#63 Cool New AI Tools and Workflows We've Been Using

video show. Hello there. What were you asking me? Hello there. [laughter] I looked down and there's just this little head. What do you mean little? Oh. Sorry, big head. It's ginormous. Mhm. Oh. Timer is My timer has gone off, so I gave Kabosu 3 minutes and I set a timer for 3 minutes whilst playing very, very tense music. It was so hard. Yeah. You got it done in half the time, literally half the time cuz the tense music was 1 minute 26 seconds, so and then the timer went off. What were you going to ask me? Um so, I want to ask about a specific video, like a segment of the live, and that could be about Well, I want to talk about it about the next Webflow tease. They're teasing us as Webflow Conf is approaching. Mhm. And I wanted to kind of like mix it with the comment section looking at the comments from our videos. Okay. Okay. What a What a a melting pot of um different creative things. You must be so creative. What was that? I don't know. What What do you want to say? Go on. Just Just Just do it and we'll see how we'll see how it goes. Yeah. We'll We'll We'll get to it. We'll get to it. Not too far. don't want to do it now? In a minute. Um we Don't we want to do the intro? People of the internet, we got to do the intro. [clears throat] Oh, I wasn't live on Instagram. Now I am. Okay. So, how do you do you want to do this intro? Hello in Instagram. Hello there. Well, the intro is just going to be This week, we have news 1 2 3. No. I haven't written it. So. [laughter] Well, this week, we have We're going to do things again a little bit differently. This is now live, by the way, just uh um you know, dispel any of the rumors that might happen. We are live. So, you can talk to us in the comments. Um [snorts] we We're going to do 1 2 3 4 5 bits of news today. And maybe I can share my share my screen on all news bits. But, we decided these are just too freaking boring, to be quite frank. So, this is our notes. Oh, I can't. Hang on. Here we go. I've got multiple screens going on there. So, we've got We've got Stripe finalizes a $7 deal to acquire Open from Replit. OpenAI halt Astra, which is their new fable model, um after model nears critical cyber capabilities. OpenAI previews ultra-fast mode with 14 times speed powered by Cerebras. So, they've moved on to Cerebras' architecture. Z AI unveils GLM 5.3. And DeepSeek uh open-sources DeepSeek Harness. And quite frankly, just decided they were just too freaking boring. I'm not excited about any of those. However, if if if you wanted us to cover them, if you want us to dig in a little bit more, and there's more to it than, you know, it may seem, then just drop us a comment. I've been Sam. [laughter] And I've been Carper. And I'm This is a week in AI design and dev. Boom. We can freestyle. Yeah? Pretty much. Uh okay, so let's me I'm preparing stuff. So, what we're going to do this week as the title has maybe, you know, suggested, we're just going to kind of go through some of the stuff we've been doing this week in terms of tools we've learned about, workflows we've been using, and it's probably going to be a bit shorter, but a lot more floaty because you guys seem to like going through the tools that we use cuz, you know, our finger is on the pulse when it comes to AI [ __ ] basically. Um Yeah. So, yeah. But, I don't mind waiting cuz we've got we've got I don't know what's going on on your Instagram, but we've got no one joining. We haven't We haven't had anyone join thus far. So, Yeah, Instagram says hi. Hi back to you. Hi back. Um people on Instagram. So, well, I'm I have the the blog from Webflow ready. I want to also I have a funny thing I will add it in a second. Um Actually, we also have that cool thing that we wanted to show, you know? No. The the real-time depth thing. Yes, yes, we do. Yeah, yeah, yeah. [clears throat] Okay, um maybe maybe I start with that. Yeah? Sounds good. Okay, now. Um, is it a white noise coming from you? Something like that. my fan. I was hoping you couldn't hear it, but I'll turn my fan off. Let me just I I guess you can hear me. sweat. Let me just No, legit Let me just sweat. Right, is that better? Yeah. It is better. Nice. Nice. Okay. Cool. So, let's show this and well, I don't want your screen there. Let's do this layout. All right. Unless you've been living under a rock, you probably have seen this video. It has almost 10 million views and look look what the Can you explain like what you see? Like what is so impressive about this video? I saw it and I immediately I was like oh, holy [ __ ] Is this real time? And then I read it. It is real time. This is happening real time. But what is it this that I'm talking about? You know what that is? You want Is that question Sorry, is that question to me? Yeah, yeah. I want you to explain like what you're seeing here. Like or like why is it so impressive? Yeah, I mean, you're it's that light is not there. Quite frankly. And through some sort of process or whatever, it's able to turn and not only the shadows well, first but the intensity of the light in the various areas in which it's being cast upon, including blocking it completely from his hand. So, to do all this in real time is is computationally very, very cool. Yeah, not to mention this is a 2D image. It's a video, but it's at the end of the day it's an image. And we are like casting a 3D uh light at a 3D feature actually into a two and a 2D image. So, we have to determine depth. We have to determine, okay, what is the closest to the camera, what is the further furthest away from it, kind of like a Z index. And then, based on that, we can spread the light into in a fake 3D space. Funnily enough, this is what I've been playing with, but for a different project. I will show you in a second, but what's really crazy, this is done locally and in real time. So, depth-aware light injection in Type GPU, we have covered it in a really like past video. I got a 448 um by 448 monocolor depth model. I'll show you what that means. I will just show you in a second. Depth model down to 8 milliseconds on my M4 Pro across 250 uh dispatches. I'm not sure what this means actually. To be to be I think this is like the intensity or like the layers. I I will show it, and maybe we can determine what that mean. Uh which is fast enough to use uh in real time. Since No, actually, that's not the layers. Anyway, uh to use in real time. Since the interface is written directly in Type GPU, I can just feed the depth buffer straight into uh the lighting pass. It never has to leave the GPU or go through any extra synchronization uh or interrupt step. So, this is really cool, and I just opened my Twitter, and I think I saw like Yeah. Like you um I saw another video just just like that, which is really cool, like right? Uh and the way it works, I can show you what I have, which is not real-time and not for the same purpose, but similarly, you see I have like this seemingly 3D um model. It's not exactly 3D. It's actually it was one image. I did cut it in multiple pieces, but it's essentially a depth map. Well, multiple depth maps, like the drink is also different depth map. And where we have the light, the the white, it's the closest to the camera and dark is the further away from it, and based at that information, we can do all sorts of stuff. We can uh distribute light intensity. We can also create animations like what I have here to create this fake effect where we are manipulating the pixels are that are closest to the camera more compared to the pixels that are more further away. So here, you see the drink looks like it's moving because the pixels are closer to the camera and I'm moving them more compared to the other pixels, and it looks also like a bit 3D. It's a fake 3D. That that's how it's done. And the way I'm getting this done is using this. So this is a a model, a machine learning model that you can run locally, made by Apple. A depth called Yes, good. Good point. Depth Pro made uh by Apple, and it's it's really good. It's insanely good. You feed it an image and it gives you another image that is a depth map essentially, and you basically put these on top of one another and frankly, I didn't do any of this. I just I was aware of this and I just told my cloud, "Hey Fable, would you do this for me?" So, basically Fable did this. And what this guy has done is much more impressive, not to compare. [laughter] I'm just showing what I was like actually working on and mine is, you know, not nearly as impressive, especially given that this is run um in real time and entirely on the GPU level. And yeah, it's really cool. People see impressive stuff and yeah, it They click like It's really cool. Yeah. Yeah, so like um cuz I I don't think it's immediately obvious what how the depth map translates to this light image and um that's the creative side, right? They've they've taken something as a view. I I'm I'm guessing that the a majority of this parallax effect is actually more so GSAP moving the individual PNG files. Um Well, I can turn that off. Yes. Yeah. Yes, but then you're also using a you're having a layer of um what do you call it? Uh you said the pixels are moving or whatever. So, there's a secondary thing that's also using it. It's separate uh what do you call it? Uh depth maps, right? So, the depth map isn't responsible for the parallax effect. Is that Is that right? Um The parallax is done in two ways, both with the depth map and with having the layers in separate layers. You could never achieve the depth the the extent of that 3D though just with the depth map. You could probably achieve a little bit but not mental. You're right. And you've done a light thing as well. Is your mouse created a light thing that Yes, I wanted to I wanted to show that. I have another one, but I have yet to run the servers because they are not running currently properly. Maybe I need a minute there because um Cuz then, you know, um the the orb, the light orb looks like it's being also cuz the thing is with with WebGL, it's just a canvas and you're just plugging more and you're plugging layers and layers and layers into that canvas of effects. And they've plugged an orb in that is also registering or is looking at that depth map that's being real-time generated from the video input, I guess. So, it's just again, it's just like a clever use of um of the the depth map. I wonder what what the Apple thing again just to break it down even more, the Apple thing is simply the tool that's able to look at an image and understand its depth from a 2D image. Now, we put it through Higgsfield before this before the show and it was able to do something with just one prompt. In theory, I think AI is comes such a way where if you can get the AI to not change anything about the image but to but interpret the image as a depth map, I think you can also do the same thing with ChatGPT probably. Um I don't know how exact it would be and actually I have the image now. Let me I'm going to save them and show them while I'm also running the the thing. So, we have this image and we have Okay, so it doesn't look like the same to me exactly. Your screen is not being shared, by the way. I know. I will I will do it in a second. Because I'm preparing them. Yeah. That's that and Cuz at the end of the day, you think of some of the features that Apple have on their cloud photos is that you can actually just click and using the AI, you can then move your phone and it will create that depth. I think the Vision Pro uses this as well, right? Uh that's probably the technology they've just given out, you know? I'm Um partially, probably, yes. Cuz Oh, okay. So, it's a GitHub repo. Maybe I can share my screen then as I'm working. I'm ready I'm ready to show this. Yeah. Okay, go on and go for it. Go for it. Yeah. So, I have two two things to show and I'll try to So, regarding That's me. regarding the question how how far can you push it without separating the layers, this is a good example. So, I can definitely push it further, but you already see some artifacts here. It's hard to see it, but now you see it. If I get the further away, you begin to see some jagged corners because you cannot like cut the image and you know, under it you need to have a you need to have something under it. In the example that I showed here, where I'm cutting things, I'm actually regenerating the image. So, the sky is not, you know, empty where the model exists. The sky has been regenerated after that. So, all of these pieces are regenerated. In this example, this is one simple image, one Claude prompt saying, "Hey, use this Apple depth pro to to to generate a depth map and then animate based on a mouse hover interaction to create a 3D effect in addition to a light uh source uh reflecting on the image on when the mouse moves over the image. So, essentially something like that. And that's the result you see. It is already like giving a really nice vibe, but it's nowhere near this depth, you know? Yeah, it's just it's just a lot. Um So, this is just has way more depth. So, that's that's just running it locally, Claude taking care of everything. All you need to know is that such a thing exists. Now, the second thing would become or like the second question is can you do this without um that model? Can you do this just by giving it to an AI? And this is one image and I asked AI to create a depth map for it. And this is the depth map. Oh, nice. Um it does look nice, but it's not a match. [snorts] I'm looking at those arches on either side as well. Yeah. And I was I was looking at the the trees and everything. Yeah. And I explicitly told it to not um change anything, regenerate the same thing, but it it's surprisingly good, but it's just not there. And that image I can actually show on that one here as well. Here I have it with Apple depth pro generated and then animated. Uh you see it's really it's really tricky. It's really nice. Yeah. So. Um before I just get into the nerdy stuff as well, I just want to shout out to everyone in the comments saying they love the content and they're subscribing. Cheers, guys. Um appreciate that. You look like some seniors. I'm I'm I'm assuming some senior developers, not senior citizens. Here is senior developers. And yeah, why not? Um we want to make a community, but that's your responsibility, Kubas. I said if you want to build a community, you take that. So, hound him to get it started and then maybe we'll have a community, but um Yes. We should do one in Discord or something like that so people can join and we can we can have a talk and can do screen share and maybe even people can can tell us what they're try to build and we can help them. That that would be really cool. If if you are into that, uh shoot me a message anywhere that is convenient. Just shoot me a message. But yeah. So, this was depth pro. And now we go off next to the next one. Should I just I'll just do a quick I'll quick I'll quickly show something. So, it's just a GitHub repo. So, this is what Claude would have done under the hood, which is go to this GitHub repo and clone it. Yes. Um create cuz it's all done in Python, so it creates a little um it creates a little environment just to kind of make sure everything's encapsulated. And then it runs that to download the models. So, I wonder get pre-trained models will be a file Here it is. I wonder what the model is. Make directory, checkpoint, get ML site. So, let's just copy that. Interesting. What's that? Oh, no, it's just downloading now. So, this is actually the model. Okay, so this is a PT is the model? dot 1.8 gig. Sorry, I can't zoom in anymore. too bad. Yeah, not too bad. So, that's the model. Not too Yeah, 1.8 Um Again, it would be interesting to try and find cuz there are some image specific models. I particularly like Kwench 3 as my image model, but all that to be said, 1.28 gig of of AI on your machine. Obviously, [snorts] and how long did it Do you Do you know how long it took for quite fast. It's quite fast. gig is not much. Yeah, I think it's done in matters of seconds if like not minutes. It's seconds. And I did talk to Claude about it. I was like, "What configurations do I have? Like can I Can I add a prompt and be like make the face to you you know, more bold or like you [clears throat] you know, add more light essentially make it more white so it's it looks closer." And Claude was Claude was like, "No, you just feed it an image and it returns you an image. There is nothing you can figure. It has a specific [clears throat] a range of pixels that it can support so the image should not be too big or it Mhm. will downscale it automatically. And that's it. That's it. I I'm not sure if it downscale it automatically or just Claude takes care of that, but yeah, it I think it's like 1,500 pixels. Uh something like that. Yeah. So, once all that Once you've got the model, once you've you know, downloaded the downloaded everything, you just run depth pro run and I is input so that means input and then you just point it to an image. So, you just drag drag an image into your terminal and then the output it will just export an an image that's black and or gray and white. Um yeah, using that model. Or you can even build applications with it. Like you can even work it into your application. So, you could in theory, well, now you can, build like a nice little UI like a web app that you just drag an image in and it will just pop it out. So, this this is what this is for. But of course, 2026 you just ask your agent to do all of this. But it's just nice to know and I think it's helpful to understand what's going on uh from a code perspective. I think it just you know, I know we're asking agents everything, but it's also just nice to know that you're you're you're along for the ride. It's not It's not this [ __ ] big wolf that you're taking out for a walk and it's just like check like dragging you down the street and you don't know where you're going or what's going on. It's just nice to know what's happening. So, pretty simple stuff there. Um yeah, cool. Um yeah, I want to show also this is how it looks. Yeah. Like the generations. Yeah. Yeah, so they were alongside. Oh, what? This is No, this was Claude. This was Claude asking. Yeah, this is this is more like how it is. So, it's kind This is because I have them in separate layers. If I didn't, it would do it all in one go and I have examples of just it being in one go. This is edited manually. And yeah, by the way, you can manually edit stuff to to do whatever you're trying to do. And I was trying different edits to see like how it affects and what surprised me, you cannot be you cannot like edit it like this. This was me like just being very vague with the brush. No, you have to It's surprisingly exact. Like it's an actual like map of the image. So, being vague with it completely ruins the experience. Like you notice it immediately. Oh, something is wrong with this. Like the 3D doesn't work. You have to be so freaking exact with it that you because you asked it prior to the show, can you do this manually? You could, but the the level of detail that there is in it, you don't appreciate it until you play with it and you notice. Like I did so many runs of like touching it with a brush, saving the image, loading it again, and noticing, oh, no, no, no, that that doesn't work like that. You cannot make the nose like pop up more by just adding more white to it. It has to be that perfect gradual Sure. uh color that it looks natural. Kudos to Apple, you know? They were open source. They were giving to Yeah. Yeah. So. They are quite small. And and again, this is probably something they've used It's 1.8 gig, right? They probably use it on your on your you've probably already got it on your phone, you know? It's able to if you've got a Well, they do that. They do that all the time. Yeah. Did you know that? Yeah, yeah, yeah. Um So, um by the way, shout out I took this image in London, you know, when where we met. Exactly. There we go. Yeah, exactly. So, that image has a depth and that is not leaving my phone. This is happening on on my phone. Apple looks at my images and thinks, okay, this one can have depth. So, I'm not sure if you can see it the top. That little button Uh oh, yeah, that too. Yeah. So, it's that's probably it. That's probably it there. So, now I can just twist that and you can kind of see it doing it, I think. I'm not really sure. see it. Yeah, it's less pronounced there, but yeah. Yeah. So, that if you have an iPhone if you have an iPhone take a picture of your friend showing you the middle finger when the hand is extended and then do this thing and you will see it looks very 3D. Mhm, but I think it can only work on iPhone 15 upwards. Cuz my phone's warm now just behind there did a little bit of computation and so like I can feel the warmth. Yeah. You got to upgrade. [clears throat] I'm just reading this right now. Two three weeks. I've used a very cool tool I think you'll like it. If you open a Discord I'll share it with you. I need to go out now. Okay, buy subscribe. Thanks for the subscription. Leave it in the comments if it's a cool tool leave it in the comments, you know, don't don't Send it anywhere you can. Yeah. Cool. Let's go to the next one. How should we tackle the Webflow thing? Let me think. If there's any evidence that Webflow behind If you need any more evidence that Webflow of you know, dug their own grave. No, no, we can't be too can't be too harsh. No, no, about not about Stripe Stacky but about the Webflow blog. Yeah, Webflow blog. Yes, I will I will I'll just talk about it in a second. So I want to think of like how how how I should do the intro. Okay. As we are approaching the Webflow Conf coming up in less than two weeks, Webflow again is trying to tease us with what's coming up and it again looks very cool and very futuristic and they are making all these like really interesting points into things that happened in the past. And this time uh in 1987, Apple shipped something strange and gave it away free with every Macintosh. It was called HyperCard. Have you heard of this? Have you been even there in 1987? No. No, [laughter] what? Okay. Built by Bill Atkinson. Uh the the same engineer who gave the world MacPaint. So, with HyperCard, you apparently you could write programs like visually, something like that. And the whole idea here is what Apple is trying to say without with not Apple, uh sorry. Webflow. W- what they're trying to say without me reading the entire blog right now for you is uh software should mold to your needs, not the way the other way around. And that says quite a lot. That's very much to me looks like this what I just showed in, you know, the previous video. Uh it it looks very much like this to me. Which is a UI that is molding to my needs. I can have pieces of this UI that I need and then pieces that I don't need, they don't show up. I build the UI for my need. For the project that I have. And this is they're not mentioning DalKit, which I is the tool that I just showed you that I'm using. Uh they're not mentioning that, but it surprisingly close to that reading through this. It looks like they're they're creating this tailor-made experience that they are talking about. And to me it again feels like it feels like um maybe Claude Design or even Cursor that have a UI. We'll We'll show you something like what it could look like, you know, Stack E made by Timothy Ricks. We'll show you that. But, yeah, in in this video we are just talking about Webflow and what they are teasing us with. And I think it's going to be this kind of like you deterministic, but yet very free UI that you can install or like add to your project based on whatever you need. And that to me is quite exciting. And this is what what they are uh trying to tell us. And we will We will know on September 1st. Another another Webflow teaser. Yeah. Um it'd be [snorts] Yeah, I mean, we don't know what they're going to be dropping and and how it's going to look. But, the only thing I'll say is um Uh I I know people don't like the look of Pinegrow and all the rest of it. But, this is something I mentioned with with Pinegrow and that uh you you put you [clears throat] you plug in a static website or a I think it works with a lot of other tools now. Um uh uh view, react, and stuff like that. No idea. But, the point is it gives you all of the tools that you might need to like make edits and change things and whatever. And my my my complaint to it was um why show me Like, I liked Webflow. I gave kudos to Webflow in this aspect. Like, why uh Webflow kind of hides things that you don't need. Uh let's say, for example, you've clicked on an uh an a child of an element that's hasn't got display flex on it. Why show me Like, so Webflow will only show you flex child like settings or controls if it's a child of an element that is flex. Flex. Pine grow showed you flex controls regardless. And I was like, why show me flex child controls if the parent isn't even flex a flex element? Yeah. Their response is and I think you sit in two camps on this one. Their response was and I I I get it is people like to work in different ways. People's mind maps work in different ways. So some people like to work from the bottom up. They like to specify their children and then the parent. Some people like to work top down. They specify the flex and whatever. And I was like, that actually makes really sense a lot of sense like the mental model these people So to insist on what controls you will need assumes that you're happy with the mental model that Webflow are insisting upon you. So long story short is he's left the room. Long story short is it will be something will be interesting to to whether they can nail or get right. You know? Maybe you want to work in a certain way. So it will be Yeah, it will just be interesting to see if they nail it. It's a bold decision. Yes, but I think they need to give you a set of tools. This is what pretty much all of these live coding tools have failed at. Uh a determined set of tools because I think it matters that uh more than ever that your controls uh should mold into your work, but also they shouldn't limit you. Um so there should be something like this. I really I recently uh found this tool and this works really well with my mental model and I was building something similar. I'm going to continue build it because I was already doing it and mine might be yeah, a bit different, but yeah, something like this. So, this tool allows you to create your own artifacts and use their UI for the control and the UI is not a fixed UI and here they show you the comparison, but I'll show you the UI first. So, they have this UI, you have like sliders, you have color select and imagine depending on the component the agent decides which one of these UIs should bind to which elements and on your notes regarding the flex control things. It's not just for that. I think you can push that concept a level even further and decide for this component the radius doesn't matter. So, we don't even show controls for the radius, even though you could have it, just doesn't matter. Or like for so many components and the padding and the um margin and even the z-index, for example, the these things don't matter. So, why do you show those UI controls at all times? So, this is kind of like the argument here and you could fall in either camp, but I'm in the camp of showing what's uh relevant and mount the UI as you need it, but the UI should be deterministic, determined like the you the design and the functionality of the UI should be determined has been determined before and here you see the like the difference. Like if you just ask Claude or Codex, whatever, to to build you an artifact, it gives you some UI controls but sometimes more than Yeah. You see this quite often. The UI itself is buggy. So, what you can do is to have at least a UI to be the kind of like uh to work in a deterministic way. So, it's not buggy and it has a nice set of controls. And this is what exactly this tool does. You can actually um play with it. toolcraft.sh This is really cool. I I love it. I think this This is cool. Yeah. Yeah. I mean, you know, you you mentioned that the UI itself can be buggy. I personally have had no issues and bearing in mind I've only done it once. Asking Claude to generate a UI that then controls the the effect or whatever it is, you know. And then it in turn then generates a prompt that I give back to um Claude and then it does it makes the necessary changes based on what I've tweaked in the panel. Did How does this work? Does this give you a prompt to give back to your agent? Or does this change the underlying code? This is something that you install. You install it and this UI library comes up um like the same way I have it here. But but yeah. So, when you change something there, does it change the code or does it give you a prompt to give to your agent to change the code? Well, I guess that mostly depends on how you want to set it up because you could bind this to the exact variables and save the variables. That's what I mean. That's what I'm asking. So, if you basically doing that then So, how do you take How do you take this? There is a button. You copy the configuration. So, this configuration is now copied and I can I can give it to the agent and the agent can do something with it. Yeah, yeah. And something really cool about Dial Kit, one of the reasons I use it, is you can have versions. So I can create version two and in version two I do this. So I can have this version and this version and I can compare them and then copy the ones that I like. And this is really useful in many places. Mhm. So, yeah. Um but yeah, I mean I I've not had much issue with with just getting Asking Claude to just generate a UI that allows me to change and then give me a prompt, but you've obviously used it more and you find that having a Dial Kit or whatever this Craft one They have limitations. I would love to know if if like Webflow go goes this way, I have so much experience in having pain with all of these tools. To the point that I started creating my own tool. Um I want So in short, I want these to be free. Like I want them to be like one for color, one for like graphs or something like this and one for gradients and not necessarily be too much intertwined so I can use them however I want because sometimes I want to use them to edit a site, sometimes to edit a component, but sometimes to create a tool with it. Like I am sometimes using these to create a tool and Dial Kit is not made for that. It actually is pretty shitty if I want to use this in a tool because then it's just not made for that. It's hard to get rid of some parts of it and then with updates you break stuff. So, yeah. I I had a lot of pain with that. Uh and they have limitations as well. I I don't want to glaze it nonstop because these are all um values that have a minimum and maximum. And a while back I started to use this design to create actually I made it even to create some sliders where you you can slide it past the minimum and you don't have a min and max number. You can have like a range but you can pass that range as well. So you can go into the negatives by dragging it further and it just stays there. So yeah, these have limitations. If you if you are not happy with the range that cloth has decided for you, let's say with with damping or whatever, now you have to ask in another prompt and this is where it gets slow and annoying because now you are just prompting very little things. And you were in this whole game of creating a UI to not prompt, to have the control but now you have to prompt the control. So you don't want to land there and I feel like to to not land in that prompt hell, you have to have a a good set of UI controls that you shouldn't you you don't need to be worried about. Um yeah. Cool. Nice. another uh another blog and we'll we'll see what they will show us, Webflow 3.0. Webflow 3.0. Next Webflow? Is there another Webflow one? Oh no, it wasn't. It was a 53X one. Yeah. Yeah. And gently related. Actually I don't know how to I'll do yours. I'm thinking like how I should explain this. So, you do yours. Yeah. Okay. Right. Well, I'm going to hope I'm going to try my best not to make this into something too boring. But, recently cuz it's going to start with story time. Recently, I I had to get I feel like it was a bit of a like a mindset shift, which I feel very strongly that we need to like adopt when it comes to using AI that it combines like all traditional automation with AI. And ultimately, it comes down to like how can we if you're if you're sat at your if if work can only get done when you're sat at your computer and you sat at your keyboard, I feel like you're doing the whole AI thing wrong. I think we need to adopt a mindset of like, "Well, what can I do or what can I get it to do when I'm not at my keyboard? What can I get to do maybe whilst I'm at my keyboard, but it's you're able to kind of do stuff, you know, without my involvement, basically." Automation, triggering that automation side. And it first came about when there there there was a bug that was happening on a website of or sorry, a bug happened on a on a on an app that I've built. And it I couldn't figure out what it was. And it was it was in a console error. And I thought to myself, "How do why how do I know what's happening in the console? Those are the hardest things." cuz I wasn't seeing it in any of the logs on the back end. It was happening only on the user's browser in the user's whatever. Stumbled across an app. Someone's already built it. It's called Sentry. Claude was like, "You can build Sentry. It will be quite easy, but this this is already existing and it's free. So, why not just use it?" So, I was like, "Okay, let's just [ __ ] use it." And lo and behold, I let me just show you, Sentry uh at the stage, Sentry would literally just give me console errors. And I think it gives me back background errors as well. But the point is it gives me console errors. And it was, you know, pretty cool. And then I thought to myself, well, how do you I maybe I can maybe they've got some sort of integration where I can create a an issue already in GitHub and then get Claude to like work on that issue. So I don't have to actually like yes, I get the email, but Claude is already on the case based on these errors. It understands the error. And lo and behold, they have an API. I can plug into that API with a GitHub action, which I set up here. And if multiple times a day, you just saw that it um it ran uh 3 minutes ago and it ran an hour ago. So multiple times a day, this just goes over to Sentry, sees sees if there's any errors, and then if it has, then it creates a GitHub issue and tags Claude in it. Mhm. Um you can have it, you have to pay for it, mind. You can have it so that Convex will literally just create a GitHub issue straight away. So as soon as an error pops up, it creates a GitHub issue. That's quite nice, but that's a pay-for feature. So I was like, well, I don't want to pay for it. [laughter] So I was like, let's build let's build you can you can use the API for free, so let's just build something that just keeps keeps plugging the API, keeps checking the API. So then that was just a that was the mindset shift of oh wow, I could I can literally just have a self-healing website that literally just I don't need to touch it. As soon as an error comes in, Claude is already on the case and whatever. We'll get into, you know, cuz you asked the question about What did you say? Managed agents, did you say or something like that? Or even just the maybe it's the GitHub thing cuz I think I think you have to pay for it. I think you have to pay for that. And I'll get to how I solved that in a sec. Okay. That that insinuated this mindset shift. And then I took it a step further. And I was like, well, what about if a client emails me asking for a a feature or they're reporting a bug that for some reason I haven't picked up through Sentry? So then I built an N8N workflow which looks at all my emails uh here. It looks at all my emails. Uh and then determine asks Claude whether this is a um I can see what you've done there, by the way. It's very off-putting. [laughter] [gasps] Why you're moving around. This is This is exciting. I [laughter] I will show it. I will show it. Um stay tuned cuz, you know, I understand it's I figured something. Um so yeah, I I This is a local version of N8N, which means I can trigger Claude locally. And you'll see how this kind of, you know, starts to connect the dots when it comes to these GitHub issues where you have to pay for it. Um so ask Claude, is this a bug? Is this a feature? Or is this just an invoice for something I bought off Amazon. And it will If if it's if it's a bug or a feature, if it's categorized as that, then it goes through this thing. It double checks whether uh a um a client of mine is an active project. So this looks in my active projects. I can't Why can't I zoom in? There we go. So this is my accountancy software, which this is my active projects. Again, this is just You don't need to do this, but this is just my little thing to um verify it, and then it matches it with a GitHub repo, and then creates the GitHub issue, and then tells Claude to get to work on that GitHub issue, and then marks the email as read. Um so, this is All this does is call a server that I've got running locally, um that in turn call call calls clash Claude.p, which is programmatic usage of Claude, which runs all under my Max subscription. So, to bring all that back together in this century issue in the GitHub, this one, this is all succeeding, but because there's no there's no new issues. See, it's not even gone to the validate. I can't I scroll down. Interesting. I can't scroll down, and I think this is because Oh, there we go. There it is. It's cuz this is a mermaid diagram. Like that. Um so, I haven't actually seen the full pipeline, but I would guess that this crashes, because if we go into pull requests and go into closed, um maybe this is it here. I think this might be it. No, it's not. Don't worry about it. Uh but, I might have got an error. Tags Claude cuz you you know, and and Claude will get to work on it, but you have to pay, you know, you use your usage credits, which, you know, if you don't know anything about Claude, manage stations, uses credits, and um it's not a way you know, you're not utilizing your Claude Max subscription. So, uh I think the long story short or the lesson learned here is um getting it running on your local machine allows you to utilize your your Claude Max subscription. Combining N A N and asking it to make a little server goes back to what you were saying, if you know it's possible, just ask your agent to do it. You don't need to know the technicalities of what I've done. You don't need to know how I've done all of this or whatever. Knowing that this is the sequence of events, you get an email or you get an a log from Century. You know that you can watch that from your local machine and trigger your local instance of Claude. Cuz in fact, whilst I was whilst we were chatting, um I actually got Claude to build my Century pipeline. Um in so I can import this now into um uh I can import this into N8N and I've got that Century pipeline that creates a GitHub issue and all the rest of it. I I can now do that from my local machine. So, I've recreated it. So, yeah. That's kind of the mind shift mindset shift is here. For me, it's just try just trying to be more proactive. Just trying to like how can I get things done in proactively and and and to finish on, I've also got a daily routine that runs through my calendar and get started on everything. Just get get started on that meeting that I've got with that person to pull in their information or do or do this or do that. You know, I let Claude just figure it out. But it starts with the it starts with the calendar um blocked out calendar section and it will just get started on stuff. Create a web page for me, a bit of research, this or that, you know. Do you Did you make that N8N yourself or just ask Claude with the MCP to make it? No, I I made it myself. I made the N8N myself. How Why? I haven't touched N8 I haven't touched N8N in a in a bit. Uh I've been using it, but now I have the N8N uh MCP. It I let Claude connect to cal.com MCP to Stripe MCP, N8N MCP, and my Webflow MCP all at the same time and be like "Okay, that that that's the page. That's the call. That's the Stripe product. And that's the N8N uh automation to send the invoices and all. Like, it all in kind of like one conversation." Yeah. So, the reason why you would use N8N the reason why you should use N8N is because one a lot of it you need to be deterministic. Well, first of all that. I meant like creating N8N. Creating N8N? What do you mean by creating N8N? Like like creating the nodes and configuring the N8N workflow. Like the flow that you have in N8N, I'm not like adding Oh, yeah. No, this this is what I'm saying. This is what This is This is what Claude's done here. So, that JSON file that I just import that into into N8N or just ask Claude to to do it. So, no no, I didn't add any N8N. Yeah, yeah. So, I didn't add any I didn't click on add node add node add node. No, no. Claude just did it all for me. I thought you meant installing N8N. That's what I thought you meant. No, but but that's a super like important point that you are making and I I feel like a lot of people have to understand this because they they mix this uh Claude being able to do stuff agentic work with N8N. And as you said, N8N is deterministic. And that's like the good thing about it. Before agents came around with things like N8N or at that time you know we had Integromat. Uh that's what I started with and make that common and Zapier. If something like so even small would not match exactly, your whole uh, flow would break. Would it would not run completely. But nowadays because you have also agents in in the nodes, you could um, ask agents to to determine a yes or no or like return a value. In my experience, so much more uh, reliably compared to extract it from the data, um, with with like programmatically. So, Mhm. there is a really good, um, mix of n8n and with agents. And if you're questioning if you need n8n in the first place, you should also consider the price because running a Claude instance every time to do something is more quite often it's more expensive, uh, talking heavy while n8n it's just faster and just does it. Yeah. It's like immediate. So, let me show you this section of the, uh, of the workflow. There you go. Um, So, let's take for example, so what I thought you were talking about is actually getting Claude cuz you've got Claude schedules, right? You've got Claude schedules that can connect to my MCPs and do all the rest of it, right? So, take take this uh, I can't I can't point. Um, take this trigger as an example. So, you've got the Claude schedule goes off and asks the email uh, what emails are available. And then it gives back JSON or whatever and then Claude's got to look through those and determine what is what is useful. In this case, this is just a server that constantly pounds Gmail. There's no tokens being used to do that. Next, you got like prepare emails. So, there's there's some code here. This one. Um there's some code here. Maybe I can actually zoom in a little bit. That knows the shape of the JSON, runs through it, and and just checks values. Instead of using tokens to do that, you just it's just code doing it. And exactly what you're talking about is like you just there's deterministic ways to do that. Claude can hallucinate. And then I chose classify classify with Claude. That's where I need that kind of like natural language sort of like thing, right? It needs to understand what they're saying, whether they're mentioning invoices or whatever, blah blah blah. So, I'm choosing when to when to throw it to Claude. Otherwise, everything else is done with REST API, like calling FreeAgent as an example. Just call the API, right? Tell me what contacts are there. You don't need You don't need to ask Claude to ask the API to then get the get the response, loop through it like no no no. So, long story short, like yeah, like determinist determinism is super super important with automation, and you need to know when and where. Like I watched a video, I think I think it was a Theo video. He was saying that people were using AI to find the odd numbers in an array. And it's like That that's dumb. That's crazy. Claude is probably Claude is probably smart enough to say, uh well, we don't need we don't need AI to do that. I've just written a a JavaScript file that does that for you, right? So, Claude is there to able to figure that [ __ ] out. I'm worried when I hear things like this that people have no freaking idea how AI works. Like these systems are Well, they know they have no idea how we are hallucination machines. They know Well, you're right, but they also don't know how systems work and how this And it's it's fair play. This why That's why developers up until today and still today get paid a lot of money because the the the mind the mental model and the logic you need to apply to build systems, to build coding things True. is is completely different to a to anything else. So, that's the hardest thing to figure out when it comes to coding. So, that's It makes sense. You know, it makes sense why they're doing it, but For for anyone listening, do the following. Like it's a fun test. Just whatever AI you are using, ask it to multiply two really big numbers with each other. Big numbers. Like think of like two nine-digit, eight-digit numbers, multiply them. It will answer you and the answer is surprisingly close. So, it will always tell you something. Yeah, yeah, surprisingly close, but it's not correct. It It will tell you something where the start is correct, the end numbers are correct, but the middle digits are incorrect. And the reason for that is the models through their trainings, they have an understanding for math, like a feeling for it, actually, not an understanding. Let me correct myself. They have a feeling, but they don't understand it completely. So, they they make a guess. And that guess is really close, the start and the end of it, but the middle of it they because they never run an operation. It's not like a mathematical operation they run. It's just a guess. So, the middle of it is always incorrect given the number is big enough. Um So, just test this and you will see how hallucinative these models are. So, next time you won't ask it with your finances. At least if you do, then you have to understand, okay, you ask it in an environment where it has access to a calculator. So, in Claude code, for example, or Codex, it can literally use the terminal to to spin up something or like do calculations. Yeah, they needed tool for Yeah. Yes. They needed tool for calculations. And I'm a bit surprised why they haven't added this because it doesn't look too complex to me, but I guess it has understanding in the middle. When was the last time The last time, I think a month or two ago. Yeah. Let me Let me Let me try. Uh I'm opening Actually, I'm going to open Claude fable five on high multiply. This is that And I'm really content. Whilst we wait for him to do that, just know that this is a bit different from our normal content. We normally just go through the news, but we thought we'd just chat about tools and things like that for the for this episode cuz the news has been pretty boring. Saying that, whilst you're on YouTube, give us a like and subscribe. That'll be good. Um but Meta have just released Meta's Pocket, TikTok style vibe coding app. Rolls out to all US users. What the heck is that? I'm getting a I'm going to share my screen whilst you do that. Meta brings Pocket, an app that lets you vibe code and share games to US users. This is yesterday. Meta's experimental vibe coding at pocket. Uh says market in Brazil allows me to generate small interactive games using AI which are published to a scrollable feed. How bloody bizarre is that? Mini games? Mini games that you can just go you know you flick through and if you want to play it I guess you start just playing with it otherwise it's just it's it's game heroin. How weird? I reckon they're going to get a lot of information on what This is interesting cuz you know what meta love? They love a bit of um data collection. Are they collecting data on what people play games or what what games engage people and then they would basically just rip that off? So I did a test and Claude came back with a correct answer and now I'm asking if it's using operation because that would be interesting operations. Last time I think it was maybe 2 month ago it didn't do it correctly or yeah. So is the answer correct that you gave it then? Yeah yeah yeah the answer is correct. I Well I did the number is not too big but um Mhm. Let's see. Uh let me Can you see tool use in in Claude? I'm just going to say multiply 764 billion 542,862 by 3,842.8. I'm going to share my screen actually so you can see. That would be good. [laughter] Uh Muhammad is saying Satya is saying, "And run workflows." All right. And this is Claude normal. Yeah, it's not high. It I mean, it's not like Claude code. No, it's not Claude code. So, it doesn't show you like tool usage. And I have no idea whether this is correct, [snorts] by the way. I [laughter] I said No, it pasted into your uh calculator or just like in your Yeah. 29 Oh, no, see? Hang on. That's not right. Yeah, it's it's correct. Oh, no, there it is. There it is. Sorry, I was looking at it is correct. It is correct, yeah. It is correct. And [laughter] it's really funny because I asked it I said, "Did you use an actual calculator or like operations to to give me this answer?" And the answer is, "Honestly, I calculated them in my head. No computer involved. [laughter] I broke them into steps mentally like Yeah, the numbers and then Yeah, it it Okay, I get I guess now it can do math it's in its head, but we know how it works. And I don't know actually if it used a computer or not. Uh, or it has it like a tool use. Um, do you Are they Are they aware of the tools that they have access to? I don't mean like the third-party tools, but like the the integrated tools that they have? I mean, ask. I mean, what tools do you have access to? Web search, search your past conversations. Yeah, it does. Oh, it says I have Yes, I Yeah, I have a code execution environment where I can run bash and Python. That that that's what we need. Yeah. Mine doesn't say it. do it. So I did it also like on my phone. Yeah. There you go. I can use use it anytime for calculations, data processing, or anything else that benefits from real computations. And that that answers us. And so basically, they do now have a calculation tool. Yeah. Based on this small test. So we can't tell for sure. But yeah, that's that's cool. right? It probably writes a short Python script to do it. I mean, like in all languages, you have operations, right? You can do it in JavaScript or anything else. You just write the operation. I remember like when uh we we were doing C++ that's what we were doing in university. We're not doing any actual programming. Well, I That's debatable what's actual programming, but we were just converting math to code. Anyway, speaking of Wow. Let me show you the This is crazy. This is bonkers. I just found it online. And look at this. Look how crazy this is. Actually, let me show you the depth. write code the app, or is this the app that you just downloaded? No, this is online. This is on GitHub pages. You're just a You're just on a website. Yeah, this is This is crazy. And I can also have Look at this. That's a normal map. Yeah, this is a normal map. This is creating kind of like a 3D map from a 2D image. This This is bonkers. That it's you know working on like feature. we had a 3D Can you Can you change the camera? Can you zoom in on the camera or something? Basically, if we had a 3D camera, it would look like you were 3D, and we'd move the camera around. Oh. Uh what what do you mean if So, normals only is You can only see normals if we could move the camera around. And you'd see a slight depth to your face. Yeah, but I see we see this if I Well, you're able to sort of No, no, that's I guess you're you're filling in the blanks with your head with the colors like Okay, I see. Yeah. Like a camera would just And it's all fake though. It's all It's all fake. Yeah, that's what makes it really cool that it's all fake. And the depth one it's pretty accurate. I mean, by the way, this is a glass. And well, it's it's not like an actual lidar. Right? It's an image. And I guess the the machine learning model With um understands that. Yeah, yeah. And with um let's talk about uh like cars, like autopilot on cars and stuff like that. It used to be that you need lidar. But actually, I think Tesla has moved on to just cameras now. Cuz this is what they do. They have been doing that This has been their biggest bet from the start. They've They've never used lidar. Yeah, I think maybe they had a phase, but for the majority, for the the biggest majority of their time, they have been this anti-lidar company saying that Elon Musk famously saying like you don't shoot laser out of your eyes when you drive. Which is not really a good argument. But if you could Well, well, not only that, but if you could shoot laser out of your eyes, you would. Duh. Like, you would totally do it. If you could see in 3D with lidar, if you had the sensory for it, you would use it. The fact that we don't have it is not an argument for let's not put it on cars. The actual argument that they have been also very clear about they are not shying away from saying it's expensive. Lidar is expensive. It requires, I believe, also very precise um What's it called? Like, at the start that the word is on top tip of my tongue. Like, at the start you need not configuration. Um Clap uh um Calibration. Calibration. Yeah, you need Yes, you need very precise calibration uh which compared to just like normal cameras, they don't need that level of calibration. And uh yeah, lidar just needs more computation, more sensors, biggy big and bulky sensors, and all of that. And that's why they are That's why Teslas suck at um just parking. Like, when you want to park near other cars, they don't even have like those proximity sensors that, you know, does the beep beep beep beep beep. It doesn't. By the way, we're continuing the show like this for the rest of the show. This is how I'm going to present. This is so fun. [laughter] I don't know. This is I I I guess I'm just 9 years old. This is just so fun. Mhm. Um I did want to mention something as well, like the aside from all of this, more about the like the automation stuff is Have you Have you used T3 code? Yes, for for quite some time. Have you used I mean, I don't know whether you've got cuz I've got the I'm pointing I'm pointing over here because there's a I've actually got a laptop that I do a lot my remote work on. Like if I'm out and about I was on a train last week and I was sending that code challenges so that when I cuz I was on a train, if I was going under I'd I'd want that to continue working and you know, converse with with with Claude back and forward. Um T3 code seems to have made it really simple just to be able to select from a remote instance of a computer running T3 code and just fire it off. And I'd love them to bring that cuz I'm a Claude code fanboy. I'd love to them to bring that into Claude code cuz right now I can only set off um you know, tasks on the machine that I'm on. I suppose I could use dispatch. Uh well, no I can't. I can't dispatch from one Mac to another Mac. I've only can only dispatch from my phone to a Mac. So there's that, but um yeah, I know. I'm I'm really trying to get into like how can I be as remote and versatile as possible cuz I was in Bulgaria last week. And I would I just had this idea the the the email triage thing where someone asked me for a bug or a feature and then Claude gets to work on it. I literally had that idea whilst I was having a drink with a mate. I was on my phone, typed it out dispatch, and it got to work on this machine here whilst I was in Bulgaria. We're currently in London. And then half an hour later or however long I check back and it was built. You know, and and and all the rest of it. So you asked like did I do it manually? No, I did it from Bulgaria. Yeah. In London. So Yeah, that it's really cool. I I haven't used dispatch by the way for some time since I'm using auto um remote control. Like every instance that I open has auto remote control by default. Mhm. So, like I don't have to even think about it. I I check my phone and if I have to leave or like just enter and I'm doing renames a lot. Like nowadays I'm renaming my It's just so helpful to to have a clear mind. We are doing so many things at the same time and you don't notice but until like later in the day that you are so exhausted and I noticed like a lot of context switching gets lost or like um I get I I I I guess I'm like getting tired by just recognizing what am I looking at and just by renaming thing, that makes it easier for me to know, "Okay, this is the chat and this is the context." So, I just continue. Well, I mean, something I've got into using recently is um is projects because projects again they take a load of files, they have their own memory, they have their own um there's my outreach one. Uh they have their own memory, they have their own everything. And their own skills even. You can even um you can even scope a skill to them. So, that means and if I click in here, I can enter an old chat that's related to this project, this this context or I can just start a new chat and again I don't have to worry about like previous conversations somewhere else in in Claude code. It's or Claude, sorry. It's literally all contained within projects. So, I would um That's cool. Yeah. I would strongly recommend projects as well cuz I I agree with you. I'm sort of getting lost and thinking I'm having like six conversations and and for most of my day now, I'm not touching like the tools. I'm not touching websites. I'm I'm using the MCPs. I'm I'm building my own MCPs. I'm I'm finding new MCPs just so I can just stay within Claude. So, it's quite It's It's It's quite a profound shift I've had the last few weeks, which is not only living inside of Claude, but also being proactive and getting it to do things before I'm even at my at my keyboard. Do you know what I mean? Yeah. Yeah, exactly. I Yeah, I feel the same. Doing Been doing the same. Everything is Claude. So, I I think this AI thing is going to really take off. I think it's It's going to be big. [snorts] [laughter] Tomorrow, the bubble bursts. And then I've said I spoke too soon now. Tomorrow, the bubbles will burst and then we're like, "Ah, shit." Yeah. It's interesting. It's going to burst. It It's interesting. Well, it will burst, but I don't think it will it will pop in the way that people expect it to. I think it will just be Yeah, people think like if it goes back, nothing is going back. No. No, you just can't. It's so funny. I was chatting to someone and they were like, "I don't really know much about AI, but I know it's bad." I'm like, "It's really interesting that you say it's bad. You know, what what like" And people is still I was in the comments section of someone's video today and they were like, "How do you feel" It was someone actually we should We maybe want to loop back to Grok bot at some point, but like it was on a Grok bot video cuz I'm still trying to understand what the [ __ ] Grok Grok bot is and why people want to pay 200 bucks a month for it. But, in the comments, obviously this was a Mac focused channel. So, they don't really talk a lot about AI. So, you get a lot of AI newbies or people who just don't know about AI in the comment section. And then someone in the comment section said, "How do you feel about um running ruining data centers or or consuming loads of water or something like this. And it's so funny cuz those are the conversations I feel like I've sort of been debunked at the moment. Like don't get me wrong, the data center thing is a bit bit crazy. Like people are still petitioning to not build data centers so close to um like communities and stuff like that. But as but I I I thought the water thing was debunked in that once you It was Once you have the water Once you have the water, it's just it's just reused. Yeah. You don't need fresh water as far as I understand. You don't need fresh water. Even like electricity is like debunked. Like pretty much everything is debunked. I've I've read reports on electricity long-term becoming cheaper in those areas even because there will be more to create. So a lot of it is being debunked and I'm like heavily Like I When when people hate AI without understanding what they are talking about, I just I don't know. I feel sorry for them. And I I understand it like these [ __ ] I have to say like they make me curse, have been saying like it it will bite your tongue. [laughter] Gosh. Pardon your French. [laughter] So like just makes me want to swear. Yeah. AI will come for your job and you can Engineering is dead. Designers are dead. Like everybody is dead. Everything is dead. Like they've been saying this for like two, three years and people are hating this. But on the same note, you shared that YouTube video with me made with AI. Oh yeah, yeah, yeah. Earlier today. I didn't watch it. I I have like some notes. Should I show it like show a frame or two. So, now that we are just touching on that, let me So, Sam shared this video with me and it's it's an AI-generated video. It has 7.4 million views in a month. That's crazy. That's like a a really high number of views, especially for an AI Some people might call it slow. I I watched a little bit of it and then I was like I didn't feel good about it. I felt like it's not cohesive in a way that I could connect with it. So, it might be an AI thing, it might not be. I'm sure it will get good enough for that. But, I have a different point. When you shared this with me and I started watching, I I thought of some like Gen Z people, especially we know Gen Z hating AI, right? It's not just Gen Z, but they notoriously hate AI. And I was like, you see two types of Gen Z people. There is some that see this and see the tools that create this and they hate it without any reasoning and they just don't get anywhere close to it. And I feel like they will get stuck in the past or like they not in the past, but it's it just feels like they get stuck in their mind. They they hate this thing so much. up. For whatever reason. they closed up. Yeah. But, I see other Gen Z or whatever, I would love to see like Gen Z's take on this and use it as a tool and create stories like never before. Like a fresh mind coming to this and create new I want to see that. Like they they it they might create some stuff that I might have a hard time to connect with, but that's kind of like the way that I want it to be. I want the fresh minds to come and like not that we are old. [laughter] Somebody just called us seniors. Not that we are old, but like there is always the younger generation with like fresh ideas and I I love to see that. I want to see the whole range. And I just feel sad that a big part of like that Gen Z doesn't see this opportunity and they are like closing their mind up and I I find that like quite sad. Either way, um it's not a you know, it's about this specific video. It just made me think about this. What was your question on this? You asked like how much did it do you think it cost to create this? And my estimate was probably 2K. 2K USD in terms of basically scenes generated with C dance 5 at 1080p um with Higgs field. I went in this is 4K to be honest. Yes. Yes, it is 4K, but I think you can easily just upscale afterwards because I don't see why you would need 4K generation. Maybe. Maybe. But yeah. Do me a favor though and I've thought about this. I've thought about this today. Just unfurl or uncurl whatever the description. Look at the Look at the Look at the team that was involved in creating this. Interesting. I didn't look this up. So, it's directed by a person Zack London and written by him. Produced by him, but then three Gossip Goblin is a company by the way. Fable Studio is a company, by the way. I want to see like what Fable Studio Yeah, okay. So, they're companies. So, they're voice cast, animation lead. Interesting. That's really interesting. So, these are people who know their [ __ ] I assume. So, when you say 2K, don't these people need salaries? No, I meant the Yeah, I meant No, I know that. I know that. I know that. This is where I'm going with this, right? So, Yeah. the the the the the top of the the decision tree here is that [snorts] we cannot determine what happened behind the scenes when it comes to AI-generated content. Obviously, the quality in this speaks for itself. Like, the details, the consistency between the characters, um and all the rest of it. Like, [snorts] to to to most people, you can see that this obviously took a lot of work. However, majority of stuff, you really don't. And I was thinking today about writing. Like, so, apparently, 80% of all posts go out on LinkedIn AI-written. It's hard to determine Thank you. hard to determine what is a one-shot and what is you know, like my my AI content is written My So, my LinkedIn content is written by AI, but it's sourced from like, a year's or two's worth of probably even 20 years worth of technological knowledge and understanding code and this and that. It's also based on 2 hours of conversation with you. It's based on Like, the amount of work that's gone into Yes, what is essentially an AI-generated post. But, not only that, I've been writing a lot of posts recently with AI. But, some Some posts take out like, you know, hours of conversation with the AI to to to sculpt that. So, you might put that post through an AI like detection test and it says, "Yep, AI." And you might dismiss it cuz like, "Oh, it was [ __ ] AI written." But, the amount of effort that it took to get to that point is completely masked by the fact that it is, for all intents and purposes, AI written content. Yeah. And there's no way to distinguish, apart from if the you know, really really shines, there's no way to distinguish that from someone who has lazily just put in one prompt to copy paste done, or someone like me, you know, who 20 years worth of development knowledge, um uh back and forward with the AI trying to sculpt it, trying to like, you know, shape it and this and that. There's no way to determine all this. So, it's kind of like it's it upsets me I don't know if it upsets me, but you know what I mean. Like, it's a bit hypo hyperbolic, but like, um it frustrates me almost that people just dismiss AI written content I agree. when you there's just no way to determine whether it's good or bad. It doesn't change the fact that it it it still it does annoy me that people leverage this and one shot it and try and pass it off as their own knowledge and this and that. It does still frustrates me, but it's it ultimately come it comes down to it's wrong to just dismiss something just cuz it's AI. If the final result is AI, do you know what I mean? Yeah, I agree. Uh especially like there are like some really good um AI like ads that I've seen. For example, the the watch there is one with a watch that um is AI generated. I think a lot of people have seen that, but yeah. Mhm. I agree. Uh with this type of content, when I notice it's AI, it's not the only reason I like it I felt often I stopped watching. It started to bother me where the physics started to break. Like anywhere in the You say that. He put So very early on in the video, we're talking within the first 5 minutes, he chucks his jacket on the on the Yeah. Yeah, I noticed And I I was like the physics look amazing on that. Yeah. So how about you bring that up and show me where you think the physics is [ __ ] versus the physics is bad? The guy walks in So this is So he's starting to walk in. Uh let's see. So he's got his jacket off here. So he's got wearing white, his jacket is off. So we want to go back. He's still wearing white. So we want to go back even further. He comes in with a blue jacket on that he chucks on a on a rack. So I reckon around Yeah, here we go. Here we go. So play from I thought the Uh it's hard to see cuz it's glitchy, but I thought the physics on that were really really good. So so the jacket moves. Yes, I I noticed that actually. I specifically noticed that and that was great. But what was not great, for example, when he walks the valley here, it does look a bit strange where how he puts his feet uh in like Look at here. It's hard to see. That to me Yeah, that to me Yeah, that to me it looks like the the feet are switching and this happens a lot. With the AI. Yeah, that I noticed and it was it it didn't work for me. Another places where where he uses that um hammer. The hammer, I don't feel the weight of the hammer in his hand. Like this looks like not a heavy hammer. Look Like you can't Yeah. That that doesn't work. The the hit maybe, but this bit Yeah. Well, if you go back even further, he's looking up. He's aiming up. This throws it this part. Then he throws it in front of him, and then it's up again. So, that that was one thing that was Well, put this way, I would strongly Well, I'm not sure. You don't how Oh, go on. Yeah. The these happen in real movies as well if there are multiple shots and these if these are not good movies, right? So, I guess it's a matter of multiple generation and and as I mentioned like two 2K, if they get like 400 shots, uh for like end up using 120 of them. Um but I think this can easily cost more if they decide to regenerate again and again and again until like the angle is correct, right? So, the thing I was going to say is right. There's a TV show called Peaky Blinders show in the UK or it's a UK-based TV show. And when it first came out and the whole concept is you're always in a first-person view. It's the The only show that's ever really done it is you're always looking from the perspective of someone. So, if you and I were talking, it would literally be my eyes looking at you. You'd be facing me, and then the camera would switch of you, you know, exactly facing me, right? When that first came out, I dismissed it cuz I was like, it was low-quality. Cameras were massive back then, and you know, it was really to get a camera in those positions, it was very very difficult. As time went on, and it was about nine seasons, cameras got really small and really good quality, and the quality of the show got better. But I dismissed it because it was poor quality like the camera whatever. That TV show is legendary. Mainly because of the the writing and the the idea and the concept and the character development and there's so much more that goes into it. But I was focused on the technology side. I missed out on that because now I love it and I can see past the technology and laugh my head off at the jokes and the the commentary and all the rest of it. Yeah. Bring it back to this video right now. If you watch it you watch it all the way through, the concept like I got lost in the concept of this video. I got lost in the idea that this is a perfect use of AI because the characters are so out there and crazy yet realistic. Like if you were to create this in 3D I think it would be really really hard to create this in 3D. Like you you look at this character on the right. Yeah, I I know what you're saying. Yeah. his eyes and whatever. So this is a all that metal around his face and whatever, right? [snorts] Very realistic. Yeah. This is really hard to both from a 3D perspective like a CGI perspective and a like a um what do you call it? A live action perspective. Like getting the makeup right and this and that. So that works this for me the the surreal kind of slightly odd like uh character design and the whole design of this works really well for AI. But then finally you watch the ending. You watch there's a monologue at the end which is fantastic which brings all of this together cuz you're lost. You don't know what's happening. You don't quite understand. It's a bit you know uncanny valley and and something not quite right and whatever and that's all intentional. Everything's intentional. And it all makes sense right at the end. And it's fantastic. So by dismissing and and so by dismissing the kind of the physics or you know, I don't like and I'm not I'm not saying you're saying this but if I don't watch AI content because of X Y Z you're missing out on so much where AI is or where it's not AI. The the the storyline is not written by AI. The script is written by people and the the idea and concept is all written by people, right? So, it all It's just the technical execution is AI. That's all that is. So, Yeah. long story short, it's like, yeah, I I've been there where I I I don't commit to something cuz I don't like some technical aspect of it, but ultimately there's a lot of human element that's gone into it, which which is great. Um but it's good. Well, I I guess it's convincing when you say, "Oh, watch it. It's worth it." Yeah. Which I I think this is a huge point that a lot of people might miss. This is the cultural aspect of content. Um like we we you watch Odyssey. I have watched it as well and it like the reason we go all we all go watch it, it's partially It's not just because it's good. It's everybody talks about it and everybody says it's good and everybody like says like you should watch it and it becomes a sensational thing that you have to do like when you are like living in 2026, you got to watch this and same with Dune part three. Like I cannot wait for that movie too. I It's just like much bigger than Odyssey for me to be honest. Uh Yeah. Yeah. By the way, for for the geeks there, they are connected. Like Dune is based on uh Odyssey. Not based on, but like they they are descendant like the House Atreides is direct descendant of uh King Agamemnon. What What is it? How How do you pronounce it? I I I I don't know how to pronounce it. Yeah. No, the the king, the the the guy with the black suit in um Odyssey. He's the grand grand grand grandfather of Paul from Dune. That's what they say and there is no like actual proof for it. So, that's part of the story that they're based on what I understand that the fact that there is no actual proof for it it's that their excuse for making that connection why they are the righteous heir to the throne. Yeah. A lot of sci-fi, you know, Star Wars, Dune. A lot of these have their roots in like Greek mythology or or just Yeah. Bible or whatever, you know, so there's a lot that a lot of them are founded in that kind of like stuff. But yeah, I love I love sci-fi. I just think it's just such an escape. Yeah, I got lost in this cuz it was just like it was it was cool to live in a live in a world which doesn't exist and um you know. Yeah, I I get the idea. We We have a comment saying loving the conversation doing this in CGI would be absolutely hard and also would take years. Um exactly. Like it would take years literally to learn all of that while now you could come up with the idea, you know, you your imagination create a cohesive story amazing characters. I mean, we all love amazing characters and then create a show. And the fact that I thought this would cost 2K which I think is still possible to do for 2K or something around that. Maybe maybe a bit more, maybe less, but not too much more. Like a CGI creation like this in a studio doing it the traditional way would cost a few hundred K minimum now probably like into the millions, right? I I I find the difficult to to do this for under a million or under 500k. I'm not sure actually, but it will probably cost that amount of money, but now a smaller team can create the same output or like similar to it. Mhm. It's cool. use you can use you can use AI in two ways really and and you can use AI as a cost-saving measure, you know, you can use it to get things done quicker or you can use it as a leverage you know, something to build upon. And I think like these guys are probably using it as a something to build upon like the fact that they didn't say, oh, you know, we're going to use AI cuz it's cheap. I think they used they thought you AI allows us to like there's a there's an artistic usage of AI. They wanted to use AI. They chose to use AI whereas I think they might even given they've got like an animation director and all the rest of it. I mean I've not looked into the the studio that actually that what's it called? Goblin Gossip Goblin. Maybe they've got some real live action things. You just click on their profile. You know, maybe they do love 3D and they do use 3D, but they've chosen AI because from a creative standpoint things some things can only be achieved or or is necessary that AI is the is the choice of of of medium, you know, but what do I know? Well, other way guys? Yeah, super cool. Um what do we have? What's what's next? What's um So [snorts] we have that last piece. I'm not sure if I can make it because I I need to go it in 10 minutes. Oh, okay. What is it? Yeah. We have the Stacky. That I wanted to show. Oof, this is a big one, too. Do you want to just see how far you get and then we'll we'll just Yeah. Yeah. Um Did you check Spline MCP? Um no, I don't is dead to me. I'm sorry. [laughter and gasps] Like, why would I use Spline if I can just, you know, use 3.js and create the model with the Blender MCP or whatever. Like Like it Spline is an abstraction and abstractions have kind of a dying a death. Yeah. Yeah. Yeah. Uh Spline was never good for production and right now 3.js just takes the win. So, it seems like the universe is telling us to leave Webflow and Framer. Here, this week we are coming with another tool that is such a strong Like, it looks so good. And it really makes me want to leave Webflow. And it looks just like Webflow and Framer. It looks like if Webflow and Framer had a baby, it would be this. Stacky made by Timothy Ricks. So, this is his tool. It's in beta. I've downloaded it and made a whole, like, actually two websites with it. Made, well, let I let Fable make two instances and I will I will show you what it can do. So, the whole idea is that you can use Astro uh the framework, Astro, which by the way, Webflow also uses underneath to build your website, and then you have this visual editor that is very much Webflow-like to to edit things deterministically and visually, literally touching the code without touching the code. And I will just show you that. So, here we have my agency website. I know it has bugs and whatnot because this is pretty much a one-time run. Can you Can you lower your screen resolutions so we can see everything? So, basically zoom in. Uh I did it I I If I zoom in too much, it messes up with the Okay, cool. Oh, that's that's cool. That's better. Okay, that It's It's fine. It messes with the Uh I guess I can also show it like this. Now, I spoiled it. So, yes, it can show all the different breakpoints next to one another, and you have like this UI that is very much like Webflow. Even here, your layers we have the pages. You can go to the blog page. You see it did make some mistakes creating it. But, yeah, as I said, this is pretty much a one-time run, and it has nothing to do with the tool exactly. This is more like just Fable getting a Figma design and creating an Astro site, and now I can edit this Astro site visually. That's the point. Uh you have CMS. You can change the You have like the layout control. And as we talked about it earlier, like I can I can add stuff. I will show you that in a second as well. But, yeah, the panel, just like the Webflow panel, and you can literally interact with it just like you would do with Webflow. And you have these components. You have the CSS classes. You have variables, and it all looks really good. And it's all code. It's literally touching the exact uh the the main source code that we have. And I will show you the the other uh example, so can And just to be clear, this is a Mac app that runs on your Mac. You download it, and this is just this you open your Astro projects inside of this Mac app. Exactly, which I will show you with this one. This is another test project. Uh I want to show it to you using this one uh because I already have it open in cursor. So, here [snorts] we have the code inside of cursor, and here we have the the UI editor. So, in the UI, we have this button, and we can change the size like from six to seven, six, seven. Uh No. [laughter] No. Uh and and you have like the text that says press me, and if I go here and say press me now, and come here, you see it has changed. So, this UI is literally reading the code and displaying the code. So, this is a true window into the code, and the way it is set up. And as you mentioned, here we have this style, um as you mentioned prior to the video. We have this style for this button that I have just added, and here we can go to our global CSS. I can look at the button. You see it's 7.4 rems. I can say 7.9, and if I save this and come here, I notice it takes a second, but yeah, it loads in, and now the style has changed. So, pretty much everything is code and you have a visual editor. It has a CMS. I'm still trying to wrap my head around it completely because it's not exactly a one-to-one to Webflow. And then we So, here we basically have MD files as you mentioned, but we can add uh collections. We We can add uh What's it called? Fields. Yeah, fields to the collections, but now I can't. Mhm. I know. I thought we I thought we couldn't we didn't figure that one out cuz that needs to be done through code. Well, I No, we There There is a UI for it, but it doesn't show up for whatever reason. Let me open the other project. I don't know how to also switch between projects, so I close it and then open it again and now I can switch. Probably something they need to um they need to [snorts] figure out how you are able to do that. See See, there is the setting. There is the setting um and if I go to So, in the collections, Oh, I can't do it in the collections. But in here in general, I can add fields. So, I can add a repeater item. So, it It is very early. I think this is If I'm not mistaken, this is just a few weeks old. And I think it has really good concepts behind it and let me show you one other thing. It has a vertical like adjustment Something that Webflow has been failing at giving us for ages. Like Mhm. We We had this with Finsweet extensions before that got depreciated. So, it's really sad to think about that like today's um tools for like visual editing, how how slow they are moving. And I really hope that Webflow comes up with something maybe like this that is opinionated, but not too much, and it allows me to do things with my agent. So, not necessarily through Webflow, and I hope not even through an MCP, because that would be another obstruction layer in a way, but like directly, maybe like how it is now with Astro. This can be any Astro project. Yeah. Just for the ones who want to add things uh manually, yes, you can add components or like elements. You have all the HTML elements that you want. You can add them manually and edit them and do all the manual work. You can do a lot of the annoying things that Webflow doesn't let you do, you can do them here. You can add media queries. You have your variables. And all of this, I let Claude set it up. I gave it um a Figma design. I gave it the tools and the links and said, "Set up this project with this design with Stacky and with Astro and with Lumos." That's it. And it did this all in pretty much half an hour. Mhm. Um and then the what the final thing to mention is that this obviously doesn't it's not Well, we the flaw is that it's not really a multi-user um experience, this this thing. Not yet, yeah. Each of your Each of your team will have to download their own version of Strappy. And to publish this, you you push it up to Git, and you'll have to have some uh integration with your GitHub and uh let's say a Versel account or something like that. And that that will need to be set up by you. So, and to get changes in you have to pull down each other's code, whatever. This to me is 90% of what I think Webflow will release. This free tool Remember what I said, I've got my prediction of there's a free tool um you can publish to Webflow. Right now this allows you to publish a GitHub, but I think they will their their software will push it to get to Webflow's servers, which is just Cloudflare Cloudfront Cloudflare. [snorts] Um and so that will be the change here. And I mean, the CMS is obviously something that lives locally on you know, in the code or whatever. It's not like some sort of remote thing. So again, clients need a way to be able to to to edit the CMS, which they can't through all of this. So again, I think they will remove the localized CMS aspect and it will be tuned to a remote CMS uh feature from Webflow, which I predict will be a a paid for bolt-on. Otherwise, you'll have to plug and play your own CMS or or whatever from it. So, I think this is I think and it's using Astro under the hood, which is well, it's using Astro, which Webflow have been a strong proponent of Astro for a little while. So, I I genuinely think this is this is what we're going to likely see. Yeah. Um I agree. I agree. I just hope that the agentic part Yes. that they won't close it. So, it's like I can use my own agent to build and then use the Webflow UI and whatever environment that they set up to interact with the build. And then I can bring in my own CMS, or I don't mind using the Webflow CMS and pay for it if it's not limited. Like, right now, seeing what other CMSs can do, Webflow CMS to me is very limited. Yeah. It It It's something from 2019 at best, if not 16. Uh that Like, I want to be able to to define components, feed those components with data dynamically, and also be able to repeat them as many times as I want, and drag and drop them in a page wherever I want. That's not what Webflow templates allow us. So, I want that. Yeah. I hope that they'll they'll improve on that. And I guess the most probable one, you mentioned the the agent thing. I think again, yeah, that'll be another panel here to agentically work with Webflow's AI, and which you pay a $30 a month subscription for if you want to use Webflow. And it will be all integrated into the app. However, you can still bring your own agent. You can still work with the raw Astro project under the hood. Um Yeah. of Webflow I call it Webflow build, but like, yeah. I think this is this is done. If not, if not, I I hate to say this, actually. [snorts] Uh but I think it will be really hard to make a point for Webflow for many, many clients, including even the the enterprise, because these tools will get better. In static, this, and probably like many more tools that I haven't covered or even seen, they will get better at all that infrastructure side of things. Mhm. They That's how they are starting, but they will get eventually good at letting you have users and permissions. I mean, Instatic has all of that, right? Yeah. So, Yeah. And then just just to again help people compartmentalize this tool and how it differs, Instatic's a good one cuz Instatic you can you host remotely and everyone can log in and change the same website at the same time, whereas this one again it it works through Git, so you're you're bound by the Git push, Git pull, um uh branches and and that whole workflow. So, it's slightly more disjointed in that respect. It's a slightly more difficult to work with. I I believe. Even though it's based on Astro, it's still less opinionated than Instatic, I believe. Um Like I don't know what's Instatic based on, but I feel like with Instatic you have um Well, I don't opinionated build. Yeah, I don't think I have to look into it, but I don't think you're building like a React project under the hood. I think Instatic is still having its own translation layer like Webflow. It's just it easy to work with. I don't know. I don't know. Um but yeah, this yeah. You just got to I think people dive into these tools and they don't they they might not know all of the use cases. So, I'm just trying to paint a picture that this is a very Yeah. uh scrappy or whatever it's called is a very solo effort. It it it this becomes difficult to work with teams if you don't know Git. Cuz you can see the publish button at the top right is is publish to Git or create a merge request or something with your changes. So, you need all of that set up for this to be a team effort or even a solo effort, to be fair. Yeah. Or even a solo effort. But yeah. But yeah, I feel like this is getting close. I can I feel like I can build my site maybe here and push it over the weekend. I don't have time. I I think you'd be better off I think you'd be better off using In static. I don't think this is the tool for you especially if you've got team members. I don't know. I feel like yeah. Um I don't know I Yeah, In static In static was also quite nice. The UI does bother me to be honest. Like this UI is just much nicer than In static. In static if you are watching like this is mixing Framer It's 90% Webflow and I think it's a little bit of Framer. If you squint if you squint it's basically Webflow, isn't it? Yeah. If you squint it is Webflow. I believe like you can learn so much from this UI. Webflow has nailed their controls when it comes to CSS and I think you should not like try to recreate this in a new way. Just use whatever Webflow has. Um and even like the concepts of pages versus like the navigator I I think this just makes so much sense. Um in terms of UX to have your pages here, navigation, components, assets, CMS even though Webflow has CMS somewhere else now. Uh which I agree with. Like the CMS could be somewhere else because it's data. It's not CMS like these things. But yeah. Um Very cool. Good job. Keep it up. Yeah, I mean it's funny because he did write that post and say like he wants to use Webflow and this and that. This doesn't look like he would use Webflow. This looks like he would just go straight to using this, you know. He built the the that what he was sick to death so he built his own tool. Yeah. So, I doubt we would be seeing much of him from a Webflow standpoint. So, go try it out. It's open source. It's free. It's on beta. Just download it and then tell Cloud, "Hey, in this folder create whatever site do you want." Play with it. Give Timothy feedback. Share it on social. We want to push this type of work. We Same with Instatic. We really want to push this type of work so people see it you support the creators and with that these tools will get better. And we will benefit from them. So. Nice. Well, that's all we got time for. That's everything. Everything will be linked all the tools we use will be down below in the description. I'll try and well, when I get around to actually doing it as well as timestamps and things like that. Everything you've seen will hopefully be released as shorts or clips even on the YouTube. So, make sure you go subscribe there. Um and that's kind of everything. I've been Sam. I [laughter] wanted to say I'm taking that away from you now. I'm taking [laughter] that away Cloud. So. Keep on vibing, babies.