# AI Agents as Marketing Engineers: Scaling Creative Production

**Podcast:** The Startup Ideas Podcast
**Published:** 2026-11-09

## Transcript

If you're anything like me, you want to figure out how you can use AI agents to go and get you customers that make you money, that build high ROI, that could just take your company and make it a successful thing.
But how do you actually do it?
What are the workflows?
How do you think about the skill files?
How do you use Cloud Code and GrokBot to actually do all these things?
Well, in today's episode, this is a full on masterclass for how you can do that.
We have Cody Schneider and he takes you through step by step in the most clear way I think I've ever heard for how you can be a marketing engineer, basically a fancy way to use AI agents to do marketing on your behalf so you can make money 24-7.
We learn how to do static ads.
We learn how to do AI UGC.
We learn just all sorts of things around paid media.
AI content, how to do it.
So by the end of this episode, you're going into 2027 understanding how to be the top 0.1% marketing engineer on the planet, because let's be real.
A lot of people don't know how to use these tools.
A lot of people don't know how to pipeline data into this.
A lot of people don't know how to put the full picture together to actually create gorgeous content, gorgeous ads that are going to get you paid.
But I got good news for you.
You're listening to this.
This is 100% free.
Like, comment, and subscribe for more.
And I'll see you at the end of the episode.
What's up, Cody?
By the end of this episode, what are people going to learn?
Totally.
I'm going to teach you how to make GrokBot the best marketer in the world.
Top 1% or your cloud code.
Doesn't matter what harness that you're using.
And then the other thing that you're going to walk away from is you're going to know how to make static ads that are on brand, that are talking to the pain points and desired outcomes of your target customer.
And then we're going to make UGC avatar ads that look real, that talk real, that have motion graphics over the top of it.
That entire workflow is going to be just...
skill files that you have access to.
They're literally in the description below.
Give it to your agent and it can go and do this work.
So that's what they're going to walk with, G.
Okay, so you're going to teach people about how to use AI to make static ads, video ads, all these crazy, you know, I've seen just like go viral, these like AI influencer type ads.
I also want to know how I can layer on human beings into these videos.
You think you can also let us know about how to do that?
Absolutely.
Yeah.
We'll talk through like, how do you basically make the coordination layer that enables, you know, a person with a GrokBot or a GrokBot itself to like be this creator manager using a tool like Tribe or SideShift to actually like do, you know, the UGC production or layer in the human.
That's the gate of like, hey, I need this human involvement before any of this goes live where that human taste is added.
to the process, to the workflow, rather than it just being this totally autonomous thing.
So how do you co-work with this?
But solve the labor problem for yourself so you don't have to have a graphic designer, you don't have to have a video editor, you don't have to have a sales development rep.
You go down the list, all these pieces can basically be handled and solved.
So this is the advice that if you had a 17-year-old son and your 17-year-old son had built a startup and they had no customers and you sat him down and you said, Son, I'm going to teach you in under an hour everything you need to know about how you can use AI agents to create ads and create content that is going to get you customer.
Is that what we're talking about right now?
That is exactly what we're talking about.
Ads are one of the most high leverage things that you can do to get initial traction and also to validate an idea if you have a business idea that you're trying to get.
like to see if the market wants that.
This is the way to to basically make the media that's necessary to sell that product or advertise that product.
Right.
This is the exact thing that I would teach somebody.
Once you do it this way, you will never go back.
You will never do it the other way because it feels archaic to do it any other way than this.
Once you are in the weeds and basically building out these solutions.
So thank you, Father.
Let's get into it.
All right, let's jump in.
So first off, I just want to talk about this a little bit deeper.
And I don't know if you want to bring up that diagram that we were kind of working on right before this, Greg, of like what is happening in this category.
So the big unlock and I think the mind shift that I want you to get into here of like what is changing is, you know, historically, if I wanted to go and make, for example, a static ad, right?
I would have to go and get a graphic designer and I would have to go and like do the copywriting, ideate off of, you know, here's these different angles, desired outcomes, pain points that I want to focus on.
Now, instead, that is just code and tokens that are being burned.
Right.
And so all marketing is just going to this.
Like all marketing is now just code.
It's just software and tokens.
And so when you look at, you know, an it static image for ads that you generate when you look at, you know, a C dance video that you create, or before I used to hire, have to hire an, like an avatar, you're sorry, hire a UGC creator to make this and then have a video editor to edit it and then have a motion graphics designer to do the motion graphics, right?
That whole thing is just stacked.
That work is now handled by a GrokBot or a ClogCode or a Dots or whatever it is that you're using.
There's all these different vehicles for this coming out.
But I think the big thing I want you to understand is that that labor that was necessary previously is no longer gone.
Like there's a renaissance that's happening in long form video editing.
I imagine your editor, Greg, is starting to already do this where like you can literally use a software called Mirage Tesseract and it will go in one shot edit.
you know, a podcast that's an hour and 30 minutes long.
It'll cut out the silence.
It'll cut out the takes.
It'll like put it in frames.
It will do the edits of like showing the screen when it's appropriate.
Does the active speaker identification, all of this work that like used to be a human on a computer, we can now get agents to do this.
I did not think it was going to get here this fast.
to be totally blunt.
Like I did not think long form video editing or like a YouTube video editing was going to get here, you know, this year, but here we are.
And I think that today, like the big thing I want you to understand is that all of the work that.
used to be a human touching keys, moving the mouse, that can now be agents that are basically doing this work.
But you have to give those agents, like when I say agents, you have to give your GrokBot, your Cloud Code, whatever it is that you're using, all of the necessary infrastructure that it needs to be able to do that work.
How I've been describing is you have to make an environment for it that is like...
healthy for it to be like to work, right?
It's like this Petri dish that's like, you know, cozy and Higa and whatever, right?
That like it can basically like thrive in.
And so to do that, what you need is you need to provide it data.
So access to your business data, you need to provide it access to tools.
So tools, when I say that, that's like.
API calls to a chat GPT image 2.5 endpoint or API calls to seed dance or API calls to, you want to scrape your competitor's ad library.
We're going to do all this today, but that's what I mean when I say tools.
And then the other piece is you need to give it access to your ad platforms.
For example, your Facebook ads, your Google ads, your websites, CMS, so that it can actually go and publish that content.
that it produces or get the final products into the wild, actually into the public.
So that's kind of the high level.
Any questions about that I can try to answer before we jump into this and actually do these builds?
I mean, the only thing I want to just say is the whole everything is for you to do that job.
Like your job now, person listening, is to basically build the system that Cody is describing, the Petri dish that Cody is talking about.
The machine around all these models, agents, the machine around context, the machine around, you know, evals, that whole thinking is the job of the person who's building a business in 2027.
If you actually care about it getting customers on day one.
Exactly.
And I think that this is, again, like the shift that's starting to happen.
is like get out of the way of the agents.
Like you are actually in the way.
Like your job is to be a conductor.
There's this running meme right on social media of like the best, like, you know, the most effective employees will be people who played Starcraft, which is this like game where you would like coordinate, you know, swarms of actions.
And like, you know, it's basically, it's that same philosophy that we're applying here.
But, okay.
So how do you actually do this?
I'm going to talk about the open source version that you can like monkey patch together if you want to basically build everything that you need for these agents.
The other thing that I'm going to talk about is then, okay, how do you get started today?
If you're like an hour from now, you want to do this exact thing, how do you go and get this basically live and into the wild?
So first off, on the open source infrastructure, let's solve the data problem first.
on the data problem side, you can use a software called Airbyte.
And so you need to build two things.
You need to build basically a data pipeline and a data warehouse.
Why do I need to build these things?
If you're just querying against the APIs of your ad channels or any of the other solutions that you're doing, you're going to run into the same problems where it's like a massive amount of data that are all going into context.
And the agent is Very likely to hallucinate on this.
Or what it's going to go do is it's going to go write its own database locally and try to stuff all of that data into there.
But then you run into these duplication errors and all these pieces.
So how do you solve this?
You use infrastructure that's built specifically for this, that exists for this exact purpose.
So the open source versions of this are a company called Airbytes.
They are the data pipeline.
So when you think about them.
Like what they're doing is they're basically going and they are like taking all of these different sources of data, right?
Your Facebook ads, your Google analytics, your HubSpot CRM, right?
Anything that you store data in.
And they are basically the pipeline that is going into this data warehouse, right?
So when I point to here, this is where Airbyte fits in.
So this is Airbyte.
And in contrast, where this all goes is a data warehouse.
And so that data warehouse, what you can use is a company called ClickHouse.
Shout out ClickHouse.
They're an awesome partner for us.
So we love them for everything that they do.
And that data warehouse.
is then where all of this data from all of these sources gets stored.
So why am I putting it all into a single place?
It's so that I can look at the data across the entire organization.
I want to be able to see the ripple effect of I ran this ad and I can connect it to my Google Analytics.
I can connect it to my HubSpot.
I can connect it to my postdoc.
I can connect it all the way down to my payment processor and say, hey, when people see ads like this, I get more revenue.
So let's go put more money into that.
The thing that gets even crazier, though, Greg, is like you can actually see who's a better customer lifetime value.
So we can tag and add all the way down to like better customers come out the other side.
Like they are a better customer lifetime value for us when they see media like this.
So not just like it made revenue.
It actually made like more lifetime revenue.
When we are doing these activities, right?
So all this goes into the data warehouse.
And then that data warehouse, you're giving access to that data warehouse for your agent.
So when you think about this, this is where GrokBot is, right?
OrcLogCode or whatever it is that you use as your driver.
And this is back and forth.
It's basically querying that data.
And why we do this is it's constantly syncing in the background.
So this data is constantly streaming in, right?
It's constantly streaming in to the data warehouse.
And you can query this data warehouse as much as you want.
You can nuke this data warehouse with queries, right?
And this is something we see happen all the time with businesses is they're like, they're going directly to these sources.
And they're running into, okay, rate limits from the API or these challenges of like collecting it or they have to have like a cloud function that's running on an hourly cadence or some monkey patch solution that they built locally to solve this.
This is how you get around this problem.
So you give it the data.
So that's the first step of this.
That's how you give it the data.
The second step is you need to give it access to these platforms.
Let's say this is Facebook ads, right?
So Facebook ads.
How you give it access is you need to create what's called a Facebook ads API token.
So it's called the marketing API.
And the marketing API, how that's used is basically to do rights to the platform.
So I'm never actually reading the data from Facebook ads.
This is only one directional.
I'm only using this Facebook ads API to upload ad creative, to turn off ads, to move ads around.
All of the data reads are only ever happening from the data warehouse.
Now, why do I do that?
What is the point?
Like, what is the reasoning for this?
The reasoning is because you're trying not to abuse the Facebook ads API or the Google ads API, which they have terms of service and agents will go and just like, do a thousand requests at once and this that is against their terms of service so you probably have seen these online where like people like okay uh grokbot got my facebook ads account banned right like don't use agents for this that's not what's happening your grokbot violated the terms of service on the api because it was sending thousands of requests per second because it was trying to do some data pool so you should just you should assume your agents are like drunk patrons at a restaurant absolutely absolutely like it's your drunk uncle at the like local irish pub he's eight guinnesses deep he split the g four in a row for the last like hour and he's basically like this is that's how he's interacting with the facebook ads api right um This same strategy, you want to do this for everyone, though, right?
You want to do it for Facebook ads.
You want to do it for Google ads.
The same thing, their API.
We have documentation on this.
I'll send this over to you, G, so that you have it in the notes.
It's just a Notion doc.
It takes you about five minutes to get access to this.
It's not super hard.
You literally can give the Notion to your agent, and it will just, like, go and walk you through the setup process for this.
So I've now solved two things, right?
I've solved the data.
How do I give that to the agent?
Let me zoom in so we can see this.
So one, how do I get it to data to how do I get it the access it needs?
And now three, this is the thing that people are just starting to discover.
How do I give it the tools that it needs to go and do the work that's necessary for it to be able to do media buying, to function as a sales development rep, to function as a video editor, etc.
And for these tools.
What you can do, again, the open source version of this is you go and get a Kai AI subscription, and you go and get a FAL subscription or put tokens in there, and you go and you get an open router credential, and you go and you get an Apollo subscription.
You can go down this list of all of these different tools that you have to go and basically get access to.
And when you have all these tools, all these APIs, and you provide them to the agent, it now, so I'm just going to take this tool piece and we'll just put this over here of a list of a handful of these.
So like KaiAI, like Val for transcription, like Open Router for LLMs, again, Apollo or actually we'll just do like Git leads because they're a great partner.
or like Git leads for your email data provider and like person data.
When you do these tools and you give it access to the agents, it now can do the middle work for you.
So for example, it can make an ad using Kai AI, a static ad, and I'm going to show you literally how to do this.
You're going to just paste the skill in and it's going to ship ads.
It's ridiculous.
When it has all these tools, it can now do this middle work.
It has the environment that it needs.
So I was talking about this earlier.
Like this is the environment that's necessary for these coding agents to do this work, to do marketing, right?
And when you have this environment, suddenly your agent becomes the most like capable marketer in the world.
I'm talking like top 0.1%.
I've been doing this 10 years.
It will come up with stuff where I'm like, holy shit.
Like I would have never thought about that.
An example of this recently is we have an agent that runs Google ads, right?
This is like a template that we have within graph.
What we found is that we can in real time audit the search terms that we're coming up for, for paid search.
And we can have an LLM be like, is this search term like.
Is the search intent buyer ready for the landing page it's sending it to?
So before we're even spending money, we're having like this, you know, labor audit whether this is a good thing for us to even be spending money on.
Now, I could do this previously, right?
I could have a human that was running that same process.
But...
Getting a human to audit 5,000 search terms in 20 seconds is impossible.
And that's the scale that we're talking about here.
And you can get it to do this work and basically function as like a team of people that was necessary.
So this is the environment that you need to make.
That's the open source way to do this is go and get all those pieces.
If you don't want to do that, you can use...
Our company, graph.com.
Go to graph.com forward slash platform.
It's literally free to get started.
Everything that I just described is built into this.
The data, all of the access pieces we handle for you.
We also make it so it's multiplayer.
So all your team members.
for example, can be five people separately using their GrokBot or their Claude code can be managing the Facebook ads account through this without having to have all their own system set up.
And then also all the tools, all the tools that are necessary to do all the work that I'm about to show you.
Again, you can set this up yourself.
That's just the way if you're like 10 minutes from now, you want to be able to do this.
This is how you go do this.
All right.
Two things, two quick things.
One, just because people always ask, I'm not affiliated with Graft at all.
He is not.
I'm here just teaching this work on how to do this.
Two, there's one piece of this pie that this marketing environment diagram that I feel is missing, which is what I would call a swipe warehouse.
So in order to get the most out of your GrokBot, your Cloud Code, whatever it is you use, you need to have examples of inspiration.
that is going to help you get the top creative.
Because a big part of this game, if you want to be the best marketer, if you actually want to get people to actually buy whatever it is you're buying, you need good creative.
And the reality is, if you have good creative, the algorithms, in terms of paid ads, the algorithms are so good now that they'll match your good creative to an ICP that'll buy your stuff.
So the question becomes, How do I, like, I buy everything.
I buy the data.
I buy the access.
I buy the tools.
I buy this whole marketing environment.
But how do you create a, you know, swipe warehouse?
And when I say swipe, I'm basically saying a swipe file.
But swipe files made sense pre-AI.
Now you need, like, thousands of swipe files.
Okay.
What am I looking at?
Just starting to build this.
Okay.
And so as an example, like, you want that Google Ads agent that I was describing?
Yeah.
Literally, here's the code.
Just paste that into your cloud code.
And out the other side will come out a Google Ads agent.
You want to, like, we're going to use this exact one here, the static ads in UGC.
You want to make static ads, same thing.
Just literally paste it in, and it's going to be able to go and, like, do these builds for you.
So, all right, let's actually build some shit.
Let's get this, like, publicly out there.
So I'm going to make some ads.
Greg, what's a product that...
Should we do Idea Browser?
We can do a bunch of them.
But rattle off some company names and we're just going to start making creative for them right now.
I would say Idea Browser and our design and strategy agency, LateCheckout.agency.
Okay, LateCheckout.
We've never actually done paid ads for it.
I'm curious to see what it does.
Okay, so first off.
What we're going to do is I'm going to take Idea Browser, right?
And I have installed this skill, the static ads.
And again, I'm using Graph for this.
You can do everything that I'm about to do with like an open source version of patching this all together.
This is just the fastest way to do it.
So, Graph is installed.
And what I mean by that is we have the Graph CLI installed, which has like all of this knowledge like stacked in this.
When I say the CLI, what I mean is it's like this tool set.
It's just a package that has all of the knowledge on how to use the platform.
Like the marketing engineering cloud is what we're calling it stored within that.
So I've already got this static ad installed.
I just do forward slash static ads.
And I'm going to drop in the brand URL and hit enter.
And what's about to happen is it's going to go to the website.
and extract the brand style guides of the website.
It's going to go and find the Facebook ads library of Idea Browser if it exists.
It's going to go and then research the pain points and desired outcomes that this product solves by going and extracting conversations from Reddit and other social media.
It's then going to propose five static ads creative.
And then it's going to go and generate the ad creative.
based on all of this information that it just extracted.
And then from there, we're going to be able to go and upload this directly into Facebook ads.
Again, going back to that diagram, because we have all of this set up, all of this access, right?
Perfect.
So again, what I did is I did, I forward slash static ads, idea browser, and it's now just working.
It's running this skill.
And when I say it's a skill, Like I'm not talking about like it moves some text around.
It's going to actually do real work.
It's going to do research.
It's going to actually pull in like what the human would do to make this creative.
It's going to go and do that process.
So let's go do this for another company now.
Let's do this for late checkout, right?
So I'm going to create another session.
So forward slash static ads.
I drop in the business.
That's all I have to do.
Out the other side is going to come the static pieces of ad creative.
So let's come back.
We'll see what's happening.
It's extracting the brand here.
What other brands, G, are you guys working or that we can do right now?
Any products that you're excited about or like, I mean, it could even be e-com brands as well, like literally anything.
I would be interested to see like how, like, you know, M.
Leon Dore.
Do you know that brand?
I don't know.
How do you spell it?
A-I-M-E.
There it is.
Cool.
All right.
Let's drop that in too.
Something like that has a strong brand and probably has some ads somewhere.
Totally.
So we'll do them.
Who else?
I would be interested to do a SaaS app probably.
What's a big one?
you guys have done recently.
We do cursor.
Yeah.
Cursor is great.
All right.
Let's do cursor.
All right.
We'll do cursor as well.
Just firing these off.
And just to come back to this.
So idea browser, this is what it came up with.
So it's going to make five statics and headlines.
Stop guessing which idea to build.
Every idea is scored on demand, pain and timing.
Get today's free idea.
Headline, build what people already search for.
Headline, your chat GPT ideas are generic.
Headline, don't waste six months on a dud.
Headline, 1,948 ideas worth building.
Okay, so these are basically the ad ideas or the concepts that it extracted from going and researching online.
Like what are the pain points, desired outcomes of somebody that would buy this product, right?
So now I said, okay, generate these ads.
And out the other side is going to come the actual ad creative, you know, based off of these concept points.
And this is how you get good like outcomes to happen is you, again, you have to give the agent what it needs, the environment that it needs to do this work, right?
So let's come back to these other ones.
We'll see where they're at.
Oh, they have zero ads.
That's fascinating.
Cursor, cool.
So Cursor, we're pulling the Facebook ads libraries.
That's great.
And Late Checkout, still building the brand style guide there.
So let's do some other brands while we're waiting.
Trying to think.
Let's do David, just like the protein bar.
What other brands are you stoked about right now, G?
Stoked about?
Or just come to mind.
Let's do an old classic maybe of like, I don't know.
I think Robin Hood's a pretty good brand.
Cool.
Let's do that.
Robin Hood as well.
All right.
These are ready.
You know what's a really cool brand and product?
Teenage Engineering.
Oh, that would be sweet actually.
Yeah.
I'm going to say, okay, make these.
And let's go to Idea Browser.
And here's the statics that are starting to come out.
Wow.
So we'll take a look at those in a second.
And same deal here.
Make these.
And same thing.
Make these.
All right.
So got a handful of these.
Let's go back to Idea Browser.
Let's look at these pieces of creative, right?
So stop guessing which idea to build.
Again, this is just based off of the brand that it's pulling.
You can modify this.
You can also get it.
the skill files that define your brand style guides.
This is the markdown file.
What are the fonts?
What are the colors?
Again, this is just extracting from the site, going and doing that research.
But I've made these ads now, right?
Once I've made these ads, I can then go, and because I've already set up the Facebook ads API, I can say something like, upload these into the graphed Facebook ads account to a new ad set, right?
And it's going to go and upload those directly into it.
I'm just going to stop this for the time being.
And now what instead I'm going to do is show you how to take these same concepts and we're going to turn this into an AI avatar UGC that's doing a script read that also has motion graphics over the top of it.
So I'm going to do forward slash UGC, write a script for this.
Based on above.
So this skill file, it understands, okay, how do I need to go and make this seed dance ad?
And I just, while this is generating, I would just do these also or make them now.
Come back to these.
Those are still rendering.
So in the example that you used, the static ads were used using chat GPT image to.
The UGC ads are using C-Dense.
Is there an opportunity to like, let's say I don't want to use Image 2.
I get better results on Nano Banana or I get better results on whatever.
Is there a way to swap models out?
Yeah, absolutely.
So you like, for example, if you get a Kai AI subscription, you could use Nano Banana, et cetera.
With the unified API, it has all those endpoints.
So you just literally say like, you know.
change this to Nano Banana from ChadGBT Image.
ChadGBT Image 2.5 is the best that we've seen.
So like I would suggest it.
But anyways, so we're going to make this UGC ad now.
There's already a prebuilt avatar in here.
It just wrote the script.
I'm going to say, okay, make the ad.
And what's about to happen is a couple things.
So people, we see a lot of people being like, I just spent $50 to make this UGC ad.
burning like tokens do this.
So this workflow that we're about to run, how we've solved for this is we've done a handful of things.
The first thing is that we're taking the script.
So imagine, you know, this is a long script and that then goes and we do an ad read using not seed dance.
So the audio is actually using a foul voice clone.
And then this makes an MP3 file.
This is what, you know, comes out the other side.
It's just the MP3.
The next piece is we send the MP3 and also an image, right?
So imagine this is like the creator image, right?
This is a reference image that we're sending at the same time with the MP3.
And also the script gets sent as well.
So all of this goes to Seed Dance at the same time.
Why are we doing this?
Why am I even talking about this?
If I just send the script straight to Seed Dance, it's going to give me just like whatever it invents at that moment.
So what I have to do is I basically have to be like, here's this walled garden that you can work within.
The script, exact script, the exact voice example, doing the ad read.
So you're not like inventing, like if you've ever had Seed Dance where it's like, it's not actually enunciating the words.
This is how you can solve for this.
And then also the reference image so that the avatar that you get out is what you're looking for.
It's not some random thing that Seed Dance thinks about or makes up.
So all of this gets sent to Seed Dance.
And then out this other side, you get this ad that is ready for you to edit.
This isn't the final form.
This ad that you get is 480p, right?
So we'll say that this is the first version, the V1 of the ad, right?
That it's this video.
This is 480p.
Why are we doing that?
It's because if I did a, you can use a higher model, but it just, if you're trying to be cost conscious, this is the way to get around this.
So you can do like 1080p, but it costs you, I think it's like $14 for that.
In contrast, I can do a 480p.
And then I upscale the video and I add blur using what's called FFmpeg.
Again, you don't need to know any of this.
All of this is happening within the skill that I just used to make this ad, right?
The other thing that it's doing is it's doing word by word transcription and it's matching the script to the transcript, right?
So it's matching the script to the transcript to confirm that it's an exact read that's happening.
Now, say it mispronounces Calendly, right?
And it says like Calendla, right?
This is an exact example of that, like we had happen.
Instead of that Calendla now being in there, what it does, because I have the word-by-word transcript, it can isolate, hey, this word is wrong.
Calendla is wrong.
Great.
Here are the timestamps, right, of where that is said.
And I'm now going to go and I'm going to replace this with the correct pronunciation of Calendly, right?
And we can swap basically those mistakes.
And this is, again, this edit, this video editing that I just described here.
This is not a human.
This is the agent that is basically doing this work, right?
Doing this within Claude Cody or GrockBot.
Everything that I'm describing as well, I think that this is kind of the meta of this whole thing.
This can all run in a cloud.
So I can host a cloud code harness in the cloud and have it do this entire workflow.
So our engineers, this is what they do.
You want 100 pieces of creative using this exact process?
This is exactly how it happens.
It's hosted in the cloud.
It's not actually on the machine locally that this is all going on.
Out the other side, what I get from this, when it does this whole stack of things on the video edit side, I get this complete ad that is doing the pronunciation right, that's upscaled, and that looks actually like a human, like a human videoed it on their phone.
And the reason for that is we use a source image.
I forgot to mention this.
We use a source image that is taking a real creator.
and remixing basically that creator, right?
I'm not saying a likeness.
I'm saying like, we're like, change the background and make this person, you know, this age, this ethnicity, you know, et cetera, is basically what this ends up being.
And if you ever get that like plasticky, like effect on Seed Dance, this is how you solve for this is you give it, you know, a real human version for it to go and do that remix.
So I now have this video that is ready.
That is, you know, a UGC read.
But there's things I want to do.
I want to add captions and I want to add motion graphics to it.
How do I go when I do that?
So this next step is you use a product called Tesseract from Raj.
This is the exact thing that we're about to do.
And what it can do is it can take that word-by-word transcription that we have up here and it can add captions or it can go and add motion graphics to the video.
So let's go and look at this now.
Video is probably done.
It takes about five minutes-ish to render.
So let's see if it's completed.
And it's completed.
So let's take a look.
I'll screen share.
Cool.
So this video is now rendered.
And this is it doing the lip read, doing the sync.
It's done those entire edits.
And I can now go and say, okay, add motion graphics to this video using Tesseract.
By the way, she looks insanely real.
So we have hundreds of these and we can make any person type.
Like you name it, we can get it.
We've got ones like you're trying to sell to the blue collar.
Awesome.
You're trying to sell to somebody who's in a corporate office.
Awesome.
You're trying to you name the demographic and like or the psychographic and we can go and hit it.
So this cost four dollars to make about four dollars.
I think it was a little bit more.
We'll ask it at the end.
It says 475 was projected.
So this motion graphic piece will be a couple more cents.
So basically, I'm getting this ad out for about $5.
We're about to work with another provider.
We'll be able to get this down to about three is where we think we'll end up be or in that vicinity.
But I've got this ad now.
So say I wanted to go and make 100 of these or we'll say 10 of these, right?
That's 50 bucks.
I want to do 100.
That's 500.
If I was to go and get this made previously and add motion graphics to it, I mean, you're easily in the range of $500 plus in the past, right?
For just that single ad.
And this is this way around this.
Now, these ads, you don't have to only use these AI avatars.
The other thing that we're starting to do is layering on tools for...
Like example is Tribe, where you basically make a UGC creator network.
What's amazing now is instead of having to have a person that was managing all these relationships, you can use the API and your GrokBot and your cloud code to basically be the social media, like social creator manager, right?
The creator management person.
So this is like literally how we are doing this.
We use that.
We also use a company called SideShift as well.
You know, just two examples of like, how do you run these campaigns?
So wait, so how does that actually work?
Because so what you're saying is Tribe is a network of UGC creators.
You got it.
So we pay them like $40 a video as an example for a company that we are working with.
I would take this script.
And I would be like, hey, guys, here's this script, you know, a variation.
Feel free to modify it.
But this is the, you know, directionally basically where to go for this.
Right.
That via the API of SideShift or Tribe gets sent to them with, you know, a prompt.
And then you pay them on like a per video basis or a percentage.
There was a video that went viral.
I think his name is Hector.
Yeah.
I mean, I don't know him, but it's super legit.
Like what he's doing is the exact.
strategy that we implement.
We do it on a per video basis and then have like a bonus based off of this, um, the spend, but there's tons of ways to basically structure these.
But, um, what this turns into is you can have the human creators.
So you have, instead of the avatar doing the read, right.
You can have the human creator and then go to, okay, I want the infograph or I want the motion graphics to be added to that human creator.
So just a way to, to make this happen.
Right.
The exact workflow I just described.
We can do that for all of these different ones.
So here's David, right?
Built on science, backed by doctors.
Cool.
For late checkout, your product is getting a second user.
Your next product is your support ticket.
Every company can have a lab now.
Fire.
Some examples.
Cursor.
Finally understand the code nobody wrote down.
Looks like it's rendering the rest of them.
I'm really curious what it's going to do here.
So here's...
May.
So this is ad one.
Pay retail, not resale.
Bought once, worn for years.
The gift they'd pick.
Try it on, stay for coffee.
See what it did for...
So still working on the Robin Hood ones.
All right.
Every one of those.
If I wanted to go and make an AI avatar ad, I could do that.
I could also have the avatar hold the product.
You can modify the prompts for that.
Again, you just tell it, I want you to hold the product.
Here's the image variation, et cetera.
And it can do that.
You just provide those as the reference images.
And once those ads are complete, so we have this final ad that's doing the read.
It sounds human.
So it's doing the read.
It's adding the motion graphics.
It's doing the pop-up.
It's looking at this frame by frame.
It's adding the call to action.
And again, didn't do this perfect on the first shot.
You can have it just like be like vision model, go through this frame by frame.
You know, we need to basically like have it be aligned, et cetera.
That's something that we actually probably need to build into this.
But anyway, we've now made those ads.
Just a question on that.
Does it make sense?
I mean, you could do this on Claude.
You can do this on GrokBot, right?
So does it make sense to have a...
creative director, essentially.
Do the first pass and be like, go and watch what comes out as the first pass for ADP and score it.
Just like how a code rabbit or a greptile does code review eval.
It says this code is ready to push to production.
Don't even show me.
the video unless like it fixes a lot of those little issues and also go and make sure that it it's at least an eight on 10 or a nine on 10 absolutely you can 100% do that um I wouldn't even like make this person like I think people think about this as like employees but like I would just think about it as like jobs to be done like what was the human doing and how am I turning that into code right like you don't need god in a box to run a facebook ads account It's actually a terrible way to run a Facebook ads account.
How do I know?
We did this.
Multiple companies and it's absolutely terrible.
The better way is like, what's the top 1% media buyer doing?
How do I go and turn that into code with a thinking loop, with a live data stream that it's making decisions on top of?
That's actually what an agent, you know.
a live autonomous agent ends up being.
But to your point, yeah, absolutely.
Like you could just add a step to the workflow that I just described.
I'm going to like ship the skill literally right after this.
I'm going to make this update where it basically will look at the motion graphics frame by frame.
And it's like, oh, here's these adjustments that we need to make so that it matches.
It doesn't, you know, you can have a gate within there as well.
Like if it doesn't meet this quality standard, then, you know, it has to go through another loop.
This is actually how we do a static creative generation for.
any like regulated markets.
Like for example, we have clients that are in financial.
And for them, like, you know, they basically have all these approval processes.
There's things they can and can't say.
So we have an agent that like is a basically an image model that's like looking at this, like here's all these things that we can, can't do, can't promise, et cetera.
And it goes in a like recursive loop, modifying the creative based off of that.
So creative, you know, one gets made, vision model looks, yes, no.
okay, send it back, make these changes.
Yes, no, and it just goes, you know, we typically get in like two to three.
So it's like you don't have a lot.
But ChatGBT Image 2.5 is like an incredible graphic designer for specific modifications.
We've just found that it's very good at keeping the original image rather than remixing it again.
And all of these things are going to get better, right?
Like, yeah, there's a random C dance generation.
One out of 50 suck and it's still garbled nonsense?
Absolutely.
Will this continue to improve?
Yes.
It doesn't matter right now because the cost is so much.
This is the discrepancy, right?
It's like, again, I can pay $500 or we'll say $300.
Maybe you get this correctly done.
It'll be $300 to have the UGC creator, a motion graphics person, actually do this.
Or in contrast, I can go and pay $4.
Yeah.
95% is good out the other side.
And the funniest part with all this is like, we'll see these AI avatars sometimes perform better.
Same script read than the human.
Right.
And it's like people are like, well, why?
And it's like, I don't know why it just hits.
And that's the game anymore, especially on Facebook ads with Andromeda.
Right.
It's like what it comes down to is like you're just feeding the beast.
You're feeding the machine.
It is smarter than you.
It is more capable than more capable than you.
It knows more than anybody else in this world.
Right.
And it will just show the right ad to the right person.
All you have to do is give it like.
creative velocity.
And there's a lot of the work that we're doing right now is like companies are like sweet.
Like we were shipping 10 ads a month.
And we're like, awesome.
Like, let's go to 300 tomorrow.
We'll just ship a cohort of 30.
Right.
And, you know, it depends on the ad budget size and all these pieces.
But like.
And your ad budget in the CPL, like there's like a formula for like how many ads should I ship based off of this that I actually get through testing in these pieces.
But I mean, we'll see this, especially for software companies where it's like, we'll ship a cohort of 30 and like an ad will get like two conversions, like two demos booked at like.
you know, $50 and maybe their average is like 150.
And then it would just stop spending on that ad because Facebook found the two people that will convert.
And then it's like, all right, cool, we're done here, right?
Let's move on to the other set pieces.
And so again, you just have to like feed this machine.
And everything that I've talked about is how you actually feed this machine.
How do you give your agent the necessary resources to be able to go and do this?
So now we ship those ads, right?
We upload those into Facebook.
Because I have the data solution, I can be like, what is working?
And all of this is just code, right?
All of this is just code and tokens.
When I'm sending, like here, we can do this right now, but like when I'm sending that prompt to seed dance, to generate the ad, right?
It's just a JSON blob.
It's just code that's being sent.
I can store that JSON blob in a database and I have literally like at, you know, the cellular level breakdown of like, what is this ad?
I can tie the data on Facebook directly to that ad, that JSON blob.
And now imagine that for a thousand ads, right?
That's the scale that this is getting to.
This is the game that's happening.
This is like why you have to do this.
This is where every company is going to be in five years.
Like you're going to have to do this to compete.
You will not be able to win unless you do this.
Basically have the infrastructure and then like have the engineering talent to be able to do this work.
The one step I would add on the creative director piece.
What I would be doing if I was creating this, and I'm going to go and build this system, is have your creative director agent go and ingest David Ogilvie's books and 100 of the best YouTube podcasts of the last 24 months of the world's best creative ad people and their podcasts and go and Now you understand all the fundamental techniques.
That's one.
And then two, the human being, going back to the swipe warehouse, how do you create a swipe warehouse of just tons of really cool examples and interesting examples that you're just constantly adding to?
And there's a bunch of tools that you can use to manage that.
You can use, I think, sublime.app is one.
you know, cause you want to make it multiplayer.
This is in graft.
Again, I'm going to do this plug.
Like, you want to ship a tool to your whole team?
Like we have a series a startup that like five people are working on the accounts across marketing and sales.
And like, as they discover new workflows and things, they just uploaded into it.
And now the whole team has access to it and all of the tool usage and all of the data usage, all of that is there.
Yes.
That like repo of that storage of knowledge, like that is one of the most valuable things that you can do.
So, yeah.
Is there anything you haven't shown that you want to show or is that it?
Maybe we do another one though, G, of like I come back and we'll do like an outbound version of this.
Like how do I build a total addressable market map?
And how do I go and find all their emails using a waterfall enrichment?
And how do I get an agent like GrokBot to go and cold email all of these people, right?
What's the environment that it needs to do that?
We can do that as well.
We can do it for SEO.
Like we can do a whole series on this if that's something that the audience is interested in.
Let Greg know below if that's what you want.
I will come and teach you literally how to be a marketing engineer anytime that you guys want me back on the show.
Amazing.
All right.
Well, thank you, Cody.
This has been incredible.
Yeah.
Gee, one other shout out, if you don't mind.
I'm in San Francisco for the next two weeks.
We're raising our seed round for grafts.
If any of what I just described resonates with you and this idea that like there needs to be a cloud for marketing engineering, that is what we are here for.
And that is what we were building.
Reach out on Twitter.
You can find me at Cody Schneider and you can literally ride along as we do this, you know.
basically fundraiser just turning it into an event so cheers thank you man love it thank you cody