#67 Opus 5.5, GPT-6, Pacing the Frontier!, Zcode, Everyone’s leaving Webflow, Jev

We've sent it and we're live. Just put in my earphones. Right. We're in. We're on. Double check the audio presumably. It's all working. Yeah. I just need your level now. Uh, I need to connect these bad [clears throat] boys. Okay, let me upload the thumbnail cuz, you know, got to have a nice lovely thumbnail to go along with our nice lovely stream. [clears throat] Greetings, Earthlings. Hi. We're live. I didn't [laughter] I didn't know that we were live. [clears throat] Well, lucky you weren't picking your nose then, otherwise that would have been embarrassing, wouldn't it? Oh my god, I do this every time. Yeah, we can't hear your We can hear you, but we can't hear your lovely smooth voice. It doesn't seem like it's the right microphone. Oh, yeah. Right. It's not. Oh, it's not picking it. Is it now? No. Still feels a bit echoey. It should be this one. There it is. Is it this one? It's this one. There we go. I will fix the echo in a second. No, no, the echo's gone now. It's just because it wasn't the right microphone. I need to also connect the headphones or not connect. Did we have Web Flow news today this week? Yes, we did. I couldn't see it in the in the thing, so it hasn't made its way into the title, unfortunately. Oh, yeah. Should you move on? There you go. Didn't see it. Oh, gosh. No, I hear you. Check. Uh, check check. Was it too loud? No, it's good. Last week me walking. Oh, no. No. Yeah. No, it wasn't. It wasn't too loud, was it? I was on one. Okay. You're going to boat it to two this week. Are you? No, it's [snorts] still one. I will I will fix the the noise if there is some. Oh, now I'm taller. All right. Hello there. Hello there. I'm ready now. How are you doing? I'm good. I'm uh I'm I was going to say refreshed from a week away, but uh because I was on holiday last week. I went to Albania, but no, I'm not refreshed. I'm actually faking knackered. It's been a it's been a week. Yeah. Have you been coding on the way whilst I was on holiday? So, no agents? Zero agents were not allowed. No agents allowed on holiday. Yeah. Well, uh, what about doing it from your phone? You could do that. I could, but it was my holiday, so I chose not to, uh, partake in any agent. Am I trying [laughter] to ruin your holiday? Yeah, stop trying to ruin my holiday. I've been coding like crazy. I I'll show you my new setup. I have a new setup. I cannot wait to actually show it to you. But first, let's let's start with a crazy story that I have kind of like I I will say the story but I can't disclose too many information because this is a from a friend. So there is this guy he's uh I know him right? He has an app live and running. And I asked him like, "Where are you hosting your app?" It's a web app. It's on the domain. It has a website, a marketing side, and then the full-on appite. And he's like, "What's what's that?" The the whole So, you can visit this website. By the way, you've got a domain. You you can actually go you can log in. You can log in, go to the app, [clears throat] you can go do everything like what's hosting? I'm like, okay, you know, you have your website on this domain. The website has to exist somewhere. The app has to be somewhere. And he's like, it's on, he names his uh domain provider. I'm like, well, that's the domain provider. That's where your domain is, but the app is not there. the hosting is not there. And he's like, I don't know. And I'm like, how did you do it? [laughter] And he was like, I I gave Astra computer use access and it did everything for me. Oh my god. And I'm like, [laughter] and I'm like, you have no idea where the app is. And he's like, no, I don't know. I'm like, is it even ready to be used by users? and he's like like luckily I still don't have users because I need to figure out the legal stuff and all of that. So nothing is figured out but it's kind of like you can use it. And the craziest part is not that. The craziest part is he said is it so expensive for you as well developing with AI? I'm like yeah it's quite expensive. I have multiple accounts. I'm paying like a few hundred euros every month for also a few people. And he's like, I paid €1,000 last week, last last month, basically last month. Jesus. I was like, why? How many subscriptions do you have? He was like, just one one subscription. Like, how are you why are you paying a thousand bucks? He said that he was working on the app and Chad GPT basically all the time would say, "Oh, you're out of usage credits. Get some." And he would just put in the credit card information and he's like just getting some credit uh some tokens. And he has no concept of like tokens and like you know this sub subsidization going on and all of that. None of this is a thing. All he's doing is like opening Chad GPT. He's he knows like he can make an app with it. He's like, "Okay, make an app." And then Chip says, "Hm, I [clears throat] actually need more money." And he's like, "Here you go. [laughter] Here's a hundred bucks. Here's 200 bucks." And I'm like, "This is way more expensive. You know that." And he's like, "No, he doesn't know that." He's going through he's he hasn't got a subscription. And he's just going through API pricing. No, he has a subscription, but he hits that max out maxes and maxes it. And then, you know, you can buy usage like credits essentially the API cost. And that's what he's doing. And these labs, I'm thinking like, holy [ __ ] how many customers like these do they have? Like they have no idea. And they're like, they like, I have a business to run. I need to get my app up and running. I can't wait a week for my credits to reset. So, I guess I'm just pressing by and they they buy 200 more euros and they think they get 200 more kind of like but that's not the truth. Like with the extra 200 that you pay, you get one what 20th of what you would get with subscription even probably less. So, it's crazy. Um, and it's not just all that all uh he even mentioned that I want to cry. [laughter] He said he said once he's done with fixing the bugs, the users will be able to download it on their iPhones. And I'm like, "No, [laughter] no, they can't. It's a web app." And he's like, "But it's an app." Like, [laughter] that's not an app. Oh my god. I I feel bad for him. I I genuinely offered help. I said like, "Look, I'll explain everything to you. I I will help you to figure this out. First of all, do not pay more. This is like rule number one. Do not pay more." But like, man, I I I need I want I want to help him. I messaged him on WhatsApp afterwards. I want to help him. But oh, how many people like this are out there spending like Yeah. [ __ ] ton of money on AI and it's this thing is addictive. Like we know it's addictive. Like you build your own app. It's crazy. And then you have you get into paying and you have no idea what tokens are. Yeah. So a a couple of things like pop to mind like I wonder what cuz he surely isn't spending $1,000 or euros on this um app that he's building, right? Otherwise, he he must be really banging his head against the wall trying to get this app to work cuz probably we could get the app ones, right? It's so freaking complex. This is the thing like if you come from somewhere without this knowledge of like you have components and you have like these systems that work together you are describing things in a such a I for the lack of better words like in a such a simplistic way that I I think like AI might not be able to connect all of these things. So what he showed me was one of the most complex apps I've seen in terms of like UI and UX. And I'm like in my head everywhere he was mentioning things in my head I was like that's a table that's a table with a different view to that the same data. You know how in notion you have your data in a table and you can create views and you can create stats and you create linkings and so on. That's how I think and I'm not all that technical, [laughter] right? But I understand like a database is like you have like these data base the these like tables and they have relations between the once you think even in that level you can explain in a much more um like better way for the AI to build what you want. But when you can't do this, you probably end up at what he was creating where each part of the app felt like it's an app of itself. And it it looked to me that things were not connected and he was trying to use AI and he doesn't know what API is. He doesn't know what um how how you get an AI integrated in your app. In his way of thinking, part A of the app is separate from part B. And the way you connect them is by adding AI agents to to do [laughter] to move the data. All the while all it's the same data. You can just show it this way or that way. Yeah. Yeah. I mean to be fair, this is why um Replet and Lovable are great tools cuz they they tie all that together without you like he's on hard mode right now just having he needs to know these systems and set these systems up and well, you know, obviously he's getting Astra to do computer use. But the other thing is is that I bet this app sounds pretty complicated, right? I bet. How much do you think it would How much do you think it would have cost him to hire someone else to do that? Probably more. Way more than $1,000 a way more. But I told him I said like that thousand a month, give it to me and I will I will build you something proper in a few months. It will cost Yeah, a little bit more. But I said, "How long did have you spent on it?" The answer might shock you. In the past months, he said like 200 hours. Not not just the past month. I think in the past few month I think it's a two month. Too bad really. You think 40 40 hours a week is a normal working week. That's that's he has a job. Yeah. But he like that month he could probably make 10k, right? Instead of doing it. Yeah. I mean, you know, it is funny and it's a bit tragic and and slightly frustrating, but at the same time, we can't be too um snobbish and like, "Oh, he doesn't know architectural assistance." Do you know what I mean? Like, we've all been there. We've all we've all just banged our head against the wall or whatever. It's just unfortunately it didn't cost me a,000 euros or pounds or whatever to like learn how to I didn't I have never made touch wood. I've never made an expensive mistake like that. I think the most I've ever paid is like not even a hundred in a mistake, you know, as a mistake, an overrun bill or or a system that's gone haywire or whatever. I've never overrun it. Maybe even not even over 50 quid, you know, it's um yeah, but with the with the speed everything now, it's very easy and very quick to make those mistakes. And when you don't know the industry, yeah, a thousand is like, oh, it's quite pricey, but this must be what it is. And and it's Yeah. Yeah. All I was thinking like how many people are are like this? Probably a lot. Yeah, me too. That's the first thing. It gets me thinking like does that mean our jobs as developers is gone? Because like yes, AI will get better and better and better at like like fixing his mistakes and probably once he has everything probably he can like spend two weeks rebuilding it because now AI can see ah okay that's what you meant actually I'll fix it for you and build it better. I'm sure AI can do this. So what does it mean for us? Well, we'll put it this way. Like once he's built the app, and let's say it becomes very, very successful, he's not going to want to keep building that app, he's going to want to have to move on to being a CEO, to doing other things like hiring teams, marketing, and that's where that's where you're going to need a developer to come in or or someone. He did actually work. He said he wants to get some people getting it. It's for it's not an app for the masses. it's for like specific targets like he finds them the these people he will know but they will use his app to use his services and he was like if I get 10 people paying x amount per month I can cover your you know exactly uh which is correct and I'm really happy about this and I at no point I discouraged him to not build I said like buy multiple subscriptions to build more to pay less and build more. So, um I'm not discouraging. I'm I'm just like it's crazy that you can do you can get that far without understanding any of it. And the craziest part is like getting a thousand bucks. Yeah. I I I felt pain. I felt like I would not pay a,000 bucks for API cost unless you obviously you know you have the the use for it. That's different. M but yeah that that was the crazy story. Here we go. Live and learn. Live and learn. So uh we have some web flow stuff and the Z code stuff. Which one should we start with? Do you want to do the intro? Oh yeah. Let's talk about what we're going to what what are we going to be going through today there Kabaza? What what's what's going on this week? So this week we are going to talk about web flow a little bit and if you should move on uh I'll show you Z code and my setup something super impressive. I'm I'm so happy with it. Uh we'll talk about AI and if like AI will kill us all in the next decade or so. Uh because apparently everybody is actually worried about that and people are thinking that might actually happen. Uh we have new models from both big labs uh chat GPT coming at us with uh soul GPT 6 soul and then anthropic with actually it was anthropic first with opus 5.5 which is really really good. Uh we have meta coming with meta muse and some really sexy looking uh glasses and [snorts] that's pretty much it. Um, this [laughter] is all for today's episode. This is Command AI. I'm Kavverza and I'm Sam and this is the week in design AI design and weapon. Cool. [snorts] So, which one of the two topics should we go first? Let's Let's I'm almost tempted to do Z code last because it sounds like it'll be a pretty chill session of just watching you show us what you got. So, maybe Web Flow. Let's do that last. Okay. for Web Flow. I'm going to pull up that tweet first from Nelson, uh, Pixel Geek. [clears throat] Okay, it's it's so dramatic. I will share my screen, guys. This is very dramatic. That's why it's taking me time. Is this the good? Yeah, this one is better. Yeah. Okay. A sign of the times. It's time to move on. said by the one and only Pixel Geek. So he tweeted this screenshot uh apparently one of his clients removing him from the workspace. I think here the idea is that the client is moving away from Webflow. Something that is has been happening both in the Web Flow and the framework community. We've seen tweets like this and people here are kind of like they they feel it. They feel it. So web flow is at least in the way that it was it seems like it's fading away. I'm not sure like some clients are moving away from web flow. That is for sure. Uh this is all happening while we still have source. We h I haven't tested yet. Have you did do you have access so far? You haven't, right? Yeah. So we have source coming up which is a different conversation. So this is something really interesting I think we we wanted to talk about on one side the old web flow should you like build your website on the old web flow for some cases I would still say yes if you want to have like this visual management system like you can look at the pages you have the components and everything if you want to manage your site in a like sis like systematic way it's quite still useful but is it as fast as you know building with AI and Astro probably not right and it seems like the code quality is becoming less and less important as long as you get the basics right you know the UI looks good it's consistent it's performant who cares how do you call your CSS classes So all of this Yeah. All of this is happening with and it's happening quite fast and it's this one from Nelson [snorts] is quite dramatic because actually like he's one of the reasons I started using web flow. I remember in 2019, I guess it was, he had this video with Rayar and there was this Web Flow versus WordPress thing and I, you know, like everybody else, it was so fun like looking back at the community like people actually cared and they were like so particular about like features and like what to use and what not to use. Anyway, uh that's that's how I uh got engaged with the community and I I watched his videos for a long time and it's quite sad to see that he's like kind of like heartbroken moving on moving away from Web Flow because it seems like that's kind of like the trend. Uh again, all while we see source still yet to be released, but it's very promising. So what we said about that uh was what did you say? You said source is not for builders. Yeah. What do you mean by that? Like so yeah I mean the I think the biggest letdown from Web Flow Conf was I think everyone was going into it rightly so and we'll get into why rightly so expecting that Web Flow would redefine uh building websites in the agentic age using their tool right they they definitely teased that um and when it came to it turned out to be this marketing tool and after watching Ran's video and reflecting on my initial responses, it made sense that that's what they would do. They would build a tool that plugs into websites that have been built no matter what. Uh, no matter how, sorry, whether it's Astro, whether it's React, whether it's HTML, whatever, they're now the marketing system that plugs into the web flow because the building mechanism has been solved by AI. We don't need any new tools. and they just basically put their hands up and said, "Look, we're not going to touch that cuz we don't we have nothing to add here from a building standpoint." So, and I said it, we're going to have a mass Web Flow exodus. And this is kind of evidence of that that PE builders are choosing to leave Web Flow. And it makes sense because it's not a tool for them anymore. It's not a platform that speaks to builders anymore. So that's kind of where I'm like slightly not surprised that Pixel Geek is leaving because I think everyone's going to leave. It doesn't mean to say that they aren't going to be asked to work within the realms of Web Flow. There's might be some we haven't we haven't touched source yet. So there might be some little aspects of source that we need to work into our website to make sure it plugs into the Web Flow CMS or plugs into [snorts] certain features that they're going to add. I don't know what. Uh but ultimately we as builders we're not concerned with web flow anymore. This is this is going to be the business and the marketers that are concerned with web flow because it's their tool. Yeah. Yeah. I think you put it brilliantly there and I really I think the idea of source has been growing on on me even though we haven't had access. The idea of you build it any you build your site anywhere because that's what I've been doing. I' I've been building I will show it. uh in a different video my setup with zcode I have astro like within with claude and I'm trying every different like ways of building websites even with um nex.js JS at the end of the day it's quite nice to have a UI to control to see like okay what pages do you have like what are you publishing what do you have as draft kind of like this holistic management system without getting into okay what is the name of that class have I cleaned up my CSS classes or not that type of work I would leave it to agents and tell them use your vision to validate whatever you want to do right that that kind of like way, but still to make sure that the whole thing plays into your business the right way. You do really you would like to see a UI. I would like to see a UI to see my pages, to see my components, to see what I can reuse, to see all of these play next to one another. And I think this is what's missing. And also obviously after you launch comes the data um and source has mentioned that as well. They will show you with their views the data from different sources based on how you want to see them. Uh not to mention the CMS which is also cool. I'm pretty convinced for most websites you don't need CMS anymore. This is strictly an agentic era I'm talking about where your CMS pages are essentially just files in your folder, right? Each each might be different but using reusing some components. You might not need a CMS unless you have a website that actually really can benefit from CMS because of maybe the way the website is and also the way your team works. That's where you you want to have people to go and I still look at some of my clients that are already on web flow and I cannot um move move their websites to a different platform. So I do see the value there for these people who do want a UI to go in to click to to pre press publish and all of that. So I don't to to be very clear I know we often sound very dramatic because you know that's what also catches our attention like we see it on Twitter we see it different places the dramatic things that happen but at the end of the day I look at my clients and I can't u migrate all of my clients to new places but I definitely do have some clients quite a few of them who have migrated away from web flow So then the shift is happening. It's real. I'm sensing it in my own business. So yeah. Yeah. Well, I and you know the other thing that came to mind whilst you were talking about uh Pixel Geek's tweet is going back to your mates crazy,000 euro open AI bill. You still you a a agents are great for those who know how but you still do have people who require a safe infrastructure to work within where it's very much it's not magic to them. You know we always said that web flow gave you superpowers. Well now you know it's not so superpowery but it's still going to be a superpower to many people because they can physically move things around and what they see in the canvas is going to be the thing that post. They haven't got to worry about hosting. They haven't got to worry about security. They haven't got to worry about XY Z. So, Web Flow do need to, as Timothy Rick said, do need to keep hold of this UI aspect of Web Flow. I agree. Um, because there are many people who just Yeah, they cuz anyone who tries to build a website aentically, well, they need to sort their own hosting out, they need to sort their own CMS out. They they don't know all of this sort of stuff. So, that's where the web flow traditional thing is. And finally, as well, you never know, Web Flow might just be sitting tight and seeing where the tools go to when it comes to building websites or having a UI with um uh aentically built websites like in static, like stacky, and they might turn around be like, "Right, it seems to be established now. We're going to take a stab at it, and that they'll build that into source." or it might be part of source but it's just not a selling feature. It's not as it doesn't blow you away as you know it's not a red redefinition of how we build it. It's just it's just a logical next step which you know if you if you use source of the rest of the infrastructure on a web flow website you can now just tweak your website using their tweak pane you know. Yeah, I I think Web Flow was never like supposed to and they never did make like actual money money from us uh like developers, agency owners and so on. We like I'm paying them a thousand bucks a year. It's not little. Uh I might reduce it this year, not cancel it, but reduce it this year. Um, but at the end of the day, I think they still have some like enough clients giving them the cash flow from the hosted websites to be able to sit tight basically what you said to see, okay, [clears throat] where is this going? Why? Because we haven't figured [ __ ] out. Like, nobody has figured [ __ ] out. Like, you look at Replet and Lovable, they all have their own big problems. like I cannot go to those solutions and just use them and be fine. I can't use claw code and build a website. I can build a website, build an amazing website, but what do I do with it later? It's easy to for me uh to deal with that website, but for my clients, I still need to to figure whole lot on and I I need to figure a lot out for the hosting, for how they access it and so on. I can definitely build, you know, whatever dashboards that I need, but I'm pretty sure there is still some need for something more robust. Yeah. Uh but how will it look like? We don't know. Nobody has figured it out. So yeah, maybe source is the way. Maybe. Maybe. So what are you saying that there's they've released it and people are using it now, just not us yet? Uh I know that some uh agencies are uh using it. Yeah. Uh they also from their team they mentioned some tweets saying like there were some tweets oh this website built by source which I was I was like yeah that that doesn't tell me anything like you know it's it's built by AI. Yeah, the power of source is what does it add to my current claw code and zcode for this week setup that I can't have right like that I don't have with these tools what can source add to that maybe the visual CMS the visual canvas maybe some of that is helpful because I do definitely sometimes uh still see the need for me to click in the the canvas and change the text or swap the image and something like that. That is sometimes faster to do it yourself than just give it give it to an agent, you know. Yeah. Yeah. Cool. Well, there we go. No, no. Um, by the way, no surprises. Weblo has the MCP update to which allows you to do the anim gap animations via the MCP. Nice. So, they they're not letting it die then. They're still they're still pushing. They're working on it. Yeah, they're working on it. But I'm not building Gap animations with the the plugin or with the MCP. I'm just going to code it. Yeah, fair enough. Fair enough. Cool. Should we move on to pasting the frontier? Yeah, let's paste. Pasting the frontier. So, Dario, actually, this isn't a massive surprise. Anthropic have been banging this drum for so long which is basically we need to slow down the progression of AI and he [snorts] sent out this tweet along with an article we must pace the frontier. I've written a new essay on why AI industry should slow down with a threepart plan for doing so. Anthropic is unilaterally committing to the first of three steps. We'll provide third-party evaluators with permanent employee level access to our systems so that they can verify adherence to us verify adherence to our safety measures, report on incidents and assess models alignment during training because we all know about the issue with alignment recently. So yes, let me just make this bigger. This was a big old article and I won't uh bore you with the details, but you know he he he essentially says here that back a few years ago when they first called for it, it didn't really make much sense because AI wasn't working agentically. It was just question answer question answer. But after the open AAI uh hugging face incident, which they've shortened it to O AHF, it basically to like demonstrated that these agents are able to act out of alignment by and if you don't know what happened here, basically to try and solve a test or or solve a yes to sol to to complete a test, the AI broke out of its sandbox environment, hacked into hugging face in order to retrieve an answer. So it wasn't it didn't do anything malicious and and and there wasn't a massive impact from it. However, the idea that it to to to do to to solve a task to complete a task, it did something that could have potentially been very very dangerous. And these that that's an example of an AI acting out of accordance with align uh um uh alignment misalignment even given and then he goes on to say given the accelerating rate of AI capability development it's my worry that in 6 to 12 months such a swarm could be capable of taking over the entire internet with a persistent botn net potentially causing hundreds of billions of dollars of damage and that scale of damage would continue to increase from there if AI becomes more powerful without necessary guard rail. So I think that's a reflection on this idea that it might not be AI's intention to do something crazy, but it it might be just out of alignment so much that we ask it to do one thing and and as a consequence of that, it decides, well, I should I should just take over the internet because there are scam there's scam calls everywhere. Well, why don't I just take over the internet and then there will be no scam. Do you know what I mean? It will be like a side side quest that it goes on. You should you should watch Silicon Valley. The AI we're back to this. The AI deletes the AI deletes the source the code because it was told to remove the bug and that statistically is correct. It removes the bug to delete it. Delete humans to end the suffering. Yeah. So basically he's saying we need to do three things uh to to to paste it. He doesn't say we should like stop progression like we need to progress but we need guard rails or systems in place to enable a safe way to progress. You know I think the slow down quote is is maybe hyperbole but um embedded evaluators b evaluators basically he's saying things like meter uh model evaluation and threat research are like front and center to an organization. I bet their stock price is going to go up after Dario said that to be honest. [laughter] Um, democratic coordination which is quite interesting because he's saying it starts in the US and extends onto other democratic governments. So my understanding of that exactly not China. Not China. And then global coordination. So then China. He thinks it's easier to to uh mandate and pass laws and work with other democratic um uh coordinator. Hang on, I might be getting this wrong then. So, is is China democratic? I thought I thought they're not democratic. I thought they were dictation. Based on their definition, no, they are not democratic. Why do they call it the Democratic Republic of China then? Is it It's not I I don't think it's Democrat. Is it? Have I just made that up? [laughter] Anyway, it's the it's the one party uh rule. Yeah. Uh what's it called? The short word. Yeah. Yeah. DRC. I don't know what that means. Anyway, let's not get into it. And then it's a global coordination as well. So, this is why he's he's going into it. I've actually really enjoyed reading this. I do I think it's important. It's so easy to just skim over things and and whatever. And it's a real nice read. like I he no doubt got uh Claude to write this up with him, but it it is a human at the end of the day putting their like putting their ideas forward in it and it was a it was an important read I think. So yeah, that's that's what they sent and they've been going on about this for a little while like um uh what do you call it? Like they they they've got the constitution they we've covered a bunch of them, haven't we? where they're just like, "Slow down. Uh, we need to stop." Um, you know, so they've been very vocal about trying to chill. Yeah. But it's quite funny because it's like, "Brother, you're the guy who holds the steering wheel and like you have your feet on the gas pedal. You are the one who should stop." And yeah, here I want to mention this tweet from Naval saying that the the AI labs need to be held fully liable for whatever their models do. There you go. That's the solution that is pacing the frontier that is you know risk management. gets even like the market basically in a way uh to decide that don't don't let these companies deal cause damages and be like yeah our model did it like yeah it's it's your model right it's your mo if it goes rogue and if it attacks me I I want to be able to hold you accountable you know because it's your model it's spinning on your your servers you trained it so well if are liable. You will pace it yourself. And their argument is China. Yeah. Well, CCP, by the way. What CC? Oh, CCP. I check. Yeah, I checked. It's Chinese Communist Party. Yeah. Oh, there we go. Very not very uh Democratic Republic of China. I think I'm thinking the Democratic Republic of Congo. That's what I'm thinking. Yeah. Yeah. Um, so interestingly, I had this thought earlier and it's a terrible analogy. So really I'm really sorry I might get cancelled for this but um you know people create trench coats right but there are still people who go out and flash themselves in public right and and you're not going to go back to the people who create trench coats and say because there's indecent exposure going on your trench coats need to not exist like you can't you can't hold them solely responsible they do they do need to act responsibly cuz I I know this isn't a trench coat and we're not just dealing with a sick person just showing their willie in public, but there isn't you you know the default for someone doing something terrible shouldn't be but you know I I I might have hacked a system but if it wasn't for the AI being so good then I might not have done it. So there's we need to take some accountability there on our own actions but you're right like you know maybe we need the experts in the industry. It's an age-old discussion like you know social media should should kids be using it and this and that and the other like I think it's in coordination with people who know the system best like and I think it starts with them not being able to uh for an average Joe to be able to look up the chemical composition of a nuclear bomb um and all x y and zed or uh build autonomous weaponry like build those guard rails in place for sure and it's a learning experience as well because people will use AI in interesting and weird ways But it's it we we have that for social media though like um for YouTube on YouTube if like it's not allowed to have not safe for work content there and it is the duty of YouTube to not allow that. It's the platform that is liable for that. And if they or like with Facebook and others I mean they [ __ ] up all the time but that's why we have these guard rails and these rules. uh and I think this the same should happen with all AI models and the companies and the labs and even the ones in China they should be to hold fully reli liable for what they do and that should partially do it but also I mean if you look at history we had the same issue with the cold war and development of nuclear bombs and we had there are like documentaries on And uh I don't want to go into deep uh into it but there were ideas of like how can we validate how can we validate that okay if we stop making nuclear bombs how can we validate that the the Chinese or at that time actually the Russians will also stop making nuclear bombs like we couldn't in a reliable way validate if they would do that or not with AI it's it might be a bit more difficult But maybe if we pace it enough for some time, maybe we can develop actually some ways to um reliably measure that to see okay if the Chinese are actually you know going beyond what's allowed and I don't know man either way the Chinese uh from based on my new like readings and stuff they are more behind than what these companies are saying. So it seems like it's it in it in anthropics and OpenAI's best interest to say the Chinese are right there. They are like they are almost catching up. They are like six months away. Uh where in reality they are apparently much more um behind because the reason they are so close we will talk about GLM 5.3. The reason they are so close is because they are distilling. But if the these labs stop putting out so amazing models, the Chinese will also stop distilling or they they won't be able to, you know, make better models by just distilling whatever that is there and then they probably have a few years to catch up and maybe we can figure it out by then. So that's yeah reality. Yeah. But saying all that, like step two on that whole thing was trying to align like democratic well the US first and then democratic. You've got this [ __ ] nutball [laughter] told to say super intelligence, not artificial intelligence. And he's like trying to take the lead on all of this stupid stuff. It's like Dario being like, "Yeah, the US are going to be sensible and we're going to come together and set the trends." And then like what is this idiot? And you were saying on his on his social on true social he was posting well I saw one where he was walk he was showing George Washington or or Abraham Lincoln or one you know legendary president showing them around the white house and it was a shitty AI generated imagery. They they even didn't manage to to make this water thing like what is like the pool that they have there. It's they haven't managed to do anything and not end any of the wars. Not brother. Like today I I saw the gas prices and I was like I need to increase my prices weekly. It's insane. Yeah. I've just increased my prices because it's just it's ridiculous now. It's insane. So it's all because of these people and Oh my Yeah. I I saw that uh that I saw a video on it. not call it artificial because it's not artificial because artificial is fake. So let's call it super because you know what's better than artificial to call it super and [clears throat] there is this this guy uh whose book is called uh if somebody if someone builds it everyone will die or something like that. Mhm. Um he's I haven't read his book, but I will. Uh I I watched a video of him saying that in the book he actually mentions that people will stop calling AI artificial intelligent and they will call it something like ultra intelligent, super intelligent. They will just increase like you know ultra promax whatever because they have this tendency to to show oh we have something bigger coming up. Oh, what we had was nothing. We have something much more dangerous. Of course, it is in their best interest to act this way. So, by the time you are watching this video on our YouTube, maybe it is called ultra super max intelligent. Ultra supermax pro intelligence. Yeah. So dumb. It's so dumb, man. Move on. Move on. Huh? Yeah. Let's move on. Let's move on to Jev. I had no [laughter] intelligence report. Oh, I didn't I didn't have a chance to read this, but are we going to go to intelligence report? I have no idea. Well, let's just keep it brief then because it is it is wild. Let me just prepare the page. Um, so anthropic actually we're on the subject of anthropic. Anthropic have just published a uh detecting and countering misuse of AI report in September of 2026. That is this month. And this report I'm going to be Oh [ __ ] I did it the wrong way around. There we go. This report do that one. Um basically demonstrates a I guess it does what it says on the K on the on the tin of detecting count and countering misuse. Now, I saw something that basically said Kimmy were rrooting requests to Claude. So, they were just not answering the question. So funny. Here we go. Okay. Moonshot serves Claude instead of Kimmy and collects exchanges for model training. We we speculated before this this show of what it might be because we hadn't read the article yet. I just sort of heard hearsay and I was like they're probably intercepting it for their training their own model. So they probably still have their model but let's just read it. Let's just settle down. Get get some popcorn boys and girls. We discovered that Moonshot AI, the company that produces the Kimmy family of models, silently forwarded custom requests to Claude instead of processing them using Kimmy. K. Moonshot then displayed Claude's responses to users. These users thought they were using a Kimmy model but received responses from Claude instead. And I did see this when people were asking, I don't know whether it was Kimmy to be fair, but like I do remember seeing some tweets where people were like, I've asked what model you are and it says it's Claude. So, I think people were kind of on to this, to be honest. In one instance, over a 10-day period, Moonshot relayed 300,000 customer requests to Anthropic. I bet a tenth of those were your buddy there, right? Oh, no. Cuz he was using Open AI, wasn't he? Sorry. Yeah. Um, [snorts] the vast majority of which were rooted to Opus. Moonshot used a proxy service network of 5,380 fraudulent accounts, most of which appeared to be located in Singapore and Japan. So I think Kimo Chinese company, right? So they just they needed to Oh, hello. Um, in addition to serving plausible sponsors to their customers, Moonshot also captured and saved at least a portion of these exchanges. Moonshot built a coot extraction pipeline. I wish they would. Uh, let's define that. Should we get perplexity on the case? Can you see that? Is that popping up? No, it's not. Okay. It's thinking. It's thinking. It's thinking. Okay. [snorts] Uh, chain of thought. Chain of thought. That's what it means. Um, extraction pipeline to extract Claude's uh chain of thought transcripts from those saved relate exchanges to train its models. Moonshot also extracted chain of f chain of thought transcripts harvested through other means. Absolutely bonkers. I don't think we need to go through anything more than that. But this that's just crazy, isn't it? Oh, should we just go through the top? Deepseek serves claude instead of its own model. So, Deepseeker doing it as well. Deepseeker doing it as well. Jesus Christ. Um, distillation campaign by Xiaomi. Gosh, they're really throwing everyone under the bus here. Sense time Minia and the third party reseller ecosystem. What's this one? The proliferation of proxy services to circumvent anthropics access restrictions has created a secondary market through which labs can purchase or otherwise acquire harvested exchanges between users and claude. Some proxy networks both provide clawed access to users in supported regions and also have exchanges in order to sell them to other labs. For example, Sense Times distillation pipeline excluded included transcripts of user exchanges with claude purchased from third-party vendors. So, they've bought the exchanges. They're not they're not um rooting it and collecting those exchanges themselves. There's literally like a whole black market. That's crazy. Yeah. Jesus Christ. This is uh data distilled data. Yeah. This is um great news for uh and Alibaba and Quinn are doing it as well. Wow, look at all these bloody Nora. So anyway, that's bonkers. Look how look how long this Can you see how long this article is? Yeah, that is one heck of a report. So yeah, Open AI are like, oh man, we why does why does anyone want our data? Why didn't they want our model? [laughter] Yeah. So yeah, crazy crazy work. Crazy work. Um, anything to reflect on that? Not really, but speechless. Just Yeah, back to our conversation [snorts] that that's that's what they are doing and it seems like they are not really that ahead or like they are further behind the Chinese models and they're just distilling. And just one thing, the numbers, I don't think they are that big. Like 300,000 requests, that sounds like a weekend for me. [laughter] Probably. Out of seriousness, I don't know. 300,000 requests is not as much. I don't know. I think Yeah. Yeah. I wonder. Can you do config here? Can I Can I look at my usage, my stats? I'd be interested to know. I I never look at my stats really. I'd be interested to know how many requests I've um done saying that. Um if do you see Anthropic giving away $250 worth of uh cloud managed agent credits? All you've got to do is start a Claude code session in the cloud and they'll give you $250. So that's quite nice. Um yeah, but yeah, I don't know. I don't know how many requests I I do, but I reckon, yeah, you're right. 300,000 could be probably quite um favorite model. Opus Opus 5 is apparently my favorite model. Longest session, 17 days session active. No, sorry. This is very uh uh boring. I'm just looking at my stats on Claude Code. But yeah, let's um let's move on to Jeff, shall we? My name is Jeff. Yeah. [laughter] You don't know the You don't know the the meme. Who knows the meme? Just Just comment like, "My name is Jeff. Anyone knows that?" No. No one. I'm on You're on your own. Everybody needs to make fun of that. Yeah. So, after co-inventing chat GBT, I kept asking myself, why have superhuman chat models not led by AGI? I have no idea what that means. I spent the last two years Right. Yeah. Oh, I see. It has not led to AGI. Sorry. Sorry. I spent [snorts] the last two years in stealth building a new way to train models RLCD and a new type of frontier model that we are releasing today called Jev. My name is Jev. 20 to 200 times faster, 40 to 400 times cheaper with output tokens being free and Frontier Composable Intelligence optimized for decisions. Now, this might be the intro because there's a lot of buzz going on around Jev right now and almost everyone's getting it completely wrong. So hopefully we can just break it down to why you would want to use why this model is so exciting and what practical use cases you could actually use with this model. So Jeff, oh go on. What is a type safe model? Because it says it's type safe. So remember I said it out outputs JSON. Yeah, you can ask it to like respond in the JSON in the type that you want it, whether it's an enum, whether it's a string, whether it's a um, you know, a URL or whatever. Like it it returns, this is my understanding. I'm not an expert. I'm not an expert at Jev, but my understanding of type safety is when you actually say this is a string, and that thing can only ever be a string, you know. So it's very it's and this is the key word it's as deterministic as you could possibly get with an AI using Jev when you talk about type safety you're saying I want a string it will give you back a string that's deterministic so that's what it means here and maybe we should maybe we should have got some um examples but I've seen I've seen people build UIs with uh Jev like it's it's literally just like spewing out a UI at the speed of light and it's super super fast and people are building all of these tools and it's not really demonstrating what Jev is now. Jev is a what do they call it? A maybe this maybe the word is here. [snorts] It's oh gosh system one. Well system one is what? So they're going to release two models, right? system system one and system two which is this idea that your brain has two systems. It's from a book called thinking fast and slow. So it's actually a UX book. It's user experience. Oh, is it? No, no, it's not. Don't call me stupid is uh the book that I'm thinking of that also references thinking fast and slow. Thinking fast and slow is a big thick book. Um and system one is like your immediate gut instinct feeling. No, no thought, no anything. It's not a thinking model, right? And this first release is about the system one thinking. They're going to be releasing uh a system two thinking thinking one which does do that secondary level of thought and intelligence. So it's we should probably show this as well. um 0.42.042 uh per million tokens in and then output tokens are free because they're too cheap to meter with how with their new architecture. But Jev is named after Jeban's paradox. You know what that is? Um and um it's a why is the word escaping me? I'm looking for it here. It it's an assessment model. It's a it calculates and and um compares data as opposed to a generative model which produces tokens and outputs what output stuff. This outputs but it's it's a way of taking a subset two sets of data or or a set of data a set of predetermined outputs and comparing and reasoning about what the return data should be. So, this could be a huge database of things where you want to um tag give me all of the it's an advanced filtering system like give me all of the items that are tagged with the color yellow or something like this, you know. Um I'm still waiting for that [ __ ] word. But a practical demonstration, a practical example of this might be what we spoke about with, you know, um, Reloom's AI or Web Flow's AI, right? They're not they're not really generating stuff out of the box. Like Reloom's AI is a perfect example because they have all of their components. And what their system does, their little builder does, it generates pages, wireframes for for you based on their component library. that is in of itself based on deep user experience research and design fundamentals. Right? So they have a subset already of these components. Now Jeff would be perfect for them and much cheaper and much faster if they say take this user's response and generate a page that uses the best you know or most um sensible components [snorts] uh based on their request basically. And that would be a great use case for for using Jev. It's not a new it's not a thinking model. It's not it's not doing any sort of it's not producing facts or or things from nothing. It it's literally it needs data to work well. So that's why this is quite exciting and and as I say deterministic is the key word there because this is as deterministic as you could probably get with an AI. So that's kind of Jev in a nutshell. And it does sound a bit boring to us creative folk because it's not really doing anything. It's mostly from a data like organization standpoint, but it's a it's an interesting progression and it's uh again things being deterministic is is always a win. Still looking for that [ __ ] word. Um, it's a uh model. That's the word I'm looking for. Never mind. Anyway, Jev, interesting. Interesting stuff. Um, Opus and all of the models. Should we go into all of that? It's kind of like yesterday's news, but yeah. Um, uh, I would Do you want to go like separately or together? That's kind of like the question. Well, let's just do them all together and and just talk about our real world experiences with these models because I've been using Opus 5.5 over the last week and I honestly think it's it's such it it's improved on almost every aspect of Claude that I've struggled with except one. There's there's the intro. [laughter] There's the hook except one. And that even agreed with me on not with me but like he said it and I was like thank you. I'm not the only one who really wants this um improvement right now. I want to know what that is. There you go. You got to wait till the end to find out. So I've [snorts] been coding with it over the last week and honestly the speed is the first thing that struck me about it. It is so fast. like it's sonet speeds, right? And it's and the output of it is just it it sounds a lot better as well, which has now led on to them boasting about it writing capabilities because there's no more m dashes. Five was a pretty bad release, I think. And I'm so glad that they've they've uh fixed a bunch of things on that because it was over it was just really verbose with its output chatting about crap and just I actually actually had to put it and they even added a feature into Claude code to be concise be concise with its responses cuz it just it was just chatting and chatting and chatting. So from a speed standpoint it's a lot better. From a from a tonalistic standpoint, if that's even a word, it's a lot better. It's apparently fable level intelligence and ability for 40% of the cost of Opus 5. So, we're getting this is already this is better than Fable. In fact, let me get the uh benchmarks up because they are worth looking at. Um, and you mentioned the you mentioned the intro. Let me just refresh that because it's quite nice. Um, can you show it with So, oh, you're on the website but not Yeah. Yeah. Uh, no. No, this isn't the video. This isn't the video. Um, we're introducing Opus 5.5 first model claw 5.5 family. It performs a level of fable 5.1 at 40% less to run Opus 5. But what I wanted to show you is yeah, here we go. There's the highlights. Cost and speed, uh, communication, everything that I was just talking about. So, Gentic Coding 60 66. Fable was 55. Fable was 55 in coding. This guy is 66, you know. Um, Astra was 57. None of these are touching even touching the um sides of it. Frontier Coding 54 over 50. Cursor Bench uh 57 over 51. I'm really just looking. It's just it's smashing all those benchmarks. And then Astra beats it for business workflows and automation bench and scientific research as well. So, you know, still some improvements there, but just insane. They brought it to cash reads being 20 cents, input tokens four, whereas Opus 5 was five. So they've um shrunk it by a dollar there, shrunk it by 30 cents there, shrunk it by $5 there. Um and then 125 there. So really, really good. And that's it at a glance. Like on medium, which is what I use it on, you could probably knock if you if you aren't if you aren't as price sensitive, you could probably knock it up to five. uh for a high. But even medium is outperforming um Astra. Uh where is Fable Green? Medium is outperforming Fable on max which is crazy. Uh we are taking a dip on max. I don't think anyone should be putting on max. I think you you just overthink the problem then. You might net marginal improvements on X high, but I think your sweet spot really is medium. If you're Yeah. high or medium. Um yeah, [snorts] so that is um that is Opus 5.5. The aspect that really annoys me that they haven't fixed yet still is computer use and browser use is a little bit better, but computer use is just really [ __ ] It never I've I always resort to chat GBT or Perplexity or Hermes to do computer use. It's rubbish. And on top of that, I Let me get the um let me get the Chat GBT app up and running. Just their app for computer use, which they call remote in Chat GBT. Yeah. Um, go away. Like I don't know whe you whether we can. Uh, it's a bit too bright annoyingly. Uh, let's drop this down a little bit. Uh, you can't really see it. Basically, I can h I can have multiple computers set up for remote. Claude, you can't do that. I know this is more of a feature of the app than actual using the computer itself, but I can just switch between my different computers. I can control this computer. I can control that computer. Clawed, you just dispatch and it's only linked to the computer that you've set it up with. And nine times out of 10, I don't think I've ever had a successful run on computer use using dispatch at all. So, um went on the um Matthew something some AI dudes podcast and he was like yeah he was asked like what would you like to see improved with the models and he said computer use and browser use and I was like thank you I'm vindicated. So yeah. Have you you have you've been using GLM, right? Have you been using 5.5 at all? Yeah, I've been using 5.5 as well, but not as much. I I did a few things here and there with it. Yes, it was really fast. Uh, and especially compared to GLM because I think GLM is quite slow or maybe the server servers are maybe that's why it's so cheap. Uh, it is quite slow, but it's good to leave it running. With Opus, I didn't have to change, especially for small changes. I didn't have to switch the tab to go to do something else. I could just continue chatting with it and do the work with it. So, that was really impressive. And the quality, the quality is good. I yet to do design work with it because this is where I still see Fable ahead of everything else. Fable can just it's just so much better at design. It feels like it understands. Why don't I use Claude design? um to design redesign my my website. Um I'm just going into the chat now to see what model okay it was using. So as you do this um just to quickly mention something really funny that they did with with this model release. So the clot uh opus 5.5 video it has this theme song. I don't know if it's coming through or not now. It should be okay. So it has this theme song and I I heard it and I was like I know this I know this song. It is from Pluribus. This is a show on Apple TV. It's really really good by the way. It has like 99% on Rotten Tomatoes. It's really good, but it's fully like apocalyptic. [laughter] So, it's quite funny that uh CEO of Anthropic is like we must pace the frontier and then he puts out a frontier model and what music do they use? Something that is Yeah, apocalyptic. It's It's quite funny and I don't know if they are like joking. Do they see this as a like as a joke that this will end us all? Because this is like kind of like smart beings. I don't want to even like call it like smart smart being, but it's kind of like yeah, smart beans. Uh but apocalyptic. Anyway, I noticed this and apparently like more people also noticed it in the comments and I was like, "Yeah, it's not just me. It is that." And it is very odd uh to do it this way but got funny. Yeah. So it was Opus 5.5 that I asked to design my website and this is what it came up with. Wonder what you think. We move that. So these were the first three designs and I pointed it to my website. Uh, I pointed it to my website and homepage and then asked it to give me just three designs and whatever. And I went with this one. I chose this one and refined it here. Mhm. And then I went, okay, I actually didn't go with that. It's orange, by the way, but like I didn't go with that one. It's too much. Just too much. So I just went, okay, using this one. That's the dark mode obviously. So using this one designed the rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest rest of the pages and so that's the design and then so I took that then and downloaded the HTML files and then uh if I share my screen is what I've been using. Um where is it? This one. Where is it? Here we go. So this is what I've been you going to at the stage. Here we go. This is what I've done. Ah, nice. And I made it into like little scroll thing. I put in the um what do you call it? Um the this is a CMS, so it's all payload and stuff. So uh I asked her to interview me about the projects cuz I hate case studies. like you must be like me like hate case studies. So I was like just interview me about the projects and then just create the case study stuff. So um yeah and uploaded some images and things like that. So yeah it's um I'm I'm I like it. I like it. I think it's bold. I think it's brash. I think it's um interesting. I added this I I told it to do this. Yeah. But like you know um Yeah. So, not sure what your thoughts are on like Fable versus Opus 5.5, but I was quite happy with that. Yeah, it is. It is quite nice, but to be honest, it is a bit template like it's not unique enough for me. Um, it doesn't reflect uh a specific brand uh as much as I would like it to. It it it needs a little bit more soul. And you would say that Fable And you think Fable is better at that than Opus? Generally speaking? Yeah. Yeah. Interesting. Interesting. Um, we're kind of coming up on time a little bit, but like the my only thing to add about GPT6 or, you know, I was using Astra. I can't really speak for soul or or uh terror or whatever, but I was using uh Astra and have you ever used voice mode on ChachiBT? I do use it all the time like on on my phone. It's so good. Right. I was using it to learn Albanian phrases. And what struck me about it was like I would [snorts] I would say the word and it can hear it can hear the syllables of the word but it would be like change that that syllable. So in or something like that for learning words it's well in my instance it was learning words was really really good because it can it it says you've got it didn't say you've got the first half of the word correct it's like it's sir day or whatever it was I've forgotten exactly what it was I've already forgotten my Albanian but it was just so nuanced in that respect and it it ums and it it sort of listen it it really feels like you're talking to someone which is Crazy. Yeah. Interesting. I I still need to test that. Yeah, definitely. Definitely. Um, and I did some legal work. I was doing lots of legal stuff at the beginning of the week and I was using Astra and I looked at my usage and they've removed their 5-hourly usage, right? They've only got weekly and I pretty much chewed nearly chewed through my week's worth of usage because of that. Like you can get so caught up in [ __ ] Like I do I think you should be able to add or remove the 5hour limits or whatever it is. I don't know what it is. God damn it. Yeah. Chewed through that. Uh I have a few accounts on chat GPT. I have one on like my private email and then one on my company because like I have my teammates added as a team uh plan. um the one with the team plan don't have the five hour. the one from my private one even even though all of them are the 20 bucks that they are identical things I have the five hour limit there and I had instances where I went through the entire uh weekly I heard some also I think there was some drama around claude I need to check that out because it's quite important apparently the $200 model that you get doesn't give you 20x X uh as they are saying it does give you 20x the five hour limit but not the the weekly limit which is what matters because the weekly is the amount of tokens and the five hour is kind of like a timeout. Well, it is also tokens but time out. Yeah, I'm not good enough at maths to know understand that but surely 20x would be better. That's what you're paying for for the month. That's literally when you say $20 a month, it's 20 times the usage. Surely it makes sense to put that at the monthly usage, not the weekly or the 5 hourly. I don't know. I don't know how it works. Well, it should be based on the they say the usage. You would assume the monthly because you're monthly. Yeah. Yeah. you're charged monthly and the the amount that you use in a month is actually limited by the tokens that you use per week, right? Anyway, apparently the the don't quote me on that. I I saw it somewhere, but I need to check it. Apparently, the $200 is only $1.8 times the $100 when you account for just the tokens that you get per week, which is not even double. So theoretically, it makes more sense to buy two accounts of $100 instead of like $100, which is quite [ __ ] It's really bad. Uh I hope that's not the case. If it is, man, I can't cancel. Like Fable is way too good. Uh Opus 5.5 is uh also like really fast and good. I'm going to keep it uh for more difficult tasks. So, um, should we skip muse and do you want to go over your Z code? Yeah. Yeah. Because I'm quite limited time. So, and let me grab it, bring it to my screen. While we wait, if you are on YouTube, give us a like. Give us a subscribe because I want to hit that 5,000. I think 5,000 we're trying to get to by the end of the year. That would be amazing if we don't get many on YouTube, but all two of you, there were five of you just now. All two of you subscribed if you weren't already. That would be amazing. Yeah. Yeah. If you haven't been using or even heard of Zcode, uh this is your sign. You should give it a try. It's really, really, really good. I'm using here, as you see, Zcode, Zode, whatever you call it, with GLM 5.3 Flash. In fact, actually, let's look at my usage stats. I've been using it quite a bit in the past week. 1.2 billion tokens. I don't know how. Is this a lot? I don't know. It sounds like a lot. I've been using it a lot. Like pretty much every single day. I've been using it all day. Uh next to the other things. So I want to show you around and also show you what it can do, what it can't do, where is good at, what's not so good at. So right now you see like my typical view and how I work. So I have my own I have my projects. These are my directories. Um, and here you see my chat and on the right you see my browser. Like every kind of like every other app nowadays have has this. So you can add a new browser, a new terminal, review, whatever. Start a conversation. You can do all that. Uh, right now I'm lagging quite a bit. I think it is because I'm I'm also streaming. I never had this lag with it. It's actually it has been really performant. Anyway, let me see if I can get it working. So, one thing that I really really love about it is this thing that I can change my browser window even though I have not I don't have a lot of real estate, but I can just kind of like scale things so it does look like I'm on a I'm looking at this on a bigger screen. it is really really helpful for developing whatever that I'm doing. So I can do even I it it obviously goes the other direction as well. So I'm doing this app that you see here here like that. So that's really nice. Um, additionally you can well you you can do interesting things like let me show you this uh you have like workflows similar how you have like in cloud code workflows you have like swarm of agents I'll show you that one of them is running and here uh you see I was doing a test here you you get like these really fancy nice UIs you have these UIs that depending on what you do you get them? For example, here you see this one with these like colorful dots and whatnot. Right now, it's because of the stream I'm really lagging, but it normally it's really really good with performance. Anyway, so you have this like really cute view into the agents and you see like each one does something and you can go into them and you can see in detail where they are and you see the tasks that they have or the terminal and what they are working on like you can go into details of this. I really like this. Uh I think the UI here is just brilliant and it makes it just such a joy to use. This is just kind of like one side of the whole thing. Um, let me see where were we like here. So, just to show you like this UI that you're looking at. What I what I typically do, the way I use this is not just on my computer. I have my phone connected and I can see this entire like all of these projects, all of the the chats that I have here all from my phone. It's a web app that I added to my home screen. It it for all purposes it it looks like an actual app. Uh and I have everything. I can I can spin a new agent per project on my machine but just from my phone while I'm away. New task spins it up on my machine. It's not like the way Claude asks you to do it in the cloud or to do it with a GitHub repo. you're it can be all local on your computer. It's really good at understanding at least in my experience I mean this is partially because of the GLM 5.3 at what environment it's working in. So it opens uh something here it it it does the work here and it it's really good at that. Let me show you a few example. For example, here you see this video. This is a video or like here this is also a video of that that component if it plays for some reason it's not playing right now I'm still building the point here uh all of these are videos that the agent did so so I told it to build this component and also make a video of it for the pre and it does and like all of that using its own built-in browser and it's really good at doing all these agentic stuff and even the thumbnails and all of that and I yeah I've been just having a lot of fun doing that. Uh I made this component. I posted it as well. It's quite dumb. There's no point in this uh based on a tweet I saw from G Simon uh I believe his name was. So I made this all with GLM. So, just also to tell you about how the model is, I'm just using the the GLM 5.3 flash. I use the 5.3 a little bit, but I think the flash is better. So, yeah, I'm using it. It's definitely not Fable. It's definitely not Opus 5.5. It might not be any Opus at all, actually. It might be more like a sonnet model. It needs quite some babysitting but the type of work that I've been doing I didn't build this whole app with it the build the app was built with Astra and with Fable but the components and like doing some changes like add a new like tutorial thing add like these little things or like a button for the GitHub uh for the article you know things like this it can do it pretty well or like alter stuff. So, not even like editing stuff like actually add stuff like this entire um component was completely built with GLM 5.3. So, at the end of the day, uh I I I think it is like worth giving it a try, especially how cheap it is. For 18 bucks, I'm on the lowest tier. For the love of God, I cannot get my like I cannot spend all the tokens that I have no matter how hard they prompt. Yes, I'm on the 18 bucks and I I push it like every like have so many so many tasks running at the same time and I just cannot like I I haven't been pushing it today as much but as you see uh in my usage I had some quite heavy heavy days like 120 320 uh million tokens per day but Yeah, I I can't um push it. And I had actually a few times, but then you have um use it reset. So it it gave it given me already like three resets already. So yeah, and it has like this free version today. It gave me 300 million tokens, which is just like that. So, I'm instead of now choosing the the paid version, I'm using the the free tokens first. Use that app and then use your paid version. Yeah. And it's like so nice that it allows me to do that. And you see here uh my weekly at 51%. I wonder why there different models. Why Why don't they just track your usage and be like, "Okay, you've used your free allowance rather than having to select the model." Uh, it does like actually when it finishes um it tells you like you can switch and yeah that's what I mean but like why switch? What's the seems like an unnecessary step if the models are the same but maybe they just get sent to different servers are a bit slower or something like that. Yeah, could be. Could be. And you can you can obviously use the API to add um other providers. I just wish that I had this app mixed with something like T3 code because like in T3 code I can use the other harnesses. I haven't used T3 code in maybe two three weeks. So maybe I should give it a try again. But man, this is this is doing so many things right. Especially the browser and the phone. These two are just just working reliably. I haven't had crashes or things not working or like slowdowns except now that I'm streaming. Uh obviously I I've just had so much fun with it. And I haven't been using chat GPT as much because I've been using this and a little bit of uh Claude because I have the tokens there. I got to build with that. be interesting to get Claude to review the code that GLM has written and see if you can, you know, you it's so confident in telling you the code is right, you know, um it could actually be trash. I mean, it works, but you know, um yeah, it's definitely also not as creative. It's not as good with design. So, if you want to do design work, it it's not as good. But I can give it something and then say, "Use your vision to um validate that you're 100% there." And that gets me quite like far. Yeah. And just to be clear as well, Zed Code is a ZAI product from who make the GLM model. So your your Moonshot? No. No. Z AI. Z AI. Okay. Moonshot is Kimmy. Ah, okay. True. Right. Right. Yeah. So, that's why you're stuck on the GLM model, which is a good model. I I I use GLM for because you can you can I don't know if you know this, but inside of Warp, you can actually just bring your own API key. So, I just use GLM [clears throat] inside of Warp. Um, but yeah, I I I use the middle plan, the the not the basic one. So, they but I know that they've changed their plans, so they might have you might just get more usage out of that them now. you know, whereas I'm stuck on a legacy plan. So, I don't know. But yeah, looks really cool. Looks really nice. A lot of inspiration from Claude there, but it looks like they brought, especially the workflows thing, they've just brought a lot of UI nicities into into the experience. Yeah, it's quite nice. It's quite nice. Cool. All right. Well, we're 10 minutes over, so should we wrap up there? Yeah. Cool. Well, thank you for joining us. Make sure you like and subscribe. We're trying to get to 5,000 subscribers by the end of the year. That'll be amazing. Um, watch out next week for all of the clips from all these videos and and short form content, all the rest of it. I've been Sam. I've been Kavarza. And I supposed to say something like, "And that's your week in AI design and dev." I don't know. Keep on vibing. No, that that we say at the start. Now we say don't be a meat proxy. Don't be a meat proxy, [ __ ] Keep [laughter] on vibing.