TLDW logo

Claude Code 2.0: Coding Has NEVER Been This Easy

By Riley Brown

Summary

## Key takeaways - **Claude 4.5 Sonnet Setup**: Install Claude Code extension in Cursor or VS Code, search for it, log in with Anthropic account for more credits, and select the 4.5 model which is almost as good as Opus but five times cheaper. [00:34], [00:46] - **Browser Face Filter App**: Prompted Claude to build a client-side webcam face effects app using TensorFlow.js face mesh, added Snapchat-style filters like sunglasses, crown, clown makeup, fisheye, and cyborg in 2-3 prompts entirely in browser. [02:40], [16:24] - **Context Anxiety Phenomenon**: Claude 4.5 shows 'context anxiety' near limit, rushing tasks roughly or doing one change at a time to avoid overflow, unlike confident multi-file edits with fresh context; check with /context command. [05:41], [07:04] - **Parallel Tool Use Speeds Up**: Claude 4.5 delegates sub-agents to read multiple files or call tools simultaneously like a manager assigning four tasks at once, making it up to four times faster especially early in context. [13:24], [13:42] - **Lex Friedman Donut Visualizer**: From Lex's 5-year-old video transcript, Claude one-shotted a spinning 3D donut landing page in JavaScript, then expanded to 10 rotating objects in 2x5 grid including Homer Simpson donut. [18:57], [20:57] - **Sandbox Beats Local Ports**: Use sandbox.dev to avoid port 3000 hassles and run Claude Code in web without local setup, as ports are annoying like sea ports you don't need to manage. [08:00], [08:42]

Topics Covered

  • Claude 4.5 Sonnet beats Opus value
  • Vibe coding learns packages effortlessly
  • Models suffer context anxiety
  • Parallel agents mimic manager delegation

Full Transcript

Anthropic just released a brand new model, Claude 4.5 Sonnet. Many people

are saying it was the best in the world and now people are saying that Codeex is better. Again, I don't really know, but

better. Again, I don't really know, but I do know they created this cool new interface which allows you to use Claude Code in Cursor or VS Code. I just think it's pretty insane that like Claude has

released this little interface that's like almost as good as Cursor's interface. In this video, that's what

interface. In this video, that's what we're going to do today. We're going to test out this new model. We're going to build an insane app using the brand new Claude Code 2.0 and we're going to test

out how well it does.

We have Cursor open. It would also work on VS Code. And you open up the extension. Why don't you show people

extension. Why don't you show people where they can get this extension?

Yeah. So, in Cursor or VS Code, you'll have this extensions tab up here. Go

into that and you can just search up Claude Claude Code and it should be like the first result here. And then there'll be install. I can uninstall it. and then

be install. I can uninstall it. and then

install it. And then you'll see this little cloud icon here and you click that and it'll open.

You just hit /lo if you're not logged in. And then you hit enter and this will

in. And then you hit enter and this will allow you to uh enter your thing and you press open. By the way, this is

press open. By the way, this is Muhammad. I forgot to introduce

Muhammad. I forgot to introduce Muhammad. This right here is Muhammad.

Muhammad. This right here is Muhammad.

We got these new pointy fingers. They're

kind of fun. Anyway,

so yeah, you be logged in and you can use like an API key or your just cl anthropic account.

Yeah, they like give you more credits than uh you would pay for if you did use your API key. So you should use CloudMax account. you have this interface and

account. you have this interface and you're like, I I've never used an IDE before. What's cursor? What the hell is

before. What's cursor? What the hell is going on? It's okay. Once you have this

going on? It's okay. Once you have this open like this, and you could have it open in VS Code, which is literally free to download. Um, we can create an app.

to download. Um, we can create an app.

And you were talking to me this morning about an idea for an app that you had and I was like, that sounds lit. You

want to talk about that idea?

Yeah. Yeah. So, if you've ever been on Snapchat, Tik Tok, Instagram, they have filters. And I remember when the

filters. And I remember when the Snapchat filter came out in like 2016, 17.

And you're talking about video filters, right?

Yeah.

I remember when they first came out, they were insane. Like you remember the one where you like open your mouth and like rainbows come out of your game. Now every place has that.

your game. Now every place has that.

Game changing, dude. It was game changing.

dude. It was game changing.

Sound like an AI influencer.

That's part of the reason why the snap like IPOed anyways. Um because it was like the filters and people could show.

No, no, filters are are awesome.

They're they're Yeah. Yeah. Yeah.

And so we can use something we can make that like we can make that in the browser pretty easily. Um we can use uh TensorFlow. We use a JavaScript version.

TensorFlow. We use a JavaScript version.

Uh have a two words I don't really understand.

That's okay.

You don't have to because I'm not even going to look at the code.

Great.

Um but we can do something similar and we could make Snapchat.

I like it.

On the web.

Let's do what you just said.

Yeah. Yeah. So I have a prompt that I could use or I could just Let's do that. Let's do it.

So create a browserbased webcam face effects application that runs entirely client side. So there'll be like no

client side. So there'll be like no server, no backend, just on the computer and then it'll run on port 3000 which is like just so you can run it locally like you don't have to like publish it to the website.

Okay.

Okay. You getting this so far?

Yeah. Yeah. It's great.

This uses real face detection but keeps effects achievable. And so like the core

effects achievable. And so like the core requirements I want the agent to do is access the user's web camera, realtime face detection with TensorFlow which is what I talked about. And then they also have something called uh face mesh which

will does a lot of the work for you.

This is like a whole project on itself.

You know what? Let's enter this prompt in and then we'll analyze it.

Okay. Are you getting bored?

I'm not getting bored. It's just you're using a lot of words I don't understand.

Let's just rip it, dude. Let's just rip it to prod.

Ship it to Project.

All right. So, Claude, here we go. We

can do slashmodel to check out which model we're using. And so, we're using the new 4.5 smartest model.

And and 4.5, their newest model's almost as good as 4.1 Opus except Opus is five times as expensive. So, definitely use 4.5 Sonnet right now.

It's not worth it. And then we have a couple modes we can choose here. So we

can do edit automatically, plan mode or ask before edits. Usually when I'm using cloud, I go to plan mode and it'll give you your plan. You accept it. In this

case, we're just going to accept whatever it does and trust it and then we can ship it.

Yeah. And so scroll up there. So

TensorFlow is a type of technology that if you spend enough time vibe coding, you'll learn these different technologies like for um you'll learn the different APIs like 11 Labs is fun

if you want audio. Gemini is fun if you want to uh like Nano Banana is fun if you want to edit images. I know like some other libraries like one that I react ski that means but that's good for

canvas based apps.

Um and there's all these little vocab things that you'll learn even if you've never written code because these agents search the internet out of the box and so a lot of times it'll just use these

package. I'm sure that it might have

package. I'm sure that it might have used this without even you telling it to like if you asked for an app and it like and that's where you kind of like learn about these different packages and then you could just use them by saying them later. At least that's how I look at

later. At least that's how I look at packages and then you sound cool. You could tell you know an or hey I use TensorFlow and they're like what they you know how to use TensorFlow and it's just like yeah bro I put in a cloud code and it just did everything for me

and so wait let's look at the UI real quick. So scroll down while it's

quick. So scroll down while it's working. They made it a lot easier for

working. They made it a lot easier for for you to to see. Now it says fliver diverting.

Um, and it's just way better than kind of the terminal view.

I think some people might like it. Me? I

mean, do you like this?

I I like I like it. I actually do get you I did get used to the terminal. I

did kind of like the terminal, but like I like how it's out the way the terminal. It's like at the bottom.

terminal. It's like at the bottom.

Yeah.

This is cool though if you just want to, you know, chat with it.

Yeah, that's true. So, it is kind of like bigger, but like it doesn't need to be big. You could still put this at the

be big. You could still put this at the bottom like you I think you could probably put it at the bottom.

I know you can put it to the side like you can with the cursor. So, if you're used to cursor, this is pretty much the same. Yeah.

same. Yeah.

So, I think I can move it here or or you could have it like as a file like that.

That's cool. And this new 4.5 model has it knows how much context it has left in its conversation.

Let's talk about that for a second. So,

you know, when you use chat GPT or Claude, it'll give you a warning as the chat gets longer. It says like, "Yo, you your conversation's too long. Start a

new chat." You know, on the first chat, it doesn't have any it doesn't have much context. It has like the system prompt

context. It has like the system prompt or whatever it gives it whatever it's preloaded into it. Obviously, it's it's like when you download your computer or when you get a new computer, it's like, "Oh, you have 80 gigabytes of storage, but really there's like some system

stuff in there by default." And that's like half of the context, right?

It's a similar thing. But what's cool about this model is not only does it like automatically remove stuff from the context as the conversation goes along,

it's like consciously aware of how much context it uses, which is unlike any model I think that's ever been created, right?

Yeah. Yeah.

Yeah. And so it kind of removes a lot of the need, not maybe not fully, like you should still clear the chat, but these I think it's the trend is the most interesting part. Like these models are

interesting part. Like these models are moving in a towards a world where no one's going to have to like clear the chat.

it'll just kind of know what the purpose it has and if it's if it'll just get rid of it.

Yeah. And that comes with like some effects cuz then it'll have like uh cognition like ran their own evaluations on the model like the the lab that made Devon and they found that 4.5 has like

context anxiety. So once it gets like

context anxiety. So once it gets like close to the limit, it'll like stop doing tasks is good and just like finish it really roughly just so it can like Oh, it'll stop doing Yeah. It'll like do it really quickly

Yeah. It'll like do it really quickly and like maybe not the best way just so it doesn't go over the context.

So an AI model gets anxious as the context fills up. I want to make a Tik Tok on that. Like this AI it's anxious.

You could see it. I mean like in the beginning it has all this confidence and it'll like edit multiple files at once in the beginning when it has like fresh context. And then if we use this a lot,

context. And then if we use this a lot, which we will, you'll see it like do one at a time, do small changes just so it doesn't go over that limit in like yeah, you should just be clearing the chat as much as you can, especially if you're moving on to a task where like

the conversation's no longer relevant.

Like if you're just doing a sequence of smaller chats, you might as well clear.

But yeah, go ahead and run that. So if

you can hit /context, you can see talk about this. What does this mean?

about this. What does this mean?

Yeah, so in this one run, Claude used about 35% of the context it has before it like clears it conversation. And then

you can see like what percentage of that like where the context went.

Yeah. Those are the pre the you know the the stuff downloaded by default. Yeah.

So yeah only like really 11,000 tokens.

So you have free space 5%.

Yeah. Yeah. Damn.

So I have 42% left.

So and then okay let's what did we try and create? Is it done?

and create? Is it done?

Yeah it's done. Face play sounds like a weird name.

Okay. Hello world is running. I don't

think I don't think this is something else that's running. I don't know why that's Oh, maybe you have something else running on port 3000.

I probably do. And I'm just going to ask Claude, please kill whatever is running on port 3000 and run this in the background.

And you know, this is one of the reasons why I like using sandbox to use this because you can use cloud code and codecs in the web. It's, you know, you don't have to deal with ports. Like what

even is a port? That's kind of dumb.

Yeah.

Why not just have someone else handle that in the web?

Yeah. When it works, it's great.

Yeah. ports.

Well, think about it like a port at sea.

You know, you have like like in San Francisco, you have port, you know, one to nine.

Know what a port is.

Um, but yeah, that's basically it.

Do you just let it celebrate?

You just never let your model celebrate.

It's crafting.

Okay, now it's crafting. Now you can talk.

Um, I do notice that it lost its color.

There's no color here in the terminal.

It's like orange and like Yeah. Yeah. I mean it did just come out

Yeah. Yeah. I mean it did just come out so true. I heard if you have used the cloud

true. I heard if you have used the cloud code SDK I heard they renamed it if you want to talk about that.

Yeah. Yeah. Yeah. So for those of you who've been following my channel I talk a lot about APIs which are ways to use this cool technology AI technology and other tech in your app. Like if you want

to create a chat GBT wrapper you can use the chat GPT open API and hook it up into your app. The same way you can do that you can also use the clawed code SDK. So the same technology you're

SDK. So the same technology you're seeing right here. you can add it to your app. You can add cloud code to any

your app. You can add cloud code to any app that you create. One thing that Enthropic did is they changed it from claude code SDK to claude agent SDK. And

the reason is that they actually found that a lot of the users, people who use this API were using it for non-coding related tasks. Ooh, something's loading.

related tasks. Ooh, something's loading.

Oh, it's asking for camera access. Dude,

get rid of this purple gradient. Every

app that you make is this nasty purple color. It's It's heinous.

color. It's It's heinous.

Well, that's Collad. But if we go to GPT, they don't do this. This is This is the difference between the mod.

I swear. Okay. So, here we go.

Okay.

We have this app. I don't see your camera. I don't know.

camera. I don't know.

Yes, I don't see. And I can't click on anything. This is fine. I'm sure I can

anything. This is fine. I'm sure I can fix in the next prompt.

Okay. Oh. Oh, there it is.

Okay. What is going on?

Maybe refresh.

Yeah, because you had to give access to your camera. Now it has access.

camera. Now it has access.

Okay. It can't detect my face.

All right. So, it has the camera into the hair here. Do you want to try your face?

here. Do you want to try your face?

Yeah, we'll try my face. Let's try my face.

No, it can't detect. Okay, we're going to have to throw this back into claw because And when you vibe code, this is this is normal right?

Yeah. Like, look at this flashing.

Also, this is normal if you're a developer, too. Like, you're going to

developer, too. Like, you're going to run into issues. You got to fix it.

Yeah. And the way you fix it is you go back to claw and you say, "Why is the network?"

network?" Just explain what's the problem.

Yeah.

Do you have whisper flow downloaded?

That's what I do. Uh it doesn't work.

It's the face detection just doesn't work. Please help me.

work. Please help me.

It's also flashing. So like I show my face. It's in the camera view and it

face. It's in the camera view and it just says no face detected and it flashes over and over.

Yes. Let's see. Do we have any Let's see the the browser tools. You probably

won't want to like do this, but I get special privileges. Oh, I do this all

special privileges. Oh, I do this all the time. I go into console

the time. I go into console and I'll copy that. If you see an error and we can throw that in. We can go over some more slash commands which are pretty cool.

Yeah, you can do new conversation or clear conversation if you just want like a brand new one. You can mention files.

You can attach files.

Yeah. And you can do MCP. I mean, do you use MCP? MCP on paper is like incredibly

MCP? MCP on paper is like incredibly useful. In practice, no one uses them

useful. In practice, no one uses them yet. And that's a meme, right? Some

yet. And that's a meme, right? Some

people actually use it. Some people use it just because they like to talk about how they use it. I think it's incredibly interesting. It's like a shared API

interesting. It's like a shared API language that they've created like a standard that you can hook up to your app, but like how do you even set up an MCP server, dude? Like I'm excited for

it to get easier to create an MCP.

I think it will. I think in the future like MCPs will be gamechanging. Right

now it's like you have to be technical to set them up kind of. Uh anyway, it's done. Let's go

kind of. Uh anyway, it's done. Let's go

take a look ski at the app. Ski.

Okay, I'll allow loading face detection.

Well, taking pictures works. It's still

not working. It's still flashing. No

face detected. Please search the internet about how to use TensorFlowJS.

Find documentation.

Yeah, find the documentation.

Don't be lazy.

How to use this face mesh? Yeah, don't

be lazy. Ultra think.

Should you clear contacts before you enter this or no?

Well, let's see how much contacts I have left. This might get what we called uh

left. This might get what we called uh context anxiety. So, we should be

context anxiety. So, we should be careful here. You are right. Oh, I still

careful here. You are right. Oh, I still have a bunch of 40%. 60%. You're

chilling.

Cool.

All right. As you can see here, it's not an earthshattering model.

I gave it a pretty nuance everyday coding. It's definitely better.

coding. It's definitely better.

It is. I mean, it's better. I mean, but like is it gamechanging? It is faster.

And maybe that was It's way faster.

Yeah. And one of the reasons it's faster is it can uh use tools in parallel. Do

you want to explain that?

Yeah. Yeah. So, uh I kind of mentioned this before um like when it has fresh context, it'll edit like multiple files at once or Yes, please. or uh read multiple files

Yes, please. or uh read multiple files at once. And so there's basically you

at once. And so there's basically you could say like sub agents like they read the files and then they they bring them back up all at once instead of like going one by one.

And so that makes it like four times faster, right?

And it it usually does that at the beginning and then towards the end of its context limit.

Yeah. I mean, if you think about it, a manager doesn't delegate to one person, wait for them to finish, and then delegate to the next person. Delegate to

four people at once. You do this, you do this, you do this, you do this, come back, we'll talk. And that's what the agents do. And so it's able to call

agents do. And so it's able to call different tools and they just accomplish things at the same time. You know, it can search the internet um and like come up with a game plan and make edits to three different Can it like it can write

code files at the same time or does it I don't know. But I think for write it can I for a read? I know definitely. I

think I've seen it write multiple files.

So read it can read the codebase to like investigate what it needs to do all at the same time.

Yeah. which I'm sure is a lot of the speed because it's if it spends more time reading than writing. Yeah. Or at

least it reads more than it writes. It

takes longer to write than to read.

Yeah. I mean, if it reads more than it writes, that's a good sign.

Yeah. It doesn't feel fast right now.

I'm not going to lie.

Cuz it's searching the web. That's like

I think it'll do that one by one.

True.

I feel like that would be useful to do it at least in like sets of three. I

think the agents are like they're like gravitating towards like what nature is already like you know if you think of an ant colony there's the queen on top and then you go down there's like the big ants and then small ants and then like

worker ants that like just die and so like I think I think agents are like they're splitting off more and more in this branch where like there's a main agent and it splits off work into like different agents and then they split off work. So, we can kind of see that here

work. So, we can kind of see that here like there's a main agent that tells like four others to like, hey, read these files. And um if you remember our

these files. And um if you remember our N video, we sort of had something similar with a CMO agent. Yeah.

Where there's one main one that like delegated tasks and then they could delegate tasks.

All right.

Looks like it's done. So, we can go back. Oh, I saw it for a second. I think

back. Oh, I saw it for a second. I think

it's working.

It looks better, too.

Okay. There's no flashing. Still like

the purple stuff, but whatever. And then

I can throw on some sunglasses. How do I look? I look like uh that one guy from

look? I look like uh that one guy from the Matrix. Um,

the Matrix. Um, the bald fat one.

Yeah, one. No, he's not fat. Is he fat?

one. No, he's not fat. Is he fat?

He's fat.

It's uh what's the guy? Morpheus.

Yeah.

Yeah, he's fat. Um, we can get a crown or a

he's fat. Um, we can get a crown or a mustache. I can't stay, you know, I

mustache. I can't stay, you know, I can't live next to schools with this mustache.

And then I have this glow, which is sick cuz it like goes around like my entire face and it does this. This isn't like on some server. It's on your computer.

some server. It's on your computer.

Yeah. Yeah. It's running locally and it's like getting my face like I'll put it on you in a second. That's a big clown nose. And then you can see like

clown nose. And then you can see like the actual mask around my face.

And then you could create a new one.

Like could we ask it to create a new one?

Yeah. Yeah, we that's what we should try next.

It's me.

Oh, that looks good on you. What up,

though? That looks good on you. What?

Give you a new eye color or a crown. King Riley.

All right, that's sick.

That's sick. So, yeah, we can like add more stuff.

I want to do one with like the masks.

That something that seems cool. Is there

a way to like do an actual mask of some kind?

Yeah. Yeah. Like uh

just see what it comes up with. Say like

search the internet.

Full clown makeup.

Full clown makeup. And then and then so and then say search the internet and find another cool effect that we could do and add that and then like So we are adding three more. Yeah, if

you find more like just add them.

I want to see what you can add here.

That's basically what I want to do. I

want to make this such a fun app to use to like add different filters that I can use and yeah, find some documentation, do some cool stuff, make it happen.

And then I'm going to do like think hard.

Um, now it's celebrating.

Okay, so added the beard clown and then it added fisheye alien sparkles cyborg.

All right, so this was before. This one

was there before.

Big beard. Let's see how I look with the beard.

All right. That is the lamest attempt at making a big beard, but that's okay.

Okay.

I look pretty weird.

A clown. Oh, okay. That one's kind of fire. I actually like it. All right. All

fire. I actually like it. All right. All

right.

Okay. Sketch.

Okay.

I can kind of see what it's I don't know. Do you want to do the last four?

know. Do you want to do the last four?

Sure. I'll do I'll do the last four. I

like it. Okay. So, we will do fisheye.

Okay. It's like a fish island. That's

pretty cool.

Yeah, I look so dumb. Alien.

That one's actually kind of fire.

Cool.

Sparkles.

All right. Cyborg.

Whoa. Ar.

That one's Wait, why do I do a pirate noise?

Eye patch.

Oh, it's an eye patch. Yeah,

it looks like I'm You're about to snipe me.

Uh, my favorite one is fisheye.

There is a slight delay on it, though.

Yeah, there's a FPS at the top of the screen.

Clown's pretty crazy. Yeah. Anyway,

that's kind of sick.

So, yeah, just like the Snapchat ones, which I'm sure took a whole team to develop. Yeah. Do this in two, three

develop. Yeah. Do this in two, three prompts.

That's pretty sick. Well, let's do one more special effect one. Before this

episode, I was inspired by this Lex Freriedman donut. So, there's a famous

Freriedman donut. So, there's a famous video that he's made. This was five years ago.

The left is a donut and on the right.

Since I use perplexity, I can hit command or command or option A. I can

say, okay, please write out the full transcript of this video. I think this will work. I'm not sure, but ideally,

will work. I'm not sure, but ideally, okay, here's the full video transcript.

And so now what I'm going to do is I'm going to go ahead and open up cursor, and we have this extension already open.

I want you to create a landing page that has the effect that Lex is talking about in this episode in full. Think hard. I'm

going to paste in that transcript. Go.

And you know, another MCP tool that I saw someone use previously is a YouTube MCP that like if you there's a link pasted, it automatically has the ability to pull the entire transcript from the

video and it can actually search for YouTube videos which means it can search then it can um pull the transcript. So

those are one example of it.

You add MCPS to Perplexity.

No, they have tools that you can use but then you can also Yeah, I don't think so. I don't know. I only use the basic

so. I don't know. I only use the basic features. But I think the agentic

features. But I think the agentic browsing stuff is overrated.

Yeah, I I could see myself going back to Chrome just because they're adding AI features and it has Gemini built into it.

Are you using like DIA or the other one?

Yeah, Comet. I was using Comet there.

What's the difference between those?

Perplexity just feels better. Dia just

never felt good. It felt good at the very beginning. I made a video on it. I

very beginning. I made a video on it. I

actually really liked some of it. Comet

just felt better. It felt more powerful.

The AI models are better. It used Yeah, Dia was worse. But anyway, back to Oh, done. Is it done already? Okay. Run

Oh, done. Is it done already? Okay. Run

it locally on port 3000. Transmuting.

I think in the original video he coded it in like C. Whoa, dude.

It just that is just oneshotted.

Okay. Please make it bigger and have three of them in different colors. It

should be 2x as big.

Go back to the browser. Can you change the code on the right?

No. Oh, that would be cool if you could cuz that is like what is making it a donut. So you can like play around with

donut. So you can like play around with that.

That's so fascinating. I wonder

that's in C. That is cool.

Wait, this is C.

Yeah, but it might have just I don't know how it's running.

It's HTML. Scroll down. Go up.

Okay. I don't know what is going on over there. No, I don't even know how it's

there. No, I don't even know how it's doing.

What if it's just like finding something online and rendering it like Oh. Oh, I see. There's a calculate

Oh. Oh, I see. There's a calculate render donut. Oh, function render.

render donut. Oh, function render.

JavaScript.

Okay, I believe in the video he did it.

I do like these views. This looks good.

Like this interface. It's kind of nice actually. Like this is just looks nice.

actually. Like this is just looks nice.

It's got this nice like little It's very simple.

It's definitely like a competitor to cloud now or not cloud cursor.

There it is.

Dude, that's crazy. So, we now have these three donuts spinning around. All

right. Using the same logic, I want you to keep only one of these donuts and then pick nine other objects to have.

So, I want a grid 2x5 10 total objects spinning around. One of them is a donut.

spinning around. One of them is a donut.

Nine other different objects. And then

at the top, I want you to put Riley Brown is the This will be my personal landing page.

What if you had it where you could like upload pictures and then it starts making that spin?

Yeah, you could. I mean, you need to get a pretty good AI model and you need to write good system instructions, right, for the AI to write it.

Oh, yeah. I guess so. Yeah.

Yeah. I guess cloud could do that. You

give the image to Yeah, you can use the cloud code SDK to do that. Cloud agent SDK, but maybe you

do that. Cloud agent SDK, but maybe you wouldn't need it.

You probably wouldn't need it. So, what

do you do you prefer 4.5 or codeex?

Yeah.

I don't know. I I did some testing this morning and I thought the difference between them was so not interesting. I

think you can't go wrong with either one, you know, like I I think right now I think it's good to get used to this new interface. I think this is genuinely

new interface. I think this is genuinely interesting to people, but I think finding the edge between things that are like really close together is an overrated thing to do. I'd much rather just have these exploratory sessions where I'm like, what can you build with

4.5? And you're like, oh, now I'm

4.5? And you're like, oh, now I'm learning something. I can use TensorFlow

learning something. I can use TensorFlow rather than like all right we're going to use these standardized prompts and we're going to calculate these benchmarks and and it's like you should just tell if one is clearly better just

focus on that one but right now I don't think either clearly better a lot of people are saying for backend harder tasks codeex is clearly better um for quick UI changes I heard 4.5 is

better just because it's fast and it just gets it done right away this is neat so this is a donut we can tell this is a donut this looks like some sort of spiral this.

Let's start. Um, that's a chair. You see

the chair?

Are you sure?

No. Maybe a rug.

All right. Right. Please make it 2x5, not 5x two like you have it. Also,

label the items in small text cuz I cannot tell what these are. I just put this in the center here. What is this? A

pretzel?

Um, a toaster.

Yeah, like Yeah, let's actually try and guess what this is. You think this is a toaster? toaster or a car.

toaster? toaster or a car.

This might be a chair, but I'm not sold that it's a chair. I mean, where are you getting that from?

I see the leg. You see the legs on top?

Legs on top. Now they're rotating.

All right, that's a theory. Yeah, I I give it a 30%.

What is that?

This looks like a pretzel to me.

Whoa whoa whoa.

That's a cone. That's a cone.

This is a cone. It

has to be. Or a clown hat.

Oh, what the hell is going on? I think

it's just an error. All right. So

clearly what it's doing here, the reason it was successful in creating the donut, it clearly just found the code online and made and just did it right. I mean

there's no way that this is just like perfect.

Yeah. I mean some of it's good. It's

like a tube. It's like a cylinder where the red one. Not that one. The red

wing is like a perfect tube.

You'll see it open up in a second. Boom.

And like that one.

Okay. Okay. I see it. I see it. It is

cool. It is. This one's good. That one's

pissing me off.

Okay, refresh the chair.

Where is it?

Oh, it's a pyramid.

Which one was it?

It's this the the orange one. That's a

yellow.

This one?

That's red. Orange.

Oh cylinder.

That's a cylinder. Below. Go.

Yeah.

Oh, pyramid. How How is that a pyramid?

Okay these What does the teal one? Go down.

What does that even say? That the one on the right.

This is a Klein bottle. A Taurus knot.

Oh, okay. It's like a wetzil pretzels.

Okay, cool. Um, anyway, yeah, you can kind of just create whatever you want, you know, like you could say change it to just one massive donut, a Homer Simpson donut.

Homer Simpson style. And the point is is you can just have fun with code. You

don't need to have any specific thing in mind. Part of vibe coding is seeing

mind. Part of vibe coding is seeing what's possible. And here you can see

what's possible. And here you can see that you can use this asy. Is that what you call it? Asky. like basically using letters to create kind of images or visuals or or v or or these like

graphics here. And yeah, you can kind of

graphics here. And yeah, you can kind of just do whatever.

You know, a good challenge would be to make it run in the terminal like a little donut in the terminal. That's

kind of technical.

Yeah, we'll we'll do that maybe when when everyone in the audience is is us, you know, once we're on vibe coding 102.

But right now, we're we're still in the basics. All right, let's see. So, we can

basics. All right, let's see. So, we can refresh this and boom, Homer Simpson. That's actually kind of

Homer Simpson. That's actually kind of dope as hell.

Maybe I shouldn't have had them all generate at once. I mean, look at the level of detail.

Thing's a unit, dude. If I was Homer Simpson, I'd want

dude. If I was Homer Simpson, I'd want to eat that.

Yeah, I'm hungry. Anyway, so this is the new interface for Claude Code that you can download and it's very easy. You can

use it for free. It'll be You have to pay for Claude, but you don't have to use cursor. You can use VS Code. It's

use cursor. You can use VS Code. It's

the same exact process. Literally the

same. And Claude is getting better. And

yeah, I just thought we'd do a little exploratory session, see what we can build. We built some two visually fun

build. We built some two visually fun things in the next video with Muhammad.

Uh, we're going to be doing some mobile app stuff. We're going to be diving into

app stuff. We're going to be diving into the weeds. We're going to get apps on

the weeds. We're going to get apps on the app store. We're going to get back to that content. If you like this style of video, let us know in the comments.

If you don't like it, also let us know in the comments so I can block you. Just

kidding. Um, yeah. Anyway, this was fun.

Yep.

Any final words?

Yeah. Thank you guys for watching. Um,

you know, what do I what do I say? I I mean, this has been a pleasure. Thank Thank you, Riley Brown, for having me down up here.

All right, thank you guys for watching.

We'll see you here in the next video.

Loading...

Loading video analysis...