How we’d build an AI-native marketing team from scratch | Eric Siu & Cody Schneider

Live. We're going to let some people come in for a second, but we got Cody Schneider today. We're going to be talking about the best AI workflows for business, for marketing, for sales. And Cody is a guy that, I mean, I'm sharing his links all the time with my team. Hey, we got to do this. We got to do this. We got to do this over here. And I think there's going to be some awesome learnings today. We're going to go back and forth on some new tech as well and what's happening with Jev. What are we doing with 5.5 Opus? What are we doing with ChatGPT-6 SOL? There's new stuff coming out all the time. I'm also going to hit go live over here on Instagram. You guys, before we get started in the next minute or 2, tell us where you're coming from. Drop your questions in the chat. Max is going to be piloting this stuff in the backend, and we'll just keep it going here. We got Max welcoming people. Tell us, guys, where are you coming from right now? What do you want to learn? I know we're talking about building AI marketing, an AI-native marketing team, but I think there's much more than that. But Cody, I mean, before we get started, I know I kind of gave a long-winded intro for you, but do you wanna add anything to your intro here?

No, you're good, man. I think you hit a lot of it. What we largely spend all of our time doing is figuring out how to do all of the middle work is what we describe it as that historically a marketer would do, whether that's ad creative research, the generation of the ad creative, the media buying, anything that you would do historically for whatever channel it is that you're working on. How do we get basically a coding harness to do that work for us and create outscaled outcomes that historically it's like, you know, and I, we run a team like this, it would be like I'd need 25 people and now one person with a Claude CodeMax subscription or a Code subscription can basically function as that entire team.

So, oh my God, dude, there's a lot to talk about here. So, but we got Jay Waters here driving from the Bay to LA, so drive safe. You know, don't look at Cody's good looking face too much. But, and then we got Stacy coming here. Thank you for that. We got Yana, we got Cleveland, Ohio, a lot of people from Ohio. You got the UK as well. Okay, so here's what we're gonna do. Um, I, I mean, maybe that's a good place to start, Cody, because you're gonna— we're gonna screen share quite a bit here. But, um, you know, one of my— I was actually just talking to my, my, um, one of my podcast co-hosts, uh, one of my podcasts— my podcast co-host Neil, right before this call, and about how, um, AI is an intelligence amplifier, right? Let's say if you're smart, you're 100 times smarter, but if you're a dumbass, you're 100 times dumber now, right? And it's just— it's, it's amplified now, and you can kind of see everything. And it's what we're just talking about from a high level, not just from a marketing standpoint, but from a mergers and acquisitions standpoint, is like there's just a lot of distressed agencies right now, and you can probably pick them up for pennies on the dollar because, you know, there might have been a co-founder buyout and they can't afford anymore, or they owe too much money to the government right now from like a tax standpoint, right? Because, you know, if you're like International, maybe your interest rates are a lot higher. So there's a lot of meta behind this. Um, and we can talk about that, but Cody, maybe your most exciting workflow right now, because you have a lot of stuff you're sharing around sales, backlinks, like, you know, running, using Cursor to run your Google Ads. Like, where do you want to start first? Let's just go there.

Yeah. Yeah. Yeah. I think the, I mean, whenever I'm doing these conversations, I'm like, first off, what even is this? And I think that helps people understand. like the market that we're entering into. And so I think starting there is probably the best part. Like basically this is being called marketing engineering. Like this is this like new role. It's also called GTM engineering if you cross over into like sales or anything that's touching a CRM. But they kind of live in the same category at this point. It's a lot of similar work and they use a lot of similar tools, right? If like if you're doing a link building outreach campaign, right? You're using a cold email sending platform like Instantly. If you're doing cold email just for lead generation, it's using the same infrastructure. So I think to start there, and I actually have a Notion document, let me share and I can talk through this of how we view and think about this new kind of world that we're living in and what this means in relationship to basically where is this going for agency owners or really any business owner. If you get leads online, this is going to become a part of this. So can you see this all right, Eric?

Sorry about that, I froze out, but I'm still here.

No worries, no worries. Can you see this all right?

Yeah, yeah, I can see it.

Cool. So just first off, what is marketing engineering? It's basically this new category, uh, that I was describing where people are basically doing all the middle work. Um, so to be able to do this, uh, and how we basically approach this, um, you can do this locally on your machine, but the more effective way is actually deploy these workflows or these agents into the cloud. And so what you need is to basically build the infrastructure to be able to facilitate all of the work that this coding agent is going to be doing for you on your behalf. You're basically giving it all the tools that it needs to do that. So over the last 18 months, this is what we found is necessary to accomplish that. You basically need to build a data pipeline, data warehouse, a cloud server to run code on, media storage to store images and videos, for example, for ad creative, databases, that the coding agent can spin up so it can report on its activities. So for example, if it goes and it creates a piece of media like a static image, it can save literally that JSON that it used to generate that image into a database so that you can come back and reference that and actually tie it to your Facebook ads data as an example. You need to be able to do recurring tasks or cron jobs and then also have some type of application authentication. So if you wanna share a dashboard as an example or an artifact as an example with your team so that, you know, that doesn't get published publicly or, you know, it's not public to the web. And then also a Git manager so that this is multiplayer and so that multiple team members can be working on this like, you know, infrastructure simultaneously. And then the last piece of this, which is like the biggest unlock, is this an API gateway of like all of the different tools that are necessary or that you need basically to do this work, right? So everything that I just described, you can do this open source. Like for example, data pipeline, data warehouse, you can use a company called Airbyte, another one called ClickHouse for the data warehouse. And you basically just stack all of these pieces together. But fundamentally under the hood, what you're trying to do is give your coding agent access to all of this. And then you're gonna, through these APIs, through the data that you're pulling back from that data pipeline data warehouse, and then all of this other tooling, you're basically capable of getting your agent to basically do any amount of marketing or any work that you're doing. And a lot of the times how I describe this for people, like when I'm initially like trying to get them to kind of wrap their mind around it, is like, I'm going to take a task that I'm doing manually right now, I'm going to co-work with Claude Code to be like, cool, like I get the outcome that I get out of that. And then once I've built that kind of like log session of me having that chat, I'm then trying to turn that into a repeatable process that I deploy into a cloud that's running on some type of occurrence, whether it's daily, weekly, monthly, it's doing some type of activity. A really perfect example to describe this is like for SEO, you research a list of target keywords, you scrape what's ranking page 1 of Google, you put that into the context window, you write an article based off of that. You add your own brand style voice based off of a transcript that you provide to it, publish that to your site via your CMS's API. So we use Strapi for our blog post hosting. And then that process that I just did and what I would do then is refresh it on a 30-day cadence based off the live data stream. But that process that I just described, I would then take that and I would say, hey, every day here's this list of keywords. I want you to research, write, and publish this new blog post. Every day on a cadence and have that automatically run. And then our job as marketers is we're basically stacking all of these things on top of each other. So suddenly we're publishing 30 new pieces of ad creative per week to Facebook. We're publishing 20 new pieces of ad creative to LinkedIn. We're automatically optimizing Google Ads in the background continuously based off of the live data stream. And we can do this and we can have confidence that it's actually working because we can unify the data To understand, hey, I bid on this Google search term. I can connect my Google Ads to my Google Analytics, to my Postdoc, to my CRM, to my payment processor. And we can know this search term, when I spend money on it, it has a greater likelihood of a higher customer lifetime value. And that's kind of the whole piece. And everything that we're doing today is built on top of that infrastructure. But I always just try to start there to give people a frame of context to think about this in.

Got it. Yeah, that's really good. And just to confirm here, I mean, It sounds like this is very much the foundational piece of what you need to do this go-to-market engineering, right? So anybody that you hire, they should be running through this infrastructure. So this, let's call it a marketing brain. This marketing brain continues to compound, right? And so you guys have all this within graph.com, is that what it is?

Yeah, yeah, yeah. So we built this platform for this and then how we work with companies is we basically will forward deploy software engineers to do these implementations, but then other people can go and basically use all this infrastructure and everything that I'm doing today is on top of that. Again, you can do this open source. You can go and burn 20 grand in Fable tokens, literally just hand it this transcript and the Notion doc that we're gonna provide you and you can go and build this yourself.

Yeah.

But again, fundamentally it's the same ideas. We're just all the infrastructure that is needed for the agent to do this marketing work for us. And the last thing to piggyback on top of this with is what our thesis and how we see the world evolving is all marketing is now just going, it's turning into code. And when you think about a Seed Dance video, we're gonna generate these today. What is that? It's literally a JSON file that's being sent to an API that's running on some GPU that's being sent back. When I upload that ad into Facebook, that is just a piece of software that is basically publishing that content that I just generated up into the platform. When I am analyzing the data, that's just an SQL query that's happening under the hood that the coding agent is writing for us, hitting the data warehouse, pulling the necessary information that it needs just in time to do the action, to do the analysis that I'm basically asking it to do. And so when you like, where does this all go and where's the puck going to, right? And how do we skate to that? It's basically this is where this is evolving to and every company that we're talking to is like, okay, how do I build this? How do I teach my team to do this, et cetera?

Yep. Yeah. And let's talk about that. So by the way, I'm gonna launch a poll real quick 'cause Cody actually shared the things that he's excited about. So I'm gonna share this poll in terms of what is most interesting to you. So there's this little poll on the side. If you guys can click that, I'll maybe end it once I have enough answers here. But the thing that we're seeing is this, right? And so if you're a good marketer, you now have everything in your hands. You can build anything that you want, right? And you can also— just because you can build anything doesn't mean you should. You can build anything, just not everything, right? But having this type of infrastructure is important because when we get people that come to us and they're looking for help, they're like, well, Eric, we can build all this. Why do we need you guys? Right? Well, Then I look at them and I'm like, well, who's going to maintain it? And who's going to optimize it? And who's going to upgrade it for you? And they're like, oh yeah, you're right. We don't want to deal with that. And so I think any of you that are on this call right now and you're in services or you're on a marketing team, just keep in mind, if you're really good at what you do, you're always going to be in demand, if not more. Now, on the flip side of that, I'll say this. I look at the work that we've been doing with my YouTube channel. I'm like, dude, It takes forever to get an edit done, okay? And even when the edit comes out, it's not that good. And then it takes forever to get a thumbnail out and it takes forever to post that. Even when we post it, there's issues with that too. And I'm just like, you know, pardon my language, but I'm like, what the fuck? Like, why don't I just run it through all these skills? And literally I cranked, I ripped out 3 videos over Saturday. I'm just so pissed off now, right? And I think the week before I did the same thing too. And then like people show up to the call and I know one of the people's on the call watching this right now, right? But I showed up to the call with the brand team and they're all like scared. I'm like, don't be scared. If you're doing this stuff, you're working in this way, how Cody's talking about right now, you're good. If you're not working this way, it's going to go 2 ways.

You're going to go up or you're going to go out.

Again, I like to say again, AI is an intelligence amplifier. It's also a truth revealer too. Maybe, Cody, it's a good place to start, I think, with maybe the Seed Dance part in terms of just giving people ideas on how Marketers actually work today, because that one has the most upvotes. Number 2, okay, let me just give it to you. Seedance is 33%. FB ad library mapping for entire category, that's number 2, tied with LinkedIn post to cold email DM, that's number 3. And then link building with cold email naturally is last because not everyone has an SEO background.

So that makes sense.

I love it.

Yeah, no, that makes sense. All right, cool. Yeah, we'll just jump into it and I'll start screen sharing and just jumping through a ton of things. Just to begin, just a quick— this is a test that we're running right now and I think it's hilarious and I'm just going to give you all a bunch of free sauce of something that we found recently. So this was published literally yesterday. We're going to do a parasite SEO strategy. So this is Facebook ads for orthodontists. We're doing a campaign for this right now for—

Do you want to explain parasite SEO real quick?

Yeah, absolutely. So parasite SEO is you basically using the domain authority of a separate website to get an article or a piece of content to rank. So this piece of content that I'm showing you right here was research written and published by AI to our Facebook, sorry, to our LinkedIn page. It's a Pulse article and I'm literally gonna show you how to do this right now. And within the time that this live call is done, we'll have written an article, published it, and have it on page 1 of Google for the target keyword. keyword. So this went in 20 minutes. This went to page 1, position 1 for this target keyword phrase yesterday. You can see it was published 17 hours ago. So anyways, I'm gonna just begin here. And this is, you're basically gonna be coworking with me today because there was a lot. I was in 30 hours of calls last week and got nothing done. And there's a ton of things that I need to get done on the actual like go-to-market for our side of our company. And I'm gonna walk you through and teach you everything that we do for that. But okay, for this first one, So how do you actually do this? So I'm gonna do it for this keyword here. It's Facebook ads for kitchen remodeling. So what we find working right now for SEO is you basically scrape what's ranking on page 1 of Google currently. You extract that content, put it into the context window, and then you write an article based off of that. And then you can modify it for your own, you know, brands based off of what you're seeing. But for example, we're running this campaign, Facebook ads for X business type. So Facebook ads for pet grooming. This article was researched, written, published, and is refreshed continually by an agent in the background. And then we have call to actions that are scattered throughout the page basically for different levels of commitment into the product. So how do we actually do this? I'm gonna start this process right now. So I'm literally, I would normally have an agent go and do this, but I'm gonna show you the manual side just to begin. You go and extract all of the content from page 1. I'm just gonna open up Claude and I'm gonna paste that in. And again, we'll just go through each of these individually. How I would normally do this is I would use a tool called serper.dev, and it has an API that enables me to scrape the front page of Google to extract out what is ranking currently. And then I would use XAI or Firecrawl to basically go and pull this out. Oh, sorry, you can't see my Claude code. So let me reshare with the correct. Okay. Full screen. So give me one second. Awesome. All right. So I'm basically just pulling in all of the content from these articles that are currently ranking. And why are we doing this? It's because Google is already signaling to us that this content is what it's looking for by it being in the search results. We basically know the shape of content that it's looking for, right? How you would modify this when you actually publish it for a brand that you're working with is you would Record a conversation with the founder of the company about their unique positioning. How do they fit into the market? What's their unique thesis? 30-minute conversation, give that transcript. And you basically take what is ranking currently, what we just extracted. You provide both what is ranking currently, their unique point of view on the industry, and then that will basically make this article that meets basically what Google is looking for. And also talk about the brand that you're trying to position. There's a lot of stuff that you can do here with iterative loops for, you're trying to, for example, there's a plugin called No AI Slap if you're trying to remove the things that it does regularly. But anyways, coming back up to this, I'm going to now go and have it write this article. So I'm gonna say, write an article for the target keyword, Uh, sorry, I need to start using transcription so we can move faster. Give me one second and I will do that. For the target keyword, what did we decide that we're actually doing today?

Hilario says it's disconnecting repeatedly, so I'm not seeing that on my side. If anybody else is having issues, you can drop it in the chat, but I don't— yeah, but, um, anyway, just so everyone knows, Cody here— it looks like Stacy might be a little confused. I think what Cody is showing here is how do you rank number 1 quickly with the content that you're publishing From an SEO, AEO standpoint, and that's very valuable.

For sure. This is largely just a very bottom of funnel tactic. I'm gonna immediately go into Seedance after this. All of this is gonna happen in parallel, so I'm just starting the initial, and then we're gonna be bouncing between multiple chats basically to push this all along. So write an article for the target keyword, Facebook ads for kitchen remodeling, based off of the following source text.

Oh, and Cody, Karina says, can we zoom in a little bit?

Yeah, absolutely. Happy to. Uh, following source text. And again, that source text that we scraped is below. So I'm going to start that writing, and then now we're going to go and we're going to start the SeedDance chat. So again, uh, everything that I have already— or that I talked about, um, all the unified APIs, uh, all the data that is necessary, that is already connected into my Claude code. Again, I'm just using the Graph CLI for this, but again, you can build this yourself, you know. But basically patching all of this together. So I'm gonna say, okay, uh, we're gonna start making ads for SeedNance. Uh, we're gonna use, uh, the AI avatars that we have saved, um, that are in the kitchen. Uh, cool. So I'm gonna get that to start running. So to, uh, begin how we actually go and make good quality outputs, um, and what we're trying to get, it's, it's really 3 things that we need to have, uh, happening. We need to have a voice that sounds realistic. We need to have an avatar that basically matches the voice and the environment that they're in. This is actually the hardest part is basically like if somebody is in a car, we need an audio track that basically sounds like they're in a car. It can't be studio-level quality, otherwise it has this discrepancy. And so today what I'm doing is I have found this voice on ElevenLabs called Natasha. That's what we're going to be using. And then what we're going to be doing as well is basically using the these AI avatars. I'll show you those real quick. So I have 5 different avatars that are different age demographics and different ethnicities. And the reasoning for this is because Andromeda, now how its targeting functions, is it's basically based off of the ad creative. And so the script that we give it and also the avatar that we use is going to basically define who it's going to show the ad to in front of them. And so this is a way for us to, like with all of our campaigns that we're running for clients at this point, unless we're doing account-based marketing, we are targeting all of Facebook and then we're making the creative basically be the positioning element. So historically you would do interest-based targeting, right? So say you wanted to market to people who have a mountain biking interest because you're trying to sell pedals to them. Instead of doing that now, what you would do is you would make the creative so that it's speaking directly to them. So This workflow that I'm gonna show you right now is what we're using for software companies. The outcomes, just to kind of show you what that has created. So this is an example of a company. This is demo requests booked using this exact strategy, basically. So this is them over the last, whatever, 12 or 3 months. So, okay, going into this to actually pull the scripts and make the scripts, what we're going to do is we're going to scrape Reddit for the pain points and desired outcomes. outcomes that people have in relationship to whatever the product is that we're trying to sell. So we're gonna do this just for Graph today for the ease of it. So I'm gonna say use the Exa AI API key to scrape Reddit for the pain points and desired outcomes that somebody would want to buy an AI agent that runs their Google Ads for them. Maybe they dislike their agency or they feel that they, The agency isn't doing anything unique or different, but again, just source from any of the Reddit threads that you can find in relationship to this. So I'm gonna let that start running, and in the background, we're gonna come back to the article that it wrote. So I'm gonna say just return the text here within the chat, and then we're gonna take that and actually go get that index. So this is how I'm working continuously in the background now is basically all of these things are happening simultaneously. And once I've figured out the system, like the Seed Dance system that we have done today, what we're actually going to go do is we can then go and publish that. So it's happening on a cadence. So every day, for example, I can have 10 new pieces of ad creative get uploaded into a CBO-level campaign in Facebook ads, and those automatically get generated, uploaded, and then I can also create a feedback loop on top of that where it's basically looking at the conversion data and it's making new creative like the ones that are winning. And we know which creative is winning because we've already saved that JSON. We've saved all of that work into that database for all of the actions that we're basically doing or working on. So I do the— use the Graph, sorry. Use the Graph CLI for this extraction. And then coming back to this Facebook piece, let's copy this over. And now I'm gonna go back to LinkedIn and I'm going to just publish this to— we'll just go to Graphs just to show you how crazy this is. I'm obsessed with this right now. So this is why I'm showing it. And I'm just trying to give you free arbitrage that exists in the market currently. So we're gonna go create a post and the post that we're going to do is write an article. I'm going to paste in that article that we just wrote, have the keyword phrase be the exact title because then that will come out as the URL, and then I'm gonna take this first intro section and I'm going to add that to the post here and hit publish. Now, once this is published, we can then take the URL that it provides. You can see here it says Facebook Ads kitchen remodeling. And I could take that URL and I'm gonna go and use a tool called Prime Indexer. And Prime Indexer basically does a drip ping to this that enables these to get seen. Do not do this for your core website. You can only do this for your basically external, these parasite sites. But this is this way. And again, I'm gonna show you this right now. This should go to page 1. We'll see what happens. Actually always live. Who knows what's going to go on, but you can see this is not indexed by Google currently, but through this process, it basically should get it indexed and we should show up in the search results. So coming back to Claude Code on the SeedDance side, it's still doing that scrape for Reddit. So it's pulling out that information.

Cody, what's the indexing tool again?

It's called Prime Indexer is the tool. And then they have an API that's accessible for this as well. So like you could hypothetically Yep.

That's really important because one of our workflows right now, the challenge is like they're not getting indexed. And even if you use GSC, like there's no clear API with that, right? So I think this is really important to have in the workflow, but continue please.

Yeah, yeah, totally. So just to expand on what's happening here in the background and like, why do I even, why am I even talking about this? You can make Facebook, you, sorry, you can make LinkedIn pages for free and you can publish this content to those LinkedIn pages. And if it's optimized appropriately and it's targeting the correct keywords, you can get this to literally show up and we'll see this like within minutes, we'll be indexed at least by Google and hypothetically on page 1. And we've done this a couple of times. You can see a handful of these here and I'll show you those. But all right, coming back to SeedDance. So this is still pulling or scraping from Reddit. So what was the next one, Eric, that they wanted to have us start on?

Yeah. So the next one. Okay. So it is a tie. You can choose here, Cody. So it can be LinkedIn post to cold email DM, which I think is probably a good one. So let's just go with that one. Then I'll tell you the third one.

Awesome.

Yeah.

Awesome. Yeah.

Go for it.

So LinkedIn posts to cold DM, how we use these is typically to extract the engagers from influencers that are in the product category that we are in. So what does that look like actually in practice? Let me just go and do AI. Facebook ads. Let's see what comes up. So I would find posts. All these are kind of garbage. Let me go to the homepage.

And by the way, Cody, do you mind if I add in for the last one as you pull this up? So the thing that Cody is doing really with the last one is he's getting a post up and it's going on LinkedIn and it goes to public and then Google can index that. But LinkedIn site is very strong from a domain authority standpoint, which is why it's going to rank better than your main site for the most part. But that's not to say that you can't test it on your main site. Cody's not saying that.

Yeah, you can write that article and publish it in the exact same way. The only difference is that do not use Prime Indexer on your core site. And the reason for that is this is basically doing a drip ping to that link. You don't want to do that to your main site. It just can have negative effects. You want to only do this bit to a large site. This is also effective if you're building listings for SMB. So for example, if you're trying to rank within the map pack for Let's say like commercial cleaning SF as an example, right? If you're trying to rank within this map pack, how do you actually do that? It's basically you want to take the name, address, and phone number and put it on as many directory websites as you can. For example, like MapQuest, like Foursquare, et cetera. But the problem with that is that these sites are massive. There's literally hundreds of thousands of pages. So how do you get Google to actually see those pages? That's where you can use again something like a Prime indexer to actually do this drip ping to it. LinkedIn is just another example of like this site that you can publish to that you can link from. So, all right, coming back. Oh my God, MapQuest is trying to nuke my RAM. All right, coming back to LinkedIn to show that process. So I'm going to see if I can find something that's about AI for marketing. Yeah, potentially Clay. So anybody that interacts with this, that could be one. So I could put that up as an observer. But the strategy here that we're gonna actually illustrate is I'm gonna find, for example, 10 different profiles on LinkedIn. And imagine this is 10 of them. And every day they're gonna do a new post, right? So they're gonna do a new post, each of them individually. And then from those new posts, I'm going to extract the 100 engagers, the 50 engagers that are coming out of them. And then once I have their LinkedIn profiles, it gets game over. I can cold LinkedIn DM them using something like HeyReach, or I can do a waterfall email enrichment to find their emails and actually cold email them as well. Again, just to talk through that process, I'm basically setting up a listener that daily checks for net new posts. And this is exactly what I'm gonna tell the agent and tell it to use Appify to do this. And then I'm gonna extract the actual engagers from the post, and then I'm gonna find the emails, and we're gonna add those emails into Instantly AI. And everything I just described, we're gonna walk through this right now. So how do we actually do this? I'm just gonna use my own profile for this just 'cause it's gonna make this fast right now. So we'll say, okay, everybody that's engaged with this, I will just give my profile. Into, uh, Cloud currently. And again, I have that Apify endpoint already established, um, to extract the LinkedIn profiles. The Apify that we like is this guy named API Maestro. He has the best LinkedIn ones in our opinion. Um, shout out him, he's a great partner. So with this, we always try to show him as much as we can. And we're using basically this LinkedIn profile extraction. And the reason that we're doing this is that We're trying to not have our core account basically like doing all of these requests to LinkedIn. And so this is this way to basically navigate around this. He is handling all the proxies. He's handling all the scraping. So I'm gonna say this now. And just a heads up.

Oh, sorry. Yeah. Just so everyone knows, the reason you don't wanna hit your main LinkedIn account is 'cause that's the type of shit that can get you banned. So that's what Cody's saying.

Totally. Yeah. And, and it's— I see this happen a lot where people like give agents like their, you know, basically like, uh, browser use, and then they have it go to these automations. I am just like, do not do that. You are just asking for something to go wrong. Like, the best way to do this— there's, there's, there's a boundary of basically like, you know, white, gray hat, right? Like where you want to live in. And everything I'm describing today is like, again, we have found that this works over all of the clients, the companies that we're working with, and also our own personal stuff and everybody that's on the platform that's using it. So I'm gonna say, all right, we're gonna use the Apify API through the Graph CLI to extract all the posts from this user from the last 7 days. And then I want to extract the LinkedIn profile URLs of those individuals and then use Get Leads to do a waterfall email enrichment. If you need to go to Lead Magic or just guess Their email as well. Feel free to do that. Validate the email with Million Verifier, and then I'm gonna give you an Instantly campaign to upload those emails into so that we can do cold outbound to them. So again, I'm just using Super Whisper to transcribe that workflow that I just described. And historically, if I wanted to go do this, right, like I am going to go and hog wrestle N8N or Zapier to basically make this workflow. Do not do that. You know, our running joke internally is friends don't let friends do N8N. The better way to do this is go straight to code. And the reason that you want that is one, it's more discreet, but second, it's also more flexible. So you can modify these workflows at any point in time. So I'm gonna get that fired off and we're gonna come back to SeedDance now.

You know what's funny, Cody? We tell people, you know, during the interview process, we actually, we tell them afterwards and it's like, oh, well, why aren't we moving forward? It's because you listed N8N. So.

That's so funny.

Yeah.

And this is taking forever for some reason. So I'm just gonna speed this up and just use Perplexity. I'm gonna say scrape Reddit for the pain points and desired outcomes of why somebody would buy an AI agent that runs their Google Ads automatically for them. All right, we're gonna get that to fire off and then I'm just gonna give that source material as the source material for the scripts that we are about to go and write. Those scripts, we're gonna create an MP3 file first. And then that MP3 file, we're going to send that to Seedance and have it lip sync using the AI avatars that we, uh, like showed previously. And again, I'm going to walk you through that whole process, but I'm just going to wait for that as it comes out. So while that's going on, let's come back to Prime Indexer now. Let's see if it got indexed. All right, cool. So it did get it indexed. See if it's live. It hasn't hit yet, so we'll keep refreshing this to see, uh, when this basically Uh, you know, get seen by Google. So typically it takes like a minute or 2 after this. So, um, all right, coming back. Uh, so we got the user post. So this is working in the background. It's scraping the reactions and comments, um, and it's adding it to a CSV. You can see all the tool calls that it's basically making, uh, for me. Um, and then while that's happening, let me just log into Instantly real quick. And I'm actually gonna just stop screen share for a second so I don't, uh, I don't, uh, show an API key or potentially—

No API keys.

Yep.

Exactly.

So All right. I just wanna make sure I'm in the correct account. Okay, perfect. All right. So I'm in the correct account. And so I'm in Instantly and I've got, I think, 1,000 inboxes that I've got connected here. The inboxes that I'm using, they are through a partner called Hypertide that we love working with. You could basically send about 10,000 cold emails a month. On top of their infrastructure for about $125 in infrastructure costs. And so if you translate that, you could send a million cold emails a month for about $8,000 is what this turns into. This is a whole other rabbit hole we can go down, Eric. We are thinking about cold email now as a one-to-one marketing channel. But anyways.

Yeah, same here. Cody, the question is, so they seem like they're a managed service, right? Hypertide, and you just pay them to manage it, right? And my whole thing is with our infrastructure internally right now, This was set up like a couple months ago, but we had Hermey set it all up. But at the same time though, at the same time we have to manage it, right? But it's not that big of a pain in the butt on our end. But Hypertide, you said it was like $120 for how many emails?

$120, you can send about 10,000 cold emails a month. Why we like them is they manage all of the Microsoft like Entra infrastructure. What does that mean is they basically isolate The domains that— it's called a tenant. But Microsoft, if you get put into a bad tenant pool, your deliverability is terrible. And so what they basically handle is the tenant management. So they— imagine we buy 20 domains and we're spinning up, I think how they do it is like 100 inboxes per domain. They put 2 domains per tenant. And so it creates this isolation so that if a tenant does get burned or a domain does get burned, it doesn't affect the entire infrastructure. infrastructure that are running on. So anyways, we just work with them. There's tons of these. InboxKit is another one.

That's great.

There's, and it's more like self-service. They're more like, they serve a lot of agencies and all partner, the stuff, the work that we do. But anyways, just service providers that we see. Instantly also offers these pre-warmed domains as well. And you could buy these from them. They actually have great deliverability. The only problem is they're not branded to your, it's not like a burner domain that you're sending from. So it just depends on what you're trying to do. But anyway, all right. So we've got those domains. They've been warmed. They're ready to send. So I'm gonna create a new campaign, and I'm just gonna call it, like, LinkedIn Engagers. And I'm gonna continue. And at this point, I'm gonna take this URL, and because it already has the Instantly API key, It knows it can automatically upload these leads. So I'll say, when you're done with the email validation, upload the leads into this Instantly campaign. And then while that's happening, I can also start working on the copy for that Instantly campaign. So I can say something like, I wanna create a 2-email sequence. So for this Instantly campaign, email 1, the subject line, have it be LinkedIn engage or or saw your comment on my LinkedIn as the subject line, and then have the text be something like, hey name, saw you commented on my LinkedIn post, took a look at your Google Ads or your Facebook Ads library and saw that you were only running a handful of creative. If we could ship 80 new pieces of creative a month for you automatically, And have a learning loop where an agent is running the media buying of the account. Would this be interesting to you and worth a call? Some variations of that. Make it extremely concise, and I want you to focus on the economy of words with the writing that you're doing for these emails.

Got it. And Cody, are you trying to share your Claude dashboard?

Oh, sorry. Yeah, I apologize. Let me— I was sharing the window. So I am back now on the Claude dashboard. And that transcription that I just did. So again, I added into that the user post. Let me just shoot this off first so that it's going. So it's gonna create that email campaign. But I added the— sorry, I didn't actually add it to this one. But I'm gonna say, add the leads to this campaign once you're done with Million Verifier.

By the way, as Cody's doing this right now, I think it's really important to emphasize that the future, like, AI-native marketers are spending the vast majority of their time within the harness. Like, it should probably be 85-90% plus, if I'm being honest. And the work that we expect people to do, like, if you're an engineer, 95% of your code should probably be written by AI. And then if it's like, if you're not, you're not an engineer, our expectation is that 70% of your work is done by AI. That's just how we are, Cody. I'm not sure how you guys think about that.

We're like literally anything that touches a keyboard at this point. If you use a computer for the work, you should be trying to figure out, can I get an agent to do this? And there's things that are still on the early days, for example, long-form video editing. But I am getting incredible results right now with DaVinci Resolve's MCP and an Astra subscription. It is ridiculous the outputs that you can get out of it. It's absolutely insane.

So yeah, even for long-form editing now, I just think there's no more excuses. Yeah, but please continue.

Totally, totally. No, no, please. Happy to talk to me in the background too. Feel free to. So I'm gonna say, here's the source material. So this is, again, I'm just, I don't know, it got in this thinking loop and I'm not gonna try to solve it right now. But I basically, again, I'm scraping from social, the conversations that people have had that have Basically, they've shown like, again, their pain or the desired outcomes that they have in relationship to the product that you, that your product can solve for them or your service can solve for them. So here's the source material and I'm gonna say, I want you, actually I'm just gonna have it save it just for now. We're gonna write the script next. We're going to write the scripts next. Uh, we're going to— there we go. And what I'm going to go do, I don't have it saved here, so I'm just going to go to my Twitter to extract this. Um, so I have a list of, um, uh, Cody Schneider posts, list of Facebook ads hooks. Um, I'm just gonna— we found it's like 10 Facebook ad hooks that works really effectively for like software, but they also can be translated to other companies or other business categories. We're actually finding them work really well for local service businesses right now too. But we'll take those hooks and then we're gonna basically do a hook and then a desired outcome, like pain point structure. And I'm gonna show you what that looks like. So here's these hooks from a post that I wrote previously. I'm then going to go and I'm gonna drop this in and I'm gonna say, Write, let's do 15. Write 15 ad creative variations, or sorry, ad script variations. Use the hooks below and then use the source material above that I just provided as for the rest of the script. The script needs to be 30 seconds long and use super simple language. Again, think on the economy of words. I wanted that like a— 12th grade reading level for this, and then the format is gonna be hook and then you know desired outcome pain point combination. So all right, so I'm gonna have it write the scripts now. Once I have those scripts, and then I'm then going to send those scripts to ElevenLabs. Once I have the MP3 files back from ElevenLabs, I'm gonna send that to we're using right now Kai AI or. SeedDance, uh, or sorry, Kai AI or Higgsfield for SeedDance. Um, Higgsfield is what we're finding we can get. We can get a 30-second video out of Higgsfield right now for about $2.50 with like the, you know, basically the pricing that we've got locked in with them. Um, and then, uh, we will generate these videos. So I'm gonna send the MP3 file and the avatar screenshot that I showed previously, and we'll bring those up again in a second. We're gonna send those. All to Seed Dance, like to Kai AI, which is hosting the Seed Dance model for us. And then that will basically we'll have that final media. Once we have that final media, we'll actually go and upload that into Facebook ads, and I can show you that whole process there. So cool. Got these email variations. Let's take a look at these.

Notice that Cody's doing multiple jobs at the same time right now. He's doing SEO. He's doing signal-based outreach. He's getting ads up at the same time. This is like. Many, many, many jobs at once.

Totally. Totally. And again, I'm not— this is not anything that a human wasn't doing previously. We are just taking the human process that used to occur, like a media buyer. What was a media buyer doing? They were publishing new ads to Facebook. They were trimming the losers. They were promoting the winners. And then they were creating a feedback loop of what does the winning content look like? How do I go and make more of that winning content? We're doing the exact same process, But again, all of that middle work, we're getting the coding agent to do that work for us because it has access to all these tools, all this infrastructure. And then once we have done that system, like I just wrote this pro— like we just did this process, right? Or sorry, this, this one here, this email enrichment process that it's still going through. We, once we've done this process, I can say, okay, I wanna turn this into a piece of software and I want you to deploy this into my cloud, right? Like into, again, I'm just using the Graph Cloud, but this would be basically stored on a server. And then I can have it where, again, daily, I give it a list of 10 different creators. It's extracting those posts, extracting the engagers from the net new posts, doing this whole enrichment. And so in the background, I'm just constantly feeding my system new leads that it's doing this outbound to. So, all right, cool. So wrote the script files. Let's take a look at those real fast just to kinda see. I'll pull you guys over here so you're not in the way. Okay, I didn't think this would actually work. I paid an agency to run my Google Ads. They checked it once a month. The rest of the month, my money went to junk searches and cheap leads that never bought. Now an AI agent watches my account every day. This is great.

Cool.

Awesome. I just kind of gut check those again. We've done this enough now that it's like I don't even really look at these anymore. I just create a walled garden for my agent to work within, and then I just let it go and figure it out for me, right? I'm letting it do all of that middle work, and This is very uncomfortable to large companies, by the way. Like we're working with a large software company right now and it's like to convince the team internally that this is the way to win, especially for account-based marketing where like ad creative velocity is the challenge. This is something very large. Yeah.

Well, now you have JEV, right? You have a classifier and the evaluator at scale and it's so cheap that like— 100%.

And the reality too is like we, We're doing a 0 to 1 right now, but in reality, I have creative most of the time that I am remixing 'cause I already have signal. Again, this outcome that I was talking about, we've shipped thousands of ads for this company at this point. We have very strong signal knowing what's actually working for them. This is just, again, that 0 to 1. So we've done that. Now I'm gonna say, I'm trying to think. What was the question? What's the name? Use the Natasha voice. We're gonna send those scripts to ElevenLabs via the ElevenLabs API. If you need the Natasha voice ID, let me know. It should be in the, like, stored somewhere within the file, but I can specify where it is if necessary. All right, so that's continuing on. We've got these emails. I'll just say, cool, are these added to the Instantly or that Instantly campaign? And the cold emails, again, still running in the background. I'm just gonna ask it, give me a status report. And then we're gonna come back and we're gonna check on this SEO strategy that's going on in the background. So let's check that again now. Where was Prime Indexer? Cool, so we got the URL. Let's see if that's online currently. Cool, so it still has an index. Which again, this is what sometimes when you're doing it live, things don't work. But the again, just to kind of show you like what can happen with this. So we did this exact same thing for like Facebook ads for pet grooming. Should be. Oh, I think that dropped off on the LinkedIn side. Maybe it's pulling in.

Oh, sorry.

It's right here. So Facebook ads for pet grooming, exact same strategy. Facebook ads for Orthodontists. Like, again, same thing. Orthodontists. Uh, this, um, and then we'll see. Let me see if it's actually, uh, just not showing on the index but it's showing up for the remodel.

Cool.

We'll keep an eye on that while it's running in the background. And actually, I might actually start another one because I'm just curious to see if we can do it while we're waiting for this other stuff.

And Cody, just to confirm, right now you have the SEO piece, you have the ads piece, you have the signal-based outreach, like the DMs as well. Those are the 3 things you're cooking right now on the screen?

Yep. Those are the 3 things that I'm cooking right now. So I've got—

I got a question.

Yeah, please.

Yeah. So on the signal-based QA, so it's like, this is coming from my team, by the way. So on personalization, right? So let's say you're You get an Instagram or a, sorry, a LinkedIn post out there, it gets like 1,000 comments or whatever, right? And then you're looking for people that are ICP fit and you're reaching out to these people. So one of the questions from our team was like, how do you QA signal-based content at scale? Are you spot checking? Is there anything, is there something that filters? And I have an answer in mind, but maybe you're doing something.

Yeah, totally. I mean, you can do an ICP filter basically on the content that it's being like, you can pull the LinkedIn content itself, um, that and basically extract the text to be like, okay, I want you to rank this, you know, 1 to 10. Does this match this ICP fit that I, you know, have defined in a markdown file? And then basically, you know, it, it kind of qualifies them on the front end for that. Um, that is a way to do this. Uh, the other, uh, way, um, that we've seen most creators— like, it's, it's a lot of it comes down to the creator selection that you're observing. Most of the creators are talking about the exact Same thing. They found a category. That's how they're basically gaming the For You page algorithm is they found a niche that's working for them. So they're not really straying away from that, to say it a different way.

And do you have a rule set here where it's like, hey, if they're ICP fit and they've engaged with you 3 times in the last 30 days, are they tier 1 versus somebody else? How do you think about that?

You could absolutely do that. Yeah. I mean, we're doing a lot more of this. This is getting into the weeds, but it's called a TAM mapping, right? So like total addressable market mapping. What this enables you to do is you basically can go— that's funny. What this basically enables you to do is you can go and find everybody that's in your target audience that you're trying to sell to. Also, this is what the best companies in software startups have been doing for years now. It's just now proliferating. Before you'd have to be incredibly technical to be able to actually accomplish this. But it's basically finding all the companies that you would wanna sell to within your industry. You find the firmographics, like the company size, how many employees they have, et cetera. And then you find the decision maker at that company and then you basically can observe that decision maker. Are they engaging with this content? You could do deanonymization of them to the website as well, using something like a lead pipe. Where it's basically like as they hit the site, like you could see that they're coming to it. Again, there's a lot of like levers that you can pull here, but basically what you're trying to trigger is you're looking for a signal. Or for example, like this person changed their job. They went from company A to company B, strong signal that they're about to go into a buying cycle. Everybody always buys new shit when they join another company.

Yeah.

And so you're basically stacking all these signals together and then From that TAM map that you've created, when a person lights up from one of those signals, you then can go and do that cold outbound motion to them.

Yeah. And I hate to keep mentioning Jev, but I'm just like, dude, now you can classify these things. Did they actually engage? Are they ICP fit? Were they active in the last 30 days? All these things that are impossible for a human, you can do it all now. And I'm not a salesperson for Jev, by the way. No, no, no.

Yeah, I think there is a lot of use case. I mean, what we're excited about with Jev is you know, we can do a TAM mapping of the entire, like, all the ads in the United States for like a category and like pull those and understand what is in them like pretty effectively using this. So for example, like we just did this recently for car detailing because we're like doing an initiative around that, like, you know, business category for some of our work. So we found like all the ads that were basically running for car detailing and then had a vision model and that's multimodal so it can look at the image and then also look at the video, basically extracts, okay, what is in the video? Writes that out within text. And then from that, we can basically source like what are the commonalities between ads that people are running, which is the signal of what's working. If they're running, if a lot of people are running similar ads, there's within different geography, strong signal that those ads are probably being successful for them on a lead generation standpoint. So yeah, other questions? I see the chat kind of blown up, so.

Yeah, so I'll get Syed and Steve have a question around what about B2C workflows or are your workflows only Yeah, B2C is the kind of same strategy.

I think the biggest thing on the B2C side that we found to be successful depending on the company type, we've only really worked with mobile applications that are on the consumer side, but ad creative volume is like, I mean, if you go and you look at Rise Superfoods as an example at this point, let me go to their Facebook ads library. I think they have something in the range of 5,000 pieces of active creative that they're testing simultaneously. This is what the best companies in the world are doing right now. is they're basically testing a ridiculous amount of creative. Yeah, they have 5,100, right? This is 5,100 active ads that they are running in parallel currently. How do you think they're doing this?

Yeah.

They are 100% using AI. You see the creative production here, right? It's literally an AI avatar that is— sorry, it's literally like an animation that they're doing. So they have a team of creatives that are pumping these out. They also are doing partnerships with creators using something like a Sideshift or Tribe. Sideshift and Tribe basically are creator marketplaces where you can pay them on a per video that they produce. So for example, we're running a campaign right now for a company where we pay $40 per UGC content that they generate. You can also structure it in ways though where it's like you pay them on a percentage of the media that is spent on the ad creative that they produce for you. So say you deploy a grant and you pay them, for example, like 3% of that $1,000 for the video that they make. So only the videos that win, basically you're paying for them for. But Sideshift, Tribe is another one. And again, I was just at a dinner in San Francisco like 2 weeks ago and a huge company whose name I cannot say, very fast growing startup. They are doing an absolutely massive amount of content production using these these creator networks like Tribe, like Sideshift. And then to manage these networks, how they're actually doing this is all like everything that we've been talking about today where you're using your coding harness. Sideshift has an API. You can basically manage these campaigns. Like what used to have, you didn't have, used to have to have like a full-time person focusing on this that was like managing the creator production pipeline, answering their questions, like actually Looking at the creative to see if it matches the requirements for the campaign. You can have literally vision model with agents and through the API do that entire management process.

I got a question. So Cody, are you primarily using Cloud Code or are you using any other harnesses and are using any autonomous agents to help you?

Yeah, yeah, yeah. So we, the company works out of Cursor and their cloud agent infrastructure is really good. So that's why I've done it. I— my daily driver is like a combination of Cursor and Claude Code. A lot of the times I have both open. Um, and then I'm just starting to use Codex again because, again, for the video editing capabilities and some of the coding, like long time horizons. I'm pretty agnostic to this. I think the biggest thing that I'm trying to like do is basically use cloud agents more and like how do I basically get those cloud agents to work in the background. Um, we don't really use— we, we had experimented with Hermes and like with these other— I would call them like almost coworker-style agents, um, and like getting them in Slack and that thing. But we just found that they're— this is probably a contradictory take, but I'm like against token maxing. I think it's really stupid. I think that's a good take. Yeah, I, I'm with you. But I think it's against like what everybody's like, token max, like you should be burning. I think it's the dumbest thing. Like, why am I paying Like, why am I paying Claude or Anthropic to schedule a social media post every time that I'm trying to, or to write a blog post? What I should be doing is I should be using my coding harness to write software that's discrete. And when I say marketing agent, like a Google Ads marketing agent, which is one of the ones that we have that we deploy very often for companies, that agent is just software with a thinking loop with a live data stream. It's not like God in a box is managing a Google Ads account. We actually found that when you hook Hermes up to God, to all of this and tell it go do work, it's kind of shit at it actually. And what is actually better is basically getting it to write software that is an algorithm that it's running. And when you take a step back, it's a lot of what we found in the past, right? How do you do the operations at your agency currently? You have a human that's running a standard operating procedure. We're just taking that standard operating procedure, we're translating it into code, and then we're giving the agent, like giving the coding agent all the data it needs so that it can write better code for that, that process. So, got it.

I got 2, 2 questions before you kind of share what else we have, and then, uh, maybe we can wrap it up with questions. So, um, the— so right now, I mean, you mentioned like you don't necessarily need a ton of headcount, right? So I'm curious, if the most— the majority of your company is working like how you work right now, like what is your headcount right now?

Yeah, we are 6 people. So it's my co-founder and I, we got 3, 4 deployed engineers and we have a head of sales. So that's kind of the structure.

Yeah, but look at how you guys are working. Like, that's crazy to me, right? It's like—

We can do a lot of—

The amount that you can do now. Yeah, exactly.

I mean, anecdotally, like, what is this? It's September. I think it was in July we published in the range, I think it was like 11,000 ads across all the companies that we're working with. And this is not like, when I say that, people are like, oh, it's slop. And I'm like, no, this is like brand style guides with iterative loops that it doesn't, it has like a gate basically where the creative can't pass through the gate unless it meets the brand style guides. Right. So we also, you can like get human in the loops for this. This is actually something just to add to this. We have found that humans are actually, they need to get outta the way for a lot of this work. Humans like want this oversight to begin with and then they become the bottleneck.

Mm-hmm.

Where they're like, I can't approve 20 new pieces of ad creative per day, right? But an agent can happily do that and at a quality bar that is 99.9% of the way there when you give it the right infrastructure, right? Like when you say, okay, again, here's this brand style guide, make the ad, look at the ad like with a different like vision model. Like does this meet the brand style guides? Oh, you're using a font that's incorrect. Cool. Send that back to ChatGPT Image 2.5. Make that change. So yeah.

And Cody, second question here. Um, Cody, um, before, before we go back to the screen share and then maybe more questions, the, um, you kind of shared some performance numbers, right? So I think it was like on $100 grand in spend, you're able to book like, I don't know, 500 meetings or something. And the cost per booked meeting was $200. Like, do you have any other performance numbers you can share around this stuff just so people can really grasp that this shit works?

Yeah. I mean, so working with a company, um, during like the wellness space, They've got their franchise. We're working like at the corporate level with them. We took their cost per lead from $70 on Facebook down to about $16 in the span of 4 weeks, like doing this exact process. We're now basically having somebody, we're jokingly saying they're like day trading the Facebook and Google Ads algorithm for this company.

Mm-hmm.

Google Ads, same, like similar outcomes where it was like, I think it was like $82 and we got it down to like $35 in the same period. So that's on the regional business side for software companies, book demos. We like trying to think about some of the other outcomes on the paid ad side, just off the top of my head. The biggest one that comes to mind is the one I showed you. We're also working with a prosumer application where we basically started, we zero to one their entire Facebook ads strategy and took it from— Like zero to, like, I think they did, it was in the range of like 20,000 signups across. We did Facebook ads, Google ads, SEO, cold email, LinkedIn DMs. We basically did that whole stack and had it out the door for them in like 3 weeks. And I think when we started with them within, it was June, I think they did about 2,000 signups in that month. Again, on that scaling, they're doubling basically month over month since we started with them is what it looks like.

So this is the— And how about, yeah. On the LinkedIn cold DMs, like what kind of performance are you seeing there in terms of like, hey, you send like 1,000 messages and like X amount of people booked? Like, do you have any numbers like that?

Yeah, yeah. It really depends on— so what we have found is that with cold outbound, the only way you make this work right now is you have to have an offer that the market wants. So the strategy that we found to work is you find an influencer in your category, you record a live class with them, And then you use that within the outbound motion where you're like, hey name, again, saw you commented on X post, reaching out because we recorded a live class with Y influencer and it's about 1, 2, 3, 4, 5. Thought you might be interested. Do you want me to send over the recording? The recording is literally, they say yes, it's literally just like a demo in disguise of your product. We found that to be like Very good on the performance side, on the response rate. I actually have a podcast with my friend Nick Abraham. Let me see if I can add the podcast. Gary Schneider, where it basically, he talks through his, he's sending like a quarter million LinkedIn DMs a month and like he explains exactly how he's doing it and how the outcomes that they're getting are like more performant than literally any other channel that he's doing on the cold outbound side. I love it. So anyways, if that's what somebody wants to go deeper on that, that's a Available there. But yeah, yeah, coming back to this.

That's killer.

So again, emails have been found. They are now in the Instantly, uh, accounts. Uh, let's see if the leads are there. Cool, leads are there. The sequences have been written. Um, everything has been mapped so that it will work appropriately. The inboxes are, you know, set up. That can, you know, literally start sending. On the SeedDance side, we'll see if these will generate in time. Uh, the scripts are— have been written. The MP3 files have been completed. It's now just with Kai AI for the Seed Dance renders. So we'll see if that, you know, how that works coming back. And then looking at— let's come back to this, see if the SEO thing ended up happening. Curious if we ended up getting it to work live. Yep. So that is now indexed by Google. And let's go see where it's ranking. So yeah, it's ranking position 1. This went live 38 minutes ago. So anyways, that's it, man. Any questions I can try to answer? Happy to dive deeper on anything.

Yeah, let's hit a few questions here and then we'll get Cody on his way. This has been phenomenal, man. So we got a question here from Syed. We'll pop it up over here. So are you deploying these systems inside the companies, for example, the ads, or are you guys running everything through your own systems?

Yeah, so for the companies that we are working with, We basically function as a platform that we build these systems on top of our platform for them. At any point, we forward deploy the engineers so that we can basically get these out the door as fast as possible. Because it's just like there's a huge gap in knowledge between this. You basically have to find somebody who has enough software engineering knowledge and also has enough marketing knowledge. And it's just like a very rare person to come across.

Almost impossible.

Yeah, it's very hard. What we're finding works, and this might be interesting for you, Eric, is It's these like 22-year-olds that have like ran an AI consulting agency and they're like, they have ran into these problems and they're looking for like a place to basically like, anyways, they're, we're finding them to be that, that's where we see this all going. But yeah, so deploying onto the platform, you could build the platform yourself. We just don't work with companies like that 'cause it's like literally we've spent the last 18 months and So much fucking money, right? To basically build out this infrastructure layer. But for the companies, the things that we build on top, build for them on top of us, at any point they can be like, cool, we don't need the 4 deployed engineer anymore. We just want this like virtual employee to continue to run on and it just lives on top of the infrastructure. And we just like, then it just becomes usage based where it's like, how many rows do you need synced? What's the credits that you need for email enrichments, for ad creative production, et cetera? Yeah.

So Peyton has a really good question that I'm about to pull up, but the, the way I see this too is, um, you know, we're all just agent managers at the end of the day now, and our edge, Cody, is living on the edge, and people are always going to need that. And so I remember there's someone listening to the, the Marketing School pod, and they're like, we just built this all ourselves, and we realized that we just can't keep up with you. And so like, therefore, if you're on top of this stuff, people are always going to need you. This agency stuff, services stuff is not going away. But let me come to Peyton's question over here. So Peyton asks, how much budget have you seen be useful to test creative on this scale?

Yeah, it's great. It depends on the cost per action, but we work with a lot of software companies where their cost per demo ranges from $50 to $200 depending on the industry that they're in. We typically start with $100 a day on Facebook ads and Google ads. And anecdotally, by week 6, we are very confident on with the data. We've had it as fast as literally a week. We have enough. You can do creative loops now where it's like every week I can do 2 to 3 creative loops where I'm like, I test a piece, a group of ads, I see what works, I go remix this. We're actually doing this for this company right now. I don't think they'll care if I share it actually. Yeah, it's called Avoid. They basically get physical people. If you're like, I want people to discover my cafe or my local business, walk-ins to happen, they get people to go to your physical store and discover you. You pay $25 and a human shows up and they literally take a photo of themselves there and it's buy a coffee, et cetera, right? It's a super fascinating business model, but for them it's been this positioning. We've been helping them on that and going through those iterative cycles of how do we position the company basically.

Cool.

Anyways, that's kind of the scale. Once you have that number, Then it's just a CAC to lifetime value ratio. And if you're an e-com company and you're listening to this, you know the game, it's just ROAS, right? But what this ends up being is you start to see signal and then you just chase it and then you just continue to double down. The problem that we face is everything has a half-life now. And so there is some type of decay that happens. So you have to think about entropy. How do I introduce new DNA to the system? The ways that we're doing that is we're basically scraping social within our category and we're finding organic social posts that go viral within whatever marketing. And that then basically introduces that new DNA. Cool.

That's great. We'll cover one last point here, but Jay had some sauce to share. I'm just going to put this on the screen. So I've never heard about p0.studio. I'm just going to share this with people. I haven't tried it myself. I don't know if Cody has, but it's amazing. Yeah, 70% less than Higgsfield API MCP, x402 for payments as well. That's cool. Okay, one last thing that I think we could touch on before we get out of here, actually. Peyton has a good question here for you. So what percentage mix do you introduce entropy?

That's a good question. We don't have probably a very solid formula for this yet. Largely what we'll see is that the CPA action will get to a threshold where we're like, okay, we think we can get cheaper. And then at that point we start to introduce, maybe it's like 20 to 30% of the creative is like Totally net new. And then as soon as, once we find that net new that is working, how we built this is we just chase that signal, right? So we're sprinting towards that. But I would say it's probably in that range. It's typically like an 80/20. I'm always just trying to think about how am I, 80% is what I know is working, 20% is always testing constantly. Everything that I'm doing is that.

That's big. And I think that's a good place to end. So the key takeaway here is this. So with the engineers, I have to constantly remind them on the team, hey guys, It's cool to build, but the challenge is if you just keep build, build, build, you're not going to go anywhere. The good thing about marketers, we know how to fail really well. So it's build, measure, iterate, which is what Cody's talking about. Once you find the signal, all in. And then once the signal starts to decrease, okay, onto something else over and over and over again. That game never ends, but that's what makes it fun. But Cody, this has been amazing. What's the best way for people to find you online?

Yeah. Thanks for having me, Eric. Pretty active on Twitter and LinkedIn. Those are kind of the 2 socials. And then YouTube as well, actually. I'm like, we're just about to go into another sprint there. And I'm literally gonna be teaching like everything that I just described in way more detail when it's not in a live setting. And you'll get like step by step because we can actually do edits and we don't have to wait to watch the inference occur. But yeah, those are the best places. And if you wanna work with us, it's graph.com. Yeah, if you wanna self-service and get started on platform stuff, go to graph.com/platform. You can self-service. There's a free trial there for 7 days and do literally everything