#53 Anthropic/SpaceX/OpenAI IPO, Apple WWDC, Webflow Dev Platform, Fable 5, Hermes Desktop

[sighs] Hello there. [laughter] We're live. We are live. And we're live. I'm back. Back and uh No, no, it's actually a lot more efficient. We got the show done in there was one actually I waffled on. I got I think I got better doing it on my own cuz I How many weekends you How many weeks were you away? Two, three. I think it was three. Well, one of them you also were on vacation and then Oh, yes. Yes. I was once in uh Croatia and once in Italy. What news do you bear? [laughter] And I couldn't see AI uh because I'm kidding. I was I was talking maxing from a boat. [laughter] From a boat. Oh, let me know if I'm peeking because this is the new This is the new setup. Am I peing? Am I Something happened. You're not peeking, but I can hear you touching the microphone. I can hear Yeah. Yeah. Okay. That That's annoying. like I might have to put it down. I might have to put it down. Yeah. Yeah. Um anyway, yeah. Sorry. So, you're token maxing from a boat. Yeah. Yeah. I was just, you know, remote control using my phone to connect to my laptop and and doing some design work done. Very good. Um yeah. Well, yeah. The I feel like the show got easier. Um the as the weeks went on because the first time I was like, you know, just get it done or whatever. The second time was actually actually I think I spoke for like two hours nearly two hours just on my own. Not on my own. Had chat. I have had chat. But you know, speaking of chat, we have chat. Me too. Um when did it get released? When was it released? Was it released today? [laughter] I think I would probably point this towards my channel. Uh because yeah, was it released today? Void. I'm curious. We've been we we normally prep for these shows since what the last couple of hours. So, anything that happens online in these last two hours, we missed which oftent times is really um annoying to be honest. Yeah, it was already I'm going to use my voice so bear with me. Kimmy K 2.7. I mean, you know, thing is we spoke about the um uh deep sweet uh deep Yeah, deep benchmarks. Um come on, I don't want to that that's a really I just saw the the popup that you got the Kimmy code popup that you can't dismiss. You have to say get started and it takes you to to another page. It's really annoying. Kim K 2.7 code is now available. Well, should we just quickly go through it whilst we're whilst we've got void on the line? Yeah. Uh, share screen. Kimmy K27. Just these these are hip shots, boys and girls. These are hip shots. Nothing I've done done any research. So, Kimmy Code, this is not what we want. Kimmy Code is actually just their Claude code. Come on. Yeah. They're called code things. So, I don't know why it's taking me to here. Let's go. Can you also zoom in? No. Well, let me find what we need to look for. I'm specifically Look Ah, they they've done some back button hijacking there as well. Oh, no. Sorry, sorry, sorry. It opened up in a new window. I am looking for like news. That's what I'm looking for. Share this tab. Moonshot AI research. Oh my god, they keep open up in new windows all the time. Kim K 2. It's not even available on their website. Look, Kim K 2.6. You're well early. That's how early we are. That's how crazy early you are. Um, let's go. Kimmy. And then it just opens up in a this. Do I have it? I don't even have it here. Um, how do we even see this? You got any um links there? uh void because I'm struggling to actually find like a like a a a uh press release on it. A press release. That's what I'm looking for. I'm on the I'm on the docs. Try Kimmy. Try now. Give me code is yeah it keeps just any any mention of it sends me through to the Kimmy code thing unless it's only Yeah, let's skip it and get if you can get us a link to a um uh thing. Hang on. Is this me? Oh no, you you looked like you were totally frozen. Then I was like, "Hang on, I haven't heard from you in a while." And you looked completely frozen, bro. Like this is this is so me. Like I get this comment all the time. I focus. What? That you freeze. Yeah, I focus so bad that people think I I don't even blink. Um I was just preparing the hyperframes um tab for the show. Okay. Well, let's uh let's skip the Kimmy stuff until Void can get us like a link or something. We'll go over it. Unless you can find something or I'll find something when you're talking or whatever. Um well yeah actually that's that was a comment we got on one of our videos that someone like we're all we're all guilty of it I think but when um we're talk you were talking about something I forgotten what you're talking about something to do with web flow and someone left a comment saying the the the disparity of enthusiasm your energy was like super high and my energy was apparently super low. Um, clearly they've only seen one video and saying that there's an issue with this and that it it should be correct. Just as a point, if you commonly see something happening over and over again, then maybe point it out. If you just see it once, don't just assume that it's just going to be this is the way it is. And I just corrected him. I said like I was actually working whilst you were taking the lead and that was why. Yeah, that's how we do it. And yeah, sometimes I also zone out when you uh talk about something [laughter] all the time. Yeah. Yeah. I'm going to eat this. What is that? I I could I [laughter] can hear every crunch. Okay, welcome to the ASMR. [laughter] Okay. So, uh we have a long list to go about and people probably think it's it's the week of only Fable. Indeed, we got more except K2.7. Do you want to do the intro and then we'll Yeah, I was writing the intro because uh there there is not like many things that are intro like crazy intro worthy. Uh but there are topics anyway. So let's just go about it. So this week uh we had WWDC uh Apple's uh developer conference. Uh there were really interesting stuff and we had a big model drop fable five from anthropic. Uh all of this and quite a few other topics in today's episode. This is command AI. I'm Kabarzo and I am Sam and this is a week in AI design and dev. I really need to get a jingle. you know, um I built an app whilst um just because I wanted to generate music and I have access to Miniaax coding stuff um because they sponsor me and I was like, you know what, I'm going to build an app that I could just generate music because they have actually a music endpoint, but their endpoint that they've released and I've tried telling them about it, but their endpoint they've released has a like one minute maximum thing that if it doesn't get a response within a minute, then it's just not going to work. I've told him about it doesn't work. When I get my little um little app working that generates music, I'm I'm going to generate us a little jingle. A little jingle. You don't want to even Google search like a jingle sound. You want to make an app that makes that. Well, yeah, cuz I want to, you know, I want specific types of jingles and I want customization. Customization, baby. customization. That's the new stash. Anyway, yeah, exactly. That's the new stash. Let's kick things off. I for you dramatically reduced the list, but let's kick things off with an article I found. [laughter] Uh it seems like can't share links here. H [clears throat] what? Maybe maybe fake the link, put like brackets and a and the word dot where there's a dot or something like that. I don't know. But um yeah, that's um how much value is AI really creating? This is a this is an article from the Financial Times which has got some really interesting points which basically goes over the fact that is AI actually of is it is there a return of on investment which I think we're all wondering about. We're all still, you know, we're all building apps with our uh claw max plans and all the rest of it, but are we actually seeing a return? So, oh, this is for the AI haters out there. Yeah. Yeah. Well, I will say I don't think the um as with all things, and this is something I've really noticed recently, AI does like to summarize things, and it doesn't do a great job at summarizing things. as Apple found out like a couple of years ago when they actually banned their AI summary from news like it you'd be surprised how bad AI is at actually yeah giving you and summarizing information when you actually go and look into it this being a point this being one of those points it's just a subtle change but we'll get into it sorry um so one particular point well the conversation about AI's usefulness has matured considerably can you zoom in please cut you off and then Start over like zoom in. Yeah. No. The conversation about AI usefulness has matured considerably over the past year. Outright denial of its capabilities has diminished as more people have seen for themselves what it can do for their line of work. The battle was now over exactly how much value it proves. Fair to say. I think that we had a lot of haters. Oh, shout out 57 people. What the hell? Where have you guys I think it's cuz your face is back. They're like, oh, we actually got to watch it now. Yeah. Um, one particular point of tension between AI's boo AI's boosters and detractors has been the disconnect between the reported increases in coders output and the apparent lack of corresponding boom in product or value creation. MIT's M Deir and co-authors tracked software developers work before and after they adopted AI tools. Coders created or edited almost 300% more files, but that boost was halved to 150% by the time they got to the number of discrete pieces of work submitted for review. And that in turn shrunk five-fold to roughly 30% uplift in the number of full software releases. This is where it starts to reveal itself as you know and let's break this down for just a second because I think this is a very misunderstood aspect to development and people who don't know how to code right code has created and edited almost 300% more files do you know this is a question to everyone do you know and I think we can all relate to this do you know how much clarity you get from actually building something and then taking a step back and being like, "Oh, okay." And the the the desire to completely like start something from scratch because it's like, "Oh, we now have a bigger picture. We know how all of these modules and everything works together. I can rebuild this in half the time and 10 times better because I just understand the architecture a lot more. I think we all can relate to the fact that when you get your hands dirty in something, you just understand it that little bit better and you can Yeah. I say for example, I mean you probably have got I don't know actually you're you're not of the Photoshop era, but have you ever worked in a Photoshop file or a frame file or whatever for for days even Photoshop for days even and then you lose all of your work and then and then and video perfect you're able to then create that video from scratch in half the time because you know exactly what you're going to do, right? You know exactly what you need to do. you've there's so much experimentation, some um uh hurdles that you've overcome, blah blah blah blah blah blah blah. It's the same in code, right? There's so many apps that I wish I could just start from scratch now. And it's like I know like the what we thought we knew at the beginning of the project is now very different towards the end of the project. Yeah. Imagine that on steroids when you can use AI to set off five, you know, variations of one app and it's it's a click of a button just to delete that work. Like you've barely lifted. There we go. There we go. Right? You do 20 versions, you barely lifted a finger. You become so detached. the half the time that half the reason why we don't rebuild stuff from scratch or we used to not rebuild stuff from scratch is merely because it's just so it's it yes you build it in half the time but it's just so much effort and also that you love what you build like you you've what what's the sunkos bias maybe something around those lines is that you know you be with AI now you don't become attached to the code that you write you literally just can just bin it, right? So when they say 300% more files and whatever, don't disregard the fact that it is quite normal to then have that harved by the eventual steady release. That's that I don't think anything has necessarily changed there. Does that make sense? Yeah, it it does. Um if you scroll down a little bit also, I think the graphic makes it a bit um easier to see. [snorts] Yeah, exactly. It's the same the same point as above just just uh yeah the the visual thing. So uh I did like one of the most ambitious web flow projects I did um especially like early in my career. It was like my first big project for the heading animations. I remember I did like [snorts] about 10 different animations in one page for the background effect. It had a background effect. I remember exactly. I did 18 I did it manually. One eight 18 different versions. The work always has been like this. like all the the great work that you that you see has that has been um put out there has been because the the creator has done it either for that specific work multiple times or in their past experience they they've done it so often that now they can output something good and now AI allows us to to do that part faster so you get this like draft uh part of the work which is is 300% much faster. But I would say uh something that hasn't changed that much for me is when I look at the the overall speed of things, AI hasn't made me much faster because the the the speed an organization uh moves at or an individual moves at is the speed of decisions. Like it depends on how fast can you make your mind up about something until you can go to actually go to the next phase that is publishing that is you know doing the next thing whatever it is. So you might create like I might create 20 different versions of a UX flow but I have to make a decision and better it to be informed decision before I can move to the next one. Does that make sense? like I I can create a lot but I can only output something like truly to the word to the world if I make that decision and you you cannot really let AI to make these decision like we are doing it but that's where we see slop uh this is where I see the I the difference between like human output and the AI output that decision making and to me this chart is about that a lot output without decisions just to explore and then the humans make the decisions to output what's good. Yeah. No, I agree. And and even more so now with AI because again it's so low effort that the desire to then fill that space with even more, you know, experimentation and learning and and yeah, experiments um is is tenfold, you know. So that being said, there's still a bit of the article to get through, but that's just all I'll say about that edited almost 300 lines per code, but then it was harved. We wouldn't do this before, but if we would, we could have like it it it's so it's so beneficial to write experiments and create experiments and make mistakes to then have that halved like this isn't to me an indicator of anything wrong. Um I mean then we get into actually what actually gets in this this funnel is is now just this is a this is a good thing this this funnel of you know more creating more files that then get reduced you know I think it's worth looking at the relative thing but again it's just not an apples for apples comparison you know it's uh it's slightly different um uh blah blah blah modest game once that work has passed through all of the human bottlenecks associated with reviewing and releasing production grade work. Now this again is just not an apples to apples. However, I do appreciate that you know [snorts] these are well this is the biggest point of um uh this article. It's it's not talking about apps that you're releasing um out into the world as an individual, right? it's talking and that aren't being used because by the way this is kind of the undertone of the whole thing is that loads of apps are being created but none of them are you know people aren't using them or whatever. It's actually more to do with in organization. How much of the work that people are creating through their code? How much of it actually gets into the production version of the app? And there's a you know any good company will have a flow of checks and balances that take place in between all that. And then if something's wrong, it gets sent back. Or if there's some code that's wrong, it will get sent back to the developer. This is kind of like normal path of course as they say when it comes to software development. It's not really Yeah. It's not talking about the influx of apps that are coming out saying because this by the way this was the graph that got selected. This this is what I saw getting shared around and stuff like that. And it says app releases again. This isn't this isn't a number of apps appearing on the app store. This isn't number of which by the way is [ __ ] huge. This is just not what we're talking about here. The mark market increase in mobile app releases over the last year has not been accompanied by any increased downloads. Most of the new apps failed to capture even the modest audience. So I think that's yeah roughly what they're kind of saying here. But the bulk of this article really leans on inside the organization. So here's where we get to the Uber stuff like Uber CEO D blah blah blah. I don't know. Actually, I'm going to try and say that. Do [laughter] you know how to say it? He's Iranian. Yes. Okay. So, I'll say his first name, you say the second name. Okay. Uber. No, you can't say at the same time. You have to say it. Yeah. But say No, no, no. Say it like I'm trying to Okay, go on. Oh, you were trying to teach me. Okay. Was it? Go on. Say it again. Roshahi [snorts] Roshahi. Uber CEO Dar Hos Rashahi revealed the company had blown through its entire AI budget for 2026 in one quarter. I'll just say something on that as well. You give people some AI to play with during the most exciting time in AI. They're literally going to do so much stuff with it. Personal, private, they're going to play around with it. This isn't indicative of much. It's like kids in a candy store basically. However, the sentiment is that it's getting very expensive. Um, so they are planning to switch much of its AI use to lowerc cost models, reserving frontier tools for special use cases. Pairing cheap open-source AI agents with top-end models acting as sporadic advisers delivered better results at much lower cost. And I think that's the that's the play to be honest. And I would even go as far as saying local AI might be the play of doing the grunt work, but then you the frontier stuff is the fables. That's creating the plans. You had something to say on this point though when we were planning out. What were you what were you going to say? Um I I have my uh my notes for the end. But basically what I said about uh Uber is that they can't point out to the features that they have been releasing lately. They can't connect their the features they are releasing to the amount of tokens that they are using. So maybe their employees are just as you said like kids in the candy store. They are just using the tokens for whatever like replying to an email putting it [laughter] or starting the dev server. Yeah. Exactly. Yeah. Yeah. And people aren't trained. You're literally giving the keys to a to a um a digger like a JCB or something like this to someone who's just been riding go-karts. Do you know what I mean? Like this is a whole new paradigm that we're all trying to figure out. The only way to figure it out is just by spending it on tokens. That's just that's just the way it is. Um yeah, that's that's what it is. Um again, this is unfair. This literally the first quarter that you could not choose a worse time for people to be like spunking their tokens on starting dev servers. Do you know what I mean? Um, so yeah, it would not be unreasonable to see if all of this as see all of this as evidence that AI's cap capacity to deliver genuine value has been being vastly exaggerated. And I do agree, however, as we've just discussed, there are plenty of other factors that they're not considering. I don't know what this guy does or if he's a de he kind of looks like a developer to be honest. Let's be honest. [laughter] But um um yeah, it's hard to it's hard to really trust someone in the media that they actually know what dayto-day development is looks like. I always love drawing parallels to history. I love spotting the cyclical events that take place that we're all standing on the shoulders of giants and this and that. So I I I highlighted this in the case of electricity and this is really interesting in late in the late 19th and ear early 20th century. Sorry, I'm going to start that again. In the case of electricity in the late 19th and early 20th century, productivity gains were modest where factories simply replaced giant steam engines with giant electric motors, but left the rest of the machinery and layout unchanged. The boom arrived decades later when engineers fitted individual workstations with their own small motors. So, you know, we've been given the key again, we've been given the keys to the JCB and we're yet to kind of figure it out. And it might be a little bit later until we actually know what it is we're dealing with and where these where where all of these fit. And you know, speaking of Uber, they're going to be maybe not so much anymore, but during these startup phases, you there's there's this desire to move fast. So, you know, you're going to just use use them um everything max all the time um because you want to get stuff done quickly and to the best quality. Like, why would you sacrifice anything as a growing startup? Again, I I I'm not saying I agree with that consumerrist um attitude, but that's just the way it is in startups. They're literally just like there's every incentive to move fast and and win the race of whatever, you know, startup they're they've they're starting. Yeah. But what what are they trying to build? So in case of Uber, I could imagine like new new features and maybe new services, new markets, things like this. But there are so many startups and apps that do I don't know like an invoicing app that you got. What do they want to do with millions of tokens? Do they want to build Netflix in the in the tool? Like what do you need so many tokens for? Um as an individual, I understand I have like multiple ideas. I want to, you know, pursue them. But as a company that has like a very clear vision and they have a tool, I don't know what how many new features you you need. And again, like on one side, you see these big firms create like using so many tokens to build whatever godly amount of features that probably nobody gonna use versus what we just talked about. um you are building an app to create you some melodies because you don't want to use like an existing app. So on one side people like big software is getting bigger on the other side you're not using big software at all because you want to have many of like composable uh little software that does ex what you want exactly how you want it and later on you you might just toss it in the bit. So it's there's two two aspects to that and and and one of them is not really for us to say. There are going to be some startups out there that out there that are going to be the Uber of you know that that old saying we're the Uber of dog veterary services or we're the Uber of Thai brides or something for horses like Yeah. Exactly right. So there there's going to be plenty of startups out there who truly believe they're on the cutting edge of innovation when it comes to their product. We don't know who they are. Whatever what you just described then in the second part which is more so and where my belief lies when I talk about software uh SAS is dead. It's about I think the future is in self like you your own software. Maybe there are some data warehouses which I think a lot of money is going to be in data warehouses and things like that but the software itself is going to be built by the company themselves by a very small dev team and they are going to you know you know um ingress that data and use it for whatever they need to use it for. what that company does. I don't know. Maybe SAS and whatever is going to be completely dead and we're going to be building these a these AI well not even AI tools, sorry. This this software to help vets run or or plumbers do their job or all of these kind of all these what do they call it? White collar. I think that's where I believe we're going and that's where my where I'm putting my bets truthfully. So yes, you're right. Um, it sounds like you're kind of in that that position, too. But there's going to be so many companies who liter who who do think they're on the cutting edge of um of innovation with with a new SAS product that they they hope that we're all going to throw money at. But truthfully, I look at a piece of software and I think, how do I build that myself? And I often just build it myself, saving myself hundreds of of dollars per per year, per month, whatever it is. That's the way I look at things. I'm just like, well, I can just build that myself now, you know, and slowly I think companies are going to be doing that, you know. Does that make sense? Like there's two aspects to what you're saying there. Yeah. Yeah, definitely. I I'm doing that that as well. Like I'm building my own software. Um I'm just looking into like uh starting from a open-source uh version of an LMS for for like a course um platform that I've been using and paying for. I'm like, I'm not going to pay for this anymore because I want to do things with it that I can't. Yeah. What should I do now? Pay them and not be able to do the things that I want. [ __ ] no. I'm going to build my own and it's going to take me I don't know. Next week I'll have it ready. Yeah. And and doesn't matter if it has bugs. Doesn't matter if it doesn't look or behave a certain way that that you don't want it. One, you'll fix those bugs in time anyway. Two, it's not a high security risk. It's just your little Yeah. And and the point is also very often, this is what I'm seeing recently so much is it doesn't like my app doesn't need to do as much as Exactly. the the the best app in the market and [ __ ] no. Like my app needs to do the things I want to do the way I want to do it. And 99.99% of the time, the apps that are out there um are just overblown with features that I will never use. So why why am I paying for those? My um my website's going to be having a massive update literally before the show. I was I'm rewriting the content all around these things. I need to I need to work on the copy a little bit more. But it is about why pay all this money for software only does part of what you need and any features that you do want take months and months and months to get fixed and you're per perpetually bound to paying for that software for the life of your company. How about pay a smallish upfront cost that's then software for the rest of your life that you own and that if you want a new feature, you get exactly what you want within days as opposed to potentially sometimes even months for a lot more money, you know. So yeah, you know, um I think we're on the same page with it. So yeah, that's interesting discussion. It's I thought I went going into this article I actually thought it was more about the amount of SAS and whatever getting released and it does touch on that but realistically it's talking about the funnel of inside of organizations from code to to release um how that is potentially being squished a little bit or or there's a concern around that. Underpinning all of that is like what the [ __ ] are they even building anyway? [laughter] Yeah. Why are they even trying? Yeah. Yeah. So uh there was this uh interesting video from Primagen saying that um they will probably like these companies will probably have something like AI credits uh for the team or like the individuals and the goal of the companies will be to save money on these credits. So you as an individ so you work for a company you have like $10,000 uh worth of AI credit uh if you use it uh you can but if you don't and save the company money you might get uh all that money that you save for for your own pocket but still get your work done. So I thought that that's also like very an interesting idea that that might come true. Yeah know very good incentivization. It really it really puts the power in the developers hands. Do they want to buy themselves time or do they want to buy themselves or do they want to put money in their pocket, you know? Um, and it comes up it's up to you. And that's all it's doing is buying you time, you know? Yeah. But like I I'm not really sure about the time like I'm not sure if AI can make you that much faster in every area. I like it. I see AI as an enabler, but I don't see it as making me that much faster in every case because at the end of the day when it comes to the like that last 10% that's where you you see oh [ __ ] like so many accumulated uh depth decision depth that's like AI so you let the AI to to build whatever you are building or design whatever ever you are designing it looks good at the start but like at the very end like at the 95% where you want to push it um that's where you you see there is a there's a human yeah there's a depth like decisions that you didn't make you let the agent to do it and now you have to deal with them and you see I see this everywhere so it's not making me faster at the end not that much but it's making me faster at the part. Yeah. And I think again that's that's probably what this article is is trying to say really like is it is that actually speeding us up? Is it just creating a debt that we have to clean up later on? Is it costing us x amount or whatever? So, but a nice analogy with electricity towards the end saying we we might just still still be figuring this out. So, you know, it's um Yeah, it's interesting. Yeah. IPO. Oh yeah, we we got an IPO to talk about. So, holy trinity. A trinity of um there's probably more actually to be fair, but Oh no, my browser doesn't let me share my screen. Oh, you got to quit. Uh oh. I hope not. What? What is Let me s my light while you do that. For some reason, it doesn't recognize um the app properly. Okay. Might be a bit better. Is it? Is it better? 60 people. How you doing? If you're on LinkedIn, if you're on LinkedIn, imagine if you're on uh YouTube, give us a like. That would be lovely. Share this with your friends and family and tell us why you're here. Tell us why you're here. We're starting, we're trying to start up a um community where we don't know what we're going to talk about, but like an insiders community where we can talk about AI obviously, but whether it's like tools, techniques, something like that. Um, have I got a banner here? Here we go. Go to that URL and let us know why you what you would theoretically do inside of a community, whether you want to join it or not, what you theoretically want to learn about. Um, yeah. [sighs] Fair play. We got a good We got a good mix of people from all over. If you are on Twitter, we primarily show this on um on YouTube and uh throughout the week we release a bunch of clips and things like that so you can yeah catch up on what you might have missed. He's back. And yeah, another um I think this is like an issue with my agents. Like there was another instance of Chrome open. So I'm not sure if the the start of the show was I I don't know if it's like in the same [snorts] file in the same if it will be uploaded or not. I don't know. I have no idea. Oh, we we have Hi, Austin. [laughter] Hi. Great to see you here. Who's Austin? Uh uh well we we are we are working together on project. So um yeah um well he is to be more precise he's giving me a bunch of like web flow work like awesome projects to work on. Yeah. Good. Yeah. Cool. Well, um, let me actually get to the strategy. So, we will be talking about the IPO, the a topic that I'm like less [clears throat] than the others to I'm not really qualified to talk about this, [laughter] but it's interesting. The reason why I gave you this one is because you're at least more interested in this stuff than what I am. Yeah. How much more interested is yet to be decided? But yeah, I I am more interested um yeah, we'll we'll talk about it about the whole thing. So, okay. So, apparently Anthropic has um filed an IPO. Well, not exactly the IPO itself. It's a draft. They are like you have to submit a draft first to the SEC from what I understand to do the IPO later this year. So they are officially doing this before open AI and if if you've been reading about this there seem to be like a race going on between u SpaceX even Google I read and open AI and Anthropic. Everybody is trying to get a lot of money like private money uh to [laughter] to keep delivering us tokens so we can burn [laughter] on nonsense software. Uh it seems like because these companies are burning money and they are now uh we see that Anthropic is uh filing this to um to go public. So uh something very interesting about it. I read that in March this year Anthropic was worth 350 billion but now they are worth almost a trillion. That's a significant grow. That's crazy. So uh what's really interesting to me is not just Anthropic going uh to do this IPO. It's also SpaceX and like what does SpaceX to do with anthropics? So SpaceX um it it was a space company uh but not so much anymore even though they they it sounds like it they are an AI company because SpaceX owns XAI which owns Twitter and all of this like Grock and everything. So, uh, Elon made a bet of buying a lot of compute, a lot of GPUs, setting up, um, data centers. Funnily enough, the guy who accelerated the electric cars for the planet, he has, uh, these giant data centers. And because there is not enough electricity to draw from the grid, he's using gasoline uh, generators to to make the electricity for these like giant data sets. It's crazy. So, SpaceX um is going to do their IPO at $1.75 trillion. It's in It's just insanity. Uh Anthropic is going next. Um and OpenAI wants to do this ASAP as well. There is a race though because there is just so much money available and like you if you are going to spend some money you have like I don't know like you have like I would guess like you have 500k in your bank account ready to spend you don't have more so you you have to choose between these companies and you probably will choose the one who goes um to do the IPO first you spend your money on that and then there won't be as much money left in the market. That's the idea. And there what's quite funny, it seems like Elon is trying to stick it to Sam Altman by working with it's just a giant loop of like companies working with each other. So Google is buying a billion dollars of compute per month from SpaceX because Google owns about a hundred billions of shares in SpaceX. So they are giving them a contract because they want to make their business look like it's booming so they can go do an a big IPO. Anthropic is also now in business with uh SpaceX because they don't have enough compute. So yeah and and at the same time they are in a race. It seems like it funnily enough uh open AAI and Sam Altman are left alone. It it seems like at least to me like that because um even like Apple as we saw in WWDC they last year they were uh in business with open AI and they had chat GPT and Apple intelligence that never came out. This year they are doing it with Google and with Gemini. Uh and Microsoft is basically getting rid of everything open AI. So, OpenAI is being left alone, left out alone, um, however you say that. And these other companies are like forming this um, ba basically a race for who can get the most money from the public. Uh, yeah. So, that's kind of like the news. U, I don't know if you have anything to add to this, but it's just a big insanity going on right now. Hm. Well, maybe you'll be able to answer this or maybe not and maybe why I'm not so interested in it, right? Can we break down what it an IPO is and what it what it means? Because to me, it feels like a bit of um in a nutshell, it feels like to me that they can't rely on investors anymore. So then they allow the public to buy a portion of their shares to effectively for for you and I to invest in the company. Yes. So IPO stands for initial public offering. So a company like SpaceX for example uh Elon Musk for example comes and forms SpaceX. it's it's his company and he he either puts money to uh to in the company from his pocket or he goes to a few private investors and he says like okay the company I'm going to give you like 5% for $10 million um whatever like and then they make an evaluation of how much the company is valued at based on like a bunch of metrics like how big the targeted uh the market that they are targeting is how much like what they can output and how much wins they can basically have in the future. But right now the company is losing money. Normally when you do this with private companies with private investors for example if you if you come to me and say hey I want you to give me some money. What I want from you is to give me some assurance that I will get my money back somehow or like you know I I will make a big profit there because you show me your numbers, you show me that you are expanding and you're building something and this is what normally happens but right now that's not the case. So I I will get I will get to your uh answer to your question in a second as well. Right now, as we know, all of these AI companies are literally burning money because they are giving away tokens. They are doing a lot of research. Uh they are um they they they are growing but they are not making profits. They are losing money because the the cost of the GPUs and the cost of like buying the the hardware and doing the research is just so high that at the end of the day they are losing money. uh there is there is a hope at least that's what they are trying to sell. So for example Elon Musk comes and says I have SpaceX and we're going to build like these data centers in the space and the the market is going to be this big and that big and you'll make your money back. The the crazy thing about it is that none of SpaceX is uh really making any money except for Starling. So all different parts of it is losing like money big time. So it's like a small division that is kind of like uh fueling all of that. So they need money. They need more money to build more data centers. How do they get this money is by doing the IPO uh public offering. So coming to you and saying like we have this big opportunity buy in buy in now and you'll be good in the future. And there from what I understand like uh SpaceX is offering like 5% if I'm not mistaken or like a really small percentage of the the the total company. So it's not like when they go public it's not like you can buy the entire company. It's like a small portion of the company but the whole company is valued at a very big number 1.75 trillion and normally that value is you know it depends on your numbers but right now it's extremely in um inflated like it's like uh I think if I'm not mistaken like 100 times uh more inflated than a normal business so that's the risk that comes with it that means means if you buy this um you are being promised this 100x growth that might never happen. Yeah, that's the thing because because at 1.5 trillion like where can it go from there and if you and if I understand this correctly if say we go in at you know 1.5 trillion but actually by the next year it goes actually down to half a million half a billion or whatever then our investment is is smaller then it's like I'm trying to think of an analogy So you pay a hundred bucks and you buy a portion of SpaceX or whatever one of these companies and then it tanks like you buy it at a 1 billion uh evaluation and then the company goes to a half a billion next year. So your hundred bucks is now worth just 50 bucks because if you want to sell it and take your money out, the amount that you sell it for is half the amount you buy it. And that's basically how it is. And there is this big worry that these companies will tank because Yeah. Yeah. And like, you know, going in uh like I say, going in at 1.5 trillion, like I just I can't see how it's going to grow from there. It's already so big. But there must be a good reason for them to want that company valued at that amount. Um um not really not like there is not that that's the point like every like every video every an like video that I watch about analyzing this is like there is no reason they are they are just like inflating the numbers it's and again like Google is buying compute from SpaceX because Elon got more GPUs but Yeah, they they have a big incentive to do this because they own 100 billion. So they they make the business look better and the the contract that they have can be uh terminated within 90 days. So yeah, that's not really like if you buy into SpaceX because they have a $1 billion contract, you need to know that that contract can be terminated within 90 days. That that's not safety for you. That's not safety for your money. And you know, all this we said and this isn't going to turn into some sort of financial thing or you know, again, we're not qualified. This is not financial information. Like the part the point of all this that interests me is personally I have a few I have I've invested in a few companies I thought would do really well. Tesla invested in Tesla cuz I thought it would do really well. They did fantastically and then I sold because I made a bit of money off of how [snorts] much how I don't know whether you remember in 2020 2020 2020 2020 2020 like Tesla went [ __ ] massive. I would not recommend investing in individual stocks because this is ex what you've just described is exactly the difficulty with investing in individual stocks. you do not know or you might you know there are so many um uh aspects outside of your control that unless you're really wellversed on the intricacies of what an impact of one thing might impact another this is the sort of uh knowledge that you should have when you're investing in an individual company like properly you know you you put all your eggs in that basket to say because of x y and zed I think they're going to do really well the reason where I'm interested in [snorts] is where this SpaceX IPO is they are forcing some of the so I'm in index funds all that to say that I invest in index funds which is a selection someone else is doing all the selection of who to invest in and why there's algorithms out there and Warren Buffett one of the greatest investors of all time says like he would put all his money in an index fund if if he you if he could put it back or if he could advise any advise anyone, index funds are the way to do it. So, you don't have control. You have where your kind of money where which industries you invest in, but you don't really have control of the individual stocks. SpaceX because they're valued so high. A lot of these different index funds, because there's tons of index funds, S&P 500, S&P 100, NASDAQ, all of these index funds, they're eager to get such a valuable company. Well, S&P 500 is different. NASDAQ are eager to get um SpaceX part of their roster because it's looking really tempting like a lot of people going to want to invest in SpaceX through their index funds and they think it looks attractive. So they're bending some of their rules that they have in place. What normally happens is if you IPO, you got your you you you join an index fund, your performance is then um managed or your performance is then uh uh reviewed after a year and then you actually get let into it, right? Then you get let into it and then they'll start putting money in because IPO is short for uh something like extremely [ __ ] overile. got over something like that. Often companies are overvalued when they go when they IPO. So that's why they have that period. Yeah. They make these companies a lot at the best time. Yeah. Exactly. Right. Exactly. When it looks good. So yeah. Um so the the these comp these um index funds are changing their rules to get SpaceX earlier down to like six months down to like 3 weeks when they haven't proven themselves to be um a worthwhile company to invest in because they haven't got that track record. They haven't got that year, six months. um and they're they're looking to bend their rules and people's money which they have no control over could be going into a company that could completely collapse, you know, and then they've lost all that money. And the reason why you go for index funds is you go into index funds for stability. You want that reliability. Luckily, the S&P 500 aren't budging from their 12 month kind of review period. So, that's good. And it's not confirmed that it's going to be the case. Other thing to rest assure is only a small percentage something like 5% or something like that actually gets bought um into the index fund. So it's going to have a very minimal impact on your overall portfolio in that index fund. However, it's a slippery slope. However, it's it's just unnerving to people who want that um uh predictability. So there's where my interest lies in all of this stuff. I'm not ignorant to uh individual stocks and and IPO and stuff like that. It's just I'm yeah I don't play that game, you know. I don't play that game. I just I I I I invest in uh index funds and I'm done with it. But it's quite funny to think about it. Uh if you buy into SpaceX, you are buying a piece of Twitter. Yeah, I know. And I saw in their own blog, maybe it was curses, but in in an official blog, SpaceX SpaceX AI, I saw the name as SpaceX AI. So, I don't know whether they changed it or whether they're going to change it or whether that's just a typo, but it's Yeah, it's crazy. It's crazy to think that. Well, I am cover AI. Gaba AI and Sam AI. AI Sam. Um Yeah. Anything else on that? Uh, no, not really. None of this is financial advice. Yeah, exactly. Exactly. Yeah. I I could talk your ear off about like I'm a big nerd when it comes to like savings and whatever and that. But yeah, startup stuff is a bit, you know, bit bit loose on that one. Um, I did find some stuff on Kimmy K2.7 whilst uh whilst you were doing that. So, let's just quickly go over that before we continue on with the show. Um, only a tweet. Only a tweet, so I wouldn't get too excited. But Kim K2.7 code. Okay. So, it's look, it's a special. It's not Kim K2.7. It's actually Kimmy K2.7 code, which makes sense why they're trying to push you to use Kimmy code. Our latest coding model is now released and open sourced, which is great. I'm going to be looking at this on hugging face. Um, improved a coding agent over 2.6, 21.8% 8% on Kimmy code bench, 11% on program. By the way, there's some benchmark comparisons we can look at in a sec, but there program bench 31.5 on MLS Ben uh MSBench liked reasoning less overthinking with 30% lower reasoning token token usage compared to chem 2.6 and long horizon coding. All of the stuff you'd expect. Six times high speed mode coming soon. Do you know I I I had a had a um brainwave today. is like when I know I'm going to have a light day coding whack on fast mode. Like why am I waiting around if I'm if I know I'm just going to be touching a few bits of app, you know, I'm not I'm not like on a full-time job when I do dip in code. Why am I wasting my time? Like [ __ ] stick on fast mode. Six times high speed. It's pretty good though, isn't it? How does it compare though? I really see this. Well, Kabaza, I'm glad you asked. Here are the benchmarks. um not not particularly Kimmy Codebench their own benchmarks which you know at least they're not saying it's up here you know compared to um you can see this right you can see this pop up of the yeah I'm looking at it on the stream um but yeah you know again we've got to remind ourselves these are open source models so these are um even though it's behind we need to give them benefit of the doubt uh program bench Um, again, I think it's just a a roughly, you know, 5 10% increase from Kimmy K 2.6. Not not beating any of the Frontier models, but it's still going pretty good. Um, those are the benchmarks that we have now. Anything else? Give me code. I might actually open that in a new tab. And well, actually, let's have a look. Oh, it takes you through to that page that we kept getting. So this is and this looks like I think before they were using Gemma Gemini CLI, but this looks bespoke. So I might actually have a little play with that. Yeah. Back to here. But yeah, so I'm going to and actually I'm going to look at on hugging face as well. Um the thing that I'm and this is a narrative that I haven't quite um discussed yet on my own channel, but we know that curses composer 2.5 was trained off of Kimmy K2.5. So they took an open source model uh trained it with re retrained it with reinforced learning and stuff like that and they got a really good coding model out of it. So that basically that tells me Kimmy is a really good starting point for a coding model and what I really want to get into. So stay tuned for this. I want to get into and see how feasible it is to train a custom agent on my product or my code stack. Maybe I've got maybe like Web Flow as an example. Let's let's train a Web Flow agent that [ __ ] knows Web Flow so well and you base the source model off of Kimmy K 2.5 cuz cursor have already done the research there. They didn't take Miniaax, they didn't take GLM, they took Kimmy and they trained compon and they got some really good results out of that. So it was a good base model for this. So I that all to say that I do want to get into pre um you know training models and and Kimmy K 2.5 seems like a or Kimmy even seems like a good model to to get that off. Um I was going to see I always forget how to look at this like the file size. Um no this all breaks it up. I just want to see how big the thing is. File info. Here we go. Can I run this on my machine is the question. Oh gosh. on the new M5 Ultra Max Pro total size. Okay, I'm guessing that's in bytes. Uh yeah, this is all What's all going on here? Should I download it? See what it says? Oh no, that's just one file. That's annoying. I will find out. 1.1 trillion parameters. So, it's probably going to be probably about 500 250 500 gigs, something like that. I want to find it. I want to find I do want to find it. Oh, there's been some quantizations already. What What about the left side? There was something. I'm not sure if it was the number you were looking for, but I Yeah, here. Just scroll a little bit down. Model summary there. um active 32 billion active parameters. This this is quite good to go into actually. Yeah. So 32 billion active parameters. I saw somewhere that um I think uh Claude has like a trillion active parameters or something [ __ ] crazy like that. Yeah. Um benchmarks. Uh, no. There is a way to look at the size of the model like the actual Why isn't this Why isn't this um showing me the total? That's what I'm uh we just do look at the we as humans we are really slow. We need to not think and ask AI to think for Exactly. Exactly. [laughter] Um I'm looking at um the quantized version how big the quantized version is right now downloads an option. What is the quantized version? So the quantized version this is the quantized version. It's quantized to 3.6 bit and it means that they're just trying to remove what they believe to be superolous uh knowledge from the model. Uh by the way there it is. What fits into 460 GB memory. Where are you looking? Oh, yeah. There we go. Yeah. So, 500. Yeah, there we go. So, I do not have enough money. But this will this will load on a Mac Studio. Do you know what I mean? So, that's uh that's pretty cool. You need to have this much memory like RAM. There we go. Yes. RAM. Yep. Yep. Yep. Does Yeah. It doesn't work if you do like share RAM thing with the SSD. No. Um, so it well no it needs to load in it need you need enough in the case of Max you need enough VRAM to support not only do you obviously need 460 gig of hard drive space that's half a terabyte. So if you've got a 1 TB hard drive that's already taking up half, then you need, you know, at least there there is no well, yeah, 512 gig. You need a 512 gig uh RAM from on a and the only one to do that is a Mac Studio. The only computer Mac, the only Mac that allows you to do that as a Mac Studio. I'm going to look it up what that costs. It's probably about 10 grand. Wait, uh, that has 96 GB. So maybe I'm looking at the Ah, okay. By the way, yes. No, so here's a point. I looked. Yeah, they're only selling 96 gig of RAM Mac Studios at the moment. So they're either looking to update or there's a weird like holding back at the moment because of um maybe they are all sold uh sold. Maybe they're all sold. Um, and as soon as I saw that, I freaked out. And that's when I put the order in for my MacBook because I'm like, how many? Yeah. How much do you have? 128. That's crazy. Yeah, that's crazy. And a 16 or 14? Uh, 14. 14 in. Why? Why do you go with a 14 inch? I don't like it. I don't like how big and bulky and heavy it is. It's so nice having a a little 14 inch that I can sit in bed with, um, do a bit of work, travel around with, like, and by the way, we've got the quantized version on the screen right now, which brings it down. So, 460 gig on the quantized version, um, you know, it's hard to even say, but on the uncquantized version, you can imagine this to be quite big. So yeah, I think the quantized version I think they had 3.6 was the biggest amount of quantization that they could have to get it onto consumer hardware because you can't really get much more like Mac at 512 is probably the best you can get on consumer hardware. Then you start to get into real like high-end [ __ ] So, and ju just to just to tell you as well, quantization, it starts at 16 floating point, then it halves to eight, then it halves to uh four, then it halves to two, and that's kind of generally what you get, but they've obviously cut it a little bit, so they've brought it down to 3.6 instead of four. So yeah, all that to say, this is the one you're likely going to be running if you got that Mac Studio. I see. Um Yeah. Yeah. Unless there must be a service out there where someone else is running the model and you're just you're just using it. Like I'm not talking about like Anthropic who are doing the same thing, but then it's their bespoke closed source model. I'm talking about like regular people just doing like a network of computers that that this model will run on or something. There must be something out there for that. Yeah, I'm pretty sure there should be some. Yeah, like that you book it in or something like this. I don't know. I don't know. Anyway, anyway, um Kimmy K2.7 code is out. You get access to it. Let's actually do this. You get access to it through Kimmy Code, uh, their new CLI, which looks like it's been built from the ground up. And it starts at 15 bucks a month. Um, oh, sorry. Yeah. Yeah. Give me code looking like this. Uh, 19 bucks per month. Sorry, that's if you spend it annually, it's 15, but yeah, 19 bucks a month. So, yeah. and you'll get a lot more usage out of it than uh than claw code for sure. Cool. Speaking of which, speaking [snorts] Oh, no. We have Apple first. Apple. Yeah. Jesus Christ. We're going slow today, aren't we? Yeah, we're going slow. Right. We need to pick up the pace or Yeah. [snorts] Or do we do we skip Apple? What do you reckon about Apple? Do we skip it? I don't know. We got We got so much to get through. Oh yeah, we have the web flow. We've only done two. [laughter] [gasps] Oh, right. Should we should we just should we just do Apple? Let's just quickly go through what I wanted to show. All right. So, we just had WWDC, Apple's conference. Everybody, you've seen the videos. Everybody has been talking about it. Something that I haven't seen anyone talk talking about is the map. I really like this update. So for now it is just for the is it called fly through something like this. Look what flyover fly over. Look what they've done. So this is the normal map that you get which is already quite impressive. And the fly over uh now gets this update for the supported cities and everything like see how good this looks. So they are I don't know using some real photos I guess to maybe map this uh create this like very very rich uh 3D again this is not for the like regular map but I guess this is something that will happen sometime in the future. Uh, and I've been actually like thinking about this be even before and I'm I'm just like I was just thinking like when why nobody is doing this basically? Well, I guess compute compute is the the reason here but like we have so many images of the world and we have these 3D maps. Can't we just, you know, just do it? Um, and Apple's it seems like has started uh doing this and it looks so [snorts] good. So, yeah. Um, yeah. So, fly over is like a specific feature that only appears in landmarks like the Eiffel Tower. I'm guessing this is New York and whatever where it just circles across the city and just navigates through the city. Look at the Grand Canyon one. I think it's Grand Canyon. It's somewhere. It's if you go back in the video. Oh, no. They haven't. Do they show it here? No. No. Go back to the video. They did. Go back to the stream. Rewind before this. It's only a little bit before this. It's literally Keep going there. Oh, maybe. Ah, that that's for the special. Yeah, spatial. Oh, okay. Yeah, it's not the map. Oh, okay. Yeah. I thought that was it. That would have been cool. So, they are creating Yeah. this fake depth thing here. They they've been doing this with the photos on like you literally have this with your photos on your phone that you take. You you can create this fake depth and now you have this in your vision OS and you can actually take panoramas with with your iPhone and then use them as your workstation whatever like uh wallpaper when you are in your vision OS. That's really nice. That's really I would like a vision OS. I feel like Yeah, I would like one just but everybody that has one and like you watch the videos, they are like right like after a year of using it or not using it so much, they just use it for some entertainment because you know it's heavy and yeah, the monitors that that you get with this is not amazing. It's great for flights. So, everybody's like, if you're on a flight, you bring your laptop with you put this on and then you have like a big monitor in front of you or like multiple. That's great. Um, yeah, honestly, I wanted as well, but uh yeah, it's it's not that useful. Yeah. Other things that we had in WWDC, what what else did like capture your imagination? I was going to ask you the same question. [laughter] So for me, I really like the the photos, the photo edit thing, especially with So you can do three things. Clean up that you could before as well, but it was so bad. There were memes about it like people doing it with uh like their hand in front of their face and doing it with Samsung and with Apple and Apple would just create something cringe. Uh and Samsung was really good. Uh, so cleanup is there and supposedly it's going to be better. Extend is really interesting and I just want to do this 50 times. I want to try it. I want to take a picture and then press extend as many times as possible. I want to see like will I get to the Mars or no at some point. you have this refframe that it has a really interesting UI that creates this like blur, but you can basically regenerate a photo that you never took. [laughter] So, the nerdy the nerdy thing about that is that it uses the ondevice AI to give you that blur and to give you a preview and then when you hit okay, that's when it sends it off to their u remote um uh what do they call it? Apple private server or whatever they call it. Then it uses the online model. Yeah. To then actually do all of the heavy duty stuff. Yeah. Another interesting point of the the whole conference or not so interesting was Siri AI. I know people are very excited about it. It's just making Siri I guess decent. But what was boring to me was that the whole when when they were like showing it, you could see that they are like saying something and just waiting for Siri to kick in and give them a response and often it was in text. I don't know why why they were not doing as much voice as they were doing text in the conference. So, I don't know. It looked boring, but I guess Siri is which has Gemini underneath is going to be better, but not for us because we are from Europe. I don't know how how like are you you're I'm waiting. No, we're not Europe, but I'm waiting on my Siri AI on this on this machine here. Yeah, my open. So, I'm not getting it. I'm not getting Siri AI or like most of these AI things I would say for now. For now, they're working with the appropriate pies. Yeah, let let's see. So, yeah, with with Siri AI, it's just like everything you expect from a modern uh assistant. What's good, like Apple has the context, so it can like search your entire messages, but I'm not sure if it can do like WhatsApp. Can it search my WhatsApp messages or does it have to be, you know, Apple product um basically data? We don't know that yet or I [snorts] don't know. Uh but yeah, that that's interesting. Uh another few things that got me uh excited, Safari extensions. I don't use Safari, but I will give it a try. Like if I want there are so many times I want to make an extension for a specific thing that I want to do. Maybe I'll try Safari extensions but I bet Google is coming up with something similar to this uh soon for uh Chrome as well. Why not? Um prompts to make shortcuts. So you know the shortcut app that you have that you can do automation. So if you arrive at the gym like I don't know like make my Apple Watch face this Apple Watch face or start something um these automations you either have to make them from scratch or you can actually download them from the internet as well based on like templates people other people have made but now you can prompt them into existence. Uh just it makes so much sense. I'm surprised that they didn't have this with iPhone 16 already last year. Um, yeah, that's pretty much it. Yeah. Yeah. Pretty good release. I mean, I I'm just overall pleased about the context, general context awareness that you can either take a screenshot of the page or not even take a screenshot, but ask Siri about the page. The thing that I and I got a video on this on my channel, but the video the thing that I was waiting for that I would like is like actual agentic capabilities on your iPhone or on your Mac so it can do things like take an attached file and email it to someone or, you know, do things across the experience, but it seems like it's it's going to be locked into the APIs that that Apple give to you. Um, so yeah, I can't say I mean I can share I can do manual processes but I can't get the agent or I can't get Siri to like do things in get put stuff in other apps and take action within other apps and use other apps. You know, it's going to be probably limited to all of the Apple ecosystem stuff. But um yeah, overall like I'm happy. I mean the uh voice dictation is better. So, RIP Whisper Flow on iOS. I was thinking about that. Yeah. Um, you can change the voice as well, but this I mean that's not a big deal, but like there'll be two variants that can run on your phone. There'll be like a small three billion parameter model. Um, my guess is just for basic spellchecking and definitions and basic lookups. There's not a lot you can really do reliably with a 3 billion parameter model. Um, and then there'll be a 21 billion parameter model that's not mixture of experts. It's a mixture of focus sparse focus attention or something like this. Oh, no, that's Miniax. Either way, it's they they've got their way of um activating three billion parameter models at a time um to with with a specific expert in the field. So, but it's not quite mixture of experts. So, but only the iPhone 17 will get that. and uh obviously you know M M5 iPads and whatever. So, uh, and then it hands it off to like a bigger model. But saying that the voice is only available on the 17 Pro because for whatever reason, see, that's how bad a parameter model does, uh, is, is that it can't even change Siri voice, you know? So, God knows how useful it's going to be. And like I said to you before the show, like it'll be interesting to know where the handoff point is, whether we can actually track that it's handing off to a bigger model, you know, whether it and then eventually it gets to the private um cloud versions of the models, which by the way are all private. This is all privacy focused, which is quite nice to be honest. But there will be limitations. Whether you're willing to sacrifice those limitations for the sake of privacy will be down to you. Basically, you vote with your wallet. Yeah. Yeah. [snorts] Um, last thing Liquid S, uh, they they gave they made an update. Now you get a slider and you can change the amount of transparency, uh, you get you can make it almost completely opaque. Uh it's very interesting because right before this happened uh I there was an article I can't find the exact article but basically a group of people uh from what I read they achieved like actual liquid glass for the web basically working everywhere. Um, so I I again like I didn't find the exact article, but I I did read it and yeah, it seems like we have that pretty useless effect. It it's not really useful, right? Uh, it's nice. It's like very impressive technically, but it's not really useable. Maybe maybe less in like as Apple use it in like little UI elements rather than actually these whatever they maybe not in functional UI elements but as an effect. Well, I'm well or I'm saying the opposite. I'm saying maybe we'll see it in functional UI. Um yeah, but it's not functional. It's bad. Like it's bad for readability. And when you crank didn't stop Apple. [laughter] Yeah. Yeah. Well, the guy the guy who who did this, we know he he's now uh ruining Facebook. So, yeah, whatever. Um, liquid glass in the browser. I think this is like the I'm not sure if that's the the article. No, this is not the article that I read. Uh, it's interesting though how they obviously came up with everything. Yeah. But but apparently it's working now because I remember at the start everybody was trying it but there was it was only working with um either filters that would not be supported in Safari or not on on mobile but now it's apparently working everywhere with good performance. Cool. Very good. Okay. Uh, platform didn't take too long, but yeah. Are you taking this one? Uh, at least the start because then you need to So, [snorts] how what is our what is the the intro? Web flow have updated their developer platform showing that they're going all in on developers. I guess developers got the [snorts] Yeah. Okay. So, Web Flow Yeah. So, Web Flow has updated their developers uh platform putting all things developers in in one page essentially like in one place uh to learn from and like I just want to share the site first and then Sam you will explain what this is because I'm [snorts] not really qualified here enough for this. The site is really nice. Look at this animation. I'm just I'm a sucker for this. Like look at this. This is so good. And this hover effect, it's really really really nice. And [snorts] what this is essentially, this is developers.webflow.com. It's a new uh home for developers to learn about everything development uh with Webflow. So from APIs to webflow claw uh cloud uh apps deling their CLI. I want you to explain this a little bit to me because I I'm interested but then I don't know enough about it what I can do with it etc. Uh anyway so that's the page and I want to show this quick thing. This is like really fun. There is like a foosball you can play. So yeah, which is actually really hard. [snorts] Yeah, the Yeah. [laughter] Well, this is what you do when you have too many tokens. I was going to say this is definitely the work of a dev powered by AI. Yeah. Yeah. You have so many tokens you don't know what to do with. This is what you do. Yeah. Anyway, so would you explain what this is? So like I can copy this npm install webflow CLI. What is this? What what can I do with this? So Webflow CLI allows you to basically manage uh aspects of your web flow website from the command line. So you can publish a website, you can um take a page down, you can uh draft a post, you can create I mean I I'm look well you can even go down if you um click on CLI on the Yeah, there you go on the left there. You can do all this sort of stuff that it shows you there. So quite a nice You can manage sites, assets, forms, things like this, CMS content. you can do all that sort of stuff. Um, which is probably s um uh not linked to but like you the API will allow you to do a lot of this stuff too. So you can not all of it but a lot of it the API will allow you to do which I was going to go on to say which in turn is what the MCP allows you to do. manage sites, take pages down, do a it's all just diff, you know, you're not you're not going to use all of these things, but there'll be certain things you just like pref you prefer the CLI or you prefer the MCP or whatever. So, there's just a a lovely choice of um tools that allow you to interact with your your uh website. The Can I tell Claude to use my the CLI? The MCP? Yes, you can. Yeah, you can ask Claude to use the CLI. Probably doesn't know much about it. You're probably going to have to use some sort of like skill or download the source code if that's available. I'm not too sure. You can copy it here. This is what really what's really interesting. You can Yeah, you can go download or like copy the everything here with at webflow at developers.webflow.com as MD files. There we go. So then you can teach your teach Claude how to use um uh the the CLI. so the CLI can can use it. Um, you'll obviously have to be aware of what the CLI can actually do. It's not magic. It's not going to be like, oh yeah, I I'm I'm able to transfer your Web Flow website to another account. Might not be able to do that. You might be able to. Point is, it's like it's not just a magic the all the agents doing is is doing stuff that you can already do either through the CLI, through the uh SDK, or through the API. That's all it's that's all the agent is doing for you. website different ways to interact with your with your website. [snorts] Um, on top of that as well, you've got things like dev link and code components and web flow cloud and and and apps. So, these are these are way these are what do you even describe them as? Like tools uh that take advantage of web flow features rather than managing your website. You you can build co- components from um the command line. So not out not in the Web Flow UI outside of the Web Flow UI. So I do wonder um what the difference is between and there is I'm sure if I sat down and scratched my head what the difference is between code components and Devink because components are React components or also Dev Link is. So are they going to are they going to replace Dev Link with code components? Are they going to keep them? There must be um there must be a reason why you would use one or the other. So, so the idea with dev link is you you create it in web flow and then export it from web flow to an external react app where the designer is um the source of the truth uh web flow designer and you or like me as a designer I control the component and you as a developer you pull it from web flow as with uh code components it's the other way around you as a developer uh you build it and then you publish it to web flow and then I can pull it from my component list um as a code component but that these are getting as you guessed right these are getting like really it's like the mix and foggy between them because we already have also uh AI component gen the naming yeah so that's just a way of creating compo code coco components but through the web flow interface, but exactly. Yeah. You'd think that they would abstract the technologies a little bit and just give you one dev link API, let's call it, and then actually if you're as a designer, um, if you're going to if you say want to push your component down, you don't necessarily care whether it's coco component or dev link, you're just like, I push my component and my developer can pick that up sort of thing. Yeah. That all being said, it's it's a part. It's um it's separate in as it currently is, but this this hub gives you the ability just to learn about it and learn how to use those sorts of things. And obviously, if you dig in, it will probably give you a good better clear definition than what we've done about what those two two uh separate why they're separate basically. Um and then of of course like the marketplace apps as well, which is the other one. So just an all-in-one place as a developer that you're not really concerned with the the Web Flow side of things. You you live in your IDE, you live in your terminal and you can do things or publish things to your Web Flow website to enable the team basically. So the the amount of effort that's gone into this hub shows that they're keen to support developers. Um definitely feels like yeah like there it felt like for a moment that they were with Web Flow Cloud it just didn't feel like it was getting the love that it needed whereas I feel like this puts a little bit of a spotlight on web flow cloud which by the way is really good. It's a really good technology. I don't know how many people are using it but it can be very very useful. They just it's just people is it mature enough? Web flow cloud. I mean it's out of beta. I'm pretty sure it's out of beta. So it just you just upload an app. You just build an app. You should try it. You should try building just an app that you push to web flow. Well, I have many apps, but I'm not sure if I want to push it to web flow because you know like if I'm making just a tiny app for myself, I'm actually putting it on Netlify or Versal for free. Yeah. Well, the idea is that it's Well, no, you obviously Yeah, of course. The idea is that the the app that you push is feels seamless with the marketing website that exists around it. So it's not the I mean theoretically yes the whole app the whole website could be your cool little app you've built but it could be that you built a marketing website um advertising your new SAS product which is going to die to 3 to six months. You've got a little dashboard icon dashboard button at the top left or top right. You click on it, you log in and that's your next app that you've built and it's yeah your dashboard or whatever it is. Do you know what I mean? But it feels seamless. You use the same components like the the well that's where you would need dev link to start using the same components and things like that or coco components and vice versa. So like you know if if you're if the designer of the app and the experience let's call it the experience they've made the marketing website and they want to use some of the similar assets in the um app then they would expose those or give you access to them with dev link saying that oftent times the marketing website it just appeals to different people than the actual dashboard and app itself. Um, but let's say for example, you're a um, what do they call it when you like a let's say you're you own a load of co-work spots and then there's an experience where you can search all of your co-work spots like that you want to feel quite integrated. So, the actual searching of co-work spots and whatever. What's what's the word I'm looking for? When when when you bring lots of stuff together like you're in Germany, so I wouldn't know any examples, but like yeah, you just bring lots of different companies together. Like, if you know what I'm talking about, Zupla, like they bring all of the different housing markets all together. And there's a word. It's a bloody word. Never mind. Do you want the listing? Yeah, like a list kind of like a there's a better word for it, but there's a Yeah, like a listing website. Yeah, like yellow pages or whatever. You're bringing everything together. So, you've got the marketing website. You want it look it to look the same as a front-end experience that you can search and do all this complicated [ __ ] with. That could be a Astro or Nex.js app that's the designer has designed all of the components that given to you through Dev Link and then it's pushed back up into the experience through Dev Link. So very very powerful, very cool if you if you've got the right use case for it. Yeah. Uh speaking of apps at Webflow, right now if you want to make a new project in Web Flow, this has been the case for like a few months now. When you click on new project, you get this site or app. They're pushing this quite a lot. Like I don't make apps, I make sites. But yeah, whatever. It's a platform. And if you do Yeah. And if you do a site, you get AI template or blank, which I like this. I like this. So, uh, we covered it last week, but codeex sites got released and does what it says on the tin. You you prompt something. It might be you're working on a report. You create a dashboard for it and you can publish that to an internal link. It's very much internal. So if you're part of an organization, so it's not a public site, internal site, probably a bandwidth thing to be honest. They're working with Web Flow and they're working with Lovable and they're working with Replet. So it's in this will be interesting to see like I feel like some of these companies like Web Flow and Replet and Lovable, they've kind of like dug their own grave because it's like by the way I don't know how you might know more, but I don't know how these things how this all integrates. But it could be that you prompt everything in chat GBT and then there's the last publish publish step in by pushing up to web flow. So begs the question why would you ever search new project new app? You just stay in chat GBT and publish it as a site on web flow. So keep saying it's like why why live in web flow when you can just do it all in chat GBT if you've already got that subscription or whatever. It's all about control like control. Yeah. So, so I have been uh testing like all of these new uh ways of building and actually I want to make a video on this. Uh I think I think I'm uh on to something. So doing a redesign using claude design and then you you know how you can export it to claw code. uh when you do the export it just gives you like a link API.claw claw something. So it's a link to that design file or like that HTML file whatever you have on cloud design. You can directly in your terminal ask claude to bring that to web flow using the web flow mcp. I tried that and it worked surprisingly well. Like under one hour of me not working just like I gave it one prompt to redesign this website and then one other prompt to bring that website to web flow. I had that website in web flow. Um obviously it's not like a it's not bad. It's well we'll see good either. Uh well I I'll I'll show you some designs I've been doing with AI. I think they are they're quite decent. They are better than average designers. Yeah, they're fine. They're fine. They get the Well, audience, you'll be the judge of that. You call me out. If the design is [ __ ] it won't be [ __ ] It will just not be great. I I've I've made a couple of videos on this. I've spoken about a lot of this and there's going to be this ship studio video that comes out probably tomorrow now. I think you should watch it where I just unleash a little bit on my thoughts on AI design and like we're I think we're thinking about it in the wrong way. I think we're thinking about AI. Don't get me wrong, it's fine. It's good and it's good enough for a lot a lot of people. Like I'm going to be I'm totally using AI design in a lot of my things totally. But it's never going to satisfy designers because they're trying to they're trying to place AI in the design workflow in the wrong areas. I think they're they're thinking that it needs to finish their design for them and that's just the most inefficient way to do it. But yeah, check out the ship studio video that's going to go out tomorrow. I I will I will Yeah, that I'm looking forward. This is a bigger discussion I think. Yeah. So, we'll get into this. Um I don't know how we cover this one. Is it separate or in the same video? Just just go into it. I mean, it's it's Well, what um do an intro and then I'll I'll figure it out in the edit. Yeah. All right. So, um I don't know. Let's think about an intro. Should have done it without sharing the screen. Yeah. Because it's a big Well, you know, we still you can control that later, right? Not to our wonderful 73 live audience viewers. You've ruined the SP. We're here for them right now. Do you know what I mean? I don't live in the moment. I live in the future [laughter] in the videos that haven't been published yet. So, what would be what would be a good intro for this? This is this is a bad news for developers uh like web flow developers or is it not not necessarily obviously well I guess so because he's obviously a great CTO. Yeah. Okay. So something big has happened to the web flow community and this is probably a bad news for webflow developers. Uh you know Alan Alan CTO am I? Yeah the the ex C2 CTO of Web Flow. He was the CTO of Web Flow. He was you know the guy who was shipping. He was all about shipping. Shipping shipping just outputting work. He's now officially the CTO. CO CEO [laughter] CEO. What? Well, okay. He is the CEO. Yeah. So he is now officially the CEO of Cloudflare. That's that's crazy that nobody expected this. And this is well CEO he is the chief engineering officer. So practically CTO of Cloudflare. He was the CTO of Webflow and now he's gone. Uh it's quite funny knowing that you know looking back Web Flow had so many um collaboration with CL Cloudflare and they just stole their CTO and now he's at Cloudflare. Um and the valuations on these companies are very different. By that I mean in favor of Cloudflare. Cloudflare is a huge for our non-dev audience. Cloudflare a huge company, a very important company that when they went down probably two times last year, maybe one time this year, one time last year. Yeah. The internet went down. Yeah. All of your whenever your web flow projects go down 404, it's because of Cloudflare. Like they are huge. Yeah. And CTO. Yeah. That's Well, he's gone. Uh, I searched on LinkedIn for CTO of Webflow. I didn't find anyone. Uh, maybe I didn't search uh enough, but not they've not found you yet. You need to put in your your job application. I hear you're looking for a CTO. Have you seen how I could start a dev server? [laughter] Yeah, just asking Fable on extra high to do it. So actually did this but I'll show why. Um so yeah not a not a long video but I don't know how to feel about it. I feel like web flow community lost like a really important person who was all about developers. He was very technical and he was very active in the community like very there were plenty of times where I said something in the community and he replied uh or he you know other community members would say something and he would send it to the team um to take care of it. So he was very active, very and it was it all happened very very suddenly and there was there is no explanation there. I didn't see a post him saying he got locked out of his laptop. [laughter] He woke up at 7 a.m. and he was, "Hey, Linda, my laptop doesn't start. Have I have you fired me?" [laughter] Yeah. Yeah. When was his last tweet? June 11th. So, so this is Cloudflare. Yeah, this is June 11th Dario. That's not this about web flow. No. Yeah, it's not directly about Web Flow. Cloudflare Green. Here we go. June 11th, 10th. I suppose he's still obviously going to support it. He He's talking about career hacks here, so that might be top of mind. Speed Cloudflare ships. There you go. Cloudflare. Yeah, he's obviously still going to support Web Flow. I do think I do think he cared about Web Flow seemingly for Yeah. Yeah, I think he did as well. This this is great for him. Like this is a big jump. Yeah. No, absolutely. That's what I mean. like they are they are big saying that you know I have noticed he's doing a lot of retweeting which may be why I hadn't noticed but I have noticed that he's been a lot more quiet on Twitter recently probably because he's probably a lot more busy to be honest probably a lot more busy yeah so next time web flood goes down or any other website on the internet we know to go to yeah you You can literally tag him when your website Web Flow website goes down and it won't come across as weird cuz it would be weird if he didn't go if he if he went to someone like I don't know Stripe and then you tagged him about your Web Flow website going. I don't [ __ ] care. [laughter] Still care. Now he can still care. Yeah. Well, uh either way there is I don't see any tweet or any official explanation that hey I'm departuring web you know some there there are normally posts like this especially for positions like this no yeah flow you know yeah yeah and Emily you remember Emily um she was the head of community when she left there there was a big goodbye they like what was it like I think a team they made a website for her like saying goodbye and and so I mean that makes sense because she was community. Yeah, head of community but even pretty pretty involved. Alan was active in the community. That's my point. It for how active he was. It's awfully quiet how he just left and it's I don't know. It feels a bit strange because he left for Cloudflare knowing that they were so close with Claude, you know? It sounds a bit like they they stole him. Yeah, it does. It does. Yeah. Anyway, uh Alan, we will miss you. [laughter] Yeah. Cool. What's next? What's next? Uh anthropic Claude. What's happening Claude? We Okay. Do you know what's happening Claude? Um, I have no idea. Apparently a new model or something like that. Uh, let's actually grab the Do they have like the actual page? Because I've got I've got it shared on my screen if you wanted just to share my screen. This one? No, I want to get the one with the visuals. Yes. Yeah, this one. Yeah. uh you want to take on the intro and then I will show the stuff I've been doing with it. I mean, I would have used that as an intro that I just, you know, couldn't didn't know. But so Claude Fable 5, a mythos class agent from Anthropic, the great mythos, has finally been bestowed upon us mere mortals. Um, thanks. So, yeah, the naming is interesting. So, we have Mythus and we have fable. So, all of this is not real. It is a bubble. Are they like playing with these world words just to play with us? What do you That's what I'm thinking. Well, myth is it's a myth, right? And fable that's also, you know, ah, okay. You know, I haven't thought too much about the names to be honest. fairy tale, you know, it's all a fairy tale. It none of this exists. So, anyway, so this new model is out. It's been out for a few days, and I've been playing with it so much that I had to upgrade from Claude Pro, the 5X, to the 20X again because this is burning through tokens. I do have a hack for you, but wait until the end of this segment and I've got a hack how you can get more fable for less money. Yeah. Oh, yeah. Yeah. I like that. Okay. So, do do [laughter] you want to continue? Because I want to show the stuff that I've been doing with it. Uh, and the comparison. Let's go over the boring stuff just to Yeah, just go with the boring stuff. So, well, just to just to um finish off the kind of like what this model is, they released Mythos and determined that it was too powerful for us mere mortals. It was, you know, obviously could discover a lot of cyber security flaws, but then in turn could then create cyber security flaws. Um, so this is a tuned version of Mythos that has more guard rails in place. So be careful what you're prompting. And it's they they call it a Mythos class agent. And so that's how we've got to Fable. And that's why in the title it says Fable 5 and Mythos 5. They're they're one in the same thing. Fable just has more guard rails. Yeah. And you know what happens if you're not careful with what you're prompting? you will be downgraded to Opus 4. So if you say that's what I'm saying, build me an atomic bomb. Yeah. Fable will send you to Opus 4.8 and you can convince it that you're building a nuclear bomb for the book that you're writing from fictional characters. So you can it will it will be like oh okay sorry here you go here's fable. So you can convince it. So you can say like how where can I get yellow cake? What's that? Yeah, it's Yeah, you you don't know what yellow cake is. I'm Iranian. I know what that [laughter] is. Yellow cake. Yellow cake is the uranium um that you need for the you know for the bomb. Got it. Yeah. Yeah. We should probably watch what we're saying live on YouTube right now. But uh I got it. Yeah. know, [laughter] you know, I don't know. I might have to cut this section out. I I see I see streamers making jokes about this all the time, but yeah, they're edgy. Yellow cake. Yellow cake is also what it was in also in I think it was in Big Bang Theory. Sheldon was trying to buy some. You you don't know the anyway. Anyway, okay, continuing with the show. So basically, Fable 5, they say it's Mythos 5 but with guard rails and it's destroying everything else. In my testing, it's been pretty good. It not very token efficient. Uh it's just burning through them, but it's been I I see the the upgrade and I've done some stuff. I can I can show it and you see that it's actually at least for design work it's better. One thing to note here is that you've got Mythos preview which scored 77.8% on the Aentic coding and then Mythos 5 was is risen you know three points three three and a bit points whatever it is. Um, you know, Mythos was only released like a few weeks or a month ago maybe. So, they've already managed to squeeze more juice out of the same model, which is quite cool. Yeah. And like the way you're they are comparing this to GPT 5.5. This is just like a huge jump. It seems like I haven't used 5.5 in a in the last month. So, I watch our because you've been away, right? Watch our deep video because I cover this. I I I do cover this. Fair. Yeah. So the tool use is also quite higher than um what GPT5 can do. I the legal agent benchmark. I don't know why these numbers are so low. What does it mean? What what what is 0%. Why is Gemini 31 Pro at zero? I don't know. So a lot of these numbers to me are I don't know we we should just uh ignore them I would say at this point to me they are not very interesting. Is there like an interesting number that you want to look into? Well, all I really care about is agentic coding and uh yeah, that's that's all I mean I like the um GDPA uh benchmark, but it doesn't look like they put it. Oh, no. There knowledge GDP val um 1932. But yeah, again, it's just it's a nice little glaze. It's nice to just look over the benchmarks and see how it's rating and then you can make relative um comparisons in your head, but beyond that, I think it's a perfect time to look into your your examples. Yeah. Okay. So, you can cut this part down how you want um in the video now. Okay. So, I want to show you a few examples of what I've been doing with Fable 5. One of them is I finally gave hyperframes a try. So this is a video made by Fable 5. So you see all of this is animated. This is hyperframes. This is hyperframes. Amazing. This is hyperframes. So this video uh is essentially this. By the way, the design is not from Claude. This design is done by us. This is like human. This is my designer putting like actual work into this and making this design. Okay. And then I gave this design to Fable and I said use hyperframes and you know make an animation and it should be this and this and that. And that's that's the output. Um I'll just play it once again. So it does this thing and then you see this hover. That's really nice. There's a click effect and there are like these really nice animations happening. And now I can export it as a video. There is a V2 version that I did and it's so bad and it doesn't like work properly. Yeah. Like for maybe I'm using hyperframes wrong. And just yeah to to get an idea about hyperframes. You see like all of these are like your layers and everything you can you could yeah you could add sound and everything you can change it manually or just ask your agent to do it and then you export it. Even the export you can do it agentically. Um there are assets there are there is a code. So all of this is built as an HTML file and then I don't know it's animated uh and you can also use hyperframe uh with um with GSAP. So you can also use GSAP and hyperframes together which I haven't yet but that can be also um I thought it was GSAP under the hood. I thought it was I'm not sure. No, I don't. Can you remember what your prompt was? Can we see? Can we Can we Can we look at your prompt, please? Uh, the thing is I restarted everything. I don't think I hyphen hyphen resume. Resume. Yeah, I can give me a second. I will. What is happening there? Uh, isn't it just Claude dash R normally? Yeah, you got it here. Okay, so here uh on the left side I installed everything I needed for hyperframes and my initial prompt. Yeah, it was like there is a floating animation. It's really bad, mate. [snorts] Um which which is kind of indicative of how you know how is this my Yeah. No, this is not my first prompt. So, that's like down the road. But it but it was at the end of the day, it was like a few prompts, not too many. Uh I'm I'm quite surprised by it. Like the first prompt was working. Like everything that you see was working. The only few things that I had to change was like this hover effect. I wanted it to be I wanted to be to have this hover effect on hover and then have a click effect. Uh and then these I wanted them to to be a in a specific way with this count up animation. Uh they existed as a a different type of transition. And then what's also really cool is you you see this text that I have here or like these pieces of text. I can just tell the agent to change them. So these in my design were mention just the word mention. I told the agent, hey, use something that makes sense. I didn't even say use web flow on LinkedIn. I just said use a word that makes more sense than mention. And it did. Anyway, so the point here being not hyperframe, it's I would say fable doing a great job here. But you might be asking, but how does it compare to other agents, to other models? I'm glad you asked because I have I didn't ask, but whatever. [laughter] So, okay. So, here I will zoom in in a second, but just so you know. So what I'm going to show you is a comparison uh of design capabilities of lovable opus 4.7 fable 5 and the Figma new update Figma agents. Okay. So this is our starting point. This is what we had in Lovable. This is an output from Lovable. [snorts] Um it's quite messy. It's at the at the at the glance you might say that's not a bad design, but if you read the text, well, it's German, but if you read the text, it's two sections put like next to one another. It shouldn't be like this. Anyway, our focus is on this left side information. And I gave my agents the task of redesigning this 4.7 did this already better. did this. Not bad. And did this also pretty good, I would say. Maybe I should change also the page um background color to make it easier on the eye. Yeah, [snorts] not bad. Um also this pretty good. Then comes Oh, I with doing that I give me a second. We you need to you need to edit this part. So that's 4.7. Now let's see fable five. I would say significantly better. Like these are usable designs. I mean look at this. These are really really nice designs. I mean my favorite being this one. This is just so good with the text in the background, the the nice shadows. I wouldn't say this looks like an AI design. I wouldn't. This this looks really nice. Uh and the prompt was keep it on brand. Use the brand overall look, but you have freedom to to do a bit more. Uh you have these two images. These are like thumbnails for the videos and design it. So yeah, I'm pretty happy with them. Then I wanted to compare them to Figma and oh my god, [snorts] these are just so bad. This is this is where you see the difference. Like I mean what the hell is this? Like this is not the same level as this. Do they choose the agent that you can use in Figma? I will I will show you. By the way, you see how there is a frame here? I didn't take anything out. This is literally this is what Figma gave me. All of these in a go is what Figma gave me. And I'm like, what? This is really low quality. So, here's how it looks. Uh you have this AI thing now. So you have your file, your pages and you have this agent thing. So you can select something and it gets selected here which is often annoying uh because if you select this and say something and then select something else that gets selected and yeah uh you send it by accident anyway. So um no here you can't do anything. You can just prompt it. And here there is like access to libraries and nothing else. You cannot even look if I do a screenshot. I cannot put it in like it just it can't do a screenshot like paste. No, you you can't paste a screenshot. Um so the the way I used it was you know that my favorite Figma um shortcut nowadays is command L. I you you link you you copy whatever you were clicking on that node and then that's the context so the agent knows what I'm talking about and then I give it the the prompt and here my prompt was pretty much identical. I was like create five redesigns of this and I gave it the exact like the links and for context this is how it's used and yeah um and use these images and so [snorts] Figma really bad I would say like the worst here uh even worse than lovable uh opus 4.7 did decent these are not designs that I would use in an uh Finnish website, but I would I would like work with them. And then Fable Five, I would say minimal effort from me to like this. I could just give it I mean there are like a few things to fix like the the borders etc. But other than that, I would feel comfortable just giving this to an agent and build it in web flow. It's that good. Cool. Um, I mean, we won't dwell on it too much because for I want to get to the dev side of things, but like this is this is great if you don't you've got like a rough idea in your head, but you don't have a specific idea in your head. If you had a specific idea in your head about how you wanted this to look, I think that you you'd have struggled or you maybe would have been bit more specific about your your um prompt. But yeah, no, I agree. It looks looks really good. Looks really clean. Um, and I'd be more than happy with that if I was trying to design something. Yeah, I did also more experimentation, but these are very very raw. So, I wouldn't go into them, but at the end of the day, uh, the takeaway is not, oh, Fable 5 is taking uh, your job. It's I know what a good design is and I can get a relatively good design out of Fable 5 better than any other model at this point. So if you are a designer just use it. You probably have a vision you know what's good what's bad and you get this out of it but it's not like automatically taking uh over your job. Mhm. Um so from my experience, Fable has been really good. I bet hearken back to um our coverage of that article where oh no where we spoke about Uber looking into cheaper models doing some stuff and Frontier models doing like the heavy duty stuff. Immediately what I did when Fable was released was I went through all of my code bases, little tools here or there or even client co tools and I got it to review the the security, review the architecture of the app, just do a do a more technical deep dive on the app and output it or output instructions specifically for Sonnet or Opus to implement. So those are the guys that are going to be implementing the code, whereas Fable is the one that really did the deep dive and and the rest of it in particular, like security, because that's going to be the thing it's it's really good at. So like I'm not I'm I wouldn't I wouldn't I mean you're you're not going to listen to this for sure, but like uh I wouldn't necessarily say like every single thing needs to be done in Fable. I think pick your battles and it will go a lot further because you don't need to use it. I mean, heaven forbid like you don't necessarily need to use Opus for everything. Sonnet is still a really good model and can can do a lot of grunt work. So, if you are balancing or if you are struggling to get the most out of Fable and you want to use it, choose choose it wisely. Um, but yeah, so like I I'm definitely using it for um [snorts] I've just used it before the show to build out like an outreach app. I just gave it to or like auto on just [ __ ] go build me an app and uh I said I'd give you a little cheat hack for for getting more out of Fable. Currently Claude Co is um double usage until the 22nd I think. No, no, 22nd. No, first 5th of July double usage. I looked in the terms and conditions. There's no, it doesn't stipulate that you can or cannot use Fable. It literally just doesn't, it just lets you have double usage. So Claw Cowwork, if you didn't know, is kind of a wrapper on clawed code. You can code with it. You can build apps with it. You can do all of this sort of stuff as long as you prompt it to it. It will be more hesitant to build an app that when you don't ask it to build an app. If you just ask it to do something, it will just build something that's really not production ready. But if you say, "Build me a Nex.js app," it will do that. And I've built a trading bot using it as well. And you can also get work done just like you would in Claude chat. I don't know why you'd use Fable just for chat. Maybe some deep research or something like that saying that. Just basically I'm saying do your stuff in Claude Co-work. leverage that 50% extra usage because Fable will go a lot further with it. So, you're welcome. [laughter] Yeah. Yeah. Yeah. Nice one. Cool. Anything else on uh Young Fable? Got two more to go. You can sit back and relax now. We've got two more to go. So, um, Prometheus Bezos, Jeffrey Bezos has started a new startup with an ex Google employee called Prometheus and it's worth already $41 billion. Let's get into it. Let's get into it. So, Prometheus is, as I say, an a an an ind this will this will create a lot of conversation. It's an industrial AI startup led by Jeff Bezos and former Google exec Vic Bajage. I'm guessing that's how you say Bajage. Bajage. Um, it's raised 12 billion in a series B funding at a 41 billion valuation. Mate, they're fresh out the gate and they're already worth 41 billion. It's interesting because how much of that is to do with the fact Jeff Bezos? How much of that to do with that is to do with AI? We'll get into that. So, [snorts] um, why it matters. It's a massive bet to rearchitect how physical things are made. So this is interesting how he's he's you know using I mean te digital technology is is kind of done for isn't it like we said at the beginning of the show like SAS is is over like why people would build why people would pay for something that sorry why people would pay for something they could just build themselves can't build physical things themselves so from jet engines to medical devices to consumer electronics and this is what Jeff Bezos For example, if you go to a current jet engine manufacturer, as you do, and say you want the exact same engine, but with 10% more thrust, it could be a 10-year program. Not because they're lazy or bad at their jobs, but because it's so complex because it's so complex. So what we're doing is building a set of tools that will empower engineers to compress that cycle time down and make that dream build loop be 10 times faster or even more. So it's using AI to optimize pre-production of machinery and processes. Um they've already got 150 employees working for them. Um and they're building what they call an artificial general engineer. There no corporate ties to Amazon or Blue Origin. This is completely separate. Um, but he could be doing the old hokeyp pokey that uh that that Elon Musk does. Um, oh, Blue Origin are hiring uh Prometheus to build us a new engine. Do you know what I mean? Um, Bezos and Bajage insist that the effort, if it works, will result in more human engineers. And I don't know whether you saw this uh interview with Jeff, but I think that's it. I don't know if you saw this interview with Jeff Bezos, but he was like his bet is that actually there will be more exactly what he said there. Uh more human engineers. I wonder if I can find it. Um it's a really interesting um finan sorry hang on cuz I am a Financial Times subscriber boys and girls. So look at me. Uh let's have a I don't even think it was with Financial Times to be honest. I think it was uh as YouTube um interview Jeff, I'm looking for a a specific I don't know. Anyway, the point is ah, here we go. This this Jeff Bezos says AI will bring golden ages, not mass Oh, no, sorry, I'm not even sharing my screen. Um, Jeff Bezos says AI will bring golden ages, not mass job losses. So, and okay, you know, this is slightly off topic, but I do believe and we spoke about this like last year that we are struggling to fathom a world we're trying to place the effects of AI in the world that we currently live in. Whereas, I think the effects of AI will have unintended side effects of many, many different industries um that we just can't predict. And I'm not claiming to be able to predict that. Whereas Jeffrey here, he's saying that there will be more jobs. And I I've got no reason to to disagree with him. Like I I think that we will turn to the physical. I I've thought a lot about how AI is going to where is the future of of commerce. Where is it? I think it's going to be in physical goods which is probably why this AI startup exists because he thinks the physical we will be doubling down on physical things because that's just not what technology can well technology can solve it but as they're doing in the pipeline now I had something in my brain that I want to get back to but do you have any initial thoughts? Not really. Okay. I had a thought I might need to just reread the article real quick. Um, I think I've lost it to be honest. I think I've lost what I was going to say. I get someone on tangents that I just [ __ ] uh forget. Um, training your own model. That's what I was going to talk about. So this is really interesting because they obvious they want to as they say uh let me quote them here. They want to where is it uh build what they call an artificial general engineer and I think this potentially could be we've got some comments as well. Um, this could be potentially a use case for again doubling down on more specific agents for specific things rather than having these gen general agents. Maybe the power and the focus is having literally maybe they're developing an agent that's literally all about just engineering. Do you know what I mean? and all about efficiencies and aerodynamics and um construction techniques and all of that stuff completely eradicating or or pre-training not pre-training but post-training and reinforced learning a model that just doesn't have any knowledge of the the star constellations or like medical well maybe medical but do you know what I mean like just information that does not need maybe this is where the future lies and it's a it's a it's a something I do want to look into is this idea of training models that actually do really well at the thing that you want them to do. Um, so Nick, uh, sorry, T- Nick 8201, physical AI is the future of manufacturing. Would you reckon physical AI means like robots? Um, or is it AI that's trained on the physical that helps manufacturing? It'll be pretty easy. it'll be pretty easy now. You just need to integrate a doc engine with a world model uh with physics and text embedding. Well, like that that's if you're suggesting well I suppose a world model is pretty good really to to have all of that. That's good for robotics, right? Physical stuff. I don't think they're getting into robotics. However, like you say, a world model would be beneficial if something needs to simulate aerodynamics or or whatever. So yeah, taking action in the real world of an application that creates work orders from telemetry. So um yeah, it's pretty interesting. I'd be interested to know how they how they get on and I'm interested partly from like a futurist prediction standpoint and indications of where these guys see because they're tapped in, right? They're tapped in. they're tapped in on what what the current state of AI really is into the in the you know um behind the scenes they're banking on uh physical engineering and stuff. Cool. Um one last one and then we will let you go. So Hermes was fresh on well actually saying that Hermes was first onto the scene when it comes to Open Claw and these agents that let loose on your machine and can do pretty much anything that you want them to. Hermes was actually first on the scene. Then Open Claw came out and then they were like, "Hang on, we got something. We're working on it." And then Hermes gets released and to in my opinion a lot more refined and a lot more well thought out. saying that I still use OpenClaw. They've just released a desktop app and I want to try and share my actual screen or even the window. I'll share the window and it looks pretty darn nice to be honest. It looks pretty clean. Um, this is kind of it. I can't Can we zoom in? Zoom in a little bit. Yeah, zoom in a little bit. So, you gave me a website to look at um and I gave it a link and you know documented it and and gave me like a content map of it all and out output it to like an MD file. Even highlighted that ladies cutting and ladies styling are returning 404s even though you've got them in the nav. So, you might want to look at that one. Interesting. Yeah. Um realistically, realistically something like this is you could ask Claude to do the same thing, right? This is not where the where the power is of um of of Hermes or Open Claw or whatever. And arguably since we've got dispatch and computer use from both Open AI and Claude, it starts to diminish the effectiveness of these agents. Saying that this does work, this is I've got this currently linked up with my Clawude Code subscription. You can you can attach it to pretty much um that's the wrong one. I want to go to sorry I am just learning this tool so bear with me. You can um link it to well we'll go over the pricing in a bit but you can link it to your news portal or you can actually just plug into uh all of your you know all of the APIs and things you've got. So they can use all those models and things like this. Um, but saying that this is open source and they're going to move a lot faster and and to be honest, they're not going to be as beholdent to security and stuff like that, which is a double-edged sword, I realize, but you're going to be able to do more stuff on these open-source agents than you are with clawed code. They just because they've got restrictions and things like this. Um, uh, and the other thing is is actually the gateway thing as well. there's a lot more uh flexibility and freedom to log it into your Telegram and have Telegram, you know, text this and it will go do something on your local machine. What I like about Hermes as well is that that I think, and don't quote me on this, but I think they encapsulate a lot more. You can run these inside of a Docker, which is a lot more safe, which by the way, we haven't even spoken about Odysius, did we? I don't think we spoke about it. I don't think we even spoke about it last week, but I released a video on Odysius, which is PewDiePie's open-source agent, which again does all of this stuff. Side tangent, you know who like PewDiePie is is um unveiled AI agents to a whole new or set of audience, right? There are kids who literally had no idea what AI could do are now interested in in AI agents. Whether that's a good thing or a bad thing, I don't know. But the point is a lot more people are interested in AI agents as I found out in the comments section of the video that I made on it. So either way, they all do one and the same thing. I set uh Odysius up with Docker and it's really restrictive. It's really good for like doing stuff like that is linked to through APIs and CLIs and stuff like that. However, it really can't take massive action on your machine. So, really nice app like I I think um let's have a look at the website as well cuz uh the I forgot I think they're called News Research. They their what I I'll ask you what you think of their design. Um, here we go. Uh, yeah, we'll start here. You know, this is their design. Aesthetically, maybe you don't like it. Wait, you're not sharing? Oh, okay. Let me let me let me go. Aesthetically, you might not be as um as into it, but there's these little touches on things that are quite nice. I like it. You like it? I thought you might not like it, but but it's well put together. nice. Like you might not like the design, but you might appreciate the Oh, what's happened here? Um, but it's uh Yeah, it's pretty cool. The one thing I wanted to show you was they have their own uh what do you call it? Well, let me actually share the uh proper proper screen. This one they have their own coding plans which give you 300 models and rates and and whatever. So, and they'll let they'll let they'll let it um roll over. So, anything under 10 pound $10, they'll roll over to the next month and whatever. So, they have their own coding plan which you can use with the Hermes agent. So, quite nice. Uh but I'm using it with core code which comes out of my usage credits which I don't know about you, but um I thought 15th of June. Oh, maybe maybe it's 15th of June, actually. But we're supposed to get an allowance equal to our subscription in extra usage. But this that search that I did to to search your website that I run through Hermes cost me not much, but it cost me a little bit and it took it out my usage credits. I thought we would be getting an additional bucket to say programmatic usage of Claude. Um, yeah, that's how I understood as well. That's how I think I think maybe we haven't hitten that date yet, but I have got an email about it. So, long story short, uh maybe hold off using your claw code thing until that's clarified. But all in all, desktop app's really clean, really nice. I'll be doing a video on how to set up local agents with it. So, go subscribe to me for that. But all in all, yeah, so now that you're there, also check the desktop app uh website um page. Just click on the link desktop app. Just where you are, you see in the navigation there is desktop app. Yep. Click on it. Yeah, that page is also quite nice. Yes. Yeah. Yeah. Yeah. That is nice. It's a whole new experience actually. It's got similar I like this. It look it's very much web 1.0 hyperlink blue. And um it's got a slight little um parallax on there as well. There was one I wanted to show you. Maybe it was on the pricing page. I think it is on the pricing page. But that that there there's just a subtle little Oh, yeah. A little effect there. It's a WebGL RGB channel. Yeah. It's a little different, you know. I think that's what we um that's what we appreciate in the age of AI. Just a little bit unpredictability, a little bit of little bit of difference, you know. Yeah. Oh, that's nice. Yeah. Yeah. Yeah. Anyway, that's Hermes desktop agent. Go check it out. Um, I did see some comments here. Looks like uh similar to open web UI. It's not really like open web UI. Open web UI is just a chat. It doesn't take action on your um computer. This will take this will do things on your computer. Open up a browser. It will I sent I sent my open chlor a text the other day saying go on this shop's website and find me all magnesium tablets that are priced £18, you know, and it just gave me a list of all magnesiums. You know, you can text it from afar and uh again, it will just use everything. It could even you might have an app on your computer that um or or whatever that it will use that app like this is very different to open web UI. So, uh, yeah. Um, why are you clicking anything? Cuz I'm sure I'm sure things are just like closing and whatever on my side. Maybe it's my Hermes agent clicking on things. Um, Hermes agent going to replace my job soon. Uh, I'm guessing you're referring to something different here. But yeah, it's uh, it's cool. I like my agents. You're not using any agents, are you though, Kabaza? No, I just downloaded the install. I just downloaded the app to make. Yeah, I was using clawed um what's it called? The no the for the automation. Oh, managed agents. No, manage agents is for like when you do it as scale, I believe. Uh [snorts] I forgot. Yeah. Yeah, I forgot. So many there's so many like like I say like clawed code has replaced a lot of what I was doing with open claw but there are just a few little use cases here or there that I I do still like sending it to openclaw plus I can bind it to my local AI and or a cheap Chinese open source agent to do stuff. So it's going to it's not going to take away from my credits. just really depends on flavor of the month I think really or that day even. Pi is killing it in AI. Didn't expect that for a timeline. I don't know what this means. Pi is killing it. Oh, PewDiePie. Pewdiepie. Um, yeah. Which one is your favorite? Claude, [snorts] Hermes, or Claude? Depends on what for really. heavy duty stuff like because you one you're talking about harnesses here like you're you're com are you comparing the harness of Hermes versus the um the model of Claude or you talking about Claude the harness cuz yeah I like I say I like one for well it depends if if I want to waste my credits to do a basic search in the background um then I'll probably choose like Hermes or OpenClaw or something like that. Whereas, if I want some, you know, I don't mind using my credits and I don't mind using my usage, then I'll choose cord. I tend to be um Matt PCO says like away from keyboard. If I'm at my keyboard, I prefer Claude. If I'm away from keyboard, I tend to prefer Hermes or Open Claw pretty much. Hope that answers the question. I see my um so I was using claude co-work schedule and apparently in the last few days it was not uh successful not running successfully for what I wanted to do because my cloud extension wasn't logged in. So at the end of the day, I'm not really happy with any agent. I haven't find anything really reliable. I find the browser for Claude is quite hit or miss. It'll be like, "Sorry, I can't do that." Especially because it does have a lot of guard rails, you know, oh, sorry, I can't go onto that certain website. Do you know? I asked it to download a torrent and it said, "Sorry, I'm not going to do that." Get China on the case. They did it straight away. [laughter] Same with building a bomb. Exactly. Right. China will do it. Um so yeah, try Hermes. See, see how you get on. But um yeah, I'd be interested to know what agent you bind to it, whether you Yeah. use a cheap Chinese one or whatever. [snorts] Should we wrap up there then? Yeah. Yeah, let's do it. Well, that's all we got time for. Thanks for tuning in. Uh, go to commandaiishow.com. We've got a lovely new website where you can watch old clips and watch old streams and contact us to say hello. Go to insiders.commandaiishow.com here to let us know what you want to do if you were to start a community with us. We're interested in start a community. We're probably going to have to draw that attention to that to an end soon and make up a mind what we're going to do. So, still some time to contact us there. Um, and catch us next week where we're going to chat to you all about in the world of AI design and dev. I hope you enjoyed this week. Don't die. [laughter] And I'm going to go play with Hermas, mate. I don't know. I don't We [laughter] don't need to know what you want to do. Keep that to yourself. All right. Jesus Christ. [laughter] Keep on vibing.