Today we are going to be building and

working on and getting our first

customers for an AI startup we've been

working on for a while in Claude Code.

We'll also be covering Claude Co-work

the brand new 247 AI agent from Claude

as well. We'll be going through use

cases on that. We got a lot on deck.

You're going to learn a ton about Claude

Code. You're going to learn a ton about

Claude Co-work, the brand new tool from

Claude. You're going to laugh. You're

going to have fun. You're going to build

things. Welcome to the live stream. Get

in the chat. Get in the reply. Say hello

to your fellow builders. This is the

place for builders to hang out, learn

things, meet new people, learn new

skills. Let's go into it. Today we will

be working on Viber, which is our app

store for vibecoded apps. This is live

right now. I will share the link out

with everyone here so they can submit

their apps, get them live on the app

store and get their first customers to

their app. We're going to be working on

improving this. We also have a couple

couple other apps we can work on and

we'll also go into Claude Co-work, the

brand new co-working tool from Claude

Code. Basically, your 247 AI employee.

I've been playing around with that for

the last 24 or so hours and that is

absolutely amazing. How is everyone

doing? Good middle of the day from rainy

Alabama. What's up modelier? Good to see

you. Coowork is sick. I checked my

entire file system on both my Macs in 25

minutes. You want to see what I did with

co-work this morning? I'm working on a

banger video for co-work by the way.

Banger video. Banger video coming up.

So, make sure to subscribe and turn on

notifications for that. That's probably

coming out first thing tomorrow morning.

Uh here's what we did.

We uh first had it organized my entire

computer. So, I had Claude Co-work go in

and uh find all the images on my

computer and put it into a new images

folder and then organize that images

folder. So, if you if you see this here,

look at this. It went in, it found all

the images on my computer and then

created this new images folder and

organized all the images on my computer

into these subfolders. So thumbnails,

screenshots, AI generated photos, GIFs,

logos, downloads from the web,

duplicates, whole bunch that it did this

in like literally three minutes.

It did it in like literally three. I

said, "Hey, I have a bunch of images on

my computer. Can you please organize

them for me?" And it did all this. How

incredible is that? This co-work man,

you guys need to be using co-work. You

need to be using co-work. This is

amazing. You want to see what You want

to see what else I did? It's done some

amazing things. And I'll go through in

the video I put out either later today

or tomorrow morning exactly how I did

all these things and share the prompts

in that. I created a media kit for my

newsletter. So I said, "Hey, here's my

newsletter, alexfin.ai. Can you create

an entire media kit for me that I can

send to potential sponsors?" And it in

like three minutes. It's so fast.

You want to see what it built in like

three minutes?

It built this entire

media kit for me. So, it went on to my

website. It researched my newsletter

and it built me this entire media kit I

can now send to sponsors. Ship it

weekly. It has a description. I didn't

tell it any of this. It just went in,

researched my newsletter. Everything I

talk about, it got all the numbers about

my newsletter. I don't know how it got

all this. It got in, figured out who's

in my audience. It created sponsorship

options for me for my newsletter.

I mean, this it built this entire media

kit presentation in like four minutes.

And I didn't give it. I just said, "Hey,

my website, my newsletter is alexfin.ai.

Research and make a media kit for me."

And it did all this.

How amazing is that?

How amazing is that?

I have a whole bunch of other use cases

I did with this. Whole bunch of other

use cases I did with this. That's going

to be in the video I put out probably

first thing tomorrow. So, subscribe,

turn notifications. Also, leave a like.

Just leave a like right now. Everyone,

just leave a like. Let's get more people

into this party. Everyone, click the

like button down below. How amazing is

that? Has anyone else before we get into

Claude Code and building out the app and

building out uh the startup and getting

customers? Real quick question for you

guys.

Has anyone been using Co-work? What have

you been using it for? If you've been

using it, let me know down below in the

replies if you're watching the replay or

uh in the chat if you're watching it

live. Let's see. I had it rename all of

my top level and second level files

using kebab case. Incredible.

Also tested a prototype for a great

markdown to PDF converter. You had

co-work up build all that. Co-work is

sick. I cleaned my entire file system on

both my Macs in 25 minutes. I'm I'm like

actually blown away and I'm seeing some

people on Twitter saying it's like

bugged and like it's not working for

them. I didn't experience that. Every

use case I threw at it, it did perfectly

fine and like way faster than I expected

to do. So it is it is amazing.

It is amazing. What's up Widget Flow? Uh

should we work on I think today we got

two options of what we can work on today

with Claude Code. We can either keep

working on Viber, our vibecoded app

store which is now live and we can work

on perfecting it and getting customers

and all that or we can do something a

little earlier stage which is we can

keep working on the 3D AI agent builder.

So like kind of a 3D AI agent manager

where you see all your agents walking

around the 3D space. We can do that as

well. So let's do this.

Let's put it up to vote. Let's put it up

to vote. Um,

let's create a poll. Start a poll.

Uh,

Vibecoded

App Store Viber or 3D AI Agent Manager.

What are we building today? Start poll.

Here we go. Boom. Vote in the poll.

polls now live. Uh, a lot of things we

can work on. A lot of things we've been

working on in last sessions or I guess

we could also uh

play around with uh co-work. We could

also play around with co-work too. Find

new use interesting use cases for that.

Uh, I also had to rename all my top

level. Oh, we went through that. Hey,

Alex. Have told you this already, but

you introduced me to vibe coding six

months ago. I launched my first SAS

yesterday. I'm already at $1,100 MR.

>> [applause]

>> That's what I'm talking about. That's

the best message I can get. That's the

best message I can get is when people

say, "Hey, you introduced me to Vibe

Coding and now I'm making money." That's

Listen, if you're watching this channel,

you're watching these live streams,

you're watching these videos, and you're

not taking action, you're a silly goose.

You're a silly goose. If you're watching

all this and you're not building

something, you don't open up Claude Code

while you watch my videos. You don't

open up Claude Code while we live

stream. and you don't build something,

you're missing out $1,100 a month.

That's an extra $12,000 a year. What

would you do with an extra 12? I mean,

that's $13,000 a year. What would you do

with an extra $13,000 a year? What would

you do if you had an extra $13,000 a

year?

Co-workers, make it even higher on your

to-do list. It's amazing. Do you

recommend us building our own website?

If so, with Claude code or how to build

and hosting? Uh, at the very least you

should make your own website. Of course.

At the absolute very least you should be

making your own website. Um, just for

the fun of it. Just as a fun side

project. Just as a fun side project. You

absolutely should.

What up, Finn? What's up, Bo Sterling?

How you doing, Bo? It builds prototypes

that you can open directly in

anti-gravity. Yeah, it writes code. It

executes code.

It is uh it's pretty amazing. I mean,

it's it's the best of Claude code and

the best of Claude mushed together,

right? It can write and execute code

just like claude code, but it can also

build presentations, go on your

computer, whole bunch of other things.

Viber. Okay, we got Viber. What's the uh

what's the poll saying? The polls are

always so close. We got 52% Vibe code

app store.

How do you know when the app is ready to

go live? Every time I ask cloud code for

a code review, you find some critical

issues. Well, you definitely don't want

to ship with critical issues. It depends

what it means by critical issues. You

definitely want you don't want to uh

ship a critical issues. But once all the

critical issues are done, you just want

to ship it and get uh testers, beta

testers, right? Start getting feedback.

Get that feedback loop going. And you

definitely want to as you go create

content, tweet, create YouTube videos,

whatever you're doing as well, so you

can start attracting people to your

brand.

I like the 3D agent space because it

feels so cool. Yeah, it seems like I see

new tweets every day of people trying to

build uh 3D agents.

I think that might be a cool idea.

There are decent amount of bugs as of

this weekend in Viber. Didn't have time

right then. Yeah, we have some bugs.

We're going to have to do some bug

fixing. We go the Viber route

out treaties is open for beta testers is

great orchestration layer for clawed

code. Let uh ambitious realism crates

know if you want to test at treaties.

What are some must prompts for Claude

and plugins?

What are some must prompts? I don't

know. That's such a broad question.

That's like saying, "What are some must-

eat foods in the world?" It's like, I

don't know. What are you into? What type

of foods you into? What do you think

about like what what meals are we

talking about here? Right? You can't

just say, "Give me a prompt for Claude."

You got to give me some more information

here. Jet, don't move. looking to hit

10,000 MR and it's on the pace to do

that too. Everest Chris, what's what did

you vibe code where you're you got,00

MRR and you're on pace for 10k MR. Give

us the link. Let's check it out. Let's

check it out live on stream. What are

you shipping, bro? What are you

building?

What are you shipping? Also, side note,

Patriots division round coming up

Sunday. Winners destroyed the Chargers.

Destroyed the Chargers. I highly

recommend having codeex 5.2 high code

review and analyze the project. They

both go way deeper than Slopus does.

Now, I've been seeing this on X people

calling it Slopus

and like and crapping on Claude

code with Opus 45.

Like, what are we talking about here?

Every test I've done

with Claude Code and Codeex 5.2 like it

like the results were just obviously

better in Claude Code like just

obviously better. So now people are call

is this like a paid is this like a paid

campaign against Opus?

Let's do this.

Let's do this. Let's open up. Should we

test? Let's test. Let's uh let's see

Opus versus Claude Code right now.

First, let's go into Viber.

I'm going to I'm going to open up Viber

and I'm going to have Codeex review it.

So, let's do this first. Everyone keeps

saying to do this. Uh let's go into the

Vibe app store here. We'll go into Viber

and have it review.

We'll have Codeex review Viber. see how

many bugs it can find. Everyone says,

"Oh, it's great at finding bugs."

Codeex. All right, let's put it to the

test.

Here we go. We're in Viber. Let's open

up Codeex.

Codeex.

Here we go. Open this puppy up.

Open it up.

What are we doing here? Open. Open.

Open. Open. Open.

Oh, no. We definitely do not want to

show. We don't. Every time I open up

files on stream, it always goes directly

to the environment variables file. Why?

Why?

Close this.

Close all. Thank you. All right. Codeex,

just open up, please, for the love of

God, open up. Just open up. Open.

Codeex's not starting off strong here.

Not starting off strong for Codeex. It

won't open.

I guess uninstall it. Install it.

Open.

I mean

I mean what are we doing here? Disable.

Restart. Enable. Open. There we go. Here

we go. We're working now. We're working.

Codeex is working. All right.

Here we go. Go through this entire

application. review it and find any bugs

and let me know what you find first.

Here we go.

People are saying codeex is better for

finding bugs. We're going to find out

right now. We're going to find out right

now. I mean, let's do this. Should we

put it headtohead with Claude?

Hopefully, he listens to what I says and

doesn't change any uh

We'll give it the exact same prompt.

Let's see what both find. People calling

it slopus.

Let's see what we find here. All right,

close all this crap. Can we just close

this? All right, here we go.

All right, let's see what uh let's see

what they find. Uh people saying slopus.

I don't know. Should we And we could

also run another test. What's a good uh

let's do this. Let's run another test

while this goes. So, I'm going to leave

this. This is going to work on Viber.

Let's run another test of codeex versus

opus.

Let's build a codeex

3D agent manager. Let's create this.

I'm going to I'm going to have codeex

build a 3D agent manager.

I can't even get like codeex to like

work.

Am I like crazy?

I'm clicking the goddamn codeex button.

Claude code opens. When I click it, I'm

clicking codeex and it's not doing

anything.

Claude code opens. I can't I mean, if I

can't even use codeex, how am I supposed

to test it?

Alex, when you work with small tools and

build for yourself, do you just use them

on your local server or do you deploy to

a random URL? If I'm just building

something and testing something, I just

do it locally

and if it turns into something real,

then I'll deploy to the internet uh you

know, GitHub and then uh Versell.

I've been working on coordinating my

Obsidian notes and my ClickUp spaces.

Claude Code walked me through the whole

process. There you go. That's great. I

mean, and Claude co-worker probably do

that even better. What is going on? Do

other people have this issues where you

click codeex and literally nothing? I'm

going to have to disable it, enable it,

restart, enable codeex. There we go. I

have to do that every single time. All

right.

I want

to build a 3D agent manager using 3JS.

I should be able to have a 3D

representation

of five different AI models. Chat, GPT,

Claude Grock

Deep Seek, and what's another one? Uh,

Gemini.

Then I can click on the 3D model and

give it a prompt and watch

it work. All right,

I'm going to send

here. So that's going to codeex. I'm

going to open up another window

and I'm going to do the same exact thing

with Opus. Opus 3D agent manager. I'm

going to give it the exact same prompt.

Uh, and we are going to see which one's

better. Everyone's saying codeex is

better now. All right, let's put it to

the test.

I just gave the exact same prompt to

Claude. Here we go. We're going to find

out once and for all on this stream.

In fact,

that's a good that's a good title for

the stream. I should probably change the

stream too which is codeex versus claude

code which should be using that. See

sometimes we just

claude code verse codeex gbt 5.2 high

which should be using

live. Save new title. There we go. All

right. Let's see what what what

experiences are people having with uh

codeex. Are people on the Codeex train

now? Do you want to proceed here? Yes.

And don't ask again.

Uh, it looks like Codeex has already

gotten to work. So, Codeex has already

gone to work on their side. Didn't ask

for any permissions or anything, which

is good. I like that. It's just going

and building code. Claude codes going in

now. This is substantial project with

multiple design decisions to make. Let

me enter plan mode to explore your code.

Oh, that's pretty sick.

I didn't know that's new.

It overwrote what I asked to do. It's

like, "Nah, bro. We're doing this in

plan mode. We're doing this in plan

mode. This is too big. We're going to

plan mode, bro." I like that. I like

that it has that taste. I like that it

has like that kind of like opinionation.

That's new.

Which framework, React, 3JS?

How should the AI agents actually

process prompts? We're going to do mock

responses for now. What visual style do

you want for the 3D agent

representations? I want robot avatar

models. What should watching it work

look like when an agent's processing?

Typing streaming text. 3D visual

effects. 3D visual effects. Submit

answers. I like that it just was like,

"Oh, no, bro. We're going to plan mode.

We're going to plan mode. Screw you." I

like that.

I like that.

Drake May all day, baby. Drake May all

day, baby. Opus is hella sloppy now. I

feel like this happens once a month.

Like once a month people will just start

tweeting and going into my chat and

saying Opus sucks now. Opus sucks now.

Like what are we talking about? I don't

experience that at all.

Like that was awesome what it just did.

I mean we're about to find out once and

for all what's going on here. If it's

Codeex or Opus which is better. They're

both billing the app. We're going to see

what comes out better. I I mean

I'll say this. If the Codeex results

better, I'm going all CodeEx.

Anthropic's not paying me. I've been

saying this since day one. The day

Codeex is better is the day I just start

making all Codeex content. All my

content is Claude Code right now because

Claude Code's the best. But listen, if

Codex gets better, I I'm not loyal to

anyone. I'm a mercenary. I'm a mercenary

to whatever tools the best.

So listen, there's a lot. This is high

stakes. This is a high stakes test. My

last 40 videos have been all Claude code

videos. If Codeex is better here in this

test, I'm switching to Codeex. We're

going to find out right now. We got

Claude code working in one window. We

got Codeex working in the other.

Right now, Codeex is scaffolding the app

and implementing the agents.

Adding animations. Okay,

Codeex, for the record, did not do any

sort of plan mode. I just sent the

prompt and it's working. Did not stop.

Did not ask me any questions. It just

started building code. Claude Code was

like, "Hold up, bro. This is aggressive.

This is an aggressive ask. You're being

aggressive right now. Reel it in."

Claude Code was like, "Reel it in, dude.

Let's go to plan mode and hang out for a

second here." I think I kind of like

that.

As a newbie, should I get the Claude

Code $100 plan or start smaller? I want

to get the best results I can from the

start.

I mean, technically, you could start at

the $20 plan. You're gonna use up your

usage within one app probably, but then

you can just upgrade from there. So, it

can't hurt to start to the $20 plan. I'm

just gonna say this, like it's awesome.

So, you're going to want to upgrade to

the $100 or $200 plan. Codeex is way

slower for coding than Opus 45. Well,

we're about to find out. We're about to

find out both the speed and the quality.

I mean, let's check in with Codeex real

quick. Let's check in with Codeex.

Oh, did I? No, I named it claude code.

Here we go. We're already four out of

five steps done with codeex. I would not

say this is moving slow. I mean, at the

same time,

we don't know how high quality this code

is. This code might stink from codecs.

Uh, it's already doing npm run builds.

It's already basically done.

Opus on the other hand has not uh

written a single line of code yet. Opus

hasn't run a single line of code. I did

start Opus after Codeex. I did start it

probably what 45 seconds after. So it

does it Codeex did get a head start, but

right now it's just doing design work.

Right now it's just doing design work.

Right now it's just doing planning.

Claude code. Codeex immediately started

writing code. I'd say that's a W for

Codeex. Well, no. I don't know. I mean,

I do like that Claude Code planned with

me. I do like that was like, wait, hold

up, bro. I got some questions for you.

Codeex is just making a crap ton of

assumptions. But at the same time,

that's my fault. I gave it a broad

prompt. So, maybe maybe that is the

expected behavior is to make

assumptions, right? Maybe that is a good

thing.

So, I'm not going to hand a win either

way just yet. Codeex 5.2 extra high is

really good. Oh, I guess that is a good

point. I didn't put it on like

extra high.

What are we on? We're just on 52.

Uh, people are going to say it doesn't

count because I'm on 52.

People are gonna say it doesn't count.

People are All right, I'm gonna start a

new window. We're going to do one with

uh the extra high. We're going to do one

with the extra high. All right. So, this

is good. We're going to compare 5.2 to

5.2 Codeex extra high to Claude Code.

This is a good thing. Uh Codeex 52 extra

high 3D agent. All right, let's let's

take the exact same prompt.

I'm opening up codeex on another window.

I have to disable the plugin, restart

the extension, enable the plugin, hit it

again. I don't know why that's so

annoying. All right,

let's open up codeex 52 extra high.

I mean, that's a win for Claude code.

The fact that I uh

I could just opus, right? I got like 20

different options with chat GPT here.

All right, choose model five. What

happened to I just saw 5.2 Codeex Max in

here.

Okay. 5.2 Codeex

extra high. We'll do extra high. Is that

what the people want?

The people want extra high. All right.

Send. There you go.

There you go. Let's check in on Claude

Code. Let's check in on cloud code to

see how it's building this 3D agent

builder. Let's check out Claude Code.

All right, we still haven't written a

single line of code yet, but we have a

plan. 3D AI agent manager, a React 3JS

application featuring five interactive

AI robots. There's the text stack. Looks

good. The agents looks good. They chose

a color, a design for each one. Pulsing

glow. So, that's a pretty amazing plan.

Yes, and auto accept. There we go. Opus

is going to work. All right. So, we have

codeex 52 is thinking. It's just doing a

lot of thinking.

And it looks like

it looks like just regular 52 is

finished.

It looks like regular 52 is finished.

Uh, what are we doing here? Opus, does

it need more? It says, "Do you want me

to write all this without any sort of

permissions?" I say yes. And then it

immediately asks me for permission for

things. Whatever. Anyone get annoyed

when Claude plans things out in weeks?

Yeah, that's really annoying. I hate

when it comes up with a plan. It's like

phase one, week three to eight, phase

two, week 19 to 49, phase three, we will

start this when you're 183 years old.

It's like, no, you won't. You'll build

all this in the next five minutes. What

are you talking about?

You won't be upset with the $100. I

mean, you won't be upset with the $200

plan. It's the best $200 you'll spend in

the entire world.

I've never used Codeex in the IDE thing

they got. So, what do you use it in? You

use it in the CLI? Is that what you use

it in? What are people using Codeex in

nowadays? I just use the extension. I

feel like Codeex is an extension tool.

I feel like the I feel like the CLI for

Codeex was just added on because

um

it was trying to match Claude code. No.

Does anyone else feel that way? I feel

like the CLI for every tool except for

Claude Code was just added on because of

Claude Code. Like when's the last time

you heard about like cursor CLI? Does

anyone use cursor CLI? Like that was

announced like a month after Claude Code

came out just because of Claude Code and

they just ditched that.

Codeex 52 Extra High has gotten to work.

It didn't ask me a single question.

Didn't do any sort of planning. Codeex

52

uh extra high just started coding just

like codeex 52 regular one.

So that's something I mean that's a

difference in behavior. Claude code

didn't give me any option but to plan.

It was like no dude this is too

complicated. We have to plan this out.

I think I like that. I think I like

that. I think I like that it had like a

strong opinion.

I had co-work organize my projects

working on because I have so many going

at once. Pretty cool to organize my

active and yeah, it's really good at

organizing files on your computer.

Really good. What about building our own

website? Claude Code to build or another

builder or better focus on applications.

Cloud code to build everything. Unless

Codeex wins this competition today. If

Codex wins this competition, I'm

switching to Codeex for everything. I am

not biased. People say, "Oh, an you're a

shill. Anthropic's paying you." No, no,

no.

I'm a shill for whatever the best tool

is. And for over a year now, well, not

over a year, Claude Code came out in

February.

Since February, almost a year now,

Claude Code's been the best. So, I'm

just a shill for whatever is the best.

If this test we run today, which this is

probably the most important AI test in

the history of mankind we're doing live

on this stream. So, leave a like down

below to show that this is the most

important AI test in the history of

mankind. Uh, if Codeex wins this test

today, I'm a I'm a Codeex guy. So,

Sam Alman, if you're watching this,

which a lot of people tell me Sam Alman

watches this, you better get there in

the back end and start tinkering with

things and turn smart mode on because

this stream has the potential to move

millions of dollars into your pocket if

Codeex wins here. Same thing to you,

Daario. If you're watching this, you

better turn on the you better turn the

opus knobs from stupid to smart as other

people are saying, because Claude Code

wins here. Claude Code continues to be

the focus all my videos. Claude is king

as widget flow. I mean, I agree. Codeex

sucks. I've Every test I've run on

Claude Code versus Codeex, I'm going to

be honest, hasn't been close. And I'm an

Open AI fan. I'm an OpenAI fan.

But every test I've done clawed code in

Codeex, it like it hasn't been close.

So when people come in the chat and say,

"Oh, Codex's way better slopus." Like,

"What the hell are you talking about?" I

mean, we're going to find out for sure

here. We're going to know for sure on

this stream. By the end of this stream,

you'll know if you need to be using

Claude code or Codeex.

You're going to know. Also, do your boy

a favor. Turn on notifications. Brand

new Clawude Co-work video. This might be

one of my best videos in months. One of

my best videos in months. Dropping

either tonight, but probably first thing

tomorrow morning. Notifications on for

that immediately. When that drops, open

it up. Watch it to the end. Leave a

like. That's my only ask. All right.

Where we at here? We're on This is 52

extra high. This is 52 extra high. Let's

see where we are. Let's see where we

are.

52 extra high. Let's see where we are.

One second.

One second.

Uhoh. Here we go.

Here we go. All right.

Is it done? It's done. It's already

done.

It's already done.

built a standalone 3JS theme with five

distinct AI cores, clickable selection,

and a control deck that lets you

dispatch prompts and watch a simulated

live feed progress while the 3D core

animates. The UI theme and motion are

custom styled

to control the room instead of default

dashboard.

Next steps,

uh, run a local server to load. Well,

why can't you just Can you just run

everything for me? Can Codeex can run

things, right? Codeex can run things.

Can it not?

Can you just run everything for like I

don't want to type all that in. You type

it in for me, bro. Claude code does

that.

Allow

planning non-blocking local server on.

Okay. Start a local server for you.

All right. Do we test this? Do we test

this? Are you guys ready? The VS Code

plugin is bugged. Use Ghosty, bro. What

for Codeex? Yes, you have to run codeex

from Explorer. Faster and Ghosty, too.

So, what people are using the Codeex

CLI? Codeex problems alone. It's not

running on Verizon, is it? Are there

ways to save tokens of Claude Opus? Are

you certain prompts that make it use

less tokens? I don't know. What are you

talking about? Is there certain prompts

that use less tokens? What are these

broad questions people are asking me?

Love using them together. One codes

other reviews. What do you use for

coding? What do you use for reviewing?

Swiss HD, captain of the Twitch

community is back. What's up, Swiss HD?

Good to see you, my friend.

Oh, let's turn on uh show comments on

stage so the Twitch community can

participate. Codeex is slower, but so

much better than Opus 405. It finished

so much faster than Claude. What are you

talking about?

It finished much faster than Claude.

All right, you guys ready to see?

Here we go. Codeex 52 extra high 3D

agent manager is complete. Here it is.

All right, let's see what we got here.

What's going on? Select an agent to

deploy a prompt. Click an agent to

connect.

I don't see any agent. Control deck.

Pick a cord and dispatch a mission. Are

you there? dispatch.

All right. I mean, there's no there's no

agent.

All right. I mean, so we're gonna have

to go back and uh fix it because there's

no agent.

Codeex is better right now for me, but

it's just so slow on extra high. It was

fast for me. Did does OpenAI and Claude

put me on like different subscription

tiers or something?

You're my favorite vibe coder of all

time. You make it so uncomplicated and

fun. A cool segment on your show would

be live critique on other people's

vibecoded projects. You always give

honest feedback, says Bo Sterling. Well,

first of all, Bo Sterling, thank you.

That's very kind of you to say. It's the

first time someone called me their

favorite vibe coder of all time. We are

now the number one vibe coding channel

on YouTube and the number one vibe

coding account on Twitter.

It's official. We've been crowned the

number one vibe coding channel on

YouTube and the number one vibe coding

account on X. So shout out you guys. I

don't do it without the FinnFam. We

don't become number one on YouTube and X

without the FinnF fam. You guys, this is

not even me being number one. This is

about you guys being number one. You

guys are number one. So congratulations

if you're watching this live stream

right now. Congratulations. You are

number one. Congratulations.

Uh what was the second thing I'm going

to say? I don't even remember the second

thing I was going to say is I got so

looped into congratulating you guys. I

forgot what the second part of my

message was. Oh, a cool segment on your

show would be live critique other

people's vibe coded projects. I think

that's a good idea. I think that'd be a

great idea for when Viber is like live

live like all the way launched. Then I

can just go through Viber and look

through all your guys apps on Viber.

Just watched a video saying Anthropic

will not be allowing code editors to use

their APIs. It has to be done in cloud

code only. Have you heard this is RG

Copelan? So this is a big misnomer. I

don't even know what misnomer means, but

it feels like the right thing to say

there. Anyone know what misnome What

does misnomer? Listen, that's a big

misnomer.

Hope I'm using that correctly. That's a

big misnomer.

Here's what happened. Everyone was

getting pissed at Anthropic

the other day because they're like, "Oh,

they're blocking other tools from using

Claude." That's not what happened.

That's not what happened.

When you subscribe to Clawed Code,

Anthropic is subsidizing your tokens,

right? You're spending $200 a month on

Claude Code, but you can use like

$10,000 worth of tokens. They subsidize

for you, right? when you use when you

subscribe to Claude Code, they're losing

money if you use more than like $200 of

tokens a month, which is very very easy

to do. What these other companies were

doing were they were allowing you to

port in your subscription to Claude Code

into their app and use your Claw Code

subscription in their app, right? And so

now you are using those subsidized

tokens in a way that Anthropic was not

intending because you're using in

someone else's app. And now all their

analytics, all the AI training they do

on your usage of Claude code, that all

gets messed up because it's some other

rapper using that subscription, right?

And now if that's like a malicious

rapper that's using your subscription

and they just like make it so you use

like $10,000 of tokens in one go,

Anthropics, you know, not only losing

tremendous amounts of money, but they're

also messing up all their analytics and

training data because it's some other

rapper using their subscription,

right? You can still use the API. If you

want to use the clawed API inside open

code or inside any other tool on planet

Earth, you can do it. They're not

outlawing using the clawed API, you can

do it. They're just outlawing you using

your clawed code subscription in other

tools. The other tools are using

workarounds to get your subscription

into there. They don't want you doing

that. If you want to use the API and

plug in your API, go ahead because now

you're paying for your tokens. But what

you're doing by importing in your Claude

code subscription is you're using

subsidized tokens in someone else's

tool. So Claude's basically paying you

to use someone else's tool, which makes

zero sense for them to do, right?

So all the are like all the people

crapping on Anthropic,

that was it was just kind of a stupid

thing. It was it was stupid. Uh I will

say this, I thought Anthropic's

explanation of the situation was

terrible. I thought it explained it

horribly. Like I think I just explained

it like a master. Like that was

incredible what I just did there. Props

to me for my explanation. That was

amazing. And all 119 people in this live

stream now understand. But the tweets

they put out explaining this were

terrible and everyone was crapping on

them.

Hire me for PR. I mean, Anthropic needs

to hire me for PR. Anthropic needs to

hire me as head of content. They need to

hire me as like head of claude code

innovate. They need to hire me at

basically every CEO. They should hire me

probably. Some people say not me. I

didn't say that. Some people say it

though. Um, put me in a PR, bro. Put me

in a PR.

What are people say? What's the reaction

to everything I'm saying? I got to

scroll down. Snoop Dog. Hi, Ray

Fernando. You back in uh you back in

California, dude. What's the story?

Ray, we're figuring out what's the

what's the goat here? Claude Code or

Codeex. People have been coming in my

chat saying slopus. Sloppus. I don't

know how you pronounce it. Sloppus.

Lopeus. I don't know how to pronounce

the uh cursor CLI is actually good now.

They rewrote it last week.

The C the cursor CLI is good. They I

haven't seen any uh content about cursor

CLI. Are they they're still updating it?

That's good to know.

Uh what are people saying about my rant

I just did on the Claude? Are people

agreeing with me? Are people agreeing

with me

on the Claude code rant I just went on

that they're not the bad guys in this

situation? Got it. Thanks for the

clarification. It's a misnomer to call

your grandmother grandfather. A wrong or

inaccurate use of the name or term.

Close enough. Misnomer means whatever I

want it to mean. Now it means what I

mean want it to mean. Is this the same

with VS Code as a wrapper? Well, no. VS

Code, you're opening up the claude code

extension. That still works fine.

It wasn't that they stopped it. Is that

they allowed it for 3 months. Yeah. I

mean, you should guys should feel lucky

that they allowed you to import your

Claude code subscription into other

tools for 3 months. It's the fact that

they allowed it is what's disappointing

people is because they got used to it.

But like it was wrong in the first

place. If Yes. If only they explained

like you people wouldn't be showing

that. Yeah. It's it's it was hard to see

on Twitter because so many people were

getting angry at anthropic.

Anyway, let's get back into it. I just I

I I just saw people getting upset and it

didn't make sense.

So, this was codeex 5.2 extra high.

Didn't work. We're going to go back and

fix it. We're going to go back and fix

it.

But, do you guys want to see how Opus

Opus just finished? Should we see how

Opus is doing here? Let's see how Opus

is doing here. Opus just finished their

work. For those just joining, here's

what we're doing.

A lot of people coming into the chat

saying Claude Code's dead. Opus is dead.

It stinks now. It's stupid now. Codeex

is a goat. People come into my chat

saying codeex 5.2 extra high GPT C++ is

now the goat.

Well, we're going to find out once and

for all, right in here, right now. We

finished it with codeex 5.2 extra high

C++ GPT C++ wrapper 3.4 and it sort of

works, but there's no 3D agents.

Opus just finished. Let's see how Opus

does. AI agent manager. Oh my god.

Look at this.

Look at this.

Hey there.

I mean, what are we talking about here?

This is I'll take my name and this

comment off here so you can see this in

full screen.

This is what codec 5.2 extra high GBT

C++ 3.84 point codec G+ C++ extra high

extra high made.

And this is what Opus 45 made. The

prompt was build us a 3D AI agent

manager where we have all the AI agents

on screen.

and codec 5.2 extra high, C++, G++ 8.5.3

extra high, extra high, extra low, extra

high made this. No 3D agents, nothing,

no functionality whatsoever.

And Claude Opus did this.

I don't know why Gemini's head's

testicles, but we'll leave it at that.

Maybe that's an hardcoded into

anthropic.

I mean,

nopiius 4.5, get the hell out of here.

The people who said slopus, the people

who said nopus,

you're banned. I'm going to ban anyone.

Like you if you came in here and said

Codeex is better than Opus,

you're paid. You're getting paid.

Like this is like this isn't even

surprising to me. This isn't even

surprising to me. This was my expected

results.

Slope is my ass. Like what are people

talking about?

Like, am I taking crazy pills? Am I on

some like sort of special tier of like

clawed code or something? Because every

month people come into my chat and say,

"Claude code sucks now. Opus is stupid.

Everything's stupid." I get that once a

month. And every time I get that, I go,

"What the hell are you talking about?" I

use this 12 hours a day. It's amazing.

It builds the most mind-blowing code

ever for me. What are you talking about?

And I just proved to you once and for

all. I gave the exact same prompt to

both.

And it's night and day. I mean, it's

night and day.

I mean, we'll go we'll build more. We're

going to build more of this. I'm going

to go back into codeex and give it a

second and third chance.

But like,

it's night and day. I mean, it's night

and day is doing a disservice to

nighttime and daytime.

Can you use claw code to implement a

minimum number of iterations inside a

Ralph Wiggum? Add the Ralph Wigum plug

and make a minimum number. I don't think

you could do minimum number, but you do

maximum number.

Like night and day. It's like it's an

insult to nighttime and daytime to say

it's night and day.

Gemini packing. Yeah,

look at that head.

It's essentially set and forget

something Cloudco can't offer in an

interactive session.

I don't know. I just gave it a prompt

and it worked. Amazing. It built exactly

what I wanted.

Like what are we talking about? Lopeus.

Burp. What's your name? Burp. Your name

is Burp. Burp. Are you being paid by

Open AI? Listen, I'm an Open AI fan. But

this is night and day. Like, this is

completely unfunctional and does

nothing.

And this is a I mean, let's give All

right, let's give Codeex a second

chance, right? Let's give Codeex a

second chance.

Share screen window. Codeex. Here we go.

Uh,

the interface appears, but I don't see

any 3D agents.

Like, what are people talking? Every

time people say, "Codeex is now better

than Opus. Opus stinks." I run a test

and literally 100% of the time Claude

code's better. Like, what are people

talking about here?

Oh, should I do the cursor CLI, too? I

don't even I haven't even installed the

cursor CLI.

How do I install the cursor CLI? We're

going to we're going to figure this out

right now.

Here's the clean correct way to install

the cursor CLI without fluff. I do

like what Chad GBT What are we doing

here, dude? Why does every question I

ask do you have to do things like here

without the fluff? Don't tell me there's

no fluff. Just don't include fluff.

I have I have cursor installed.

Install a cursor command from inside

cursor. Open side curs open cursor.

Press command shiftp to open the command

pallet.

All right, I guess I'll do all this.

Whatever. We have to find out if cursor

CLI Ray Fernando comes in here and says

cursor CLI is legit. I'm listening to

Ray Fernando.

All right. All right. What's it telling?

Open cursor.app. Press commandshiftp.

Doesn't do diddly squat. Okay.

Didn't do anything.

All right. Let's uh I'm going to Google

it. Google cursor CLI.

Cursor CLI.

Cursor CLI. Are people using the cursor

CLI? If you're here, go in the chat and

let me know if you're using the All

right, here we go. It's just a curl.

What are we doing here? Chat GPT. Chat

GPT has no idea what it's talking about.

It's just a curl command. Let's not over

complicate this. File.

New window.

CD. CD. Let's do this. Uh, make direct

cursor

3D agent cd cursor 3D agent. Paste this

in. All right, here we go. Let's see.

Cursor CLI. Let's see how it compares.

Is cursor CLI back? Is cursor CLI back?

Start using the cursor agent. Type in

agent. Okay.

Is cursor CLI back? Oh, that's cool.

That's cool.

I'll give I'll give cursor the win for

animations in the terminal. That's

probably the best animation. Press any

key to sign in. All right, I'm logging

in.

It win. It wins for uh animations in the

terminal for sure.

If we stop the competition right here,

cursor wins right now. If we stop the

competition right there, it's the

winner. Trust this workspace. All right.

So, by default, it uses four five opus

thinking

for the record for those wondering at

home.

All right, let's see. I do I'll say

this. The animations, the blinking

lights, the way this is structured. I

like this the best from a user interface

perspective. This is cool. This is neat.

This is neat.

Shout out cursor team. That's neat.

What's the result of the normal codeex

one? Good call.

Good call out. All right, let's see.

run the install commands for me and run

the app. We'll we'll find out in a

second what the uh codeex 5.2 just

regular codeex 5.2 not the extra high

whatever

uh in a second.

Does it work? Codeex

says burp. Now burp's just crapping on

every one of them. Opus clearly the

winner. Let's test if he even works.

Well, the AI functionality is not built

in yet, right? I had to put in the API

key. So, that the AI functionality

wasn't built into any of these. Right

now, we're just testing the interface,

but the API key part shouldn't be the

difficult part. All right, dev server is

running.

All right, you guys ready?

So, cursor CLI is doing its thing. Looks

great. What I'm about to show you

is

codeex 5.2. No, no, no, not even codeex

5. It's just Chad GBT 5.2. So, this is

codeex. It used the Chad GBT 5.2

model.

So, let's let's we'll review it again.

So, let's review this again.

This was codeex 5.2 extra high. what's

supposedly OpenAI's best coding model.

This is it. This is Chad GBT 5.2 codecs

extra high thinking. This is what it

made.

Total failure, right? This is Opus 45.

Looks awesome,

right? Remember, we're building is a 3D

AI agent manager. We want to manage

multiple AI 3D AI agents. The next

screen I'm about to show you is just

chat GBT 5.2. It's just chat GBT 5.2. No

codecs, no extra high, nothing. Just

chat GBT 5.2.

I mean this is this is significantly

better than codec 5.2 2 extra high C++

A++ GPT 5 extra high extra high chat GPT

5.2x codeex 5.2 too, right? Way better.

I mean, I think there's an argument to

be made this is better than

Opus.

Just from the kind of layout of chatting

with the agents,

I still think Opus is probably the

winner so far just because the 3D models

look so cool.

But I think there's arguments you can

make that some of these UI elements are

better than Opus.

I mean it's obvious that the codeex 5.2

extra high GPT+ C++ is not even in any

of these.

It may be because you have a very good

MD file for Opus. Not everyone here

probably maybe do with your setup. Maybe

do a video on the dot. So my clawed

rules file. I've shown my claude rules

in a couple videos, but I'll do a claude

rules specific.

If you I als I shared my claude rules

file in the vibe coding academy last

week. So if you want my claw rules file,

it's available right now in vibe coding

academy link down below if you want to

see that. Now

codeex is better than opus for detailed

code review and debugging. That's why

the correct answer is use the model for

the task. Some are better than others at

certain things.

I mean, you're probably I I've

experienced that where it's it's been

able to fix some bugs Opus could not

figure out. So, code review codecs might

be better, but for my new workflow where

I'm just straight up using the terminal,

I don't want to use VS Code with five

different things open.

I mean, I guess I could open codeex in

one terminal and that's like my code

review terminal.

Yeah, I mean, that that's the answer,

right? Opus is the daily driver. Codeex

is the code reviewer. But the people

coming in saying Opus is stupid. What

are you talking about?

What are you talking about?

Opus is the goat. I built three iOS apps

in a week with Opus. Yeah, it's I I I

build three iOS apps in my sleep with

Opus.

I build three iOS apps while I'm on the

toilet with Opus. Like, what are we

talking about here?

They're just rage baiting. I see it all

over

X. All right, let's let's uh let's let's

give CodeEx a second chance. We're not

done here. We still got cursor CLI

going. We still got cursor CLI going.

I'll say this.

I think the cursor CLI has the best look

and feel to it.

I think this looks cool. I like the

structure of the It's showing too much

code. Too much showing me too much code

here. Like just show me what you

changed. Like show me like a description

of what you changed. You don't need to

show me all this code. Um features five

unique AI agents, interactive controls,

chat interface, visual effects, npm

rundev.

Run the server for me.

What is this percent? Is that the

percent? Is that the

Is that the context window? What's the

percent right there? I assume that's the

context window and not like my usage or

something. The server is running. All

right.

Are you guys ready?

You guys ready to see the the cursor CLI

results?

You use clawed code to edit the plugin.

The reason Opus produces bad code is

because people tend to either over

complicate everything in their setup or

they don't give it enough info to build

what they have in mind. I never I don't

get bad results from Opus. What are

people talking about? Burpus. This guy's

name is Burp. It isn't Opus' fault, but

they just need someone to blame. The

reality is Frontier models are

non-deterministic. So, a two-s sentence

prompt won't reveal as much as a

thoughtout multi-step plan. In fact,

yeah, I mean, that's what made Claude

code great, though, is that it was like,

hold up, this prompt is too broad. Let's

go into plan mode and let me ask you a

few questions. I thought that was great

behavior by Claude Code.

Can you build the databases using cloud

code? I mean, how do you integrate

database and backend into your front

end? I mean, all I basically want is a

shout out.

What?

Aravest Ahab, is this like a trick

question? Did you just want your cat's

face on the screen? What? What do you I

mean, all I basically want is a shout

out. So, do you not mean the question

you asked me? Was this all a troll to

get me to say your name and show your

cat on screen?

Is that all this was? Aravis the Ahab.

You just say integrate a back end. Like,

you just say integrate a back end. Boom.

There you go. Does Opus agent manager

work oneshotted? Yeah, it was all one

shot. Everything you saw there was one

shot.

[cough]

[clears throat]

Excuse me. I made myself uh spicy

buffalo wings during the Patriots game

Sunday and I've had a cough since then.

I ate I made like really spicy buffalo

wings during the football game Sunday

during the Patriots game. And since I

had those wings, I've had like a really

really violent cough. What's that about?

I don't feel sick at all. I don't feel I

don't have any other symptoms. I eat the

really spicy chicken wings and now I've

had a violent violent cough for like

four days straight. Did I like

permanently damage myself?

I mean, I guess I could technically ask

AI like what are we talking about here?

Anti-gravity with Gemini 3. See what

builds? Should we add anti-gravity to

this test? Should we add anti-gravity to

this test? What's the difference between

the CLI? Isn't it should be the same.

Stake a stake a crypto casino has slot

GBT. It is interesting. I'm not doing

your slop casino online.

What's that browser? Chad GBT atlas.

That's another one I see on Twitter all

the time. Oh, no one uses Atlas anymore.

All I use is Atlas. Atlas is amazing.

Atlas is amazing. You can make a game

and then play it for real money. I have

no interest in gambling. Anyone here who

spends their time gambling,

you're wasting your time. Especially

online casinos, I mean, you're you're

going to lose money. Why would you waste

money?

Open code is even better. Ambitious

realism creates. Are you What's wrong

with you? So, same model, different

rapper. Let's see how it's going to come

up with the result. All right. Oh, yeah.

So, you guys want to see the uh

the cursor CLI results. Here we go.

Cursor CLI. You ready? Ray Fernando came

in here talking a big talk. He said,

"Oh, cursor CLI is where it's at." He

also happened to be posting uh uh

pictures from the cursor office the last

week on X. So, a little bit of bias

might be here. But hey, listen. If this

result I'm about to show you

is the best, then I'm going to be It

doesn't matter if he's biased or not.

Doesn't matter.

Quality is quality. You ready? Here we

go.

3D AI8. Three 3D AI agent manager test.

Cursor CLI. Here we go.

That's pretty good.

Okay. How do I talk to it? Claude. I

click it. Hello. I'm Claude.

Hi Claude.

Oh, wow. Look at that. It has like the

little sparks coming out of it when it

talks.

This is pretty good. I can zoom in and

out.

I mean, this might be better.

Is this better than clawed code?

Here's

the here's the I think the graphics are

better on the cursor CLI,

but I like the models better on Claude

Code. Like I like how they made a

different That's very thoughtful, right?

The different models, the testicles on

Gemini's head.

That's very thoughtful and tastefully

done. They just made them all just

different colors on Cursor, but I think

the graphics are actually better.

Rio is Cursor's lead design. He's a

monster at design. That's That terminal

is I think the most beautiful terminal

experience out of them all.

It would be hilarious if Vanilla Codex

did better than Codex Extra Super. It

did. It did do better.

All right. So, I mean, what what do you

guys think? What do you guys We're going

to do a poll at the end of all this to

see what people think's better, but what

do you guys think's the best one so far?

Is it is cursor CLI the king right now

over Claude Code?

Codeex is Patriots with Gerrod Mayo as

coach. Opus is Patriots with variable.

That's a fact. That's a fact.

Codeex is not good with anything to do

with UI to be fair. Well, that I mean UI

is a pretty important part of any app.

Cringe blocks blocks black swan. What

are you cringing at? I'm cringing at

you. What are you cringing? looks clear

to me. Yeah, I think I think the UI is

cleaner on cursor.

I like the agents better on clawed code,

like the different designs, but I think

this overall aesthetic is nicer on the

cursor one.

I mean, we have more we need to do here.

We need to actually implement the AI

functionality.

And I like to have them move around and

do things. So, we're going to implement

that.

Yes, Codeex and Claude both in ghosty

tabs.

You going to do the C++ thing every

time, says Black Swan. The answer to

your question, Black Swan, is yes, I am

going to do the C++ thing. In fact,

Black Swan, the fact that you even

called out the fact that I'm saying C++

at the end of every time I say codeex,

I'm going to do it even more now. I'm

going to go out of my way even more to

say codeex 5.2G. 2, GBT, C++, A++, uh,

extra high G++, C++. The fact that you

called it out, went into my stream,

said, "Hey, Alex, I don't like that

you're saying this thing

is now going to cause me to say it 10

times more." So, yes, Black Swan, I am

going to say codeex 5.2G C++ every

single time now. Every single time.

All right, let's let's open up

anti-gravity. Let's do this in

anti-gravity. I like this little uh

benchmarking test. This is a pretty good

benchmark. Oh, I'm going to do I'm going

to do anti-gravity. New folder.

Anti-gravity

3D agent manager.

Open. All right, here we go.

Anti-gravity.

Leave a like. Hey, if any if you've used

anti-gravity, leave a like down below.

I'm curious how many of you guys use

anti-gravity. Leave a like down below if

you've tried anti-gravity.

All right, it's loading.

Introducing Gemini 3 Flash. What are we

going to use here?

We're going to use Gemini 3 Pro. I feel

like we got to use Gemini 3 Pro.

Jesus, it's like whisper down the lane.

Every time I copy and paste a prompt

from another window, the prompt gets

more and more unstructured. All right, I

got to open up another window here.

Um,

so I got to grab the prompt from a

window that's not going to screw up all

the structure of this.

Copy.

Anti-gravity.

Paste. Here we go. Fast. I don't know if

I want fat. Yeah, I guess fast. Send.

[clears throat]

Just use T-Mox. You can have as many PES

in one window as you want. Even let them

talk to each other. I don't know. I like

Ghosty. You like T-M? I like Ghosty. Uh,

anthropic banning third party harnesses

while opening I goes full open source.

William Crannle, were you here earlier?

Were you here earlier? William Crannle.

William Crannle. They didn't ban third

party harnesses. They didn't ban third

party harnesses. William Crannle, I

explained this in detail. You're

spreading misinformation, William

Crannle.

They banned importing your Clawed Code

subscription into third party harnesses.

You can put in your Claw API key and pay

for your tokens that way, but when you

put in your Claw Code subscription,

you're importing

subsidized tokens. Anthropic loses money

on your Claude Code subscription. So,

they don't want you taking that

subscription and putting it onto other

tools and paying for tokens and other

tools. If you're go if they're going to

lose money on you, they want you to lose

money in their tool. But you can take

the Open AI API and put it into the

third party tools and pay for your own

tokens.

Yeah, I get that. So then why did you

say your comment before

maybe was late? Yeah, I'm like 45

minutes behind on chat. William Crannle,

I'm like 45 minutes behind on chat. Chat

goes quick. We got the number one vibe

coding live stream in the world. I can't

be up to date on the chat every second.

This is the number one vibe coding live

stream in the world on the number one

vibe coding YouTube channel with the

number one vibe coding X account.

Alex, you're supposed to take the seeds

out of the peppers before making the

wing sauce.

Drink ginger shots. It should help. Have

your tonsils been removed. I didn't get

my tonsils removed.

Get well, Suman. Well, I don't think I'm

sick. I just ate hot wings.

Hey, what do you think of Gemini 3

frontend skills? I'm going to be honest

and I'm a Google fan.

I like the company. Does that I don't

know. Does it make me evil? I like

Google. I like Google. I like the people

there. The people there reach out to me

all the time looking for feedback.

They're out of all the companies in AI,

Google reaches out to me the most.

Anthropic doesn't reach out to me at

all. Open AI has never acknowledged my

existence. What other AI companies are

there? Cursor never acknowledged my

existence. Google constantly DMing me

and reaching out to me.

But I'll say this, I I Gemini 3 is not

in my workflow.

I don't think it's the best at front

end. I don't think it's the best at

backend. I don't think it's the best at

writing code. I mean, it's the best with

media. Like Nano Banana Pro or whatever

is like that's the best. I make all my

all my thumbnails are Nano Banana Pro.

Anything media related, uh, I'm going

Google. But from front end, back end,

I'm I'm coding all the cloud code. But

shout out Google. They acknowledge my

existence. Number one vibe coding

channel in the world. Number one vibe

coding live streaming X and Open AI and

cursor never acknowledge my existence.

Not that I have an ego and I'm expecting

them to. I'm going to win no matter

what. Even if they never messaged me

ever. I'm just gonna say Google, biggest

company of all of them. Biggest company

out of all of them. Constantly messaging

me.

But I will say this cloud code is the

best at the moment. So it is what it is.

Um

Droid is better than cursor CLI.

You you certain you certain about that?

You sure about that? You sure about

that?

Hey guys, I need your opinion. My Clash

of Clans subscription just expired.

Should I renew my X20 subscription or

switch to open code? Is this a troll?

Is that a troll message?

Google is like that. I can easily that

Google's more proactive with their

users. Yeah, Google Google called me

yesterday.

Google called me and was like, "Yo, do

you want like a bunch of free credits?"

And I'm like, "Yeah." They're like,

"Okay, I'll email you." Then hung up and

that was the conversation.

That was it. That was the entire

conversation with Google.

Literally like one of their BDRs, one of

their business development guys

sits there, makes cold calls all day.

Cold called me. I pick up, he goes,

"Hey, you want a bunch of free credits?"

I go, "Yes." He's like, "Okay, I'll send

you an email." I'm like, "All right."

He's like, "Bye. Bye." That was it.

I don't think one person from cursor

follows me on X.

Honest criteria. What's honest question?

What's the criteria to being number one?

Being the best

then the capas stuck in your tonsils.

You get a seed stuck. So you think what

a pepper seed stuck in my tonsils? Is

that what's going on here? Anti-gravity

is still working here. Anti-gravity is

still working here.

I could not afford the API cost because

I use the $200 plan, not API, thus it

matters to me. But did you get my

explanation, William Crannle, that

if Open AI is subsidizing tens of

thousands of dollars worth of credits

for you every month, they're not going

to want you to put their subscription in

another tool, which they have no idea

how it runs or how many credits it's

going to use, you know?

All right, we might have anti-gravity up

and running here.

Bro, why are you mad? Explain again. Am

I mad?

I did use anti-gravity since the quad

code has been absolutely cooking in

ghosty. I literally wake up out of sleep

and code. It's just me like sleep code.

Now I'm in Twilight Stage. Uh, so you

you watched the new workflow video, my

dude?

You like it? You like using ghosty?

All those promoting codecs over opus.

Maybe because you're on Windows and Opus

works better on I don't think the models

work differently on different computers.

Tried anti-gravity cursor VS Code

terminal ghosty guy now strictly dbtech.

That's what I'm talking about. Drake May

guy. Ghosty guy. He knows what he knows

what he's doing.

The new uh workflow video is taking off.

Everyone's switching their workflows.

Try T-Mux with Ghosty. Great experience.

All right. What about building our own

website? Cloud code to build or another?

I just I've answered that question 20

times. JD, use cloud code for it all.

The development server is now running.

All right.

All right. You guys ready for

anti-gravity?

Who's ready for anti-gravity?

Who's ready for anti-gravity?

is ghosty in Windows. It isn't,

unfortunately.

If you're a vibe coder, you have to

switch to Mac. That's just like the

state of things. If you're a vibe coder,

you just have to switch to Mac. I know

it stinks,

but all the vibe coding tools come out

for Mac first. It is what it is. Like,

it just is what it is.

All right. Who wants to see

anti-gravity? I kind of have to pee.

Open AAI subsidizes their $200 a month

plan and they actually contact Open Code

to help with the OOTH login. Google

allows Open Code to use their

anti-gravity models. Google allows

Google allows Open Code to use their

anti-gravity models, including Opus.

Yeah, but I mean like but you can't

blame a company for

not wanting another company to burn all

their tokens, right? You can't be mad at

Anthropic because Open AI is like,

"Yeah, take advantage of us." Right?

Obviously, Open AI only allowed Open

Code to do that because they're behind

cloud code now, right? If Codeex was

more popular than Claude Code,

OpenAI wouldn't have done that. But that

maneuver of OpenAI swooping in an hour

after Anthropic was like, "No, Open

Code, you can't use our stuff." is a

maneuver that someone who's like in

second place does, right? You know what

I'm saying?

It's like what Meta is doing right now,

paying all these researchers like a

billion dollars a year. Like if Meta was

in first place, they wouldn't be

stealing all these researchers from

other companies. Not that I'm saying

it's a wrong thing to do. I'm just

saying it's the type it's a catch-up

move. You got to do catch-up moves

sometimes, right? I got to pee real

quick. Then I'm going to show you the

anti-gravity. Stick around. I'll be back

in 30 seconds. I got to pee. Then I'm

going to show you anti-gravity, the

results from the anti-gravity test. Be

right back. Everyone introduce

yourselves in the chat. Everyone go in

the chat. Say, "Hey, here's who I am and

this is what I'm working on." Be right

back.

All

right, you ready for anti-gravity? Here

we go. Did anyone actually go in and

introduce themselves in the chat? No,

you didn't. You didn't follow my

instructions. You didn't follow my

instructions. I said f I said go in the

chat and introduce yourself and no one

did it. No one did it.

Hey, I work on collecting thirdparty

data ethically and running ML. Oh,

Topshooters did it. Top shooters went

in. He collects third party data

ethically.

All right. You collecting data right

now?

Top shooter. You collecting data right

now? Enthropic dominates enterprise API

usage. That is what subsidizes the plans

along with the $20 plan because 85% of

$20 plan users only use 35% of their

quota monthly. Hasheconomics.

Little economics lesson there from uh

Ambitious Realism Creates. Thank you.

Ambitious Realism Creates. I didn't hear

that introduction instruction. My bad.

Well, listen closer next time. Beware of

Attack Chef. Listen closer next time.

Hey guys, I'm Jason. Check out me out.

Taylor Man77 or Jason Beans. There you

go. All right, here we go.

Anti-gravity test. How is their 3D AI

agent builder manager?

What do you guys think of that?

Oh,

hey there, Gemini.

I like that. That isn't English, but I

like it.

Claude.

Hey Claude.

Oh, I like that.

I like that little animation while it

types. I mean, this is I mean, it's it's

this is pretty neat. This is pretty

neat.

Oh, you turn it around. Everything's

backwards. I like that. That's pretty

cool.

This might be number one. Is this number

one? All right. So let's go back through

these.

This is codeex 5.2 extra high C++.

This is Claude Opus in claude code.

This is just chat GBT 5.2 vanilla.

This is cursor CLI. I think they use

opus in cursor CLI.

And this is anti-gravity using Gemini 3.

I think I'm going to be honest and I'm

trying to be unbiased.

I think from a spirit of what we're

trying to do with this project

perspective, I think Claude codes number

one. From a spirit of what we're trying

to do, we were trying to do 3D

representations of like bots

representing the AI models and Opus went

for that pretty much precisely, right?

And it differentiated each one. It came

up with a new design for each one. Like

that's the spirit of what we were

looking for with our prompt.

From a just pure aesthetics perspective,

I don't know. Gemini might win. Cursor

CLI is really If we're going pure

aesthetics,

I think I go Gemini number one, Cursor

CLI number two, then Claude number three

from if we're going pure aesthetics. But

I think from a spirit of what we wanted

to build perspective, I think Claude

Code's number one.

Can you try Google AI Studio? It seems

to be really good and can show you.

All right, Google AI Studio. This is a

good little test we're doing here, huh?

We're going to know by the end of this

live stream what's the best.

Here we go. Google AI Studio. People

want Google AI Studio. We're going to do

Google AS mod. I don't want Gemini 3

Flash. I want the the real deal 3 Pro.

All right, there we go. Build.

The reality is Anthropic made an

unforced error because Open Code put out

a limited amount of $200 Open Code black

plans and sold out twice in 24 hours.

How many of those have claude code

cancels attached? I think the only error

anthropic made was the PR. Just the

explanation. Their explanation wasn't

good. That's all. I don't think their

action was bad. I think their action was

totally fine.

I have no issue with what they did and I

think it was smart.

I just think the messaging was bad. It

was bad messaging. They didn't explain

it well. The tweets didn't explain it

well of why they cut off open code.

Just say, "Hey, we lose a lot of money

on the subscriptions."

The reason why we have the subscriptions

in clawed code is so that we can train

our next models so that we can run

analytics. And if you're taking these

subscriptions and you're hacking them

into other tools, it ruins all of that.

It just it it corrupts all that data.

So, you're still welcome to use open

code with Claude. Just use your API.

That's all.

If that was the messaging, I don't think

anyone has an issue. I don't think

anyone has an issue.

Maybe just you ambition realism creates.

But the messaging was it was just it was

too corporate. Their messaging was too

corporate and I think confused a lot of

people.

Godson still likes cursor. All right.

All right. Here we go. Google. Google AI

Studio. It's all set. Done. It's already

done. Here we go. Click an agent to

begin. There's no agents.

All right. So, first test, I'd say it's

the worst. I don't see any 3D agents.

We'll do a poll. We'll do a poll when uh

we've tested all of them out. We'll do a

poll when we tested them all out.

I just think a little warning would have

been prudent.

I don't think they need a warn. It's

their product. It's their subscription.

They're not taking the product away from

you. They're just saying, "Hey, you

can't break our terms of service and

hack our subscription into another

tool." That's all they're saying. You

can still put the API just as intended.

I think people are pissed it stops with

no real warning. Well, what are you

pissed about? Just get your API key. It

takes 5 seconds. Go claw API, hit enter,

go in, copy and paste your API key in,

and you're good to go. There's no

disruption of service. They're just

saying, "Hey, instead of using your sub,

use your API.

Instead of using your sub, which was

hacked into this product, and we're

losing tens of thousands of dollars off

of, instead just type in your API key.

That's all.

No disruption of service.

I have zero issue with what Anthropic

did. I My only issues with their

messaging. I thought their messaging was

terrible.

Does anyone have anti-gravity Gemini 3

over the other ones? We need some

rankings. Let's do this. Well, let's do

this. Let's fix Let's fix 5.2 codecs

C++ High.

Between clawed and open code, which one

should I choose at the $200 price point?

I mean, I think it's incredibly obvious

to be frank with you.

Claude code with Opus 45 is the best I

think the best AI tool ever made. So

it's up to you.

Please hard refresh. Okay, let me see if

the hard refresh fixes this.

Okay, so the hard refresh

makes it so we can see a ground. We

still don't see any agents.

We don't see the 3D representation.

Okay. Now I see a ground and the agents

listed in the side panel, but I still

see no 3D representations

of the agent. So now we're on prompt

number three to get Chad GBT extra high

C++ to work.

What are we cooking today? This is

Conrad Major. Conrad Major, we're doing

the ultimate test. We're testing. We We

start off the stream. People come in go,

"Oh, Opus is dead. Opus is dead. Codeex

is king now. Codeex 5.2 extra high. C++

is the best. Go with that." So, we're

doing a test. We're testing all of the

AI IDs. Which are the best? We're

building three a 3D AI agent manager.

This is what Opus came up with where we

have our five AI agents we can talk to.

This is what f Chad GBT 5.2 came up with

vanilla. No codecs. Chad GBT 5.2

vanilla.

This was cursor CLI. Cursor CLI. Very

impressive.

This is anti-gravity with Gemini 3,

which I actually think might be the most

visually and aesthetically pleasing of

them all.

And this is uh next. This is uh Google

AI Studio, which I think's probably the

worst. Not usable.

Not usable.

Still see. I mean, you're not supposed

to be using this for anything of

serious. Like, this is a serious app and

you're not supposed to be using Google

AI Studio for anything serious.

Uh Claude Code Jet Brains Unlimited

Chrome extension is my preference. I

have not liked using Gravity.

Why haven't you liked using

anti-gravity? Out of curiosity, William.

I mean, I I don't use I'm gonna be

honest. I don't use anti-gravity because

Claude code's better. I think in a

pretty by a pretty sizable margin, but I

do like the direction anti-gravity is

going in. I do like the direction

anti-gravity is going in, like from an

agent manager perspective. I like their

product vision. The execution so far is

not good.

How is Ralph Wiggum? I have a video that

came out a few days ago on Ralph Wigum.

I like it. Is it my daily driver plugin?

No. But in some specific circumstances,

I like using it

with open code band. Is moving to Claude

Code CLI the best option now? Yes,

I have a video out on that workflow came

out yesterday.

Was it yesterday or was it two days ago?

I think it was yesterday. video on my

cloud code workflow on using the CLI

yesterday.

There's a Verizon outage right now.

Verizon's down. I saw someone make a

Verizon joke earlier

and then I just got a text message from

my dad saying all of Verizon's down and

then I looked at my service and I have

no service.

How can an entire network go down?

All of Verizon is down right now.

How's that possible?

Verizon. How is it possible an entire

network is down?

Looks like half the country's Verizon

service is down.

How is that possible?

Did someone hack Verizon?

I'm seeing activity now on Verizon.

Verizon is awful and that's why my phone

lost service.

Okay, you play let's play a game.

How quickly can you go on X and see a

political conspiracy theory?

Like how quickly can you go on X and see

a political conspiracy theory? Jesus

Christ. All Verizon. All right.

Looks like Codeex uh let's see.

All right. We gave Codeex 5.2. 2 C++

extra high. Another prompt to try to fix

it and uh still not working. I'm

surprised that Gemini did as well as it

did. Opus is always my first go-to AI,

but have heard many say Gemini is good

for UI. I mean, that UI is really nice.

I was surprised.

The clawed opus 4.5 quality has degraded

a lot for me in the past days.

I get this so much. I get so many

messages. Oh, Opus is degraded. It's

getting worse. They made it stupider.

They made it dumber.

I never experienced that. Why don't I

ever experience that? I've never

experienced that.

Opus has never felt stupider for me. And

I feel like these results tell that

story, right? Like I think this is the

this was the best one out of all the

ones built.

Claude and then anti-gravity. Clawed for

the spear and overall layout and future

usability. Anti-gravity is purely

aesthetic. Yeah, I think Claude's the

best. Should we do a vote?

I think we could do a poll.

Let's end this poll. Let's do a new new

poll. Which was the best?

clawed code

cursor CLI

um

anti-gravity.

Can I add more options? Oh, I can add

more options. Uh and then what else? Uh

codeex.

What else? What other options do we have

here? Codeex 5.2.

There we go. There's the poll. vote.

This is clawed code.

This is well, no, we're not going to

count that one. This is 5.2 codeex.

This is cursor CLI.

And this is Gemini 3 on anti-gravity.

Oh, for those that use cloud codes

Chrome extension, I got a pro tip.

What's the pro tip you got, William

Crannle? You should know that web

content accessibility guidelines is is

and how it works in screen readers

matter. All right. Well, there you go.

There's your pro tip.

Verizon was down on East Coast and

Southeast earlier. Wasn't aware. I'm on

the West Coast, baby. I'm in California.

I have SOS mode.

But basically, I see many people use the

visual part of the Chrome extension to

view and navigate a website.

By the way, everyone, subscribe, turn on

notifications. New video on Clawude

Co-work coming out first thing tomorrow

morning. Potentially today, potentially

this afternoon, but probably tomorrow

morning. Subscribe, turn on

notifications. You want the best way to

support this channel for free,

the most free way to support this

channel ever, when I post a new video,

open it up the moment you get the

notification, leave a like on it, watch

it to the end. That's the best way to

support this channel for free. If you

want to support this channel not for

free, Vibe Coding Academy link down

below. Full Claude code community master

class all that link down below.

Next call on Friday. I just I just saw

that Gemini CLI has its own version of

Ralph Wigum. People using Gemini CLI. Is

that a thing?

I used that back in the day. It wasn't

great. It was the most free out of all

them. In fact, Opus 45 was trained on

TPUs. I read that this morning.

I'm uh I'm putting a lot of money into

Google, investing in Google.

Opus has issues after compacting, but

other than that, I have not had any

issues, but that is much more compacting

than Opus. Some lonely vibe coder hit

the wrong button and took down Verizon

by accident. Yes. Right. All right,

let's see what the poll's at.

I mean, the poll is not even close.

People have clawed code over the others

easily. Anti-gravity in second and

cursor and codeex in dead last. Claude

code 61%.

Anti-gravity 22%. Codeex and CLI 8%.

It's not even close.

Does that put it to rest?

Most people saying Opus has performed

worse are usually having context issues

are just haters. That's why his multi-

aent workflows and Ralph are getting so

much traction. Separate context windows.

Every time someone says Opus is getting

stupider, I never understand it. And I

just proved it. I just proved it right

now. I proved it to all the haters.

I just proved it, did I not?

There we go. Viber. Oh, let's see what

uh errors they come up with Viber.

Let's see what codeex we told Codeex

find all the errors.

We told Codex, hey Codex, find all the

errors in Viviver. What do we got here?

Uh so codeex on the left claw in the

middle

high impact likely breaking core

functionality broken superbase off

cookie handling an API route overwrite

set cookie for each cookie and formats

cookie attributes incorrectly which can

break login session refresh. I've never

had that issue. Multiple API routes

authenticate with a server client and

then perform DB rights with browser anon

superbase clients.

Okay, that's a good one. Some security

issues found

duplicate app URL handling is

inconsistent. The DB enforces normalized

uniqueness but API pre-check won't

reliably match stored links.

Let's see what Claude came up with.

Critical issues fix immediately. Prompt

injection.

User controlled app URL is interpolated

directly into Gemini prompts without

sanitization.

Race condition for username change.

Check then act pattern allows two users

to claim same username simultaneously.

Race condition and voting. Voting logic

checks if v uh vote exists then updates.

Concurrent votes can duplicate.

These are a lot. These this found a lot.

I I'd argue Claude code found more

issues and better issues.

Prompt injection.

Uh I'm not going to say better.

They found different ones.

anti-gravity even showing in that poll

is encouraging because I have been

anti-gravity for a few months now and

I'm loving it. The anti-gravity one was

great. The results from anti-gravity

were really good, but to be fair, you

should do it with standard.mmarkdown

files and no skills. I don't think I

used any skills for this test.

I didn't use any skills for this test.

It would have asked me permission to use

the skills. I don't I didn't give it

unlimited permission for skills.

Output both the markdown and have one of

them merge the two files.

Oh, that's a good call. Model here.

Can you output all of these bugs to a

new bugs markdown

file?

I'm gonna have them both do it. Then

merging. That's a good call out,

Monolier. Very smart. Very smart. I feel

like there's going to Fox Bro and asking

what team is the best ASE. A little

biased, but I'm Claude guy. I mean, I'm

only a Claude guy because Claude codes

the best. If we ran these tests right

now and Codeex was the best, I'd be a

Codeex guy. If I ran these tests right

now and Google AI Studio is the best,

I'd be a Google AI Studio guy.

I'm only biased for whatever tools the

best.

And to me, me and the polls all said

Claude Code was the best. I mean, it's

is it subjective? Yeah, I guess. But

like I think visually I thought it was

pretty obvious Claude Code was the best.

Was it not?

AI ho. Am I an AI ho? Is that what I am?

William Crannle. An AI ho.

Is that what I am? That's what you think

of me? That's what you think of me,

William Kandle? You think I'm an AI ho?

That's what you think of me? Okay,

that's fine.

I think after all these tests,

it's still clear

code's king. It's why I make that's why

my last hundred videos are all Claude

Code videos. If Codeex were to get

better, I'd make Codeex videos at the

wazoo. What are your thoughts on GLM

4.7?

Uh, I think it's great.

If I, you know, I I do a project I do

want to do this year. I do want to do

dood do that I do want to do this year

is I want to build my own AI lab

and run GLM 4.7 locally and have my own

local private setup but I have to be

honest like

it's fantastic and it's incredible that

it's open source and you can run it

locally that's amazing props to them but

like at the end of the day I want to use

what's best in order to get the best

efficiency I want to use what's best and

Claude Opus 45 is the best. You know

what I'm saying? Like props to GLM47.

It's open source. That's amazing. It's

great for humanity, but I need the best

technology at the moment. And Opus45 is

the best technology.

But I do want to build my own AI lab. If

there's any uh AI companies watching

this right now

and want to send me a new Mac Studio

with the M5 Ultra

and sponsor a video,

hit me up. You can build my AI lab for

me. I'd be very happy if you built my AI

lab for me.

Whatever you do, don't get a Spark. Just

got one and it's frustrating on AR. Why

is it frustrating?

because that was one of my options was

getting a DGX Spark. Why is it

frustrating?

I was either going to buy DGX Spark or

like uh

build a computer with um

the RTX 6000 or whatever in it.

Did they create two separate bugs files

or all in one?

Did you delete the other lines that were

in there or make a new file?

Limitations. Bleeding Edge does not

support some programs. Okay, that's good

to know. armbbased binaries and things

just haven't caught up yet with the x86

options. Okay, I won't get the spark.

I'm talking about the people voting.

What made you switch from cursor to claw

says nostalgia games? So, I started on

Cursor

August 2024 started on Cursor. That's

when I started this YouTube channel

because I was so blown away by Cursor.

Uh, and then that winter, so like

November 2024, cursor just got so

stupid. Like it all it would do was

create errors for me. That's all it did

was create errors for me. November and

December 2024, all it would do was

create errors is using Sonnet 3.5 at the

time.

Then I switched to Windsurf. Windsurf

was fantastic. And then I started

getting errors just like cursor. So many

errors. And then Claude Code drops

February or March 2025, right when

Windsor was starting to really piss me

off.

Switched to Claude Code, haven't looked

back since.

Been the best tool I've ever used ever

since then. for almost a year now, which

is like

10 years in AI terms, you know,

and it's been the best since. If cursor

becomes better than Claude Code today,

I'm switching to cursor.

If cursor becomes better than Claude

Code, I mean, if the Teletubbies has

come out with a vibe coding tool and

it's better than Claude Code, I'll use

the Teletubby vibe coding tool. You

know, I'm not loyal to any of these

companies.

I'm just loyal to what's the best tool.

And for 10, 11 months now, Claude Code's

been the best tool.

You can connect four Macs together. And

the numbers were pretty good. Just saw a

video on the Mac cluster. Yeah. If

there's any AI companies watching this

and want to send me four Mac Studios,

hit me up.

I'll sponsor. You can sponsor my video.

Send me four Mac Studios so I can build

an AI lab. I hear you, man. That's how I

felt about Claude when I first started

back in October. Cloud was great until

it kept giving me errors after errors

and switched to cursor. Haven't looked

back. Okay, so you like cursor better.

Everyone everyone has their own

preferences. Listen, I'm not going to

argue anyone.

I only argue with you when you come in

here and say opus is stupid all of a

sudden. It's like no it's not. The spark

looks awesome. The spark is sick.

Uh but apparently it has its issues.

Hey brother, does anyone ever told you

you have the cadence and mannerisms of

Trump? Uh, I get this reply quite a bit

on my videos and then I go to people I

know and I'm like I get all these

replies on my videos saying I sound like

Trump and I'm like do I sound like

Trump? And they all go, "No, what the

hell are you talking about?"

I mean, he's a New Yorker and I was born

in New York and lived in New York for a

while,

so maybe that's it.

He's got a New York accent.

What frameworks do you recommend, GSD or

Ralph Wigghams? I put out a video on

Ralph Wigs a few days ago. Ralph is

great if you use it in the right

situations.

We got to fix Viber man. Such cool

potential. A future tools.io vibe. Uh,

let's fix it. We'll fix on uh Friday.

Should we fix it on Friday? DBT. We got

distracted today. People came in and

said Claude sucked and I got distracted.

We'll fix Viber on Friday. Does that

work?

How do you feel about anti-gravity at

the moment? It's a really cool concept.

The execution isn't there yet.

Really cool concept, but the execution

isn't there yet.

I'm from New York and you don't give me

Trump vibes. I get that. I get the reply

a lot that I sound like him. Do I? I

don't think I sound like him.

Try to use Gemini 3 Pro and Claude Code.

It gives better results in anti-gravity.

Does a Gemini 3? I'm sure. I'm sure. I

mean, Claude Code's the best vibe coding

harness in the game. Cursor or Claude

Code for building mobile apps? I think

Claude codes perfect for building mobile

apps.

I think cursor they do they did ship

just recently the ability to like

preview mobile apps in cursor

which is cool but that's not my

workflow. My workflow is claude code and

uh you open up Xcode and then you run

the simulator in Xcode and you can look

at your

you know your app there. I have a I have

a whole workflow video if you want to

build an iOS app. I have a whole

workflow video in my channel from like

three weeks ago or something.

It's the facial expressions, the

exaggerations, too. No hate, bro. Love

the vid. Just subscribe. Well, thanks

for subscribing.

It's the facial expressions, the

exaggerations, too. You got to

exaggerate. You got to keep it

entertaining. What if I sat here and I

didn't exaggerate at all? This video,

these videos would stink.

Showbiz baby.

When do you think Claude Code gets

dethroned?

I mean, that's obviously impossible for

me to say,

but I will say this, and I truly believe

this.

Claude Code has been the best since it

came out in March of 2025. So that's

what, nine or 10 months? 10 months. For

10 months,

it has been the best vibe coding tool.

And then you go even further back than

that. Sonnet 35 has been the best coding

model since that came out a year and a

half ago. Sonnet 35 has been the best

coding model

and then all the obviously follow-ups

have been the best coding model since it

came out a year and a half ago.

So in AI terms, a year in AI terms is

like

a decade in regular terms, right? So

it's just it's just like imagine a

company having like a decade being on

top. That's a huge amount of time. And

so all I'll say is like if I had to

place a bet on an AI company, it would

have to be anthropic just because

they've figured out how to maintain a

lead for a year now. That's a tremendous

amount of time. That's a tremendous

amount of time.

So I wouldn't bet on them getting

dethroned,

right? Since Claude Code came out, we've

seen 10 versions of codecs, 10 versions

of Google, you know, platforms, 10

versions of cursor, and none of them for

even a millisecond have been able to

dethrone Claude for not even a

millisecond. That's pretty amazing.

That's pretty amazing.

Yeah, I like using Gemini and Claude

with the free preview still running.

Yeah, get get your free tokens. Nothing

wrong with that.

Guess I'll switch to Claude codes since

I've never used it, but cursor is so

creative. Listen, I'm not saying you

have to. I'm just saying based on my

experience, Claude Code's the best. I

don't think there's any uh thing that

can go wrong from you testing it, right?

Maybe you just test Claude Code. Pay

$20, test it. If you like it, stick with

it. If you don't, stick to cursor.

Everyone has their preference. Just my

preference is strongly been cloud code

for a year now. I'd say Lovable is best

for apps. Lovable's price structure is

garbage. I've been saying that forever.

Um,

I don't think the lovables of the world

can win. I don't think economically it

makes any sense.

I think over it's like the people who

say, "Oh, I win at sports gambling."

Like, you might be winning right now,

but over a long enough time period,

there's no way you can keep winning

because the economics don't make sense

for lovable.

They're taking already a a a poor

economic model of AI tokens which are

too expensive right now and layering on

their own margin on top. You can't do

that for a long time period. Everyone

who uses lovable and these other similar

tools always complain about the pricing.

Replet, they always complain about the

pricing eventually because the pricing

just doesn't make sense. And and so the

only way you can win is if you do what

Claude's doing, which is you own the

model and you subsidize the tokens.

That's the only way you can potentially

win.

Agreed. But don't you think the compute

of Open AI will overtake at some point?

Well, if it's purely a compute game,

then XAI is going to win. If it's purely

a compute game, no one has more compute

than Grock, but it's clearly not a

purely compute game. If it was like

OpenAI's had more compute than Anthropic

this entire time, but Anthropics had a

better coding model.

So the game is clearly not just compute.

If it's if the game was just compute and

that's it, Grock will Grock should be

winning.

But I'd argue Grock is not top three or

even top four models right now.

And don't get me wrong, I'm a huge Elon

Musk fan. Huge Elon Musk fan. But at

this moment in time, January 14, 2026.

The fact that Grock's number one proves

it's not a compute game.

Are you using Max times 20? Yes, I'm on

the $200 monthly plan. I have a couple

of the $200 monthly plans.

If Claude Code gets beaten, I think it

will be XAI just because the pace at

which they iterate and catch up.

I think they were there for a while XAI

with the iteration and and and speed,

but when's the last time Grock has

released an update? It's been like five

months now. I feel like we've gotten

multiple Cloud updates, multiple chat

GPT updates since the last Grock update,

right?

What's Koshi saying?

What's Khi saying for the next uh

I bet that we were going to get rock 4.2

by the end of the month. That's tanking.

That might be my first loss bet. That

might be my first loss bet ever. I might

lose all my money on Kashi because of

this bet. I bet they were going to get

Grock 4.2 by the end of the month. It's

tanking. I got it like 80%. It's down to

60%.

We're tanking here.

They haven't released an update in like

five months now. How can we say XAI is

the highest pace of iteration? We

haven't got an update in like five

months and we're saying we're not even

get an update this month. I'm going to

lose money on this.

Elon, if you're watching, I'm losing

money on this. Release an update so I

can get my money back. Elon,

also give my Cybert truck actually smart

summon, please. When am I gonna get

actually smart summon on my Cyber Truck,

Elon?

Claude Code's a money printing machine.

My guess would be they're losing money.

My guess would be they're losing money

on it on Claude Code.

I wonder what they think about Claude

Code at Google and Open AI.

My guess would be

more.

This is a pure guess. This is not based

on any information at all.

My guess would be a major, this is pure

guess speculation.

I would guess a majority of the

developers at Google and OpenAI and XAI

use claude code. That'd be my guess.

Pure speculation,

but I think it would be stupid if