Four AI bots just built and launched an

entire ad campaign for me. 50 creatives,

nine ads launched and ready in my ad

account, and I didn't have to build a

single thing myself. I was out getting

lunch. This is Grobot, brand new from

XAI, and I got early access. Every bot

gets a name, a job, and its very own

computer in the cloud. So, I built an

entire team. Marge runs my ads. Ditto

posts my content. Velcro manages my

school community and Jenny generates all

the ad creative. Then I wire them all to

my Mac Mini via Tailcale so that every

automation I've ever built with Cloud

Code, they can just use right off the

bat and I didn't have to reinvent the

wheel. And then the real test, one group

chat with all four bots. Literally

working together to build an evergreen

ad campaign for my school community.

Parts of this genuinely blew my mind.

And while there's a couple things that

Claude Code still does better, I'll give

you my honest answer at the end of this

video on whether or not this is ready to

move my entire business operations into.

Let's get into it. Today, um I have

three different use cases I'm going to

be setting up. One is a bot called

Marge, whose whole um purpose is

basically to manage all of my ad

accounts. So that's like creating ads,

managing the media buying, giving giving

me insights. Um, you know, this is her

her life. Marge is short for margin. So

that's kind of fun. The second bot here

is called Ditto. And basically its whole

purpose is to post my content everywhere

for me and edit my videos and do

everything from like um setting up many

chat automations, email flows, LinkedIn

uh DM automations, everything that I

normally would do um for my content. So

its job is to take a video that I give

it, post it everywhere, write captions,

set up the backend stuff. Um so yeah, I

gave it this prompt. Um, you know,

basically this is after a lot of like

brainstorming and Claude, this is what I

came up with. So, it looks like now it's

uh asking me a couple questions. Where

should I pull raw videos from? I'd like

it to have connection to my own

computer. So, I'm going to ask it um can

you SSH into my Mac Mini to view files

there?

And now it's thinking. While Ditto works

through some tasks I've given it, I'm

going to go ahead and get my third bot

set up here. It's called Velcro because

it's going to help people stick around

in my school community. So the whole

purpose of this bot is basically to

serve as a surface um you know any

questions coming up coming up in the

group help me you know find you know if

somebody asks a question and the the

answer exists somewhere else like kind

of draft replies for me obviously I edit

everything um and also like grab course

material from different parts of the

school and basically just as act as a

community manager and like assistant for

So that's the whole concept with this

one. So part of this will be I I'll need

to make sure that it's connected into my

school. And the way it will do that is

all of these have their own um computer

which is pretty crazy. So each bot has

its own isolated machine that it runs

everything in the cloud. So looks like

out of the box we've got Chrome, a file

manager, and like a terminal. So, um,

yeah, I'm I'm really curious what we can

kind of like get figured out with this.

Okay, this is actually pretty wild. So,

um, when we hit the, you know, how is it

going to get files? It was like, okay,

maybe we'll grab them from Google Drive

or I can give them files manually or

whatever. But I already have a Mac Mini

set up with like dev setups, like my

repos, everything's already set up on my

Mac Mini, so I don't want to like

completely reinvent the wheel. So, I

asked my Mac Mini if it can just connect

to Tailcale and Tail Scale is what I use

to connect all of my machines together

and allows them to like basically always

be able to search in other computers.

So, when I'm on my Mac, a my MacBook Pro

from anywhere in the world, I can like

remote into my Mac Mini or if I'm my

phone, I can connect to my Mac Mini.

Everything's always connected, which

makes it so I can be just anywhere and

always have AI, have access to anything

on across all of my different devices.

So, this is super cool because I just

asked Grockbot to connect to my Mac Mini

via Tail Scale and I just added it to my

Tail Scale account. So, I have Ditto

agent now as a device within my Tail

Scale and it shows as connected. Um, so

now it's saying we're connected. I can

reach your Mac Mini from anywhere over

tail scale. So then I said great. Now I

want you to and I gave it like the

location of my like dev folder for all

of these um content automations that

I've already built out in cloud code.

So, I gave it the link to the

destination of that folder, and I'm

basically going to I'm basically asking

it, I want you to do a deep dive, fully

understand everything in this folder,

how it's set up, how you can use it to

do all the things I'm going to ask you

to do, like post my content, edit my

videos, all this stuff. Because again,

I've already done a lot of the work here

inside of Claude Code. So, I don't want

to completely have to like reinvent the

wheel inside of Grockbot. I want to just

point Grockbot to the work that I've

already done and then allow it to use

all of those tools inside of this

Grockbot computer environment, which it

can do now because it's directly

connected to my Mac Mini. So, this this

like cloud computer that Grockbot always

has access to now can fire off um

scripts and automations from my Mac Mini

for me. And Grockbot also has an app,

which is like this is what I'm really

excited to test and see how well it

works is if I can basically just use

Grockbot as like my execution layer for

everything that I'm doing in my

business. Okay, so now basically

um Claudebot is going to town. It's

asking me which raw um video should it

edit first and I pointed it to uh a

video about editing videos with Claude.

So, this is the video. It's probing it,

checking for an existing edit. Um, and

then it's going to start just like going

through a process that I already have

set from Cloud Code to edit the video.

And this is actually pretty cool because

it was able to just like to see that it

already edited like I already edited

this video once inside of Claude Code

and that it's posted live on Instagram

and everything. Um, it's not going to

post it for me. Okay. Okay, so here I'm

now I'm telling it that basically I want

it to still edit the video. Um, so we

can just do like a dry run together and

that I want it to go through the motions

of what it would do to post this video

on all my channels, but don't actually

post it yet cuz obviously it's already

been posted. And we're kind of treat

treating this as a dry run to make sure

that all the kinks are worked out.

Popping back over to Velcro. Looks like

it has my school community up on its

computer and it's asking me to sign in.

So, I'm going to go ahead and go into

the computer and log into my school.

So, I went ahead and logged in and told

it that I am done. So, now it's going to

continue. And now I'm popping back over

to Marge and I'm saying I want you to

install tail scale like what we just did

with Velcro. And I'm saying we just I'm

telling it we just did this with Velcro.

So talk to her to see exactly how it was

done. So now you can see this is this is

so crazy. It says messaged Velcro and

it's asking how to get this done. Okay,

this is pretty dope. It's actually

saying that um we already have tail

scale connected so I don't have to do it

in for every single bot. So one bot

having tail scale works with everything

which is awesome. All right, so I'm back

in Velcro and it's kind of wild. that

did like a dry run of this thing and

basically it's like here's some people

who have joined with DMs, here's um

questions that people are asking with,

you know, some draft responses.

So, it's just nice to see that. And then

um people who have joined with zero

engagement, so like offering a DM to

send people uh to nudge them to get

started in the community. Some at risk

people for cancelling. Pretty cool.

Okay, this is pretty cool. It was able

to take the video um and do some like

light edits on it and then write me I

think some captions and have it all

staged to go live in all my channels.

So, it did what I would normally have it

do inside of Cloud Code but just inside

of Grockbot, which is pretty sweet.

Okay, so now I'm over at Marge who I

pointed to my folder in my on my Mac

Mini that has like all of my information

about like running my ads on Meta. Um,

and we've got now now she's asking me

some questions like what should the

target CPA be? I'm going to say 140. It

also pulled some of my historical

information from my ad account. Okay.

So, I'm going to try something and tell

it that.

So now I'm going to tell it, I want you

to draft a post to my community talking

all about how I have early access to

Grockbot and built out some pretty sweet

bots. I want you to talk with the other

bots like Marge and Ditto just so you

get full context around what we've been

building today and use all that

information to write the perfect school

community post. Make sure you use the

humanizer skill that's in the school

developer folder so that we make sure

that the text reads like a human wrote

it. [sighs]

All right, let's see how that does. All

right, here Marge was able to give me a

overview of the top ads. I just like

literally launched these ads like an

hour ago, so they've only spent $13 so

far today. But I was like, I want to

know which ones are just getting spent

so far. And so it gave me like little

screenshots of the ads. And then I said,

I want to actually have the preview

links. So then it gave me actual links

to every single ad with like the ad copy

and comments and all that good stuff. So

I could see and it showed how much spend

each one of these ads has gotten so far.

So like pretty sweet. If I was out and

about and then I just wanted to like

check in really quick without like

trying to open the ads manager app and

do all this stuff like this would

actually be pretty dope. All right,

coming back to Velcro the community

manager for school. And this is

[clears throat] what it wrote for me for

the like Grockbot update. Um, pretty

cool. Like you know sounds like

something I would write because of the

human ad copy scanner that it used.

Also, um, you can see here it got a

message from Ditto. And if I click on

that, it'll show exactly what Ditto was

able to provide to Velcro to help it

write this post. So, here's what's like

just kind of wild to me is all of these

chat bots can can like they can act

independently, but then they can also

work together on certain tasks. So, like

these are all focused on different um

outcomes, right? Like Velcro's whole

goal is being a community manager and

making sure people are getting great in

info on my school community and like I'm

able to do things in the school

community. Ditto is the content manager,

so taking my content, distributing it,

like handling all that good stuff. Marge

is like running all the ads, giving me

all the information for the ads. But

like really all three of these, it's

kind of like my flywheel, my content

flywheel, right? So, I've got content, I

got my school, I've got my ads. Um, I

could probably have another one in here

for like brand partnerships if I wanted

to have one that like manages all of my

brand partnerships. Um, but what's so

cool is like they can all work together

and like I don't know, that's just

that's like such a cool concept to me.

All right, so I'm making a fourth bot

and I'm going to call it Jenny because

it is going to handle all things Gen AI.

And so basically I'm doing the same

thing I did with the other three bots.

I'm pointing it to my folder genai

folder. So inside of um my AI projects,

this is another folder that I have

claude code working all the time. It has

like my arc ads. Everything that I do

with Genai lives in this folder. So, um,

basically I am pointing Jenny. I'm

pointing Jenny at that folder so that it

can get a full understanding of

everything that we've built with AI and

how it can use all these tools to

fulfill its purpose, which is

essentially be, you know, the designer

or like the AI designer. So, right off

the bat, I noticed that it was like

trying to find this on my local

computer, which is so I told it I told

Jenny access or to access that folder,

you must use tail scale and to ask Marge

how to get that set up. So, now it's

asking Marge how to get the tail scale

set up. Um, her I'll pick the deep dive

back up as soon as she replies with the

tail scale steps. Marge came through.

tail scales already on this computer as

Ditto agent I'll SSH into the mini and

keep reading from there. So like it's

like shared context across multiple

different bots you know like and you can

see them messaging back and forth like

it is so cool. Okay so it looks like it

got all the information from Marge the

gen AI folder four pieces basically

explaining how this whole folder works

together and everything that exists

inside of it. So now it's doing a

readiness check on ARC ads and key to

make sure that they're both connected.

It's found the keys. It's hitting ARC

ads and key with a readonly health call

now. Just basically making sure that the

connection is established. Like right

here, I just started a group message

with all four of them. And I asked, "Hey

everyone, just want to make sure

everyone's connected to Mac Mini via

Tailscale and knows how to access their

corresponding dev folders." Um, so Jenny

is giving me a yep, good to go on my

end.

>> [laughter]

>> And here we go. We've got all four

checking in. Yep. From Jenny, uh, Ditto,

same here. Marge, Velcro, they all said

that they are connected and pointed to

the folder that they should be operating

inside of. Okay. So, now I want to try

something that is going to like put all

this to the test. So, I'm going to do

another group chat here um with Jenny

Velcro. I don't think I need Ditto, but

I need Marge is I want to ask it to

basically build a new ad campaign for

me. So, Jenny is going to create the AI

creatives. So, I'm just going to do some

static image ads. Um, Velcro is going to

grab all the information from my school

community and then Marge is going to

take all that information to actually

build the ads for me in ads manager. So,

let's give this a try. I'm just going to

use Whisper Flow to write my prompt for

this. So, I'm going to say, "Hey

everybody, I have a task that I would

like you to tackle together. I need some

new ads built for my school community,

promoting people to sign up. Um, so I

need you all to work together on this

one. Uh, the first step here is I want

Jenny to create some new AI statics.

Marge, can you pull some historical data

from my ad account around which image

ads have performed the best and pass

that over to Jenny so she has some

references? And I'm also going to

include an air table database full of

hundreds and hundreds of AI statics or I

guess they're not created with AI,

they're real statics, but there's

already prompts in there for you to

rebuild them with AI. So, I want you to,

you know, take the information from

Marge so you understand what's working,

but then I want you to go and grab some

new templates. Maybe create 50 ad uh

concepts for me using the creatives from

the Air Table. And I want to make sure

that you're working with Velcro on this

so that you have all the information you

need about the school community um to

make sure that the text on the statics

is super accurate and up to-date and

relevant. Um I want Jenny I want you to

create all 50 static image ads. I want

you to host them on a herenow canvas and

then have the ability for me to go

through and like select my favorites

from that canvas. And then that will

tell Marge which of those to start

building the ads with. And Marge, I want

to make sure that you write new ad copy

for all these static images. And again,

review the ad account, understand which

pieces of ad copy are doing well. Don't

let the flash sale, an anniversary sale,

um, skew your data because this is this

is meant to be an evergreen campaign.

So, make sure you study the historical

evergreen creatives and not necessarily

the flash sale ones because, uh, we're

not doing a flash sale right now and

we'll make sure these ads work as

evergreen ads. All of you are free to

ask me any questions you have about this

process. Actually, maybe we can start

there. If any of you have questions,

maybe think about them and ask me five

questions uh just to make sure that we

fully get this perfect on the first try.

Okay, so that was with Whisper Flow. It

was able to put this all into a

beautiful prompt for me. So, I'm going

to submit it and let's see how my my

bots get to work for me. I'm super

curious. Okay, this is so cool. Jenny

wrote five questions for me and then uh

it's like pinging the other two like add

anything I'm missing. Um and you can

just see like all the other bots just

getting to work. So Velcro has got its

five questions and it looks like Marge

is now working on her five questions. So

once all these questions are laid out,

I'll do another whisper flow again and

just respond with my answers. Okay, so

we've got all basically 15 questions. So

I'm going to go ahead and do Whisper

Flow again and just answer them. So let

me do this and I'll start Whisper Flow.

I just gave you the Air Table link, so

that should be in the chat for you to

use. For the offer on the creatives, you

can include the price of $150 a month.

Definitely don't invent any claims. For

the format, you can do one by one or

4x5. Um, you know, if the source

template makes more sense to do 9x6,

basically just follow like the source.

But I would lean towards 4x5 as the

primary for all of these. GPT image 2.

Let's do that for anything that is

textheavy. You can use nano banana for

photore. That's fine. For the here. Now,

I want you to create a new canvas. and

just use a new canvas there. The authent

the authentication for herenow should be

in the dev folder, but let me know if

you need me to add it in the chat for

you. A Velcro, I'm answering your

questions. Now, the price on the

creative, yes, $150 a month. Don't show

the annual and don't use $97 anywhere

cuz that's not the price. You can use

the member count, but make sure you use

the most up-to-date for the offer

promise. Primary CTA is join school.

Yeah, there's no free trial. Just

school.com/mrpaidsocial

for the URL is fine. The current in the

product, uh I want primarily these to be

focused on the Claude code vibe coding

and less on the master class. So, make

sure you comb through the community on

like everything that I've been posting

as it relates to AI and using claude and

claude code. And then also referencing

any of the information from the actual

courses that I've put together in in the

class as well. Also, I put together that

start here course that has like the full

overview. maybe use that as your

knowledge base for this so that you have

full visibility into everything that is

in the course in one sheet. And I'm

wondering if it makes sense to even like

host that sheet as like a Google doc

that we can have linked in the ad copy.

I think that could actually be pretty

cool where we could show like here is

the full outline of the entire school

community that you can preview before

you even want to sign up. So maybe

create a Google doc version of that

course. The start here course shows

everything in the community and then we

can link to that in the ad copy um of

our ads. I think that would be I think

that'd be really cool. Um yeah, banned

on the image. Don't talk about refunds.

It's not a free community. Don't

guarantee anything. There's no

anniversary sale. And then Marge getting

to your questions now. Uh evergreen

scope. Which campaigns adsets count as

evergreen? They have evergreen in the

campaign name. So just focus on those.

Winner definition 3090 days or lifetime.

You can do lifetime. I think that's

fine. Build target. Drop new ads into

existing evergreen ad set. No. What I

want you to do is actually build a new

ad set in the campaign that I just

launched today. It's the school

evergreen campaign. So, just create

a new ad set within that campaign and

have it set to $100 a day. You can

review the ad set that I already

launched there today for how it's all

set up. Um, copy structure. I want five

primary text, five headlines. You can

reuse a couple evergreen bodies, but I

do want you to primarily write fresh new

primary texts. So, glean information

from what's worked in old primary texts,

but maybe only use one. Maybe use one

top performer. And again, don't include

any old pricing. So, update anything

that is a top performer with uh correct

pricing there. Uh let's see what else we

got here. Selection handoff. When you

star favorites on the canvas, is that my

green light to build those 50? Let's

I'll the the ones that I select in the

canvas will be what we build with, but I

wait for my confirmation that the those

are ready to build ads with because I

might have some that I select that I are

favorites, but maybe there's some that I

might have feedback on. So, actually,

that's a good point on the here.now

canvas that you make, have a comment box

underneath it so that I can select the

favorite but also leave a comment for

you. And if there is a comment,

basically that just means I need you to

regenerate that image and apply the

changes that I'm requesting. Okay, so

we've got a bunch of prompts. Like this

is a huge prompt. I'm answering all

their questions. And like again, Whisper

Flow is so so helpful for stuff like

this because I can just ramble and it's

like I'm actually talking to a team

member and giving them feedback instead.

I didn't have to sit down here and write

everything. And I feel like you just get

way better um giving clear like clear

instructions to your AI agents. So I'm

going to go ahead and submit this to the

group. While I was talking, I saw that

Jenny got the Air Table. She already

started working on looking at that. Um I

don't have Air Table connected yet.

Everyone, okay, if I install the Air

Table connector, so it it it like tagged

everybody to ask if it could install the

Air Table connector. Um which is just so

funny. It's like, let me clear this with

the group before I do it. It's so crazy.

So, quick update here. I've been just

kind of like chatting uh for a while.

Not that long, maybe like 5 minutes. The

the difference. And they went ahead and

they built like a full school overview

doc that it's going to link in the ad

creative for me, which is pretty sick.

I'm probably going to add it to the

about page for people that are looking

at the community as well. Crow do that

for me, which I maybe I'll ask her to do

that. And then um we got Jenny access to

Air Table and there was already like an

API key that I pointed it to and I had

like it's just really cool to see all

the different bots kind of like

interacting with each other and helping

each other solve the problems. So, I

just asked Jenny, "How are the new ad

statics coming?" And right now, it's um

it showed me that it was uh looking in

Air Table for it. And I think if I click

in Jenny individually, I can see kind of

what it's doing on its own outside of

the group chat. So, here it was asking

me um do we want a concept list first or

just generate? I'm going to do a concept

list first for B. Yeah. So now it's

going to like create a concept list um

from the air table for me. Okay. Uh and

we've got a status update here now. So

it's saying I've got Marg's evergreen

winners plus Velcro's facts, but I

haven't locked the 50 templates yet.

Short list is locked. 50 templates

pulled. Um Jenny is saying or Velcro is

saying look solid from my side. Hooks

match the facts pack. one check before

you burn credits on image blah blah blah

never yeah okay Marge is saying give

Jenny the ad side thumbs up so like

they're all cross-checking Jenny's short

list of templates that it's wanting to

use so here's that short list it looks

like so um you know it's got a bunch of

different templates u that's one of the

beautiful things about generating these

ads with AI in mass is I can have it do

50 and then I can choose like my

favorite five, you know, I don't have to

like mindlessly just have it create all

of them. Um, so we'll see what this

comes up with. So, it's kind of

interesting. It looks like there's kind

of some bouncing back and forth you need

to do from like a group chat into the

individual chat. Um, it's giving me an

estimate on credits cost for each.

Um, I'm waiting on Caleb's final

approval. So, it's messaging me here.

The batch now is running on my Mac Mini

and it's going to ping me when the first

wave lands. So, now it's getting to work

and it's probably going to create like

the canvas on here now for me

separately. While it was working on

that, I asked Velcro or sorry, I asked

um I asked Jenny up here to create some

like avatar images for each of the

thing. I just I had it just randomly

like use uh like a claimation style. So

this is the one for Jenny. It's got like

a paintbrush cuz it's like the creative

one. This is Velcro. It's like a hug cuz

it's a community manager. This one I

believe is Ditto cuz it's like copying

things over. And Marge, we've got like

the data analyst person because they're

managing all the margin and everything.

And then it was kind of cool because I

asked it to um you know hand these

images over to everybody and then each

person or sorry not person each bot

individually was like uh the avatar

thanks Jenny and they like implemented

it themselves. So now they all have

these little avatars which is pretty

cool. Okay, so I asked everybody in the

chat basically like, "Hey, is there

anything else we can be doing while

we're waiting on the AI statics to be

done and everyone was basically like,

"Nope, we're all set. We're just waiting

on the images." And then I tagged Marge

here and I said, "Go ahead and build out

the ad set while we wait." And so it

went and it built out the ad set that

we're going to be using to launch these

ads into. I popped over to Jenny

directly and asked basically like,

"How's how's it going?" Um, and it gave

me a status update. Two out of 50 done.

So, it's trying to give me an image

preview, but it's not working. So, I

wanted to see like, okay, can I see this

stuff in clawed or sorry, in my um Mac

Mini folder. So, I went into my Mac Mini

and I saw that it created a school

evergreen statics folder and there's an

image folders folder here. Um, and I can

see like some of the ads that it's

created for me so far.

Um, pretty solid. I don't know who this

guy is. Um, so we'll we'll I'll do like

a a more more of like a review when it

gives me all this in here. Now, but I

can at least see that the ad images are

populating now from Arc Ads. Um, it's

estimating 30 to 60 minutes, but this

shouldn't take that long. So, I'm going

to tell it like, "Hey, Jenny, really

quick. This shouldn't take 30 to 60

minutes. You should be able to fire all

of these prompts to Arc Ads via the API

in parallel and have all the images

generated at once. It seems like maybe

you are doing this one at a time and

that is unacceptable. So, please adjust

the way that you are getting all these

creatives made from Arcads. So, it said

you're right. Um, it's going to stop the

the slow five at a time runner and fire

all of them in parallel. But then I also

said I wanted it to use my likeness for

any of the ads cuz I noticed in some of

these that it's using like like this

guy, right? Some random person. Um but

I'd rather have that be me if it's going

to be using that. So I'm asking it I

have like a folder of images on my Mac

Mini of me just like in different

settings that it can use whenever it's

using my likeness and generating it uh

with AI. So I'm pointing it to that

folder or I didn't even point it to the

folder. I just said I basically find the

folder. Um, and it looks like it found

it, which is great. I didn't have to

give it a URL or anything. I think

that's the beauty of getting this all

set up where these bots can remote into

my Mac Mini via tail scale and

automatically like search through

everything because I don't have to be

super specific. I can just say, "Hey,

there's a folder of this in on my Mac

Mini. Like, go find it." And then, you

know, this time it found it right away.

I didn't have to give it a bunch of

detailed instructions. Um, which is

super cool. And here's something that's

super crazy about this method is in the

folder here, you can see there's like a

folder for prompts. And I can look and

see all the prompts that it wrote it

wrote to generate all these ads. So like

I didn't have to sit here and like write

all the and some of these are like

really uh you know thorough. Um, so I

didn't have to write in any of these,

which is just amazing because this is

already trained on prompting structure

for each model. It already knows like

how to put all this stuff together. So

that's super cool. Okay, so here's

something that's pretty cool is this is

like, you know, it's students thing. I'm

going to go get coffee and I have the

Grockbot on my phone. So I can literally

just like have the same exact session.

Yeah. So, I can have the same exact

session on my phone in real time as

what's going on on Grockbot. So, I'm

going to go keep tabs on this and come

back. Okay. So, really quick, I just um

I was able to connect my iPhone, but I

just want to highlight [clears throat]

this is the app view right here. Okay.

And I've got desktop open as well. And

you can see like they're exactly in

sync. Like it's showing everything on my

phone that it's showing on my computer,

right? Um, so that makes it really easy.

Like I can go I can select any of my

bots while I'm out and about and they

have full control over their own

computer environment and my dev computer

over SSH um through tail scale. So like

really this is like the perfect

execution layer for anything that I've

built already with cloud code. Okay, so

to finish this up, looks like we have 50

creatives to review from Jenny. They

whipped up the school evergreen statics

v2. So I'm going to click on that. This

was created in the here now link. So

it's right here, here. And now I can

choose my favorites. I can leave

comments. And then all of this will be

sent to my bot for review. So let's take

a look at some of these. This one looks

pretty I'm good with that. So, I'll go

ahead and favorite it. This one's kind

of cool. Shows the process of using

Cloud Code to whip up a bunch of

creatives. So, I'll go ahead and start

that. I'm not going to show the process

of me like looking at every single

creative and, you know, favoring them.

But, like obviously there's like this

one looks pretty weird. I'm not going to

do that one. Um, that one looks weird as

well. Um, you know, so the whole point

with this is like you're getting a lot

of creative like possibilities at very

like cost-effective rates. Like I think

each one of these images maybe costs

like 10 to50 cents depending on like the

model you're using and all that good

stuff. Um, there is a lot of text on a

handful of these which you know could be

good or bad. Um, so yeah, I'm going to

go ahead and choose my favorites and

then also if I care enough to go through

an iteration process, like if I really

like one of these, I will leave a

comment on the regen note. Um, but I

think for the most part, I'll probably

just choose my favorites. Okay, [snorts]

so I favored nine out of the 50 of

these. There are quite a few that didn't

quite make the mark. I in general I'm

noticing that this round of creative is

not quite as good as what I would

normally get from just using claude code

and specifically if I was using

something like Fable 5. And so part of

this might be the issue or part of the

issue with this might be that Grock bot

I don't know the model being used

doesn't allow you to select the model

but I think it's probably using Grock

4.5 which um you know is just not going

to be as good as like a top tier model

like Fable 5 at writing prompts. So,

there were nine though that I was pretty

happy with. There was a couple that I

gave it feedback on. So, I'm going to

see if it can regenerate them. Uh, let's

see if I can Yeah, right here. Like, I

was like, I like this concept. Some

parts didn't work. I explained what

didn't work and what I wanted to fix.

So, now I'm going to copy the feedback

and then I'm going to pop back over to

Jenny, paste the feedback in, and then

it's like giving it, you know, this was

the creative, this is my feedback, this

is the number, blah blah blah. Um, and

now it's going to get to work with the

next versions. And then I think it's

probably going to save the favorites and

start passing them over to the others.

So, in the group chat here, let's see.

We've got Velcro saying, "Canvas noted.

When the star picks, I'll spot check

winners and for claim drift blah blah

blah." Uh, before Marge writes copy

until then, I'm clear. Okay. I just uh

harded or starred my favorites in the

canvas. I gave a couple with feedback

for uh but go ahead and start building

the ones that I approved.

And so here we've got got your nine

favorites fixing the three regens now.

So working on that. Okay. So here Marge

confirmed the winners. Uh crow is set to

like check for claim. So, a reminder,

Velcro is my school manager, so she's

checking against claims made in the ads,

making sure they're factual. All good

there. Marge is good on product facts

for the copy. So, you can see we're just

kind of like passing things from one

person or like one bot to the next bot.

I'm keeping it moving forward. Okay, so

update here. We got some ads uploaded.

Um, I just went through my normal QA to

see, you know, how it did and overall it

did pretty good, but when I came down to

the ad copy section, there were

definitely some like m dashes and

whatnot. Um, so it did not write the ad

copy as I would have liked. Everything

else looks good though on my So, one

thing about this that I asked it to do

was run it through my AI detection skill

because I have an AI writing skill that

I run everything through just to make

sure there's no like AI tells. So, I

asked it to run through that. Um, it

also uploaded 14 ads instead of the

eight that I had approved. So, I asked

it what was up with that. And basically,

it said like it attempted and failed,

but like some of them went through or

whatever. So, um, some of them are just

duplic. So, at the end of this, I'll

probably ask it to like for me, but, uh,

it's working on fixing the ad copy.

While that's doing that, Jenny is

working on the regens. So, I'm going to

ask it what's the update on those

regens. Are we done? And just see if uh,

that's ready to go. But, we're getting

closer to having this thing fully built

out. I'm Grockbot, which is kind of

cool. Okay. I just thought of something

else that I don't really love about this

setup with Grockbot is like um sorry one

sec is right now I have Marge doing this

task for me and it says it's working but

I can't see what it's doing or like what

it's thinking and one thing I love about

claude code is when it's working on

something I can see how it's thinking

and how it's moving through a task and

then if it's like if I know that it's

going the wrong direction I can just

interject and correct it so I don't have

to sit and wait for it to do something

wrong and then have to say exactly what

it did wrong and then redirect that way.

So, it's not my favorite. Um, you know,

it says clean ads set with nine approved

ads is ready. So, let's see if it

actually did that. This was what I I

asked it to basically create a new ad

set. Um, it renamed the one do not use

that it made originally. So, let's click

into this one. Looks like we've got nine

clean. So, it did it did the task right.

Um, let's take a look at the ad copy

here and make sure that everything looks

better than it did before. I'm not

seeing any like M dashes or anything

like no no no tells here that I'm able

to spot in terms of how it's written.

Um, this looks pretty good, I think. So,

here we are. I mean, we got nine ads in

here, all generated with AI from start

to finish. I've got the images here all

uploaded with five variations of primary

text and headlines for every single one.

I'm going to let these rip uh and my

we'll set these live and we're just

going to see how they do for a while.

But um you know overall the process of

of running this with um with Claw or

sorry with Grockbot um it was it was

fine. It was fun. You know, I liked

getting to see the different bots

interact with each other. It wasn't

quite as like polished as my cloud code

system typically is. And I think that's

just a difference in the model being

used because again, I think Grockbot, my

guess is is using Grock 4.5 as the LLM

under the hood for everything. Cloud

code I can, you know, do Opus 5 or Fable

5 or some of these top models and, you

know, I get a really good output. So,

I'm going to keep testing this. I I

really do like the concept of different

like goto bots for specific things. So,

you know, I think it would be really

cool if Anthropic or OpenAI came out

with their own kind of like version of

this. Um thing I think would be really

sleek and I hope that they add is the

ability to select the um model being

used under the hood for the bot. That's

what you get inside of cursor. So, you

know, I think um there's some

improvements that can be made here, but

there's a lot that you can do. I mean, I

I was able to point it to my Mac Mini. I

was able to get it into my like dev

folder um setups for everything for

everything I'm doing in my business. We

got the job done and we were able to

delegate across multiple different bots

to get it done. So, like overall pretty

happy with it. Um, I think there's just

some things some things that compared to

my cloud code setup I'm kind of missing

with with Grockbot, but uh literally

just came out I [clears throat] think

yesterday. I'm sure they're going to

make improvements to it. Let me know

what your thoughts are in the comments

if you uh think that you give this a try

and let me know if you've got any

questions about any of this stuff. Um,

you know, I'm going to keep playing

around with it and let you know if

there's anything that uh, you know, I

find that's really interesting. All

right. Thanks everybody.