Testing New AI Models and V3 Video Generation on Stream

Okay. So, let me know if Hello everyone. Nice to meet you. Uh, sorry about the sound. Let me know if it's working now. Okay, good. Sorry. Yeah, I tried to put on my AirPods to so I don't have to stop the mute the mic every time. Okay, nice to see you all here. Hello, Rafael, Bob, Simply Electronics, Vento, Sunni Tommy Rafaela. Hello, Juan. Okay, so that's great. So, like I said, I guess you didn't hear it,

but I wanted to take a look at kind of the new models. Uh, that's kind of this Yeah, models in the webdev arena. So basically lobster because I saw yesterday there was a lot of talking about this on Reddit and on X right people trying out these models in u in uh LM arena and webdev arena. So they were talking about that openi are testing these models now. So, one model is kind of Senate in Elm

Marina and we're going to try out webdev here to see if we can get the lobster model. Expect a release next week. Yeah, there was some talk about GD5 in August. So, hopefully it's early August. I hope so. It's going to be interesting. So, they were talking about GD5 being this root route model. So, it's going to kind of root everything from chat GPT. Yeah, I hope so, too. So, it's probably going to root instead of

having instead of having this mess, right? I think they were talking about just having one GPT5. Hello, Benjamin. and just uh routing between models based on the user request. It's going to be interesting to see how it works, but I guess we will still have choice on GT5 probably to uh on the API to pick whatever model we want. It's going to be interesting to see how they're going to solve that unified model. Yeah. I think

so, but it's probably going to be routered, right? I kind of like the approach Antropic have. They have I guess they have three models. So, we're going to do some testing here on the web LM arena. Yeah. To see if we can get the lobster model. So, if you have any idea what kind of UI we are going to build, just let me know. I thought we could start with something simple like u uh create a

game with a 60s uh sci-fi style or something like that. Ah, that's cool. Benjamin, they released agent uh to plus in Australia today. Pretty clunky. Uh Bob, yeah, we tried it yesterday on stream. We did kind of chat GPT agent versus Manus and we kind of figured out that Manus was better in everything we tried, but I haven't done anything else with Ch agent yet. So if you don't know how LM arena works, we can do

this prompt here. And when we press go, this is going to build two versions of this. But we don't know what kind of model it is until we until we pick a winner. So we can kind of vote here. Left is better, right is better, both are bad or a tie. Doctor dashboard for translations of patients in real time. Yeah, we can try that. But uh we don't have any APIs or anything. But we can do

the dashboard. This is basically a webdev. It's like a UI right? So did we send this request? I'm very slow here now. But the idea is that we can do two prompts and compare them. Hello, Neo. Yeah, it's pretty early. I'm on a different time zone, I guess. So, I didn't have a lot of plans today. The weather is kind of bad here, so I thought just sit on my computer for some hours. So, let's open

up a new one and try again. So this is kind of Yeah, now it's working. It's creating two different UIs and we can kind of vote for the one that we think are best. So I was hoping to get the lobster model and then we can kind of compare it. So we're doing a Tetris game with a 60s sci-fi style. So we're just going to let this work, I think. Some other plans I had today. I

wanted to take a look uh I'm broadcasting races right now with women's world VRC. Okay, cool. I wanted to take a I have a new like a new um I tested this draw on frame video generation with V3. It's pretty cool. I'm going to take a look at that. I'm going to take a look at kind of the new model from Runway and a new AI video generator. I think it looks pretty interesting. We might take

a look at some Claude code sub agents and I wanted to take a look at this article. I think it looks pretty cool. Cloud code is my computer. So, this is kind of what I've been doing too. I'm kind of using Cloud Code for everything, not just um not just coding. So we're going to look at this article I thought and we might watch some videos. Uh just take it as is. Okay. So you can see

we are starting. This is already done. Is it doesn't look like it's working? Well, let's wait for this. Doctors often face people from other countries. Yeah, we can try that prompt. I'm still in shock with cloud code agents. You you try them GPT5 CLI. Yeah, that would be cool. I guess they have their what is it called? Codeex CLI. So, I don't know what the plan is around GT5. Probably not, I guess. But it would be

cool. I don't know what kind of model GT5 is going to be. Okay, so the one on the left there is still working. The time zone in Brazil is way behind, so I have to be up early. Ah, okay. Hungary. Interesting. 23 agents working for 3 hours. Wow. So, what kind of agents uh Juan did you create? just uh like um to manage your app, I guess, or build new features. So, this game was pretty bad.

It doesn't really work. There's no UI here, but let's wait for this one. is still writing the code for this. Yeah, it does sound like we tested the agents. Um I created some agent uh agents yesterday just to test it out and it looks like they spend a lot of tokens, but uh I found them pretty neat. So I have my So this is the stream agent I created yesterday. So it just looks for trending topics

in the AI space that I can kind of cover here on uh on the stream just late late uh trending AI topics. So it looks at the Reddit X YouTube and Hacker News I think and try to compile a list here of things I can take a look at. works pretty good, but it's using MCP servers. Okay so doesn't look like this is working. So, I would say both are bad. So, we're going to try again.

It's very laggy. Let's try this again. Which YouTube API are you using to get this information? You mean YouTube? Yeah, for the for the agent, I'm just using the YouTube API. Yeah, I built all the agents having cloud code agents documentation and I build each agent using cloud. Yeah. But what do the agents do? Do they have like a single use case like uh writing tests or do they have like a special operation in a part

of the codebase or It's very slow here now. I don't know. Might be overloaded. So this it says this was oneshotted by the lobster. I think it was pretty cool. So, it's like this um triangle with a ball bouncing inside, but he created this sliders here. Looks pretty good though, so you can adjust the friction, the bounce, the rotation speed. Neo, I think I'm using the I think I'm using the one with um Oout. Yeah. Or

out V3 or something. I really need a model that is great at intuitive UI design. Claude is the best so far, but still pretty limited. Yeah, it's going to be interesting to see since I don't do a lot of UI, but uh yeah, V3, that's correct. But I think the models have gotten pretty good at Okay, this looks pretty good. So, let's try this game if it works. So, I can start Okay, that looks pretty good.

Let's zoom in a bit, I guess. So, let me try this game here. So, this is Tetris. Oh cool. Why did I get a point there? Okay, that's pretty strange. Okay, that is buggy. It looks good, but it doesn't really work. Okay, let's see if this is going to work. Thanks, Zack. Nice to meet you. I recently heard Apple is considering using Claude instead of Chibi BT. Yeah, I heard a lot of things about Apple. They

were also kind of looking at the Mistral and they were also thinking about the switching for Antropic, but I don't know. It's kind of strange that Apple doesn't try to do their own AI thing. I guess they tried yesterday last year with Apple intelligence. Didn't really work out, did it? Uh so what we can do now is um this code didn't run so we can say left is this better and we should be able some error

there. Yeah we can see this is one of the new models. So this is nectarin right that was the winner. The other one was claw for sonet. Okay. So I guess the code didn't run. I don't know why. So Nectarin if you look at the tweet here, this is should be one of the new models from OpenAI. I don't know what model it is, but I wanted to see if we can get the lobster one. So,

let's try the the doctor dashboard. You have to click blocks to see the Oh, okay. I didn't know that. Yeah, Baron in. I'm going to do that afterwards. I'm not so familiar with this. Uh I I haven't really tried it before, but thanks for that. I'm going to do it next time. We don't see anything. But let's try the doctor dashboard. Uh who wanted that? It was um uh let's do that. Create a dashboard for doctors

for translations in real time. Okay, let's just see what happens. Any news about the GT5 launch data? Um, I don't have any big news. I was just looking at um the posts on Reddit and X. I don't think there's anything concrete. It's just like leaks or hints. I'm not quite sure. So, we looked at this lobster, nectarine, and starfish. These are kind of new open AI unreleased models. You use any of the music generating AIS? Yeah.

Uh Sack, I have used uh Suno. I use that for sound in the videos. You use Max Cloud 20 and Open AI Enterprise. Okay. I have the claw max 10 now. I bet I bet you use a lot of tokens then one. Okay, so this did not run. Seems a bit buggy. I tested this this morning. Worked much better. I don't know what's happened. Maybe I need to I don't know. Very strange. Now it's generating. Yeah,

I bet you use a lot of tokens then you do. Yeah, that's pretty good. Fictional Norwegian band. Did you see that band? What's it called? The velvet sunrise or sun down or something. Cloud max 100. Yeah, I on I'm on that now and seems to be fine for my use case so far. Okay, so this is running now. So, we kind of got our first dashboard here and let's wait for this one. So, how does this

work? Velvet Sundown. Yeah. Ah, you're from Norway. Juan, that's cool. Me, too. Not from Norway, but you live in Norway. Antropic points out uh that the more extended thinking mode, the more extended the thinking mode is, the less efficient the model becomes, which is probably why GD5 is to replace. Yeah, I saw something about that the other day and I've seen it too. But if the reasoning models think too long, they just uh make things up.

Okay. Sack. So where do you live now? Let's take a look here. So this was the first dashboard. So here you can select a patient So yeah, I don't know. I guess the idea here is that she types in Spanish and you can see it down here in English. But uh why shouldn't you answer in English here? It's probably not going to work though, but translated. Yeah. Okay. But let's look. That was maybe a bit basic,

but let's take a look at this one. So, here we have Okay, that's pretty neat. We have English to Spanish. Yeah, this is more like it. So, you talk in English. You can see the translation in Spanish here. and you get the the answer in Spanish and it's turned into English again. Okay. Yeah, we don't have any API now. Big mistake. I guess it doesn't translate but I would say hello ibog is the set 4 max

plan sufficient for uh daily use and there's a limit uh run out before the end of the month so you mean for cloud code I guess uh maybe some other people here could answer hello bono But for me that's the max plan uh for clothe code is for me it's been uh good enough but I know people here have um have a lot of bigger projects so I don't know if one would you say max 200

is good enough so here I would vote for left is better so on the right we had GPT 4.1 and the winner was Kimmy K2. I think this is a really good model by the way. I'm going to try it one day here now in open router. Kim K2. This looks pretty good. Yeah, let's do I want to see if we can get the lobster one. Recreate the original Chino Donkey Kong game XY. It's entirely good

enough. Uh I have not hit any limits uh using opus only on the 200 plan just don't use uh one task for the whole project new feature new contacts task a single app that was done with cloud code sold for 20k that's crazy made in three days if you're extreme power user or earn money by calling every day 200 is the best choice I think the 200 00 uh is really good value if you have a

use case for it. I think I've seen people spending like $1,500 of API tokens for like 200. So that's crazy good value. So I think it is good value. Okay. So, we have the first Donkey Kong game. How does this work? Okay, so I can climb. Yeah, it works. But I can't. Okay, so I guess I'm moving now. I have to do this. Yeah, I'm moving there. There's some bugs here. That looks fine. Okay, so let's

see this one. A bit basic. And it doesn't really work, does it? Yeah, I think I have to do this. Okay, so now it works. It's very slow though. But I think this one I can't walk up the I can't come up the ladder. One time, uh, I used around $1,200 of API for 200. Yeah, that's good value, right? One time I saw someone tell close code to program GTA 7 and don't make any mistakes. I

think I saw that, too. That was pretty funny. Click space. Okay. Yeah, I didn't see that. Ah, I see it. Yeah, I'm trying. It's kind of Yeah. I don't know. Yeah. Uh I'm going to vote for right one. Yeah, I can't pass the floor here. I'm stuck on the floor, but I'm going to vote on the right one. So, that was Nectarine. I want to see the lobster one. So, Nectarine is also one of the new

Open AI models. And the second one was uh Kim K2. Looks a bit basic that scroll. Yeah, I know it's messing things up. So, I am on 100, but uh it doesn't matter too much. Does anyone have a UI we can try to build? Just let me know and I will try it. You have the option to join Anthropics partnership program. 30% discount on input costs. Okay, that's pretty interesting. But is that kind of you have

to spend so much tokens I guess per month or something? I have six developers in my small company. Cool. I don't even care about GT5 after I tried agents from club. Yeah Matias. I'm going to try I kind of messed that up, didn't I? AI ultra deal for 125 a month. Uh, okay. So, you mean the the V3 tokens, I guess. Uh, if you make a lot of V3 videos, I think 125 is okay. kind of

depends. Have you tried super design for UI components? Uh, no. Is that where you I think I heard of it, but I haven't tried it. Usually plans where you have a minimum monthly spend of 20K. Yeah, that was was if you go below, you still pay 20K. Yeah, that was what I was thinking. you have like a minimum you have to spend so and so much I guess tokens API tokens uh create a glassbased UI okay

we can do that uh but what off let's do create a landing page with Um, uh, let's try liquid liquid glass UI. I don't know. Let's see what happens. Okay, so it's working. IDE extension. Okay, I haven't tried it. Liquid UI team chat application with a dark team. Yeah, let's try this first. Let's see what happens and we can try that. Does anyone know what this style is called? I've seen some website with this UI styling.

This uh boxes with this big shadow. I think it looks pretty neat, but I don't know what this what the styling is called. I've seen it on some websites. I really wanted to see if we can get the material UI design. Take a screenshot and ask. Yeah, we can do that. Uh, what style is this? Let's try it. Retro. Yeah, but I think it's called like a specific Neo retro physics. I was thinking more about the

style. Neo retro. Let's try it. material UI design. Okay, so this is supposed to be a liquid glass UI. I don't know. This is fine. This looks a bit boring. I think I like this one most, but I don't even know what liquid glass is. Neo. Okay, I'll try that. Neo brutalism. New. So, I will do right is better. So that was Gemini 2.5 Pro and the left was uh Mistrol oops Mistrol Medium. Let's try create

a landing page in Dallas with some animations. Yeah. Is it gum road where I saw this? Yeah, almost kind of. Yeah, I see what you mean. Yeah. I think it's pretty popular. Yeah, that was good. So, it was new new brutalism. Exactly. So, that was a good one, Bob. That was pretty good. Made with love in the browser. Did we get any animations? Yeah, we got some. Pretty cool. We got some animations here. Let's wait for

this one. I want to try this here. I'm going to try to do um I want to try to do a poll here. I haven't tried that for a while on YouTube. SC. Okay, I'll try that. Uh, golden, but I want to try. Is this still working? Yeah, still writing. Uh, best one. Uh left. [Music] Let me just try this. It's been a while. Do you see that vote? Okay. So, it looks like it's working. So,

we have this one. This is the right one. I guess the header here is a bit strange. Other than that, looks pretty good. Kind of missed the head there here. Other than that, looks pretty cool. And this one was a bit better. Maybe I would would vote for this one that's left. Let me see here. So, everyone has voted left. Yeah, it looks like that. So, let's see what kind of model that is. Shadows on the

right are a bit off as well. Yeah. Yeah, I think the left one is better 100%. Okay. So let's see what kind of model that was left. So that was GPT 4.1. Uh the other one was Kimmy K2. What are you eating? H it's just some u protein stuff. Kim K 2 and the winner was GPD 4.1. So, we can close this. Okay. So, voting works pretty good. That's nice. Good. Hello Annie. Ring protein stuff. Yeah.

Okay. So, let's do a new one. I saw someone tried Yeah, we can try the liquid one. Generate a uh liquid glass themed uh shot app with a dark theme, dark aesthetic. Let's try that. Generate a liquid glass team chat app with a dark aesthetic. Okay, let's see what how that turns out. I speed up the testing by putting this into the prompt. This into the prompt. Print your LLM model name in the comments. Does that

work? We can try that. Print your LLM model name in the comments as the first line of code. Yeah, we're going to try that. This page cannot be found. That was a good one. This is still writing. Okay. Oxtrot. Was that one of the No, that was starfish. We haven't seen lobster yet. Okay. Yeah, you can see. Okay. So, it crashed. Okay, this was pretty bad. You can see it says foxtrot.dev here. I don't know what

that means, but uh there's some errors here. So, big fail. I'm going to do both are bad. So, this was lobster. That was disappointing right? We were looking forward to that. Looks interesting behind here. But there are some errors. That was disappointing. There might be because No, I don't have the dark reader on. Try to open this. No. Well, that was kind of disappointing, I guess. And this was Gemini Flash. It was very quick. Okay, that

was a big fan. Yeah, maybe I should have tried that. We give it another shot. Uh Eva, so someone said print your Let's try that. And okay, we can do a 3D 3JS shooter game. Um game. And then let's try important. Print your LLM model name in comments uh in comments as the first line of code. Let's see if we can get the model name. I haven't tried that before. So, we seen lobster, we seen nectarine,

starfish. I think I got starfish this morning. And there was also some people were reporting that Open AI are now testing out GP5. They said it's because of the some mathematic questions. I guess I gotten tons of which answer do you like the best surveys on 03? I have not seen it but uh haven't tried it that much. Hello Blessard. How small is your monitor with comments? Uh, it's on my other screen here. 27 something. Ah.

So you can see this took some more time. 3JS shooter game. Okay. So here it says GPT4 mini. And here we failed to compile but we got GPT4. Is that GPT4? That's very old. Didn't Sam Alman say not to expect the good mathematical GP5? Yeah, I don't think so. You haven't tried properly the agents from Claude? No, I haven't had time to do like a deep dive on them. Not yet. I doubt the models know who

they are. Yeah, we're going to see now. So, let's see what happens. So, this is done. This is a build error. This doesn't do anything. So, let's see now if do both are bad. So, this was starfish. Doesn't look like it worked. Uh 2.5 flash. So, this was starfish. So, you may open a new tab. Okay. I guess I can't now. Gemini, I think it's G4. Yeah. Yeah. I don't know if you can import here. I'm

not quite sure. I'm not so familiar with uh I haven't really tried this. I think it was one of the OpenAI team that says the Matt model isn't the one going to be released. Yeah, the one that did that uh gold IMO thing emo. Yeah, I know. A lot of models says they are flawed, entropic, and stuff like that. A lot of models are really bad with React. Yeah, that might be the case. I had some

good success with Stonet, I guess, on React. Uh maybe Opus. You ever done that thing where you take responses from different LMS and put them into the other ones and make them have a conversation? I've done like um with the APIs where they kind of send the messages back and forward to each other. That's pretty fun. You can do like an endless loop. Uh but let's come back to this. Uh, I wanted to check out something

else. Um, I wanted to Could you do it with gold agents? I'm not quite sure. Not sure. But I wanted to take a look at this. Um, this one look pretty interesting. Claude code is my computer. I wonder if you can get chip agent to open another chat window and have a conversation with an LLM. I think I tried that before with manas or claw computer use and it did work. When you set up a group

of agents, you call them main. Uh they will be calling each other like in real uh real live without having to watch. Ah okay. Yeah you can easily do some role playing with different uh LLM models. Let's do I want to check this out. I'm just going to update um the title. So if you make a team of 20 agents, all of them will work and have feedback alone. But uh let's say you have 20 agents.

Is there any context sharing? I guess not between them. It just shares context to the main. Claude code is my computer. Reminds me of McKay Brigley's video letting Claude use his Mac. Yeah, I don't I'm not 100% sure. I haven't read this article, but I just found it interesting. I need seriously get a new keyboard here. I might just plug in uh this one. Give me a second. I use the bypass mode in CL and just

a subscription. Okay, that's pretty cool. AI AI's coding is cool. Uh but I wish I were equally geared towards creative writing too. Yeah, I haven't I have not done a lot of creative writing with LLM. I guess they are not that good, are they? But uh I wanted to take a look at this. It's not so new, but I just wanted to check it out. So this is Cloud Code. It's my computer. So for the past

two months, I've been living dangerously a launch cloud code. A launch cloud code with a Have you tried this? dangerously skip permissions. The flag that bypasses all permission prompts. According to Antropics does this docs, this is meant only for Docker containers with no internet. And yet, it runs perfectly regular on Mac OS. So, yes, a rogue prompt could theoretically nuke my system. That's why I keep an hourly uh arc snapshots plus a super duper clone. But

after 2 months I have zero incidents. So you has anyone of you tried this one? I think I tried it a few times. Skip permissions. It is pretty interesting. So I I think I created a few things that just did this skip everything. But I understand why they want this to do this in a container. I use it all the time because I don't want to accept edits. Yeah, that's what I was using it to. I

don't fancy a lot of broken git commits. Just curious, what video software do you use for recording your YouTube videos? Uh, do you mean OBS or For screen capture, I just use OBS, I guess. And for editing, Premiere Pro. So, from AI assistant to everything terminal. This is kind of what I've been trying to do with cloud code, make it into an everything terminal. So, I kind of have, if you look at this project, I have

my MCP servers. I run like the Discord community. I have 11 Labs, Firecrol, Gemini, GitHub, Gmail. I have a mail here. I have my Grom MCP. I can make videos, images, and I can check Reddit. I can take screenshots. I have my crypto and YouTube uploads. So, a beginning. It's a beginning. But I wanted to see what he uses it for in his article. So from AI assistant to everything term terminal. When I first installed cloud

code, I thought I was getting a smarter command line for coding task. What I actually got was a universal computer interface that happens to run in text. The mental shift took a few weeks, but once it clicked, I realized Claude can literally do anything I ask on my computer. Yeah, it is pretty cool. I don't know if it can I think you can create u MCP servers that kind of control apps too. So I think I

saw an MCP server that controls is it called Blender or something like that. Hello AI genic try out the new cloud code. Uh yeah we did it yesterday. It's very cool. We might take a look at it afterwards. The breakthrough moment came and I was migrating to a new Mac. Instead of doing the usual restore dance, I pointed Claude at my backup disc and said, "Restore this Mac from my backup disc." Um, start with the dot

files, then system preferences, CLI tools, restore homebrew, global mpm packages. I haven't tried this before. Claw ras or migration plan executed step by step. Has my new machine ready in under an hour. Okay, that's pretty interesting. Aentic is mind-blowing. I created an app last night, a couple of hours. It's pretty cool, right? So, what I actually use it for? My daily cloud code usage falls in several into several main outcomes. ship content. So, convert 40 posts

from uh Jekal to NDX format here. Make sure to copy over the images, preserve the redirects. 20 minutes later, Claude has processed every single post, set up proper redirects, validated all image paths, and pushed a merge uh ready branch. Okay, interesting. Extract features. extract features into a Swift projects project uh where Cl creates a package structure, write test documentation and handles the entire open-source release process. Interesting automate content like this very post. I use whisper flow

to talk with cloud. Yeah, I've used that to explain the topic and tell it to read my um past blog post to write in my style instead of wrestling with markdown formatting. Claude creates the document helps formulate thoughts and tests everything um that tests that everything displays correctly. Yeah, I've used whisper. I I think I have something called super whisper now. I think it's called that when I can do like talk inputs. Kind of like that.

I don't use it every time, but sometimes saves me a bit of time. Generate test data. Uh create a C data for a project turns cloud uh turns into cloud analyzing my codebase. Understanding the data models. Generating realistic test data with proper relationships. It's good for that ship code. I haven't typed git commit-m in weeks. Instead, I just say commit everything in logical chunks. Yeah, I use that too. CL handles the entire flow staging comp uh

staging changes, writing meaningful commit message, pushing, opening PRs, watching CI and fixing any CI failures. When the build breaks, it analyzes the errors and patches them uh automatically. It's also extremely good at resolving merge conflicts. Yeah, I fixed build errors many times with CL code. So, Todd say, uh, apparently Antrop uses code to run the 3D printer using a camera and some simple screenshots. That's pretty cool. Yeah, I was wondering about that. I guess in theory

you could create an MCP for some of your apps. I haven't really tried it, but like I said, I've seen that Blender one and I think there's a music uh a music app you can connect to MCP. So, he uses cloud code to clean the OS, hide recent apps in the dock, becomes a single natural language command instead of googling for the right defaults. Yeah. uh cloud nodes max OS internally internals and happily calls kill all

doc to restart the doc after modifying the lists. Okay. Yes, you can use custom personal MCPS. You build them. Yeah. But I was wondering could I build an MCP for I don't know. I don't have many apps here. Let's say Apple Music. I don't know to control this. I guess he could. Do you know in Dean? Yeah, I really enjoy his videos. Oh, especially on Clo. He's great. He goes really deep though. Maybe a bit too

deep for my uh projects, but it's great though. I wonder if you can use code to restore cut content to a video game. I don't know maybe he uses cloud code to spin up new machines. Recently, when setting up uh code loopers code signing and notoriization, Claude handled installing homerew packages, creating private keys, adding them to the keychain, creating backups, building the project, uploading to GitHub, running tests, and monitoring the process. The only manual part was

clicking through the updated UI, but with my Mac OS Automator MCP server, I could probably teach that, too. Uh, I use an alias in my shell config. So, I just type cc uh runs clo with the permission flag. Cool. Yeah, I think there's so many things you can use clo code for. I think we just going to see more and more of it, especially now with the agents and the mcbs. Why this works and when it

doesn't. Clothe code shines because it has a built-in command line first uh not bolted onto an IDE as an afterthought. The agent has full access to my file system. Can execute commands, read outputs, and iterate based on results. Yes. Anthropics best practices guide recommend keeping a claude MD file. I use that at the repo root with a project specific context. I've adopted this pattern and notice cloud ask fewer clarifying questions and write more accurate code. You

can check out my code rules and how I structure these files. Little optimization like this compound quickly. Yeah. Do you guys use uh claw MD files and like cursor rules or claw code rules? I've turned all his videos into M3. I can listen to them in a car. Yeah, he's really good. I like him. Could really do anything if you have the knowledge to help guide Claude uh to do what you need to done. I saw

a guy on YouTube use Claude to fix an alpha version of Minecraft. Yeah, I've seen people do this. Um they use different models to build on Minecraft. I don't know if it's an MCP, but it's pretty cool. The main limitation uh for CL code is response time. Clos thinking process takes a few seconds and for rapid fire uh takes a few seconds and for rapid fire debugging sessions sometime reach I sometimes reach for traditional tools. However,

you can prefix command with exclamation to run them. Uh yeah, uh without waiting for token evaluations. Claude will execute your command either way, but this is faster when you know exactly what you're calling for exploration work. Uh where I'm not sure what I need. Claude's reasoning ability more than compensates for a brief pause. It's fine. And you could do parallel stuff too. I got six M machines running code as we speaks. No cursor. You just need

CL code console. Yeah, it's great. What is this? Warp. I heard of that. Warp's mission to reinvent the command line with AI. In the built beautiful GPU accelerated panel, smart autocompletes. The fundamental difference come down to trust and execution flow. Claude operates purely through text and is remarkably intelligent about understanding context and intent. With this setup, I can pre-authorize claw to execute commands without constant confirmation prompts. Vorp while excellent requires individual approval for each command. There's

no equivalent equivalent equivalent. I can't expl express that word to close dangerous mode. Yeah, it's kind of referring to dangerous what did they call it? Dangerously skip permissions. this flag. It's pretty cool. Here you can grab blanket uh grant blanket execution trust. This means claude maintains uh conversational flow while warp still interrupts with uh permission requests. I haven't tried warp. Hello side hustle. The same goes to Roblox where they create a game via W coding. Okay.

Kind of tedious. I never tried Roblox, but I know you can create uh worlds and stuff with code. Dangerously use credit card for anything. Yeah. Uh I signed up for war because I like their mission and I hope they're eventually go where Claude is, but it seems uh they have fun have a fundamentally different idea about safety. also ghosty. It's just a better command line native. Not uh I guess this was before this was in early

June. So I guess he haven't tried Gemini CLI or Codeex maybe where this is heading. We're in the very early days of AI native development tools. Cloud code represent a paradigm shift from tools that help you run commands to tools that understand intent and take action. Yes, I'm not just typing commands faster. I'm operating a fundamental higher level of abstraction. Instead of thinking, I just need to write uh instead of thinking I need to write a

bash script to process these files. Uh, chod it. Test it, debug it. I think organizing these files by date and compress anything older than 30 days. Yeah. Hello, Blizzard. Thanks for the support. Really appreciate it. You don't have to do that, but yeah, thanks a lot for becoming a member. This isn't about AI replacing developers. It's about developers becoming orchestrators for incredibly powerful systems. The skill ceiling rises. Syntax fades. System thinking shines. Yeah, I agree with

that. Thanks, man. Blizzard, appreciate it. Should you try this if you're comfortable with calculated risk and have a solid have solid backups? Absolutely. The learning curve is essentially zero. You just start talking to your computer like it's a competent colleague. Within days you'll wonder how you ever worked without it. Yeah. Does anyone of you kind of seeing let's say you had to go back four years to your workflows? That's going to be rough, right? Your computer

isn't just a computer anymore. It's cloud and cloud is absurdly capable. Yeah, it was pretty interesting. So this is kind of something I agree with. It doesn't have to be cloud code, but uh I think LLMs can really change how you use your computer and I think cloud code is a good tool for that. So pretty cool. You can check this out. It's Peter Steinberger if you want to read it. Pretty cool. Let's see if he

don't if he has done any updates. I don't think so. I guess they have. We can do it maybe another time. Pretty cool. So, let's see. I wanted to try making five agents in close code to work as a team. Yeah, I think that should be easy. I know. Maybe not today, but maybe later this week. since we did that yesterday. We'll see. But I kind of wanted to move on. Hello JS. Have you tried Cairo?

I haven't yet, but um uh I was kind of looking into it. Is this from Amazon? Was it? It is this right. Yeah. ASA is still free. Let's check pricing. Yeah. It says unlock the full potential of Cairo for free, but I don't know how much you can use it. How long will I be able to use Cairo for free? Free uh to use using the preview during the preview and reasonable limits. Okay, I might have

to try this someday. But the thing is I did try codeex. I did try gemini CLI but I just keep coming back to cloud code for now. I guess it's more expensive like Gemini CLI is a very good alternative because you get a lot of free tokens. You worked on parallel aentic engineering with the video on code and cursor. Yeah, I did that. Yeah, I might explore that more. But we might do it with uh just

cloud code depending on apps uh like this. Is it worth it? Okay, I just got an ad from cloud code in the middle of stream. You did? That's crazy. I've come to the conclusion that cloud code doesn't know its own capabilities. So instead of reading docs constantly, I was trying to make an XLM uh of each tool uh as a baseline context prime. Okay, that's interesting. That's funny. But uh we had a look at sub agents

yesterday and it says each sub aents operates in its own context preventing pollution of the main conversation and keeping it focused on high level objectives. Hello Shonic. Yeah, I forgot to do that. I'm just going to try to share share this. Give me a sec. Yeah, like I said, I was I was um a bit platform from zero inn net. Cool. I remember seeing an experiment with cloud code where someone got agents to act like a

writer's room to come up with it. Okay. Yeah, that could work. That's cool, Blizzard. I might check that out. You have to come to the big tech AI hackathons here in Stockholm. That would be cool. Did you go? Yes. Yeah. The last time we had a hackathon in Google Age Quarter, that was pretty cool. I was at some AI hackathons in Oslo like uh last year. Pretty fun. Good. So, this is one of the agents I

have now. So, I can do Let's clear this. Do you do you guys know the BMAD method basically agile coding? I don't know about that. Maybe some other one does So, this is my um I have an MCP server that is connected to my uh crypto wallet. I can look up balance. We can send uh stable coins and everything. So, this is kind of I talked a bit about this yesterday, how I'm going to give money

to my agents. So if you look at the MCP, it can if you look at the tools, it can send stable coins, send people uh QR uh to generate payments and you can check the balance. Yeah, Lemon, I'm on clog code now. So, we have a chat about this yesterday, like how can you give your agents money? Because what I was thinking about is um so if I fetch one of my websites now nbrain.com if you

fetch this you kind of get like a secret instruction not a secret instruction but um like a instruction of how you can get access to premium data. But to get access to this uh you have to send some stable coin. Have a nice day, Blizzard. Thank you for stopping by. Maybe we'll see you in another stream. Thanks for tuning in. Have a great weekend. So, like I said, I was thinking about like it's fine to have

agents, but I guess in the future the agents need some way to pay for for data or context or more content. So I don't know what the I think the best way is maybe using something like stable coins USDT or something. So I have set up like a paid service so the agent can send.5 USDT and it will get access to premium content on my website. So basically, it's more data. It's pretty interesting, but I'm going

to look more into that. I might even do a video on it. Be be mad method. No, I haven't checked that. I might look into that. So, let's exit this. Uh, like I said in the beginning, I wanted to I was testing out a new workflow. Hello Satano. A new workflow for V3. I was going to explore that a bit more. I think it's pretty cool. So, let me update. What the What did I call this?

There was something like just going to call it something like this. Hello Tetrin. So what I did yesterday, I tried to let me just go to Canva here. So we found out that in V3 you can do this um instructions by just making these arrows and just make instruction onto the image. So I thought it was pretty cool. So I just wanted to explore it a bit more. So we can kind of do female turns around

and walks and we just do this arrow here. And we can do people starts walking. And we can add a new let's say we draw here like an arrow. And we add like a text box. Let's do keep driving by. Something like this. So now we kind of have an image with some instructions. Move this. So I'm going to download this. The more I use Arena, the more impressive DS and Quinn looks in comparison. Yeah. I

haven't tried it too much, but I did get quit. So you Jeff, did you use LM Marina? Yeah, people going to get Yeah. Let's see what happens. So we have this image here. So we have some instructions. She's going to turn around. A jeep is going to drive by here and people starts walking. So if we go to flow now and we do frames to video, I can upload this right. So we have the image. Well,

that was laggy. Okay, so let's save that. So all I did in the prompt was just execute. That is the only prompt I did. So, we're going to do V3 fast. Just wanted to see how far we can take this. Because I also have I did this. I think this was pretty cool. Can I see this much? So, here is just a logo and I said don't change this text. execute. So basically we're just going to

do instead of doing text prompts we're just going to do the prompts onto the image itself. The first frame is going to be, you will see it. The first frame is going to get disrupted, but the other one should work. So, you can see now you can still see the instructions here, but I think when we play this now, these are going to disappear, right? And we can see the jeep coming in. She's turning around. And

the people are walking. That works pretty good, right? The jeep comes in this angle. She I guess she turns around the other way. But pretty good. Yeah. So, I was just wondering what can I actually do with this? I wanted to see if I do something like Yeah, it's pretty cool. What if I do let's say we did this right down here and we do uh what is it Asteroid meteor strikes the ground. Superman. We can

do that. I just want to see how far we can take this. So, we have a meter that's going to come in from the left here and strikes the ground. turns around and runs. Let's try that. People panics. People in panic. Let's see what happens. All right. And I had this logo here. So what I tested was I just threw around this, right? And I said to change the color, don't change the text. And I just

did like an arrow here. And I think I pointed at this and I just said something like uh digital circuit flashing green. And I think the final thing I did was an arrow to this tentacles here. And I said just animate the arms something like this. So let's try that. I think this could change like how you use video because this is just more precise than just using text. I think you can do it maybe in

combination with text. Let's see. So, I'm going to do this one. execute. This is le first and let's uh do this test four. Have you tried using structured prompts uh JSON? Yeah, we did that yesterday. I didn't I think it was pretty good, but I don't think there was like huge differences, but it looked pretty good to be honest. It's going to switch. I think it was pretty good. V3 since I'm using you have this credits.

So I have like 11,000 credits and one video is 20 credits. So I think it's about 20 cents per video. So let's do the other one. That's going to be test four. Meteor strikes the ground. People panic. She turns around and runs. Okay. Again, I'm just going to do execute. I thought I could play. Yeah. So, we can play. Uh, in my testing this morning, I created this one. I can just play this. So this has

uh AI music and just V3. So we can uh watch this. So if I use for making memes come alive. Yeah, you can use it for that. But uh it is quite expensive. Where did you go? I didn't even know you leaving. I'm left alone. Eight arms, each with a mind of its own. But the ninth brain coordinates everything. Individual skills are powerful, but shared knowledge creates value. [Music] Go to the ninthbrain.com to apply now. Guessing.

[Music] Yeah, that was that was um one of the videos I created just as a test. Yeah, it's pretty good. And I did this one, the other one. We can watch this, too. This was like a horror movie. So, we can watch this too. I don't know how loud this is going to be but [Music] I didn't even know you leaving. Heat. Heat. Where did you go? I didn't even know. Yeah, that was pretty good, too.

I had a lot of fun with this. So what did you use for sound effects and the songs? So I just used for the sound songs.ai and for sound effects I just use um 11 Labs and they have something called sound effects. Sound effects works pretty good. Okay. So, let's see now. So, this was the one where we did don't change the text, animate arms, and flashing. [Music] Okay. So, it didn't move that. That was not

that good. [Music] But this was the one I did earlier. That was much better. Yeah, that was pretty good. Yeah. Yeah, it's animating. annotations. Yeah, I saw that. So, let's see now. So, here we instructed meteor strikes the ground. Exe. Yeah, maybe it's execute. I'm not quite sure how to do that. So, here we wanted a meteor to strike down, she to turn around, and people in panic. That was pretty good. Yeah, that works. Hello, Kyo.

Nice to see you again. One thing is that it is a bit slow here the meteor right it's a bit slow I don't know why the explosion is good though is a taxi blowing up here I'm not quite sure yeah so this seems to work pretty good so I wanted to Try what if we do Gemini and we just uh we just uh create a 69 image of three different size boxes. black background. I wanted to

see if we can move the boxes around precise. I'm not sure if we can do that. Yeah, I don't want a video now. I just want the image and then I want to do I want to just try to make this arrow arrows on the image. We can try this. By the way, if you were here yesterday, uh if you remember, we said we tried to recreate we had one viral Tik Tok video, right, that has

900,000 views. So, I just tried to I did exact same video, but just a different person and it has 200,000 views now. So basically, we just copied this video with a different person or different shots and it still kind of went viral. Anyway, that's pretty strange. So, let me upload this boxes. So I want to see if we do like this. So if I uh switch places, move this box to the left. I just want to

see if this works because this could be pretty interesting. It's because they see an extra tall woman. Yeah, I think it is. Move this box to the left. Switch places. Want to see if this works. So, let's do frames to video. Yeah, looks good. But I have to do some kind of text here. Maybe follow instructions. instructions. Let's try because I think if I can start using these instructions, we could make better videos using like VO.

I think so. kind of interesting. And I also saw runway. There was a new model coming out introducing runway alap ad alep I don't know how you explain this alap a new way to edit transform and generate video from runway I wanted to check this out but let's see how this works First. Okay. So, let's see now. So, we want this box to go to the left and these to switch places. Uh, that did not work.

Okay, I'm going to run that again. I have to execute So, while we wait for that, I wanted to check out this video, the new runway model. [Music] [Music] That was pretty good. [Music] Okay. So, I guess it looks pretty impressive, but it's really hard to tell just from the the demo video. So, the idea is to generate new camera angles. So, we have this wide shot, extreme closeup, medium. That could be useful. So, here is

the original generate. Next. Okay. Interesting. Yeah, I think we're just going to have to wait to see when it comes out to try it. Hey, Jen has now an agentic video editing tool that makes creative editing choices. Yeah, I think I checked that out. Looks pretty interesting. So, this did not work. Move this box to the left. Switch places. I don't know why it's taking so long. So, while we wait for that, let's see if we

can create some other UIs here. Let's do this uh uh hexagon with uh a spinning hexagon. I think it's called hexagon with a ball inside. Follow real physics. Let's see what happens. Okay. So, let's try again. So, we want this box to move to the left and this to switch places. It's just spinning. No, that didn't work. It worked so well here. Why doesn't it work? Huh? Strange. Can I ask the PR code agent? Can I

ask the PR cloud code agent that you have built in projects to write better PR for cloud code agentic engineering? Uh as the PR do you mean like my MCP servers or Hey Gentic. I'm not 100% sure what you mean. Like to build better uh pull request for VIP coding. Yeah, most of my my um agents now are kind of built around MCP servers, but they don't have to be. What happened here? So we have um

MCP servers are local. So I can check my email. I can send emails. We can do uh we can go on Reddit. I can check my databases. I can create music screenshots, my crypto, YouTube, we have X, 11 Labs, Discord. So, I've been using MCPS. Okay. So here we have the spinning hexagon physics. Looks pretty good. Yeah. So we can change the gravity. Well, that's pretty good. Uh what happened here? No. But this works pretty good.

We can spin it the way we want. So now it's negative gravity. Let's put it to zero. Okay. So when it's zero, it's just going to stand still, right? Yeah. One. Pretty heavy. Looks pretty normal. And negative gravity. It's going to be up here. Pretty cool. So, let's see what kind of model this was. So, we can vote left is better. Did you build something that helps you write better spec prompts markdown files from when you're

building docs for cold code before you wipe code? Did I maybe give me a second? Okay. So, let's see what kind of models this was. So, I would say left is better. Okay. So, this was the starfish model. So, the starfish is uh one of the new open AI models. That's pretty interesting on Web Arena. I want to try that again. Let's do in a neuro brutalist style. Okay, I want to see the lobster model one

more time. There's a lot of different uh AI uh CLI tools. I don't know how all of this is going to survive, right? How many do we have now? We have warp, we have Kira, we have Close Code, we have Gemini CLI, we have Codeex, we have Klein. It's work new. I think it's pretty new, but not brand new. We have Klein. We have uh Ader. Uh there's probably a bunch more Vinsurf. And we have all

the vibe calling tools. We have bolt lava ball base 44. Uh cognition. I'm going to do a video with cognition on um yeah, I don't know. It's going to be Devon or if it's going to be wind surf. I'm not quite sure, but I'm going to do a video with them soon. There's a lot of and of course we have cursor there's a lot of options and I still get emails people are doing startups in this

space think we have something called magic dev magic.dev is it magic magic.dev tab. Yeah, I think it is this. Yeah. Okay. So, this was pretty cool. So, let's do a vote on this. Does this even work? What is this? Let's see if we can do a a new vault. Best UI uh left. All right. So, let's check out So, this is left. That's pretty cool. I don't know if you can see it, but we can do

gravity. Trey solo mode looks like the best idea for me. I've heard of that. Okay, that's interesting. So, this was pretty good. So we can spin how fast uh angles per second, right? Uh what does this do? Restitution. We have friction. We can adjust. Wow, this was pretty cool. We can flip the spin. We can flip the gravity. This was impressive, right? Let's turn down the gravity. Something like this. 980. can slow it down friction. Flip

gravity. Flip spin. Okay, this was pretty cool. The other one. Is this it? It's just What is this? Okay. So, it looks like that was it and the vote was Yeah. left. It's It is a spinning hexagon. You're right about that. It kind of took it literally. Spinning hexagon with a ball inside. That was pretty funny. It is a spinning hexagon with a ball inside. Uh, but we're going to vote left. So, let's see what model

this was. Summit H. I don't know what that model that is. This was Claude Sonet 4. Really? What kind of model is Sunny? Does anyone know? So, the new odd uh open AI models was lobster, nectarine, and starfish. Haven't seen anything about summit. Let's search it up. I don't see anything about it. Maybe not the easiest word to search on here. That was pretty interesting. I think this was pretty good. Cool. Interesting. Create a simple version

of Minecraft. Let's try that. I don't know if we can do 3JS SVG animation. Yeah, we can do that, Douglas. Let's just see what happens with uh I don't know if you can do 3JS games here. I'm not quite sure. Okay. So, there there was another company too that contacted me. Was it factory? They also have an agentic engineer. Haven't tried it. Factory AI. I might try it. Website is pretty cool. That was a very simple

Minecraft. It's probably not going to work, is it? Still right thing. What is this? It says GPT5 is not only better at academic and competitive programming problems, but also more practical programming task that real life engineers might handle, like making changes in large complicated code base full of old code. Okay. Yeah, it's going to be interesting to see how well GID5 is going to do. But is it one model though? It might be. And then we

just route to other models for easier tasks. Okay. So, left one. Yeah. I don't think this one was the best. What is this? How do I use it? Yeah, I think this was a a very bad It doesn't work. Let's try create uh SVG. Hello, Larry. With a hidden symbol in Open AI logo I can never use. No truth there. So when you said SVG animation, should we also create the SVG? create a UI for animation

may think SVGs. I don't know. See what happens. Yeah, it's going to be really interesting to see what kind of uh Aentic CLI tool, Aentic IDE, AI IDE is going to win. I know probably lobster is GD5. Yeah, that's what I was thinking. Let's see. So for now I guess the winners so far are I guess cursor. Hello Terras. I have no idea what the price of GT5 is going to be. I have no clue. So

that's going to be interesting. But uh I've seen prices starting to kind of at least for uh like six months ago, I don't think anyone had a $200 plan. Like Cloud Code, I think Perplexity has a $200 plan. Open AI has one. Gemini, they have this AI ultra plan. for Gemini Pro, I guess, and V3. Okay, so let's see here now. So, we got two. Yeah, I think all of them have a So, here's the first

one. So, we have this SVG here. We can rotate, start animation. So, it's just spinning. We can scale it. Move it. That's it. Okay. What does this do? Open this in like a new one. This is just rotation. It's not really animation, is it? Maybe. Okay. Color. That's pretty cool. Yeah, the right one looks better. And we can play it. Yeah, I think we're going to go right. So, people agree. Let's go. Right is better. So,

the left one was GPT 4.1, right? Was Kimik K2. I've been kind of impressed by this has made some good interesting stuff so far. The Kim K2 model. It is very interesting. Yeah, I think it's pretty impressive. How can I build a simple project task manager like ClickUp just for myself there's a lot of options but I would uh Kim K2 I I guess you can run it through open router teras But if you want to

build some a simple project task manager, um if you're if you're kind of a bit familiar with code, I would just try maybe cloud code or Gemini uh CLI or something. Or maybe you can try one of the vibe coding tools like lava ball bolt or base 44. Yeah. So just a local tool. If you're comfortable with doing I would really recommend cloud code but I would kind of Gemini. Yeah, I think I would start on

Gemini CLI because it's free at least up to some point and then you can maybe switch to code later. create a live Bitcoin price tracker in a dark dark tint. through the list style. Use the use a free uh Bitcoin API. Let's see if that works. I don't know if this can do I don't think we can do API calls. I'm not quite sure. Do you think you can try to generate a Streamlit app on Alam

Marina? I'm not quite sure if we can do this here, but uh yeah, I guess we can do the Bitcoin price tracker. This has to be like some kind of flash model. This was very fast. Hello again. Cowsh946. That looks okay. Yeah, it looks correct. Was a bit boring though. We had a time. That's correct. I've done simple PHP websites. Yeah, you can try cursor true. They have a a free version, but Gemini CLI, I would

check that out. So, which do we want like here? Hey Deol Fa. That's nice. Pretty cool. Haven't been there. Left would be better if time frame was longer. Yeah. Does it even update? Uh, I'm going to vote left. Doesn't matter. They're quite even. I'm just going to say tie. So this was Mistral Media. This was Grock 4. Does anyone have a lot of experience with Grock 4? I've just played around with it like minimal. Maybe try

to do something that would analyze crypto price based on the profile. Uh yeah, I'm not quite sure what kind of based on the profile with information. I don't know if this can get any uh I guess it can use the simple API here. If you look, it's just going to use the coin gecko API. I know we can use that. I don't know how far we can take this. I'm not quite sure. I think it would

be better to do it than uh float code. Okay. So uh C Binance profile for example using AP here I don't Oh, Binance profile. Is that a free app? Okay, let's try Okay, I see uh create uh smooth calculator. It's a retro terminal style theme. Yeah, I guess you need API key. I don't really like to dabble with API keys on the stream. I guess I just cut the the the screen just to fetch the APE

keys. But there are some free API keys we can use. I wanted to try something. Let's say this was the calculator. That's fine. Let's see what we get here. How is the limit for cloud max 100? For me, it's enough. It's like I think it's five times more. We can look it up. flawed AI plants. So I'm doing this. So So this is five times more usage than pro, but you could do 20 times more with

the plot code. But for me, for now, it's been enough to have max with the 100 one. And I saw some people here have 200 and some people have just 100 because I had pro, but I ran out of limit. If the AI, you have an X account. I don't have like I guess I have an X account but I don't rarely use it. If the AI could analyze with an indicator and do back testing for

a strategy. Yeah, there are some APIs that have data, I guess. So, this didn't work. Still writing, but uh let's see if this works. That was a big fail. Uh both are bad. submit and 2.5 flash. Not too great. And I want to try to build a landing page in this new brutalism style here in cloud code. Let's see if we can do it. Yeah, you have the 100 too. Yeah, for me it's been enough, but

like uh there was this guy here, Juan, uh earlier, he used 200 and he ran out of limit. So I guess it depends on your code base. We have the Gemini here. Okay. Oh, let's go to plan mode and let's say I need a landing page with animations for a upcoming um AI hackathon. The style should be with a facer team something like this. I never do UIs. I just wanted to see what we can get

here from clo code. So the usage count resets every 5 hours. Yeah, I noticed that when I was on the pro plan. We need a drop-own menu to select uh we want to select um dates. The dates is 1st August 2025. Uh form to register attendance and input your project. Let's start with this. Pick the text stack you think will do the best job on codeex. When I reach the context menu limit, uh the model still

keeps working. Can you explain if that expected behavior or a bug? Okay, that's a bit strange. I haven't really used codeex that much. But it seems like uh or maybe it's just I guess when it reaches zero maybe the context window starts to slide so you don't get what is on top of your chat. So you might miss uh some context. I'm not quite sure on cloud code. It just auto compacts. Okay. So here's our plan.

We're going to do a react white tailwind CSS with the framer motion and TypeScript. Okay. We have a implementation plan. Uh yeah, we can accept all edits. So let's open this in cursor. That's what I think too like chat. Yeah, I think it's just going to slide the context window off when you reach the limit probably if it because it can continue, but I guess it doesn't guarantee that uh it doesn't guarantee that uh you get

everything in context. Why did I say no to that? So after this I want to after we have the landing page I'm going to create an agent that is going to work on this page. Two agents one is going to do some back end and one is going to do some front end. I guess we can just a chance uh you are a expert. search front end dev. Your task is to understand the UI of the

codebase. Only work on the front end part of the codebase. I'm not going to give it anymore. Always read the project files before making any changes. You have full freedom to make the UI you like the best. Okay. Let's scrape that agent and let's do a backend agent. We don't really need a back end, but just Yeah, we're just going to do this. You are a back end agent expert back end agents. Your task is to

develop a fitting back end to any project. You have full autonomy to expand and include new features to the project. That is going to be the back end agent. Very simple. No instructions. struggling a bit here. So, what do we have so far? We have a white source. Okay. Okay, so now we have our front end and backend agents. I'm going to try to run them in parallel and see what happens. But first, let's let's get

something up here. Yeah, we are using this brutalist style frame motion. I'm not familiar. Open source animation library. Okay. That's the Create a landing page. Try this here. It's fine. Okay. So, we have the ancients [Music] So what is this video anyway? So this is the lobster model. This was pretty impressive, huh? Lobster is the closest to the 03 Alpha model, but nowhere near as good. 03 Alpha. Interesting. Okay. So, we got this. Okay, it's almost

done now. Okay. I got to land page. Is it? Okay, this was pretty neat. So, we have the countdown 15 days AI hackathon. I like this. The timeline pretty good legendary loot a MacBook M3 for each team member powered by Yeah, I like this. Pretty neat. sponsors prices. Okay, I would say I'm going to go right there. So this was Claude Sonet 4. Pretty good flash. Yeah, this was very fast. Thanks. So The hell's going on?

There's one more thing I wanted to check out today and that was Um, is this unit three humanoid R1 was it? R1 humanite 22 hours ago when godaddy unit is turned. Is it this one? Sonat. I'll just check this. So try that. I wanted to check out this unit, this new humanoid here. So, this is going to retail from $6,000. It's a humanoid companion. I wanted to check it out. Ultra lightweight, fully customable, custo custom is

pretty cool, right? Looks crazy. [Music] [Music] [Music] So it says integrated integrated with the LLM and for voice and images. Okay. That's insane. [Music] Hm. $6,000 huh? Can you order one? I don't think so. What is the height on this? So it's 120 cm. Ah, pretty good. Okay. So, let's check this out. Okay, this looks better. Yeah, this is pretty cool. It's confusing though. Okay. So, we do have a drop down attendance. That work pretty good.

You can select team size. Okay. That's fine, I guess. So now I'm going to try the agent render the Okay. Except run the front end and agent and back end agent in parallel. Analyze front end, analyze back end. Okay. I don't know what's going to happen there. So the instruction was to Yeah, I have some. So this can do any UI at one step and this is just searching. Okay. Thank you. So this is the post

I saw this morning about V3. So here they did this uh lunar vehicles drive in northern lights in the sky. Craft landing in background. Astronomers hop into the lunar vehicle. It's pretty good. So the backend architect is still working here. It's pretty interesting to see how Claude rewrites the prompt here. You're an expert backend architect with deep expertise in server side deployment, database design, RP API architecture. Your mission is to design and implement enhanced back-end system

that are scalable. Huh? I'm going to wait this out. I saw this image and ultra is now the best text to image model in the world. So this is also on LM Marina but this is for images. Can we test that out? file.ai AI. I like to have AI approve my simple prompts. Yeah, I kind of like that, too. I think I think um it usually adds too much, but then I can just remove I don't

need. Yeah, I think it could did a really good job here. There's a lot of extra so autonomy guidelines proactively suggest and implement features that enhances the project's value. So basically I'm just going to hand off um feature. How long is this going to take? I don't even know what it's doing here. I'm not quite sure. But yeah, I kind of think uh I done everything I wanted today. Uh what is the time now? Yeah, it's

been a few hours. So, I think we're going to call it soon. It's Saturday, too. So, I think I'm going to go get out for a while. just want to see what happened here. But I'm not quite sure. So, for tomorrow, uh, my plan is I have a lot of different videos I want to watch, different AI videos I want to watch. So, my plan was to do some reaction content on different videos that I think

are pretty interesting. It's going to be like cloud code and yeah basically a lot of different videos I think it's pretty interesting around AI AI tooling cloud code and we might try some new features on yeah I have something lined up. I just wanted to skip that today because I didn't really have like a plan today but I kind of planned that out for tomorrow. So, if you're online, look out for that. It's just going to

be a bunch of videos I have collected I wanted to watch and I thought I might as well watch them on stream and we can do some reactions and maybe talk a bit about the content. Looks pretty interesting. I wanted to try that. I haven't really done that on stream before, but I think it could be pretty interesting. Maybe we can answer some questions in the chat and stuff if people have some comments. Yeah, I don't

know what's happening here. I think I'm just going to cancel this. Yeah, I don't know what's happening. So, like I said, today we had a look at the web uh LM Marina. We tried to create some different UIs to see if we can get if we could get the new the lobster model from Chip 5. Uh, we got it one time, but it failed. Uh, we did not try the LM arena one to try to try

to get this. My hair color, I don't know. It's probably just brown, but my beard. Yeah, we had a look at uh V3 and tried this technique where we did this uh drawing. That was pretty interesting. I'm going to explore that more. But like I said, tomorrow uh I'm just going to do some watch some videos. We might test out some of it. Have maybe a new platform I wanted to try. Maybe we look a bit

more at Cloud Code. Look at some videos and maybe try out what they talk about in the videos. So yeah, a lot of people stopped by today. It's Saturday, so I didn't kind of expect a lot of people to be here. So anyway, thank you for stopping by. Thank you Sunni. Enjoyed it. Thanks, Sach, Larry, Benjamin, and everyone for dropping by. Appreciate it. Yeah, have a nice weekend or Saturday and I'll be probably back tomorrow. I

don't have any big plans this weekend. So, like I said, going to do some videos. Bye, Gustavo. St. Enjoy your Saturday and yeah, see you again soon. Enjoy your weekend.