to crawl. So, I put my company website here. I put my API key here and I put my account ID here. Hit enter. It takes a little bit and it then returns a um a an ID and then you just say you get you put that job ID that you get back there. You put your account ID there and you put your API key token there. And that will just that just dumps you a [ __ ] ton of HTML in your terminals. all in your terminal. But then you start to pass it some particular bits of data um such as uh where have we got here? JSON options. You can even prompt it. Look, it's an AI endpoint. So you can prompt it. Get me product name, price, description, and availability. And it will just do that. It will just return that into the a into the uh interesting API. So this is a cleaner way of doing it. might be a great way to do um website migrations. I'll tell you that or extracting the data from one website to input in another one. I I wanted to ask like this is also going to be great if you want to create some like health checking tools that that would grab the content of a website and then you could run some tests. You could do a lot. It's it's a handy bloody tool, but I think as you're going to get into it's the memes that are quite funny. So yeah, like get given to me in HTML, HTML and markdown. Um, you can even authenticate look uh all of this. So it's really useful. Yeah. But I think it's the it's the manner in which it was done. I think is the funny thing, right? Yeah. I I don't know if it's that being the issue or just people Well, I've got this if I I'll show you this tweet that I found. Cloudflare spent years fighting scrapers and selling anti-brot protection. Now they've launched their crawl API that can scrape an entire website with one request. Sending a single AP. So I could I could crawl your website and I could do whatever I want. I could prompt crawl your website with this. It's not like it protect like it's not like it protects you or anything like that or or that you can only crawl websites that you have hosted on any website. any website. So very interesting and it's a very different approach from a company that its [snorts] whole entire job is to protect websites. Yeah. And prevent crawling. So it's just quite ironic there really. Yeah. It's this uh it's this meme, you know it, you either die a hero Yeah. or live long enough to see yourself become [laughter] Yeah. Exactly. Uh yeah. So but the memes aside, I I saw so many memes similar to this and you know uh internet is funny about these things but I find it really useful for potentially like doing client research. Uh like you you get a new lead or you want to reach out to somebody. Uh you want to build an outreach tool. Maybe somebody maybe wants to build an outreach tool for themsel and maybe I can dump a bunch of websites that I found a bunch of leads with their websites and I can run some automated uh tests and yeah even the results that I get I could use them to write them the outreach materials like to to kind of like okay I understand what this tool because um I I see what the issues are. Yeah, that that could be very handy for that and as you mentioned to migrate content to Yeah. Yeah. Well, I um I had my website um downloaded. Uh it's not downloaded. I'm just tired. I had my my company website was on WordPress and I literally just sent warp to I said just crawl this website and download all the assets. Right? So it just it was just done through regular old it wrote some Python did some regular old crawling techniques. It's not impossible to do. So, impossible to do, but I think one I reckon this might bypass some like it does it in a way where you're not going to trigger any rate limits or whatever because it's already cached your website or something. Um, and two, well, no, that's it. It would have been handy for that. I think it would have been nice to But how does it But it wouldn't work for anything, you know, behind a a login. So I can't log into my own Instagram for example. You can log in. You can log in. Yeah. Yeah. So can I crawl my let's say for example I want to make a database of my followers or whatever and I can I want to search through something. Yeah. So I um remove other media from stage that one. Um, so I mean you'll have to try it out. But look, if you, you know, pass this through, go to instagram.com and authenticate with that. Give it a give it a whirl. Give it a while. Crawl with authentication. I didn't see that. That's interesting. It would be interesting to see if uh Meta will block this or not. Yeah. Is it is it like a a playright thing that's actually navigating around and it's figuring out what that is or is this more API driven? Yeah. Because for example um let me try actually let me try it. If you want to see who have seen your story for example you you can't you can't search it. This is something you you can't search specifically who have seen your story. Mhm. Um there there are a bunch of things that you can't search. So or like likes uh who has liked you can't it's quite dumb because you can see the full list but if you want to search it you have to manually search it by hand. You there is no search bar for it. But this could potentially create tools because people are I know this because people have asked me [laughter] so often like this could be tools that could make some money if if you can go uh past Instagram and Meta security, not security, but like the the limits or maybe they have some uh limitations there for privacy whatever. But if you could pass them, you could build easily a tool where people could could search against against these things because people people are curious. Well, Instagram seems interesting because I'm just trying to navigate around and I'm just trying to get to the notifications page, but it's all it's all um Oh, okay. I've got it. I've got it. So, I'm on my I'm on my notifications page. I'm going to run a crawl on on notifications. Bear with me. So, if I uh Okay. Okay. I'm going to [snorts] Yeah, but the thing is uh I don't think it will work cuz you cuz you first need to crawl the website and I can't just put in Instagram notifications. Like I can't Yeah, it's like I I need I'd need to look into it like uh post. Oh, no, no, no. This is Yeah, this is good. Yeah. So, I crawl notifications, but then I'm going to give it a Yeah. Okay, cool. I think I think we got it. Um Oh, balls. I messed up. I bet this is well exciting, guys. I bet this is really exciting. Paste authenticate. Um, [clears throat] Instagram. Let's go copy username. I can't show you this because it's got I'm putting my username and password in here. So I keep trying to make you expose these things, but yeah. Okay. You're well trained. [snorts] Success false code thousand authentication error. Yeah, just reading. Yeah, I know. I can't I can't. Yeah, I I tried to crawl Instagram notifications and then authenticated with my username and password, but it just didn't didn't do it. Um anyway, interesting. Anyway, yeah, I don't think it's Well, yeah, who knows what it it is for me some sort of like agentic test. By the way, maybe we should make this a test. make this a standard test for any tool coming out like even open claw can it go to a specific Instagram page and perform some task I have like many ideas for what like you know all the reals and screenshotting them like a specific page has and then put them in a notion I haven't managed managed to do this with my open claw reliably. It can do all the pieces and bits but it can yeah it can't do it reliably even though though it has its own Instagram account my open claw u but yeah interesting interesting anyway this should be a test we should make a test like these companies say our agent can do computer use and this and that but like can it do some research that I need a human to do it today cannot do it because what I said is [laughter] don't by the way it's not for spying on people it's for like brands with the content that they have and the views that they get and the the likes and you know you could grab so much content from uh so so much data from that but maybe the old um coding with Python and uh scraping the data is the way the old school way yeah anyway [snorts] yeah that's that. Let's uh let's move on to uh visual visualizations. We've got a double whammy here from Chat GBT and Anthropic both offering quite similar features but not quite. But I bundled them into one anyway cuz why not? But OpenAI have launched a dynamic visual explanations in chat GBT, letting users interact with formulas and variables in real time for 70 plus math and science topics. So this is Oh, sorry. I was I didn't realize I was sharing my screen. New ways to learn math in chat in in math and science and chatbt. So, you basically have these little I don't know if that's moving too fast for you, but you have these things that you can slide around and help visualize stuff. Um, I was going to give it a go in chat GBT, but I think you get the picture to be honest. But how can we be sure that these are correct because there must be some tools built in because we know that LLMs cannot do math if you didn't know. Mhm. So I reckon I reckon because they've been so specific on the number of you know topics 70 plus that a lot of these are potentially hardcoded in yeah there are tools like tool calls that these LLMs have and this is potentially something like hey if it's this specific law in physics or chemistry or whatever M here's the formula for it and there is like a calculator uh that can do the actual calculation because LLMs can't calculate uh and the visuals also look very on brand so it's not like AI is not generating the visuals necessarily so there are like a brand and like a specific ific UI going on here. This is interesting like with the controls and the inputs and the output I see consistency. So so they have built around this idea and now this is as a as a tool um available in chat GP which is welcome and this is the way it should be uh because that's how it can be reliable becoming more of a learning tool as well. So in the same breath Claude have also introduced you can see this there is more less interactive more um well I guess interactive but like more just I guess canvas like and and whatever so they're also pushing yeah exactly showing and not telling um so yeah similar sort of thing but pushing people to learn I think this is another reason why people get so confused with AI because it's like, well, it can seemingly do everything. I don't know where to start, you know? Not that it's a problem. It's like you get stuck in instead of like being paralyzed by the the [snorts] uh what do they call it? The paradox of choice. Just get stuck in and see what it can do for you. Um but yeah, quite cool. Uh where are we? Uh okay, this is this is the this is the big one from the from the stream and then we can get chatting now. So, have you seen uh magic layers from Canva? No, not I'm not sure. Probably not. Yeah. So, uh Canva have launched magic layers, a new AI feature that transforms flat AI generated images into fully editable layered designs, analyzing structure, separating objects, and restoring editable text and maintaining element relationships. So, this is the landing page. So you can see here, oh, you scroll down, it just this is obviously begins as an image and then it breaks it all out. It's kind of weird how it says AI generated images. So does it need to be AI generated to begin with or something? No. No. So you start with your flat design, you upload it, and then you can then actually edit the text and elements, which is quite a nice thing because we're going to try it now. And um but unfortunately I don't have any text on there, but let's let's give it a go. But what is the diff? So any tool that I see nowadays, I'm like what is the difference between this and just giving it to ChatgPT or Claude uh code and saying separate it out. No, just say like build this. This is an image. This is your idea for a website and just literally build this um website and it will and you just tell it okay well animated and separated, right? M ah but I guess with images you need a image also model that creates separate PGs so there has to be some guard going on I see that you're selecting an image okay yeah so nice so it's doing something here the living room hopefully it's going to separate the couch and whatever I've got no text cuz I wanted to try that like editable text thing. I mean, you could you you'd see how far you'd get with it, but I imagine this to be a for loop with nano banana. Do you know what I mean? Select the So, okay. So, we've got some selection going on here. There it is. Oh, that's nice. That's nice. That's I can even look I can even move the stuff around on the shelf. Yeah. Yeah, that's that's well put together. Okay. So, this is this is actually really nice because the this is this is for a job that I'm doing right now and um the client often wants to like add stuff around. They might say, "Oh, look, we don't want the laptop anymore." Cool. So, can you move the TV again, please? Okay. So, this is for people who haven't done this kind of thing. something that is really impressive that that the the wall behind when when you take it okay it's not maybe as impressive because you kind of see that it's not perfect especially on the left side of the TV you see like between the clock and the wall you see the wall is not consistent anymore if you do this in the old way with Photoshop you have to fill the place of what you take out and obviously You need to make that consistent looking like nice and here I I can but yeah it's it's it's quite good. It's quite good. I mean I it's Canva in it. It's just turning it. I mean it's already is like let's be clear but it's uh it's definitely an all-in-one tool. In my latest video which is coming out hopefully this week um you and I discussed it. you suggested Canva as a a website developer and um you know it's a lot of people like it and it's not not that expensive either. So it converts uh PNG JPEG images into multi-layered editable designs with one click as you just saw. Um and like I say the text gets restored and you can type it. I wonder if what if it's like a paid for font. Do you know what I mean? What if you've p you you're nicking a paid font that will work? Well, technically you can build any fonts now with with AI. Oh, yeah. Bet. Yeah. And what? And they're and they're like, you know, but I don't No, I don't think so. So, this is where nobody knows. But you could take a paid font and rebuild it with AI, but then you rebuild it and what if it's not perfect? What if the AI That's probably That's probably the loophole, isn't it? That's probably It's not It's not the same font. So, it's not the same font. It's worse. Yeah. Yeah. Yeah. Or it could be better. You know, it could be better. It's just not the same. It's not the same. Uh and man, it it's kind of difficult. But but because what if I like just look at a bunch of fonts and make my own? Yeah. I don't think there can be any case made against like small creators and big ones won't do it like Apple won't just take a font and steal it and for small people I I I don't think there will be a case against it like nobody's going to come after you if you take a nice font and mix it with another and with a website that receives a 100red visitors a sort of thing. And if you if you grow too big then you pay the fine and that's called um you know this is just your business cost like cost of doing business and that's how all of these AI companies are doing it you know clawed and I'm not saying this is moral it's not to be very clear but this is how this is done in the industry they they stole literally every piece of content out there. And I'm not just saying like they soft stole it. They literally pirated the books. Like actually something you can get [snorts] fined for here in Germany easily. Like easily you can get fined for pirating through the like these torrent for example. They they literally torrented the books and they're paying the fines and it's just because you know what they might they make it they've made that money back. Do you know like a thousandfold? Yeah. Yeah. This is it. This is this is why law uh has not kept up with um private enterprise really like private enterprise was not as powerful as it as it was as it is now. It's uh it's people are getting just ridiculously rich just the regular well not regular obviously but you know it used to be that um yeah the the power and the the money was reserved for politics and I don't know I don't know it was before before our time but um it's it's clear that law has not kept up. Yeah. Oh, anyway. Yeah. Do you have more to show? No, but it was cool. It was cool. I That's it. It surprised me like you're surprised. You're flabbergasted. It I'll show you something. So, okay, we done with that. I'll show you the the tool. So, we have also this comment uh from earlier. Why can't I show the comment? Maybe because we already did. No, it is. Yeah. Yeah. Yeah. Have you built any apps or tools recently? Can you share any learning? I I can't speak about the learnings, but I've been building this tool for us and I want to show it. So here I'm building a full like YouTube helper with AI. Yeah. And these are thumbnails for our one of our videos that is not out yet. By the way, we've done like actually maybe I should show the the Figma. We've done like hund we have like close to 500 uh videos and we've done pretty much like every thumbnail manually except for one. One of them is made with [laughter] AI. Uh, let me actually guess which one. Guess which one. Find it. Go. Go find it. Am I Am I lag lagging? I think I'm lagging. Well, audio wise you're fine. So, okay. So, I'm lagging because I opened Figma. Oh gosh. Oh gosh. I'll just show it and close it. Yeah. This is this is the amount of like thumbnail building we've been doing. So, it's quite a lot. It's a lot. Now, let me close this. So, this tool uh I know the UI is bad. I've been building other things um mostly functionality and I migrated the UI. I know there are like inconsistency here and this banner top is bad and this thing here not really good that yeah I know this I'm I'm fixing them but the whole point is here for example I'm giving like a source image this is like a screenshot and then another source image this is like an example thumbnail and It's it's making, you know, it's making some of you you see like it's taking your image and putting it there. Not all of them are good, but that's kind of like where I'm in. So, you're choosing the image. You're choosing which screenshot. It's not it's not taking that from of No, no, it's not fully automated. So, this is this is just like instead of using any of these uh what's it called? the one that that I subscribed to um the mockup one graphics. No, the which I still can't log into by the way. Yeah. Oh, that's crazy. Hicksfield. So like instead of like going to Hicksfields or like midjourney, I built my own you know midjourney with you know the API from Google and the whole point of this is the workflow is the way I want. So here we have packaging. Here I drop the transcript of a video which I want to I'm building some more tools that can grab the transcript automatically. But now I'm just copying the transcript, pasting it here, sending it to AI. It sends it to our next tab which is an AI chat. This is a full-on AI chat with projects with again the UI part partially is broken but partially is also really nice. I mean look at this chat search nice interaction. I was very specific about the prompt and it turned out really good. Um so now I have all of this and what's specific about this is like these like copy buttons. You see how I have these? All of this is under my control here with the packaging builder. So I can let me just slow down and just simply say this is a visual way of me making a prompt. This is just a visual prompt builder. Why a visual one? Because instead of typing, I can just click a few things and change how the prompt is set up. Then we send it to AI. then the AI takes it and then here I can have my own way of you know copy pasting it's making it much easier for me uh so the formatting essentially uh and obviously I can see the cost and a few things here and there and I I have a canvas to work in and we'll explain that and then now I have the image so it's going from the video transcript to meta data that we need for the video to the image thumbnail that is not quite there yet. But then thinking about building kind of like a brand manager around it. So, so you would have photos that you always have your disposal that you can add and you can just like drag and drop things. And then I want to bring this canvas to the mix. So I can explain here with just like drawing where you would be. This is like the person and this is the two lines of text. And then this becomes the context for the model to create the thumbnails based on. So this is the answer to that question. What I've been building has been this. And my biggest learning has been that AI is giving me superpower but it's not making me much faster. Like I think the amount of time that it's taking me to build this would roughly take for a senior developer to build this. It's may maybe a senior developer with AI would be faster, but I'm just not seeing myself to be extremely fast. I add a bunch of features, but there is a lot of time in refining the UX and the things that I want to do and the the details and not let's not talk about the broken UI [laughter] which also needs another uh pass to go through. Yeah, I think that that's the perfect use case for the Figma MCP. Like fine-tuned design is just it's not something you can really get done with AI. Same with dev. Fine-tuned dev is very hard with AI. Once you start really fine-tuning and being very explicit with things starts to fall apart then so let's talk about that. That's why I'm building a UI for a UI uh that builds UIs. So tweak pane is what you've probably seen. This is some UI controls that you can use to tweak something. Um this is similar to you know how web flow has things. You have a UI to to manage stuff like it's much nicer because you see it that uh where you're working. Same goes with like twigping and so all the tools and cursor added you know their own UI select tool and where you can change stuff and um just out of fun and curiosity I'm building on top of tweak pane this so you see this thing that I have here look how fun it looks and by the way this is inspired by another tweet that I saw and I've been having so much fun building this UI control. Like look how nice and gooey and how like stretchy it is. It's just Yeah. Um we also have this toggle which is it's it's literally just it could be this but it just I'm adding the light to it. Uh, and I'm trying to bundle all of these UI control pieces. These are like input types. And then build them um in a way that I can click on any element that I'm building. for example, just just like the way a cursor does it that I can click here and change the UI, change the the focus state, the active state and a bunch of things like this and then in the return I get a JSON that I can give to my claw and say, hey, the current thing is this, the new one is this. Change it. So it goes in and changes it the right in the right place hopefully. I mean it's it's a prompt. Uh we know how it works. So a UI to modify the UI. But for now I'm just having a little bit fun with making these just nice and I'm I don't know. It's I'm I'm a designer at heart. So I'm I'm playing with that. Nice. Um I think the I am I've been working on some um client projects and I think the the the speed obviously depends on the actual project itself but I think the speed at which I'm doing things is just crazy and like uh the client I think it's their first introduction to what it's like to have someone work with AI and what I'm finding difficult is not not to do with the the actual development of it. It's going amazing. Is is the consideration on the price of things and I'm torn between well I'm shooting myself in the foot by doing two weeks worth of work in an hour. So what what's the it's the mental kind of uh gymnastics I'm playing with myself on how do I charge? Do I do I double my rate? Do I Do you know what I mean? So, it's that kind of stuff that's that's I'm contending with at the moment. Yeah. But just just to correct myself, I didn't mean like the speed is not faster in that way. I mean for completing a product when it comes to building like the whole thing I think this and we see this in studies as well but I think for building like a prototype or an MVP the speed is like 100x a thousandx in cases. Mhm. But yeah I agree you should double your rate. You you should triple your rates. That's what that's what I was thinking. That's what I was thinking. Which is annoying because it's a scary number. It's a really scary number if I was to do that. Well, well, you you can say it like this. Hey, I will build you I will build you this prototype that you want it. I will build you that within a week, you know, or depends on, you know, their timeline. If they want it in two days, you say you you build it within two days and it's going to cost X amount. Mhm. Um, and I've seen, you know, what you do. So, I'm sure like not many people can do that. So, if they want the time and it's going to be a win for them. Oh, yeah. Oh, yeah. Yeah. Something that would take like months to build now you can do it all in a week or even less. That's a win for them. M um the other thing the only only thing I've built for myself recently is more of a fun thing and I took it through a through through a friend the other night is I've got this app that I built in Replet. Mhm. And it essentially allows me to upload recipes that I have cooked and that I like that I know that I like. Um okay. In theory, you could use it as a recipe store, but that's not really the point. The point is like I want to be able to choose from a bunch of recipes that I know that I like. And I scan them. I I take a photo of the recipe thing. Um, or I can upload, you know, if I've taken a screenshot or whatever, scan it and it puts it into my database. And then I can, that's what page you just saw. Now I can tell it like, "Oh, I've got garlic. I've got chopped tomatoes. I've got peppers. I've got butternut squash. I've all of these ingredients from the recipes that I've uploaded to you. What can I cook tonight? What's a what's a what's a bunch of recipes that I can cook tonight?" And uh you know, it will tell me it might tell me what I might need because it's a close match but not exact match. So, it's quite handy from that perspective. And then I added a feature that when I'm in the in the supermarket, I can tell it to give me a shopping list and a set of recipes where that shopping list will cook a bunch. So I can I don't have, you know, I'm not buying new recipes every single day or whatever or making new recipes. I'm reusing ingredients and they're all flowing between each other sort of thing. So that was quite a cool feature. So it was just to kind of help me out with that. I haven't I haven't used it all that much to be honest, but it's something I've really really wanted for a while. What is the difference? I mean, yeah, having your own app, obviously it's cool, right? the cool part uh stays. But cool, [laughter] but what what is about you know why not just use the notion MCP and Claude and to just tell Claude, hey, take a picture, add this to my notion recipe database and then you can query against that as well for based on a new picture of your fridge. Um, yeah, but you could The thing is I want on my phone, right? So when you say claude, are you talking about claw code? No, you can use the claw uh can't you use the claude MCP for notion on the app on your phone as well? Can I use the what's say that again? What the the MCP? Like can't you notion MCP? Uh yeah, but I need an app to send. Yeah, you can't. You You just use No, you can just use the Claude app. Oh, yes. That's why I was That's why I asked what what do you mean by Claude? Like what what are you referring Claude app? Um could have done. Yeah, I could have done. Um I wanted the inspire feature and and like a cleaner UI so it is literally just like one or two clicks away. not muddying the whole, you know, otherwise everything will be in in Claude. But I reckon, yeah, with a bit of consideration, I could probably build it into uh Claude and then use notion as the as the By the way, I would I would do also build an app because it's cooler and it's, you know, separated and Yeah. Yeah. Yeah. I think I think that what you're saying there and uh we didn't cover it but it's the whole acquisition of um Cal AI by My Fitness Pal. They do you do you see about this? Did you hear about this? No. So My Fitness Pal which is as you can probably tell a fitness app. They acquired a company which you take a photo of the food that you're eating and it gives you the rough calorie intake for said piece of food. Nano banana rap. Well, not even a no banana rapper. Like you don't even need n banana. It's it's literally any any app will do that for you. Chacht will do that for you. Claude will do that for you. The app Claude um the app chatbt yet they acquired them for something like 3 million or I don't know something crazy. Um a lot of money and it's like a tool that you could vibe code in a two in like 4 hours. Like it's Yeah. Literally absolutely nothing. Crazy how we we feel so empowered. But if you don't know how to vibe code it, it's a system that you could set up with zero time, right? It's literally something you can do within the app. I think that's a great way to prototype things is to to see what you can achieve in that. And this is a good conversation because it's like I spoke to a guy who who does wrestling who who monitors wrestling and stuff like that and he has all of this data inside of notion and [snorts] he was able to pull all of that data and show it in in graphs and things like this. I've never done that with notion but he he managed to do it um to visualize all of that data and to control it, search it, you know, whatever. Of course, this would be great for a dashboard to to, you know, to log into and this and that and the other, but the prototype is just is is proof if proof of concept just using notion. Everything is in notion. So, yeah. And it also reminds me of the conversation that I'm continuing to have to this day on my LinkedIn about designers who are um with the whole Figma MCP thing who are like, you know, you're just going to have to design it again anyway. And I'm I'm I'm trying to say to them like design uh is not first anymore. It doesn't need to be first anymore, if at all. And it some of them are just not getting it whatsoever that you can get away with something built in notion. You might get away with something that you have not designed, you have not even developed. You've just prototyped it in notion and you know what that's good enough. So, and it's still just as it's just a tool. So yeah, pick your pick your battles really on on how much you can um invest in it and see if you can get away with it inside of the Claude or Chat GBT app or in notion or something and then build from there. Pro prove the concept works first before committing to any real work. Unless of course you're just having fun, which is what I did. Yeah, I I can relate to that. Like I'm building like 20 apps. Uh the other day my brother came by and we were in the garden having some tea and he's into like styling like with clothes and stuff. Yeah. With fashion. So he was like, "What can you do with your AIS?" I was like, "Okay, so what do you want to do with fashion?" And he was like, "What if I uh what if I have an Instagram account and you know show fashion?" I I was like, "You know what? You could if given you have the API, of course, the money for the API, you could make a thousand piece of content like a thousand images and videos per day, not buying any fashion piece. you just you would you know get um the the clothing from these like clothing companies. You could build an app to match things together. You can do it manually or you could do it with AI or like semi or like mix it whatever and then give those to nano banana to generate an image based on your face and you're you you wear those things and then you make a video off of it and then you put it on Instagram and then you would get views and then you get people to buy these clothing um on your website and let me build you the website for it. So, I just opened clot code. I don't know. I I chose claw code maybe because it's the best uh inside of warp the the terminal uh that we both love with whisperflow. So I just we were sitting having it here and then I was I started recording and I just said hey I want to have like different products shown in an image of you know like person and I said go take stock images from the web that are relevant with fashion and stuff. So it did it did like in five minutes I had a website where you can go click on a person and then you see all the clothing all the the items they have in that single picture and you can go to those and buy them presumably you know with affiliate link. This is I would say this is quite a valid business idea especially if you generate with AI and if you do it u cleanly and you know pay a bit attention to the output you could actually make money with this and I didn't need any technical capability for all of this stack and there is pretty much I don't see like issues with you know with security and things like this because you have a bunch of affiliate links that that you are linking on your website and you're producing content and putting it on Instagram. So, it's crazy how you can be enabled if you want to to to build a valid business around just a simple idea. Mhm. Um yeah. So, well, I mean, say speaking about enablement, should we should we chat about uh what we were chatting about before the show then about the questions you're getting asked and whatever. Well, the the question we made Yeah, sure. Uh even though we we took a lot of time to get here, I suppose. Yeah, I suppose. Yeah. How long have you got? Not long. Uh preferably. Okay. Well, maybe we can just summarize then. So you were given the question of like where to start, right? Yeah. I see this all the time. By the way, this is not new today. Back in the day when I was showcasing a dumb thing I was building in web flow, people would ask where should I start? And I see this so much like should I start with J? That that was the question at the time. Should I start with Python or JavaScript? Pal just start. M and today is like should [snorts] I use chat GPT or cloud code? Should I I' I've heard you know this does this and Miniax and you know should I go free? Should I go p paid lovable web flow? Is it still good? I I genuinely I I don't think like people would lose anything if they would start with any of them. But actually start and and when I say start like start learning like you experiment, you learn and be conscious about it. That's the point. That's the the the the conversation we had at the very start. be conscious, be curious about learning and just go on and just repeat it. It It blows me away how many people procrastinate by like over ridiculous silly little things. is I was talking to someone who wants to get into content creation the other night and I was saying to her like I mean she's started like it's good that she started but she's she's taking me through all of her shots that she had and like experimentation and this and that and it's like this is so cool but learn it by doing it like get it out there rather than waste loads of time doing test shots or whatever. Get a video out there watch it back at the end of it. be observant about and be like, "Oh, I quite like that." You know, and and and learn by doing. You You're so right. You're not going to waste any time by they all contain a lot of the fundamental aspects to building stuff. I've got a video on my channel um which I uh what do I It's the Verdant. It's a sponsored video. It's a Verdant for the Verdant IDE, but I take you through all of the different aspects to it. But it's doesn't matter whether it's Verdant, VS Code, Cursor, whatever. Um, it's just about what pieces of that technology. And you don't need to know all right from the beginning. You'll pick that up as you go along. The point is is they all are. We're all amalgamating with the with what's working right now. And uh you're going to have you're just everyone's just going to have their own preferences. And I love I love claw code. I really like Claude as a model. I like core code as a CLI just because it's it gives me focus. Um I don't gain much at all from codeex, you know. Uh I love that it's a bit cheaper though, so I can do a bunch of things uh through codeex and and beyond what what Claude can do, but and it seems to be doing fine for me um on my current project, but I wouldn't say my current project is particularly uh difficult. But um yeah. Yeah. And a note on that as I mentioned it seems like almost free to me codeex app. Well go to your usage like what's your usage say? It it says weekly 82%. And I never It says 18th of March, but it I guarantee I guarantee in two days it will say [snorts] it will say 20 of March and then it's refilled. It seems like it says weekly, but it's refilled every day or every other day because I never have managed in the past 3 weeks of working with it to get it below 70%. And this is like genuinely the amount that I'm building with. Like these are all different projects that I'm actively either researching or building right now. This has been going on. Maybe there is a bug or something here because this has been going on for hours. But I did something different here which was giving the description to chat prd. This is I don't have you ever run into any limits whatsoever? No with [snorts] no because on my first day of using codeex I ran into my limits very quickly and I had four agents going off and I was doing a bunch of work. However, this was me using the CLI and I and I I I have no evidence of this. It was literally something I saw, but it it was a message inside of Codeex that said, "Use Codeex app for double limits." Yeah, let me show you. And so, I started to use Codeex for that very reason. Have you ever used Codeex as a CLI? Because you might see that message. In fact, I might see that message. I haven't even opened it for a while. Look, I have I have it. I have it. I have it. Actually, I just didn't want to show my other uh chats. Um I you see I still use uh Warp. I love Warp. And if we go to codeex, uh it says actually updated. Okay. Let's let's update it. But it does say what what you just said. Okay. So, it's just Yeah. So, let's So, is your is yours loading? Have you got it on your screen? Yeah. I don't know how but I have not got code. Um, [snorts] no it doesn't say it right now [clears throat] our fastest uh inference at 2 explain usage but that's different. Um, no but I'm sure I've seen the message. So the message you're talking about, it tells you use codeex through the codeex app and get extra uh usage and that has been apparently the case for me. Um this one that I said it's probably stuck has been you see it was it started two hours ago and it's still been going. So maybe I will send something like continue or stop it. What would you do, by the way? It's stuck. So, this was a I would just say continue. Oh, I'd say what's the status while it's running? Yeah, but I would steer it. So, press command and enter, which steers it. [snorts] Um, that's my question. I don't know if if that's something in the settings. No, no, no. Just hover over the enter button and it'll show you. No, no. I look. Okay. Followup behavior. I have it by default on steer. Oh, nice. That's cool. I didn't know that. I I changed that. Yeah. So, yeah, it just says uh I got the same message as you. Um which is trying to uh sell me fast mode. Um, either way, I think we're getting increased usage just from using that app because they want people to use the app and get hooked on it. So, and then I saw uh Sam Alman's message to say uh we're extending the trial for free users or something which I can imagine extends onto pro users and this but I can only speculate. But um yeah, you'll definitely get more out of it. But saying all of that, you know, you've got the CLI which is in the terminal. you might prefer that you've got the app here which has been praised a lot for the organization and clarity of of using it. It is a bit nicer to look at. Um yeah, you know, codeex is a lot better now, but you'll be looking at code. It really like it stems from what your background is and your comfortability and what you want to achieve. If you just don't feel comfortable doing any of this, then yeah, you might push mush more towards like lovable or any of the vibe code apps that do take care of a lot of the database setup because with with codeex you are responsible for the database. You are responsible for the authentication. You're responsible for knowing how all of these pieces get put together. Whereas the harnesses like lovable, they probably have some sort of system prompt or some sort of um boilerplate code. Like all replet projects look the same. So I'm sure all projects start with a boilerplate folder structure and tool and uh technologies already installed that then the agent is put to work on which is why probably gets a good result. So you're definitely you're definitely required to know a bit about technology and how it all fits together if you're going to go down the codec route. Let's call let's call web flow beginner but then we are talking about building apps which is a complete different ball game but web flow for building apps because everything is taken care of you don't need to care about the code um repletable is like medium um and then codeex is like expert you know. Yeah. Yeah, super expert. Old school super expert would be literally writing it in VS Code, but there's there's there's much there's a lot of benefit to learning code, but in terms of creating something, you know, with AI, you'll end up using AI. My today's workflow that might change tomorrow is going to chat PRD. I haven't used it too much, but the two instances I've used um has been really good. So I'm going to continue using chat PRD. You could do this also with chat GPT. Chat PRD just has maybe the system prompt. Maybe it is just a system prompt. But uh at the end of the day I go the idea goes to chat prd there I get a complete prompt. Then I paste that where into codeex app because the UI is nice and because it seems like I have unlimited tokens. I don't know how but it is. So today it might change tomorrow. Um I'm not in love with the model like GPT 5.4. I have it on extra high all the time but yeah it's it's okay. It's okay. Uh then if I have to you know go like for invite uh for yeah for like editing API or like adding my API and just random edits I I still use um cursor but that's my usage for it and I will cancel my subscription because I I don't need it. Uh and then I have obviously uh the warp terminal to here I fire up mostly just clot code for bunch of like refinement or sometimes I have multiple agents doing different things or one does research and the other one does implementation but I I have these like essentially like four things or mainly like these two doing the heavy heavy lifting. Yeah. Um, cool. I'm um, yeah, codeex app just because of the extra usage. Clawed code. Uh, I do use co-work. I'm starting to push more towards co-work for, you know, uh, stuff doing stuff on my computer, but you know, um, and then and then warp. Yeah. Just to Yeah. I introduced Claude co-work um to my assistant and she's in love with it. She uses more Claude um tokens than me just through Claude co-work. So any task that I give her, uh, I give it to her like through a long voice message and she throws it at Claude and either creates like a mini app, you know, one of those like kind of like artifacts with to-dos or um, if it needs to, we push that into our notion. So we are using less and less of notion directly but more as a database. you you throw things at it at it and then you ask Claude to also return results from it. So you instead of like going to notion and looking at it. That has been one of my biggest issues with notion because the more the the the cleaner you get with your structure, the more difficult it gets to navigate because then you have things separated properly with filters in place and things like that. M it gets difficult to go back to and use it. It becomes something where you store everything in it and it's you know people have been saying your second brain it's no notion like building your second brain it it has become very difficult to retrieve content from it. Uh and with cloud it's it's just like you know or in general with MCPS or API however you do it just so easy add these tasks or what are the tasks about this and a bunch of these have context there so it can also write that email it can do that research and plot uh co-work so yeah and there's a lot of chatter at the moment about Obsidian and all that is is just a it's just a a a local database made up of markdown down files and and the issue is is uh that's great if you're completely on your own and you're just building this second brain or some some organization and skill. That's great if you're on your own, but it's the moment you need to share that with a team or you need to, you know, share access or or have a common uh source of truth. That's where Obsidian falls down. I do want to do a video on Obsidian because I think people are getting hyped up about it. I mean for you know if you are there's plenty of people just working completely on your own but I think anyone who has to work in a a team are quickly dismissing the idea of of obsidian. Um yeah but yeah it's kind of it's well well equipped for LLMs just because uh for well yeah for LLM just because everything is created it's a markdown um note thing you know it's all made up of markdown files. So anyway right let's should we wrap it up there then? Yeah. Yeah. All right, boys and girls. Well, that's all we've got time for. Make sure to give us a like on YouTube if you're watching us on YouTube right now. Uh, give us a follow at Commandai Show. We've got a website now which just links through to the YouTube. So, if it's any easier to remember, command ashow.com. Um, give us a subscribe and all that good stuff. Join us next week for all the news in AI design and dev. I've been Sam and I've been Kar. Peace out. No, keep out vibing. Keep on vibing. No, no, no, no, no. Keep on vibing. Peace out.