# AI Agents for Autonomous Marketing Operations

**Podcast:** The Startup Ideas Podcast
**Published:** 2026-03-02

## Transcript

How can you use AI agents, MCPs, and a bunch of different tools to make money on the internet?
Today we walk through it all.
Yes, you can vibe code anything and right now, and that's great, but how can you actually use AI agents to get you customers 24 7?
Well, today we live build it.
We actually spun up 10 cloud code instances, and we show you how you can do it to help you get customers on repeat.
I loved this episode.
It's my friend Cody Schneider.
He's an absolute legend when it comes to vibe marketing and growth marketing.
This episode is saw C.
And by the end of the episode, you're gonna feel pretty confident you know what to do.
You are in for a treat.
Enjoy the episode, and I can't wait to see you in there.
The startup bag is podcast.
Cody, by the end of this episode, what are we gonna learn?
You're gonna learn how to build your first agents that allow for you to go and build personal software to do marketing, sales, growth, customer experience for yourself.
And by the end of this, you're gonna come out of it with this whole new tool set that allows for you to do all of the middlework without touching a keyboard.
You're just gonna use your voice and have agents do work for you in the background, man.
It's gonna be crazy.
Okay, and can you list off a few of the piece tools and pieces of software we're gonna use?
Like, yeah, absolutely.
We're gonna touch Phantom Buster, we're gonna use instantly AI, we're going to use Vrafonic, we're going to use Railway.com.
Uh, we're also going to use um a bunch of different other tooling that's in my go-to-market stack.
So we're also just gonna use like the Facebook ads API as an example, uh, as another uh uh just like, you know, way that we're gonna interact via uh this uh this agent harness cloud code.
All right, and we're gonna live build it and everyone, well, you're gonna watch, you're gonna watch the whole thing.
So let's get into it.
Cool.
Sweet man.
Um, so just to begin with, gee, do you do you know like GTM engineering or like what it even means or like where it comes from?
No, honestly, I don't.
That's what I'm saying.
It feels like a buzz, it's just a buzzword, right?
So this is it's actually like made up by clay.com, which is hilarious.
Um and they originally did it as like uh a way to explain somebody that like does basically like cascading workflows for like data enrichment to do outbound sales motions over email or Slack or you know, it could be like cold calling.
Um, so that was kind of the origin of this was like it was just basically this term that was given to it, but it's quickly evolving into something entirely different.
Um, and so let me screen share and I can just like show you like what we're seeing this work as now.
Um, but basically, like what we're like seeing is that the can you see this all right?
Yeah.
Cool.
So how I'm thinking about it now is like basically everything that used to be the middle work that we would do, like all of anything that I would do to touch the keyboard, I'm now passing it on to some type of agent hardness, um, whether it's claud code or it's codecs or any of these tools.
And so my job suddenly turns into like I have ideas, I pass them on to Claud Code, and then I'm basically polishing the end product.
And it enables me to do like things at scale that that were just previously impossible.
And to just to give you like a taste of like what I'm talking about, we're gonna do this today.
Like build a hundred Facebook ads, publish them to Facebook, build a dashboard to track that, analyze the data within Claude Code, have it turn off the Facebook ads that are the low performers, have it bump up the Facebook ads or that are the best performers to a new ad set with its own dedicated budget, and everything that I just described, that happening in like literally you know 30 minutes.
So I'm anyways, again, not really sleeping.
Um, so this is kind of where it's at now.
And I'm gonna talk through like this whole setup process and actually how to do this.
Um, and then I'm gonna talk about where it's going, like how agents are the natural evolution from this.
Basically, as soon as you like start, you have this epiphany of like I can get this thing to do work for me, then you suddenly have this uh like you come to come to Jesus moment of like, oh, I can just deploy this onto a server and now it's doing this task for me in the background.
And I'm building out this personal software for myself, for my job, for my my you know, tasks, et cetera.
And this isn't some like hype thing of like go do open claw and give it access to everything.
I'm talking about like specific, like jobs to be done workflows that are custom made for how you want to operate in your day-to-day.
So that's kind of the high-level man.
I any questions I can try to answer, have to go deeper on anything.
No, um if you can teach me this by the end of that episode, I mean, that's sort of the that's I think the question that a lot of people have in their heads right now.
Like, how do I how can I do that, right?
Because that's gonna be an unfair advantage.
Um, so yeah, let's let's go, let's go through it.
Perfect.
Let's jump into it, man.
All right.
So, first off, what you need to do if you're watching right now is I want you to go and I need you to create a folder that you're gonna start living out of.
So the one I live out of is called uh graft growth agents.
So everything I do now where I start my work, it all exists within here.
And the first thing that I'm gonna have you do is you're gonna set up an environment file.
And this environment file, it just holds all of your API keys that you're basically going to be working with.
So what I'm doing is I'm basically having, and I'm just not gonna show this just because it has literally all of our API keys for everything, but it has uh I can open up this example one.
Um, so it has uh like intercom, it has our send grid API, my HubSpot API, my cow.com API, my perplexity API, my Facebook ads API, uh million verifier, instantly everything that I live on top of that is a part of my day-to-day growth stack.
This is like what I'm working with, basically.
Um and so what what this translates into, or like why I am I'm why you start here is you're basically starting to interact with everything that you do on a daily basis via the APIs.
And this is actually how I'm thinking about everything I do now, and like how I buy software in particular, is how robust the API is.
It's funny, I was talking to a friend recently, and he's like, if you're looking at Salesforce versus HubSpot right now, Salesforce, even though it's like historically a more clunk like clunky CRM, it's actually the better product for this AI foundation because it has a more robust API, so you can do more with it, basically.
Um, and this is what this like turns into is you're all of the work that you're doing, and we're gonna do this together today, is going to be happening from this like repository.
When I say repo, all I mean is just this like folder that we're living in that has all of these files.
And I'm gonna be using Cloud Code like throughout the rest of the session to basically be building out this personal software and be building out uh, you know, actually doing work, if that makes sense.
So that's kind of one component of it.
The last piece um is then I would strongly suggest get suggest getting something like a super whisper or uh any of these other transcription uh softwares because it enables you to just so quickly go through the process of building out like what you're trying to do on the distribution side.
And then optional is just installing the Claud Code front end design skill.
I've just found this to be like one of those things that it, you know, if we're gonna generate a UI, it's nice to have it look pretty.
So, all right, that is kind of the foundational pieces.
Now let's actually like great, that's cool.
You've just built this.
What do you actually do to go uh get started on this?
So the first thing that I'm gonna do to get started is I'm going to go and I'm gonna have uh Claud Code start responding to people on LinkedIn for me that have asked for an asset.
So I've been doing all of these like giveaways basically.
Here's one that's an email triage.
Um I wrote this uh giveaway doc, you know, notion document.
I'm now gonna go in and I'm gonna get this agent to start running for me in the background while I have other work going on.
So I've got the Claude Code or sorry, I've got the Claude uh Chrome extension installed.
And I'm gonna go and I'm gonna say, um, so I'm working out of this uh that directory, right?
That I've already been in, and I already have built this uh basically skill and it's a piece of software that will go and comment on everybody that asked for this uh asset.
So I'm gonna say this right now.
I'm gonna say uh uh let's run or we're gonna transcribe it.
So let's run the LinkedIn uh respond uh uh software.
Uh the uh keyword that you're looking for is triage.
I'm gonna provide the notion documents and the LinkedIn post URL.
And then I'm gonna select the uh post URL and I'm gonna put that in.
And then I'm also gonna select uh the notion document that I want it to do as well.
So I'm gonna give it that.
Uh it's gonna start running.
So what it's gonna do right now is it's gonna basically open this up.
I'm gonna babysit it for a moment while it can while it starts this process and just make sure that it starts on the right path.
And then once it's on the right path, then I'm gonna go and basically start on the other things.
And actually, while it's thinking, let's just go to these other uh places.
So the next thing I'm gonna do uh is I'm going to uh go or sorry, it that just opened the LinkedIn profile.
Let's bring that back over here.
Um so this is now running.
Uh perfect.
And so this should now start commenting on those responding back to those people.
I'm just gonna change this to most recent just so that it works backwards on this.
And then we're gonna just let that run in the background.
So while that's happening, um, what I'm going to do now is we're gonna build a Facebook ads generator.
So um I've been doing this where it's basically a and I'll show you an example of what this looks like.
Let's just go over to uh LinkedIn and I can give you an example of the output that we're going to actually create today.
So it's basically a bulk generator of ad creative.
We're gonna create this template, and then I'm gonna go and do research based off of uh Reddit and other social media posts for the pain points that people experience, and then we're gonna go and bulk generate all these uh variations.
So let's get that started right now.
So I've got again those API keys are stored within here, and I've also made that uh that skill.
So I'm gonna tell it right now I want you, uh I want you to create a bulk Facebook ad generator.
It's gonna be a 1080 by 1080 pixel image.
Um, what's gonna happen is uh I'll give you an example of what one of these ads looks like, uh, and then uh we're gonna go and build a template around that, and then we'll I'll basically uh create a uh or give you variations of text, uh both titles and paragraphs that we I want to be generated.
It'll be a zip file that we download um uh for the beginning.
Um can you make this into a UI as well so that we can visually see uh the uh creative?
Um uh for the first thing I just want to be able to see is like what the actual creative will look like.
Um for this, you're going to use uh just React components.
Um so I don't want you or like I just purely build it with React components and then uh also uh to actually change this React components into a PNG that's downloadable.
We're going to use um HTML to Canvas, it's just uh uh uh you know resource that you have available for you on that.
Uh ask questions if you need.
So I've just transcribed that.
I'm now going to put collad uh into uh plan mode, and I'm just gonna let that start running in the background.
All right, so while that's running in the background, let's come back here and let's see the work that it's doing.
Uh so it's going through these, and I believe it's now commenting.
So that's happening.
So we'll just let that run uh in perpetuity and respond back to them.
All right, uh, so next uh I'm just gonna just click through this quickly.
Um I'll share it now.
I'll share it uh I'll share it after setup and then input method, form-based UI.
Let's do both, and then I'm gonna hit submit.
All right, so now that's working on that in the background.
I'm gonna open up another folder and I'm going to start claud code again within an entirely uh another differ or another window.
So I'm gonna do documents, forge slash graphs.
Uh, let's go to uh agents uh and then demo.
All right, so the next thing uh that I want to build uh as an example um is uh I'm going to uh basically pull information um for so I just I I just did this actually so we could like talk through um and uh about what this ends up looking like, but basically scraped all of the podcasts that were within the marketing category and then built a workflow that goes and cold emails them and then an agent that responds back to book me on the podcast.
This ends up turning into way better performing than I expected.
This is what my week looks like.
That's crazy.
Sorry, what is instantly?
Yeah, so instantly is a cold email software.
Um this is just a part of my stack.
So it's one of the things uh that like is within that environment file that allows for me to build on top of.
And so what I'm uh like how how I can think about this, like on the uh is it's basically like my manual workflows that I would do previously.
We're just like daisy chaining those together, like using this software.
So I'm gonna bring this into a new desktop and let's just rebuild that whole thing.
I'm gonna say um, you have the Raphonic API key.
Um I want you to build a software that scrapes uh podcast uh host emails from Raphonic.
It then sends it to uh million verifier to verify the emails, and then it also uh will then send it to an instantly um uh uh a campaign.
I'll provide the instantly campaign that I want it to send to.
All right, so I've got that now.
Um I'm gonna put that in the plan mode and let that run as well.
Um, and then we have that as its own window.
And so while these two are working again in the background, um, we'll we can then go and actually do uh like some other work.
Uh so let me get this going and I'll just say cool.
So that's the plan.
All right, so now this is my we're in this folder.
This is like my demo folder, which I just like every time I give this presentation, I just nuke.
This is kind of to show you like how to start it from zero to one.
This folder that we're in now is my actual folder that I live out of.
Um, and I'm just gonna show you some of the things that like are like capable with this.
Um, so for example, I have it attached to Notion, and I've basically given it a uh um an example of like how do we write a uh uh like a notion like giveaway, right?
So I'm gonna go and we're gonna create one of these together right now because I need to actually accomplish this.
So I'm gonna give take this URL and then copy this over.
And I'm gonna say, okay, um write a uh or create a notion document.
Uh based off are you typing with your hands.
I know the problem here we can do it in the the transcription of so create a notion document based off of uh our like current our you know our our structure look for the skill that has this.
I'm gonna provide context on what that should include.
Um you should uh uh incorporate uh stuff that we have we haven't within the repo, like the documentation that I have in the repo on how to do this.
All right, and so I'm gonna copy this over, let that run.
And now it's gonna go and create me a notion document, just like the one that we have being sent in the background here currently.
All right, um, so in that folder, I've already created the uh bulk ad generator.
So I'm just gonna go in there just to show you like what you can do with this once that it's it's like you've actually gone through this process of like zero to one making this.
So this is the bulk uh the creator as an example.
So um that's gonna continue working on that bulk Facebook ad generator in the background.
While that's happening, let's go to graphs uh growth agents, and then I'm gonna start clog code within there.
And now I'm going to uh start locally the bulk Facebook ad generator.
Uh I just wanna uh bring that up uh within um uh my uh uh local uh so that I can uh create some ads.
So again, it's already created the software for me.
I'm now coming back to it, and I'm gonna talk through the whole process of the actual creation of this because it'll just like make more sense uh momentarily.
Um but the goal here is basically what we're trying to do is I'm uh trying to create as many different variations of these ads as possible.
So, how did I make this ad?
Um, this is entirely code, and I think this is something to like double click on.
Um, everything in here is a just it's just React components, like that.
This is this is just a React component and entirely built by code.
And so I can make an infinite amount of these at scale, and I've done this already.
So you can see this bulk forward slash or forward slash bulk forward slash HTML.
Um, so we're gonna go through this together though.
How do we actually like do this process?
So I I would go and I would give it an example, and we'll do this over here while this one is like working on it.
We'll come back to that.
But I would give it an example of what uh like ad I'm trying to build.
A way to do this is uh if you don't know where to start, go to Facebook ads library and you can see what your uh like competitors or other software companies are doing in your category.
This is actually how I made this initial one was I found like this before and after format, and then I basically had it build off of uh that before and after.
Um, but this whole thing, like everything you see here is just code, it's entirely code.
Um, the other way you can do this is with something like nano banana, where you're like going and bulk generating these, right?
Um, but once I found and I built that template, I can now make all those variations.
So let's go back to Cloud Code and uh let's say okay, I want you to use the Facebook ads API, and I want you to go and scrape uh the pain points that you see, or sorry, uh actually let's restart that.
I want you to use the uh perplexity API and go and scrape Reddit for the pain points and the outcomes that growth marketers wish they could have uh from like something like a Looker Studio um or any of these other business intelligence softwares that uh they're using.
Uh we've uh focusing on like the the data analyst component of it of how they can't get bandwidth or it's too complicated to get started, or they can't unify their data all into one location.
You can also source from YouTube, you can also source from uh Twitter if necessary.
So I'm gonna have it go do research and those pain points.
For the ad, sorry, for the ad itself, wouldn't we want to use Nano Banana Pro, like the best image model that exists?
Like, why are we using code when we, you know, when we couldn't do that?
Yeah, I'm just doing this purely like this was just a way I thought about doing it.
Like, like absolutely the nanobanana thing.
Um the the only thing I found with like nano banana is that I I sometimes have trouble like getting it to stay on brand.
And if I'm trying to just like figure out the messaging variations that I'm trying to go after, this can be a faster way to do that.
Again, there's like a million different ways to do this exact same thing.
Uh I I think yeah, if you're going to use nano banana, you should look at something like um I think it's like Kai AI, I believe.
Uh nano banana.
Or yeah, chi uhai.
Um, and you can just like basically it's one of these bulk buys.
Um, but we've been using that for these bulk generations.
This that I'm doing costs me nothing.
Like it's literally on, you know, maybe a thousand tokens to do all of these, like these generations.
And so this that's a reason for it.
So it's like I can go, we could go and create a thousand ad variations right now, G, and like this literally costs nothing.
Um, that's that's a part of it too, right?
Is like your goal is going to be come up with the best ad creative that's going to actually, you know, you put in a dollar, you get three dollars out.
Once you get that, uh, you can make it.
Well, there's there's two schools of thought.
One school of thought is you need the best creative.
So you need to send it to Nanabanana Pro to get scroll stopping creative.
Another school of thought is like, well, you actually have some pretty quote unquote ugly ads that just speak to the IC the pain points that you can kind of get a good understanding of this is gonna bring you a dollar fifty uh when you put in a dollar, and then you can get it from a dollar fifty to three dollars once you figure out the ad.
So you're kind of saying maybe it's best to like get as many ads as possible, start using the those, figure out the one or two or three creative that actually crushes, and then go ahead and go crazy with spending tokens.
Totally.
Like the the I guess a different, like I'm just trying to find like the the format or the angle that's going to be most receptive.
I'm gonna remix that a thousand times, you know, after it.
Um the this is like the big piece of this, especially with like how much we can do um on the uh like the the anybody can go and generate as many of these as they want, right?
Like it's literally infinite.
But identifying those winners, like you're talking about now becomes the challenge that you're going to face with all of this.
And like we're gonna get to that in a second, but the the the main thing I want to emphasize with it is like this is malleable and flexible, and the spit the pace, like just think about you manually having to go create 50 ad variations in Figma, and like you can just now make those and get those live and test those.
And as soon as I find a winning like format from a uh uh a language that's being said perspective, I can then go and remix that into all these other different templates.
I can go and find like what are different winning ad formats that I can now port this to, but this is a way to just like start immediately and then get that basically out into in in in public.
I also think that like the same ideas can go into different formats.
Like we're all you know already seeing this where it's like cool.
I made you know a static format, I'm now gonna port that over to a UGC format, and I'm sending that to the Hey Gen API to pull in that uh um, you know, like to make that creative, like pull in, uh pull it in uh as a video and then bulk upload that to Facebook, if that makes sense.
So I I like where I'm going with this or where I'm seeing this head personally is like I'm gonna build these tools that an agent is going to have, and then it's gonna be able to run this process in the background where it's basically has the ability to make new creative.
It can publish that creative to directly to Facebook, which I'm gonna show you in a second.
Um, it's gonna then analyze what is working, and then it's gonna like basically turn off what isn't and promote what is, right?
And this is like everything I'm gonna show you today is at like a small scale.
Like, this is literally like weeks of realization that's starting to happen, like with this.
But I just again I I want to plant the seed of like what is possible using this tooling to like do your again, that middle work that historically like you wouldn't be able to do.
So yeah, it does is there questions about that I'm gonna try to answer?
No, let's let's keep cooking.
Awesome.
All right, so this keeps like trying to scrape things.
I'm just gonna be like um instead.
I want you to just brainstorm uh pain points that people have uh with data reporting, specifically uh the unification of the data into multiple locations.
Cool.
So we're gonna do that.
Um once I have all those variations, I can then say, okay, now go bulk generate every one of those.
And then at that point, I can download these as a CSV.
And what I will do is just wait for that to uh while that's happening.
We'll just get this to start downloading.
And that Facebook ads API is gonna allow for us to bulk upload all of those pieces of creative that we just downloaded.
So here's all those variations.
I'm gonna say now, okay, now uh use the bulk or let's do the transcription.
Okay, now use the bulk uh Facebook ads generator uh to go and create these variations.
Um put them in the forward slash bulk bulk.html uh page when you're done with this.
So this is now happening.
I'm gonna go back over to see what the other things are going off of.
All right, so let's see this notion document based off of the current.
We'll continue to let that happen.
Um I want to build this uh perfect.
We're gonna let that go as well.
So I'm looking at the scaffolding to build the bulk ad generator, and we'll check in on this one to see where it's like basically at in its process of responding.
And it might have completed, and it did complete.
So that's done.
It ran for 15 minutes uh on its own in the background.
These are coming back.
Um so while these are all happening and I'm waiting on them, or they're waiting on me, I can then open up another one of these.
So I'll just click through this uh to just kind of get it moving.
Um, but I would then open up another one of these tabs and I would start on the next project.
So the next thing that I want to do is I want to build a LinkedIn um engagement scraper.
Um, so I'm going, so basically at people that engage on the LinkedIn Pro or the LinkedIn posts, I'm gonna pull them out and then send them to a um uh basically add them into instantly.
So we're gonna go, we're gonna find their LinkedIn profile using Phantom Buster.
We're then going to uh you know do that whole flow.
So I'll I'll do that in a second.
Let me just get this up.
Give that its own section.
By the end of this podcast, you're gonna have like a hundred desktop.
This is literally how I'm working now.
This is like I'm just jocking agents across, and then if I can automate them and get them to do like, like if I can figure out, okay, this is the specific lane that you can focus on.
Then I'm spinning that up onto a server on Railway, and I'll talk about that in a second, on like how you can on demand create databases and on demand create these servers so that this software starts running in perpetuity.
So we're gonna do this demo one.
All right, so I want to make a workflow where it's a uh I basically within Slack, um, you'll do forward slash LinkedIn post, and uh anybody in Slack will be able to just drop in a LinkedIn post that they think is a good fit, and then that's gonna go and it's gonna use the Phantom Buster API to extract all of the engagers, and then it's gonna take those LinkedIn profiles.
We're gonna go and enrich those with the Apollo API, and then from there, we're gonna send it to the million verifier API.
And then finally, we're gonna add them to an instantly uh uh campaign.
Ask me questions if you need.
All right, so I'm gonna turn on plan mode for that, let that start running in the background as well.
We'll come back to these, let them continue to cook, publish the landing page.
Well, let me get context on what this one is again.
This is the craziest part when you're going from screen like uh screen to screen and realizing that you're an agent jockey, yeah, and that you're trying to get context on each one.
You're like, okay, what was this one doing again?
And I I find that the context switching is actually difficult.
I I did as well, but now it's like now it feels like I like that has expanded.
Like, and again, this is just how I've been working for the last like six weeks.
And it was like maybe I could have like two or three of them in the beginning, and now it's like I'm comfortable with like we could have 15 windows open.
I'm about to literally go buy a new computer because I'm like, I need more RAM, I need more like ability to do this in the background.
Which just sounds so stupid.
Totally.
Yeah.
Like and comment this video so that you know I could send some YouTube AdSense revenue to code.
Yeah, I didn't know.
Or you can get some more RAM.
Uh it's all ridiculous.
This guy needs some RAM.
No, man, we I it's like I'm just realizing like what this turns into.
So okay, we bit, we made all those pieces of ad creative, right?
I've got those variations, and it's just text variations.
All right.
So now I'm gonna go back to Claude and I'm gonna be like, I want uh now I want to buckle a bulk upload all of these ads as drafts into um a Facebook ad set.
Um here is the you or here's where the um folder is locally for the um creative, and I'm gonna provide the Facebook ads, uh ad set URL to you in a second.
All right, so I'm now going to go back to Finder.
And I'm going to copy this, paste that in, and then let's go back to Facebook.
And I have this uh ad set that I've already created for this demo.
It's basically just here.
And I'm just gonna paste in this URL.
And so now here's that URL.
It's gonna uh basically uh bulk upload all of those pieces of creative into that ad set.
Um so while that's happening, I'm now gonna go and I'm gonna create a dashboard about this.
Um so let's just pull out the ad set ID.
Let's do the ad set ID, and I'm gonna go over to Graft and I'm gonna be uh pull up Facebook ads as the data source, and then I'm gonna be saying uh this is the ad set ID.
Make a dashboard showing clicks over time.
Um also have a scorecard that or sorry, I didn't do the transcription, make a dashboard showing clicks over time uh as a line chart.
Also uh within that line chart, can you include uh um the cost and the CPC as lines as well?
And then uh add a scorecard also that has total spend, total traffic or total clicks as another scorecard.
So those are two separate ones.
Um, and then I want you to also show demographic data as a bar chart um of showing the ages.
Uh so that that's its own separate chart.
That's a bar chart basically showing uh the impressions by the age uh categories.
So I'm gonna let that run in the background.
Uh we'll come back to that in a second.
Um, but now that I'm have this campaign that's running and I'm trying to track what's happening within it, I can basically go and uh like build out a tracking dashboard for this.
Um the other thing that I can do, so once it's uh got the ad set, what destination URL would you like?
Um just put them uh as a draft.
So uh while that's working, I can also analyze what is happening in that specific ad campaign.
Um, and I can turn off the losers of that ad campaign.
So I'm gonna show you how to do that.
So I do documents forward slash graft, forward slash uh growth agents, and then claw, and then I'm gonna get that URL again, and I'm gonna say um use the graft MCP to pull in the data for this ad set that I'm about to provide from Facebook ads.
Uh I want to look at uh the CPM data uh to see which ones are the lowest performing, like the highest uh CPM price.
All right, um, and then let's provide the ad set URL again.
Uh let's go ad set, and then while that's happening, we can come in and check on the other ones.
All right, so it's now built that entire bulk Facebook ad generator.
Um, or sorry, which one is this?
This is the look one.
Okay, so it's made the updates uh to these ads.
So these are an entirely new uh ad set that it's basically pulling in.
Um so instead, I already did this, right?
Of like bulk like downloading these as a zip and bulk uploading them.
So we won't go through that process again.
But this is how easy it is to basically make those variations.
And not just variations, these are variations based on pain points that people have said publicly.
Yes, exactly.
So it has pulled in basically the social dialogue that's happened.
So the best ads that I'm seeing perform right now are basically your selling outcomes or you're talking to the pain points, right?
And so I'm I'm just guiding it to focus on those things, pull me that information, and then build the ad sets around those.
So while that just happened, um, this in the background, just use the graft MCP to pull in all of the low performers.
So uh these are the let's look at the ones that have the highest CPM.
So these all have high CPM.
So I'm just gonna tell it turn these off.
Uh I'm gonna say uh use the Facebook ads API uh to turn off uh these ads with the this ad name.
And so now it's just pulled in this live data uh from my data warehouse.
And this isn't an MCP that's interacting with the Facebook ads API.
I just want to like emphasize this.
So it's not have you're not running into rate limits.
Like I again, if you go and publish like a thousand ads and are running those variations.
There is literally no way, like if you're spending enough, like there is literally no way that you're gonna be able to analyze this data without a data pipeline and a data warehouse.
And so this is like what we've built, right?
At Graft.
Um, so anyways, uh the just to get back to the MCP thing, there's like this page nation problem.
So like we see this all the time where people are like, yeah, I plugged in a Facebook ads MCP, I'm interacting with it.
And then I realize that I'm only seeing like 5% of the data that I think I'm actually seeing, right?
Um, but so coming back to this, I've now said, hey, turn these off.
So it went and it paused those ads for me.
And so it's just just to walk through what we just did, just kind of reiterate this.
We just did ideation, we just did bulk ad creation.
We just analyze the data for the performers.
We just turn those off and on based on that.
And at this point, you're probably starting to have the epiphany like, oh, I can just turn this into a repeatable process.
And this is where I see all of this going, basically, is you're going to going to have these agents that are running on top of your live data.
They're analyzing it, making decisions based off of like the model.
So, like, for example, how I would run this is I would have a test campaign where I'm basically testing new creative constantly.
I would have a cron job that's on a daily basis, basically going and turning off the low performers, and then the high performers, they get bumped up into their own ad sets with their own dedicated ad budget for a CPA action.
And then that whole thing could just run automatically in the background.
And then to track that, I'd build out a dashboard that basically is showing me, you know, that information.
So I can come back to that like later on and see what's occurring there.
And then the other thing that I would then go do uh is uh uh like potentially have a conversation in the morning.
Say for example, um, so I've got the graft MCP uh in uh my cloud I um uh like chat.
Um so this is also technically on my phone.
Um so like in the morning, I'll wake up and be like, how much traffic?
Let's just do this.
How much traffic went to uh the or how many new users went to the home page of the website yesterday?
Uh and I'll just say use um the graft MCP and uh Google Analytics for.
And we'll let that run.
And so I can basically get a brief each morning, whatever those KPI metrics are that I care about, and have a conversation like with my data that's live and being seen continuously within the background.
And you can give this to your whole team as well, um, so that everybody on your team also has access to this, both from within cloud code, within their like whatever their harnesses that they use, whether that's chats GBT or cloud, and then also the ability to like do that tracking within like dashboardings or conversations.
Now, coming back to all of this, right?
We've just built out basically this whole like cycle of funnel.
How would I now go deploy this?
So, this is where it gets like the most interesting.
So, what I'm doing right now, say I wanted to turn this into an agent.
Um, I'm using uh Railway for this, and the only reason is just because I saw a tutorial and that's how I've like figured out how to do this.
So Railway has a really robust API key.
And say I wanted to spin up, for example, this bulk ad generator, so that my other team members could use this, right?
They could come back and basically like use this software that I've created.
I can just tell Railway via Cloud Code, hey, spin this up into a um uh uh uh uh like a server that I can access, um, or I can just deploy this directly to Vercell.
Um, or any of these workflows that I have.
So, say for example, we were talking about that LinkedIn um uh funnel.
Uh, let's uh go see which one of this it is.
All right, so this is the podcast software.
Um this is the image generator.
This is the LinkedIn.
So say I wanted I want this to be accessible and perpetuity in the background.
I can then take the software that I co-work on with Claude, and I say, okay, deploy this to a server on Railway so that my whole team can basically use this action or always be adding, like whenever they come across a LinkedIn post as an example, they could be adding that into the queue so that it just automatically goes into the email like filtering.
Uh or sorry, the email, like cold email process.
Um, and this even goes further.
So, like how I'm starting to use this G, and I'm curious, like to get your thoughts on this.
So basically, like I had to do some data analysis work the other day.
Historically, I would have like downloaded the information, put it into Excel, and then I do a bunch of pivot tables.
Now instead, what I did was I down, I went to directly to the URL.
I had it push it into a Postgres database that I on the fly created using the railway API.
It just pumped everything in there.
I then did the analysis together with Claude.
And then at that point, I basically pushed from that Postgres database the outputs uh to the location that I wanted.
What have used it would have it would have taken me probably five hours historically to like clean the data appropriately.
And I smashed that out in like probably 20 to 30 minutes.
Um, and then as soon as I got done with that database, I just spun it down.
And this was the most like interesting part of this was that like it was basically like on the fly UI, on the fly, uh or the the epiphany I had was like on the fly UIs, on the fly databases, like on the fly software is going to become the standard for these people that are like you know, working at the forefront of this.
So um, yeah, man, I I we could probably, you know, sit here and watch me work for hours if you wanted, but that's I kind of everything I had that I wanted to show you today.
The only other thing uh I'm just like keep getting asked, like, how do I do this?
How like show me more technical details?
Um I bought the uh domain GTMengineering course.com.
I'm gonna give this thing away for free to everybody who wants it.
Um it'll be entirely public.
Um, we've already got a wait list of a hundred.
Um, but basically I'm just in the process of building this out with like step by step, and it's everything that I do, I'm just gonna document into one place.
But anyways, uh just throwing that out there as the last thing.
But any I I'd be curious on like okay, you see you just watch this and you're in like a role in some company.
Like how do you how do you defend against this?
Like with your job, or do you is it just like you need to learn this now?
And like I'm I I want to hear your thoughts because I you're seeing way more than I am with everything.
I'll tell you my I'll answer a question, but I want to start by saying like my biggest takeaway from all of this.
Yeah.
So my biggest takeaway of all this is when you connected it with Railway.
And I I it's a glimpse into the future of autonomous marketing.
So marketing, you know, all you you basically what you've done, like all those sort of uh jobs to be done were were jobs to be done that were literally done by human beings, right?
Um, and then you kind of stitched together, you know.
You know, if you've ever run ad campaigns before, you know how painful some of these things are.
Like just uploading the ads alone.
I was like, I have literally spent like I I mean, I'm just imagining uploading a thousand ad variations, like dude.
Like it just GTSD on the pod.
Straight up.
I I did this early in my career, it was absolutely painful.
It was painful, right?
It was painful, and and it's not fun, it's not fun at all.
Uh removing and figuring out low performers, uh, at a creative, not fun, not fun.
And you need to be on it.
Um, so the idea that you can, you know, make this an agent that's working 24-7 and that's managing all these different things, is the dream.
It is absolutely the dream.
Autonomous marketing, the dream.
I think uh who are the winners and who are the losers of this?
The winners are gonna be uh you know, one person businesses, small teams.
Um and then maybe you're your head of marketing that currently you're getting paid $100,000 a year.
Now all of a sudden, if you can figure out how to do all these things, and this is where I'm answering your question, if you can figure out how to do all these things, um, you know, you could make the case like, hey, triple, you know, triple my salary easily.
Right.
Like from a value perspective.
Like, like if you can do all these jobs to be done, you're one person instead of 10.
There is a case to be made that you've made, you know, you've added a tremendous amount of value to your role.
So I think uh, and then the unfortunate thing is I think a lot of a lot of these jobs to be done, and this is where I disagree with a lot of people, is I think that there are is going to be a lot of job loss.
Uh real job loss.
Like it just who anyone is.
I think it's going to be extremely rapid job loss.
And then like I'm just thinking about like the early days of like what we saw, you know, in the industrial revolution and like United Kingdom.
Like they I mean, you basically have this displacement and then new roles get created, but like in that interim, still a lot of turmoil.
It's gonna be chaos.
It's gonna be chaos.
And I yeah, like I have a friend who runs a startup, and he texted me yesterday and he's like, I think I'm gonna fire 50 people, and that's like 70% of his team, right?
And I'm just over here and I'm like, how, why?
What you know, tell me the reason.
He's like, I think I can automate all of their jobs right now with like agent swarms.
And I'm like, okay, what's an agent swarm?
You know, because that's just this like throw like term that gets thrown around right now, and is like, oh, it's just an agent that does like a specific thing, and then there's an agent that manages like that whole system.
And then, like, imagine like five pillars under like another agent.
And I'm like, oh, I've built that.
That's what an agent swarm is.
And I think that this is the thing that like people aren't really because now it just runs in the background.
See, like I have one that's just like crawling LinkedIn, like as we speak, and it's like looking for like ICP, and then it enriches them, it writes a personalized email and it cold emails them.
Yeah, and like I don't think people understand like what's about to happen in like the next 12 months.
So and I'm excited about it because I think there's like again, it what if you can build, like it is incredible, like you're so capable right now, especially if you have domain knowledge.
The other thing that I'm finding is like just because you can like it can be built, doesn't mean that you can build it because you don't have the vocabulary.
Like when I look at like my co-founder Max, right, and his technical vocabulary, how he can describe the problem to a coding agent, is so much more sophisticated than I'll ever be able to do it.
And so the output quality that he can get from this is at a like a level that's in the top 10, top one percent.
So if we translate that to something else, like say you study graphic design for 20 years, and like you've been working in the industry for 20 years, the vocabulary you have to describe something is gonna be so much more sophisticated than what I have.
So I'm like, this happened the other day where I'm like, I wanted to put texture on the back of an ad.
And I was like, how the fuck do I do that in the background on a cat trip trying to describe it?
It came out terrible.
And then I found this like person giving a description of like how do you make it have a TV type texture, right?
And it was like all of these, you know, specific, like it was like a specific lexicon to describe it to quality, literally one-shots it, you know, immediately, like what I was looking for after that.
And I had you have that realization that this actually becomes like the superpower.
If you can incorporate these tools into what you're doing for work and have that domain expertise, that knowledge that's like on top of that, that actually is what makes you like incredible.
And so it's the same thing that we've always seen where it's like, oh, you have one or two skills with like a deep T, and then you like that what makes you valuable.
This is like that, but like you know, times a thousand, where it's like if you have the vocabulary, you know, and six things, and you come to these tooling and can basically express and explain like what you're needing or what you're looking for, it changes the entire system in my mind.
But I I I again I'd love to hear your thoughts on like where you like, and also just like what you're seeing within your own companies that you own and all like you know, within the market as well.
Well, I I think the reality is there's a lot of people, even if they have a ton of domain expertise, they don't know the tools and how to use the tools optimally yet.
Now, I think that the tools are gonna get so good that it the UX is gonna be so easy at some point.
But for right now, like, for example, if we if we counted off all the tools that you mention in this in this podcast, you probably mention 17, no more, 20 tools.
I'll include some of those in the show notes.
Yeah, I'll give you a list of them so that you can pass them on.
Phantom Blaster, instantly.
Like, like I'm not even talking about like Clawed code and stuff like that.
You you like you've done your research, you found the tools, and you know, I think that's why a lot of people listen to this podcast.
They try to, you know, it's a way to for them to learn a lot of that stuff.
But the point is, I have a lot of respect for the people that understand they have domain knowledge, know that that domain knowledge is super valuable, and who are going out out there and trying things.
Um, but yeah, this, you know, the other sort of takeaway I have from this podcast is your insight around APIs, which I thought was really interesting.
So, like in the old way, in the old way of like SaaS tools and stuff like that, you you didn't, I mean, it was nice.
An API was a nice to have, you know, it was about how good is the software that you're using.
Um, how good is the UX?
How good is the brand?
How good is, you know, when you press this feature, how quick is it, how instant is it?
But now when you're living in a terminal, for example, and you're using MCPs to talk to LLMs, um, you kinda, you know, the nice to have is actually the this the UI.
The nice to have is the SaaS.
The nice to have is going to this website and look pretty.
Ultimately, what you care about is the output and the thing is running, the agents are running, you know, 24-7.
They aren't, you know, hogging tokens.
The output is high quality.
It's doing the thing that it's says it, you know, uh it it should do.
And I think uh Sam Altman said recently something about APIs.
I think he said the every company is going to be an API company.
Totally.
I align with this.
Like now doing this, like I there's a software, like I just won't put them on blast because I know how big your audience is.
But like, like there's a thing you can do in their UI I can't do in their API, and I'm literally about to churn because I'm just like, this is critical for me.
And now it feels archaic for me to go and interact with your fucking A like UI to do this outcome, like output that I'm that I'm like I need.
And I think that this is going to be like I think there's going to be companies that are entirely just like what it I mean, we literally have this conversation internally.
Like, do we build an AP like a UI?
Because that even a thing that we do, or do we just build the tooling that enables you to see like where we see the puck going?
And I think that like, you know, I we had to come, you know, have a come to Jesus moment of like, oh, this is uh like where we know the puck is going is entirely different than like where the normie like is right now and like the adoption cycle of this, and so like having to like meet there to ride this.
But like, I mean, it's very clear, right?
Like, for example, the like with the graft MCP, that's a live data feed.
So I may I'm like when I'm like, show me my Google ads and Facebook ads, paid ad spend, right?
That's a live data feed that's happening underneath the hood.
It's like a it's like an endpoint that I'm hitting.
It's literally on the fly generating a live data endpoint that I can pull from my data warehouse from.
So with that, I can basically build whatever I want on top of that.
And I'm like, okay, well, let's just go build a custom dashboard like for what we need.
But what we're finding is that like there's like kind of different use cases.
Like the I guess what I'm trying to say is like you're basically making your agent so that or whatever it is your tooling is so that it fits into any harness.
So whether you're working from Claude iOS or Chat GPT desktop or Claude Code on, you know, in your terminal, or like cursor or you know, even the UI, it's unified across that.
And people can basically take that wherever they want and get the same outputs.
And so from a product standpoint, this is how we're like, you know, focused on it and kind of moving forward.
And I think it again, what it comes down to though, is like everything that like we're talking about this tooling piece, like, unless you know what to do, it's very hard to get it to like if I if I didn't know this these exact tool sets to use to like go and do these actions, like here's how to do this LinkedIn thing as an example, there would be like a very low chance that it would be able to figure this out.
It's totally possible, it can.
It's just like gonna take longer.
And you can do this now with like Claude and with Perplexity, where you're like, I'm trying to do X list five APIs that can help me do that.
But I think this is how people are gonna start building this.
And like I'm finding myself doing this, where I'm like, I have this like vision of a workflow, and I'm starting with the final product, and then I'm like working back and like you know, basically piecing together how does this work and then having the agent go and build that for me.
And so then this comes back to like how do agents discover the the necessary infrastructure for and like how do you be the thing that it picks when it's going to build, you know, XYZ thing.
And like again, these are the things I'm losing sleep over now as we're like getting deeper and deeper into this.
Well, Cody, I appreciate you being so saucy with sharing all the stuff with us.
We do appreciate it.
I need you know, it's been too long, it's been too long.
You need to come back on again and and uh share some ideas and stuff like that.
So we'd love to have you.
Um and uh have a ton of them, man.
You can go for days about Chrome extensions right now.
You can literally have Claude Code one shot them and just turn on Facebook ads in the background automatically.
I also have an agent that was running an Etsy shop for a little bit.
That was crazy.
It just got banned two days ago, which was hilarious.
So, people, please, you know, let's beg Cody to come back on the pod and uh have him on tune.
This is an open invite, Cody.
You can come on whenever you'd like, share ideas.
You get you definitely get my creative juices phone.
So I appreciate you so much.
And appreciate it, G.
Thanks for for your time as always, man.
I always love coming on.
Thank you.