How To Build AI Agents That Do the Work For You | Cody Schneider

Welcome to the Growth Boss Podcast, where we explore the tools, strategies, and AI workflows helping businesses grow faster and smarter. I'm your host, Elena, and today I'm joined by Corey Ganim. He will be sharing with us how to build AI agents that do the work of four marketing hires. So, Cody, thanks so much for being here today. I'm really excited to dive into this topic. Yeah, thanks for listening. Super excited to be here. I'm looking forward to this conversation all week. Absolutely. So, for anyone who might not be familiar with you, can you give us just a 60-second overview of who you are and what you do? Yeah, absolutely. So, for the last six years, I've spent kind of my life in early-stage startups, specifically in the B2B space. My specialty is like, how do I get my first, you know, 10,000 customers? We've been building AI marketing products since October of 2022, like kind of right before ChatGPT launched. Yeah, we're, you know, a bunch of different iterations that finally end up what we're building now, which is graph.com. Graph is basically deployed marketing agents for businesses. So, agent examples would be like an agent that entirely runs your Facebook ads, your Google ads, your search engine optimization, your cold email outbound. We've done them for social media scheduling. We've done them for real farms. I have an endless list of kind of things that we're like building out. One that I'm pretty excited about right now, actually, is like a Pinterest agent. It's basically for a landscape design, like a software company. But anyway, yeah, that's kind of the high level. Our specialty is how do I build distribution and how do I basically implement these agents to, you know, do the jobs of what a traditional marketing organization would do. Yeah, that's amazing. How did you specifically get into the AI agent world? Yeah, so I was working at a company called Rupa Health and we had just scaled that. I joined as my co-founder and I, Max, actually met there. We joined as employee like 10 and 11. But we had just helped scale Rupa from like a $20 million valuation to 110 in about six months. Rapidly hired a bunch of team. I think at the end of it, we were at like 25 team members that were just on the content side alone from a production standpoint. Yeah, we were doing a podcast, email newsletters, live classes, you know, posts across all social influencer marketing, etc. And we honestly, like how it all originated is we were basically like, okay, what are the operations process that are super time consuming that a human's doing? We started to see that these AI tooling could automate a lot of these processes. And this would have been in like July of 2022. Yeah, we were just kind of experimenting on the side. And then it really in that fall, like something changed and it became actually usable, like the outputs for good enough where it was like, this is a step function above what it was previously. So yeah, that was really the origin. We kind of the products that we built previously was like a long form content repurposing tool. We also built like early SEO agents really rudimentary in comparison to what like what we're seeing work now built a tool that was like basically used by ecommerce stores to like programmatically generate landing pages based off their product catalog. And then it had like a feedback loop based off of the live click stream data that was coming from the pages. So just how you'd see like an Amazon or an Etsy or you know, how's any of these large marketplaces how they function as they basically dynamically generate these landing pages. And we were doing that but for you know, brands that were, you know, 10th the size, right? But yeah, and then, you know, really from that, we kind of saw the writing on the wall where everybody was trying to build agents or go to that we initially built, we actually tried to build, you know, an agent, just like everybody to go with or to start by initially like doing a Google Ads agent. And then we ran into all these issues where you basically like if you're trying to call the data that you needed from the Google Ads API, you would run into rate limits, or you'd run into truncation errors, or you'd run into the agent just full stop hallucinating what it was saying. It's actually like make, you know, an agent run for a company, you basically have to have like a live data stream of what's actually driving revenue. And so the initial version of the product was a BI tool where we solved the data pipeline in the data warehouse. And then recently, we've just like in the last three months, we've added basically the agent infrastructure as well. So we deploy what's called a Hermes agent into the cloud. So it's the data pipeline, the warehouse, and then a Hermes agent. And then that all works in this like cohesive loop where you're like, okay, I want to drive, you know, leads for X business type, here's the shape of what that lead looks like the agent gun, go and research those leads to qualification on the ICP, find the email addresses, like actually cold email them, and then manage the inbox, like responding back and forth. That's amazing. Honestly, crazy. Like even saying it now. Yeah, it absolutely is. So I can't wait to kind of break it down a little bit more. So our topic today is how to build AI agents that do the work of four marketing hires. So just to kind of level set and start at the foundational level, can you give us your definition of an AI agent? And how does that differ from someone who just uses ChatGBT or like Cloud Cowork? Yeah, totally. I think of an agent as like some job function, there's some job to be done that it's doing it on like a repeated cycle, right? So I imagine like, say you have like a standard operating procedure within your organization. So we'll say like you're a social media manager, right? When you look at the job function of a social media manager, it's basically like, what content is kind of going viral within my category? Okay, how can I remix that for our brand? Let's go now create that content. We're going to schedule it out, you know, to post across all social, and then I'm going to look at the best performing content, like what is the winners? What are the losers and let that influence it like my circle. An agent in my mind is basically, you know, piece of software that's running that process, right? In all reality, like whenever you hear agent, it's really just software under the hood with like some type of thinking loop. I think everybody is trying to sell a bunch of snake oil in this category, especially in our category, like we have these conversations all the time where it's in particular in like the last like three weeks, there's been multiple companies that have come to market where it's like, we'll entirely run your ad account. And it's like, no, you won't. This is total. Like the only way to actually get these things to function and work is you basically take like, what is the process that a human was doing? Okay, how do I go and basically, you know, repeat that process, like with having an agent run that so in comparison to like a chat GPT or one of these like, generative tools. So say, for example, like a great way to explain this is like, you're trying to write a blog post. And so you're like, okay, let's, you know, find target keywords, research those target keywords, like what's ranking on page one of Google currently for that keyword. And then I'm going to put that into the context of, you know, the like chat GPT, I'd like literally just copy and paste it all in and then be like, okay, write a, you know, 2000 word blog post based off of this, you know, for this target keyword based off of the source material that I provided you. So the human is basically going through that like prompt chain to get that output. What an agent is doing is largely that exact same process, but instead of, you know, it basically a human pushing it step by step through, it's just running that process for you. And then typically what we'd like to see is like when it has some type of like larger thinking loop, I think that's where it's probably more sophisticated. So it's like, yeah, it made the content, but then it published it. It's looking at the live data stream. Okay. Which of these posts are working most effectively from like customer acquisition? Can we go make more content like that? Can we do a content gap analysis based off of like what we've written and what everybody else has written to see like how we can improve the contents of that content is like refreshing itself. And so I think that's how this fits like into this is largely it's, it's typically like a linear workflow that some type of like, you know, agent is basically like initiating with, with thinking loop like in it. So it's more of the full cycle that the agent is able to take on versus you saying, okay, do this and then do this and then do that. It's able to do that. It's able to do the entire process with, it sounds like more of the human overseeing it versus managing it and running it. Exactly. Exactly. I think about it more like I'm a curator, like my job, you know, to compete in this new world is to have like a ton of domain knowledge about whatever my expertise is. And then basically be able to, you know, explain that expertise or like what I'm looking for to whatever this agent harnesses and then have some measurements of success that I'm like, you know, moving towards. And for every business, it's the exact same thing, right? It's like, we want to make more money. So it's like start there. And then we work back basically into, okay, we want to make more money. What is, what are those, you know, what does that actually look like? Okay. We need more qualified leads. Okay. How do we get more qualified leads? Where do our, like our target customers spend time online? Okay. They spend time on Instagram, Google Maps and, you know, like Facebook, right? Okay, cool. Let's, what are our, what are our strategies to show up in those places and basically, you know, be in front of the people that are looking for this, the service or product that we're trying to sell. So. Yeah. The goal is always save time, make more money. And that sounds like something we can do both with what we're talking about today. So you mentioned that in the headline topic for today, four marketing roles. So just kind of high level, what are those four marketing roles that we talked about? Yeah. The ones we're deploying the most for companies is, is Facebook ads management, Google ads management, search engine optimization. And really that kind of falls into the same category now with like AI search, whether that's like chat GPT, you know, or clot or perplexity, all of them. Honestly, Google Gemini has been like exploding lately from a referral traffic standpoint. So we're seeing more and more of a focus there. And then the other one that we deploy a lot is like cold outbound. So like cold email for organizations. Those are kind of like the big four. And then there's a bunch of other, you know, things that, that we're seeing companies want where they're like, Hey, I need social media management or I need like influencer outreach and negotiation where they're like, Hey, I don't, you know, I want this thing to go and contact a thousand influencers. Find a hundred of them that are underpricing themselves. And that's like the, you know, the 10 that are most underpricing themselves. Let's work with those, et cetera. But yeah, I think that's the biggest thing, especially it depends on the business type. Like we're having software companies where they're basically like, can we just work with you instead of hiring, you know, an entire marketing team. We're having like local businesses where they're like, I hate my agency. I know they're robbing me. You know, can we basically pay you to just like run, you know, the Google ads agent for us, right. Where it's like just a flat fee. It's like optimizing continuously in the background. And when you look at like a lot of the agencies that people are, and we're, we're starting to partner with agencies too. Like on these products where it's like, Hey, they have a, you know, 150 Google ads accounts that they run for appliance repair companies across the United States. Right. And then we're basically like the white labeled back of house solution for them. So they handle all the client management. And then we are just like operate the agents for them that are running the actual ad campaigns. But, but yeah, those are the four that we're seeing like people most desire. Absolutely. And it's largely just cause like, that's like when you look at the majority of business growth, it's kind of across those four categories. It's called outbound, it's paid ads and it's, it's organic, you know, ranking. So, but yeah, happy to dive deeper into each of those and like how they actually operate. So, yeah. So I'd love to talk through kind of the process for a business owner or, you know, marketing professional who's listening and they're like, this sounds incredible. I'm not quite sure I'm ready to work with someone to do it for me yet, but how could I kind of start to get in it? What would the process be to just kind of play around with building an agent? Where would we start? Yeah, absolutely. I'll talk through the Facebook ads one. Cause I think for a lot of people, this is like one of the channels that they've interacted with most in Google ads as well. But I can go down this list on the Facebook ad side. Again, depends on the business like size and structure, but I'll just use an example of a company that we're working with. There are landscape design AI products. Basically it's like you give your address, it uses Google maps to like map your backyard. And then basically it's like, here's what it can be. And it gives you the material list for you to like physically hand to a, like a landscaper in your geography. And so it just makes it so that it's like, you know exactly what it's going to cost. There's not some like cost plus pricing that's occurring, you know, for that production. So for them, what we did, we scraped like Reddit to find what are all the pain points and the outcomes that my target customer was using. So for the person that's listening, how you can do this is with something like Codex or with Claude Code. And there's a tool called Firecrawl that, or XAI. They both have the ability to basically go and extract information from the web. But you can like scrape Reddit, scrape YouTube, really any social media of people complaining or like talking about like what they wish they could have that your product solves. And then, so we aggregated that information for them. And then we use a tool called Nano Banana to do a static image generation. So based off of that, like all of that, that corpus of information that we extracted from the web, again, their pain points and the outcomes that the person wants. We generate 10 new pieces of ad creative daily. Those ad, like pieces of ad creative, like we basically gave it a brand style guide, like here's the fonts, here's the colors, you know, here's kind of the composition, et cetera. And then those ads automatically get uploaded to a Facebook ads account via the Facebook ads API. All this is like pretty technical, but like for the person listening, you would create what's called a developer API key. You can do this. Anybody can do this for any Facebook ads account. And then again, within Claude Code Codex, you basically are giving that, they call it a coding agent harness, but you're giving it that API key. And you can interact directly with your ad accounts via this. Like you can literally control them entirely. Like you have to know nothing about how to actually use them to get them to, you know, operate. But how we do this then is, go ahead, please. The actual agent creation, what is that done in? Yeah. So we do our agent creation with three things. So it's, we have a data pipeline, a data warehouse, and then the agent harness. So data pipeline is just a stream of data from all of your data sources. So imagine like you're running Facebook ads that probably looks something like Facebook ads, you know, Google analytics for your CRM, say like go high level. As an example. And then maybe you have like your payment processor that's connected as well, like Stripe or something like that. So the reason that we have to connect those is so that the agent has an understanding of like what is actually working. Right. So to give you an example of this, of when it's kind of failed in the past, there's all these like AI SDRs or BDRs, right? It's like sales development reps that came out in the first cohort of kind of all this AI tooling. And when you look at it, they're like, you know, they're kind of written off as they weren't successful or they like they couldn't actually make impact. When you look at the postmortems of like how they failed, they didn't actually fail. They created a lot of leads. Just the leads were really terrible quality. And it's largely because they didn't have like the observability of is this lead qualified? Like is it actually turning into revenue? And so you have to basically give the agent like the data that it needs. It's just like a human, right? Like if the human is making decisions off of, you know, poor data, it's very likely that it's they're going to just not do well. And we see this within like all sizes of orgs that, you know, I've ever worked with, right? Where it's like, you know, sales want something. They're like, give me whales. Marketing's like measurement is like give me leads, which are contrasting like different things, right? And so you kind of have to have this like meeting of the minds of like, okay, what does a good customer look like? How do we actually go and like find that good customer? How do we track that? And this would be like, you know, what the human would do traditionally. And then, you know, based off that alignment, that's that measurement of success that we're going towards. We're just taking that same concept and basically giving, you know, the agent that. But so when I say an agent, what we're actually using is a harness that's called Hermes, similar to something like an open claw or there's a clawed agent SDK is another one of these. All it is, is basically like an agent that's so an agent like at its core is basically a piece of software that's living on a machine that's in the cloud. That's just like always on. And so you always like have access. This, this agent has full access to that computer basically. And so you can be like, Hey, like go research this thing. Like, like literally I can text it. Right. I can be like, go research this thing. And, you know, it will respond back. Like here's everything that I found based off of, you know, the last 15 minutes of research that I just did. And the reason that it can do that is because it has that computer that it can use. And it knows how to like navigate that computer. Right. It's a lot of it's just like from what's called the terminal. So like it's writing code to go and navigate the internet, which is this whole other kind of worms of like, what does the internet turn into if it's just billions of agents that are navigating it. But yeah, at its core though, it's, it's just those fundamentals. It's basically like, does the agent have data that it can, you know, that it needs to make decisions on. Like I put an agent on a cloud. So it's on a computer that's just like running constantly. And then it's like giving it some type of standard, you know, operating procedure where it's like your job is this, your measurement of success is this, you know, go and run these activities basically for us. So, so you said Hermes, that's the platform. I don't know if that's the right word that you use. Yeah. We like Hermes a lot. We've tried OpenClaw on some of these other like options and I think they'll get better as time goes on. But like OpenClaw was really the first version of this where it felt like, you know, I had like, I don't even know what to call it, a digital human in a box that I can like interact with and like delegate tasks to. But it just is very brittle. Like it wasn't enterprise ready, which again, for our customers, like they're, you know, I need my Google ads to always run appropriately. And like, you know, if we set a budget of whatever, five grand a month, it's not going to go and spend 50 on accident because it's just like, you know, failing. And so we kind of tested everything. We found Hermes to be like at a quality level that is just way above, you know, anything else that we could find. And so that's why we've like focused on it. We also like know the founders a little bit and like we've talked with them on social. And so just, it just feels like it's like, it's a very serious like solution rather than like, okay, what's possible. Yeah. Now for someone who's getting started with this, is that what you would recommend for them too? Or do you think something more like an OpenClaw would be all they need to get started? Honestly, like for the majority of people, I think it's even a layer down from that, like using something like a CloudCode or a Codex and then have the agent go do work for you. Like have it go and I try to like more and more how I work is I am using some type of transcription software. I use one called Super Whisper. I just know the founder and like it's a great product. I'm literally working in Codex or CloudCode and I'm like telling the agent, okay, we're going to go do this together. And I'm like, you know, physically like with words saying that. So I'm not typing. And then I am like trying to delegate like any of the work that is me either touching the mouse or the computer. I'm trying to delegate that, you know, to that. And I think for a lot of people, like as soon as they realize that that's even possible, it's like this like light bulb comes on of, oh my God, like everything that I do is on a computer. And like you're telling me that I can basically have, you know, this and it is an agent. I think the difference is like when you look at CloudCode or Codex, it's an agent that's on your machine and like it can't really like effectively run tasks. Like say, for example, it's like every day at 9 a.m. I want you to do something. It's not really like built for that. It's more like you're the human. You're kind of driving its actions, but it's doing all of like the I call it the middle work. So it's like it's doing all the middle work for you. So like, you know, again, touching the keyboard, clicking the mouse, you know, et cetera. And then your job is like I need to come up with like what does it need to do? I need to be able to explain it really well. And then I need to be able to like audit its actions, right? Like the outputs that it's generating. But for most people, like and I've taught these like classes, like live classes in a webinar setting before. Like this is like this like total like change of mindset. Because and once you get to there, then you start to think about, okay, cool. What are all the processes? Like I just co-worked with it. What are all the processes now that I do on a daily basis that I despise? And how can I go and basically turn that into a system that this runs? And then what naturally evolves from that is like it's typically like, you know, a human does it entirely using something like a codex or a cloud code to have like, you know, the human is driving. And it's like, you know, kind of some like a co-work situation. And then the next evolution is like, okay, now like I'm going to try to make this entirely autonomous. And I think that's the best adoption for most people is to kind of go through that. And again, I don't know, whoever, you know, whoever's listening right now, whatever industry you're in, you can have this positively affect your work that you're doing. Like if you own a med spa or you own an HVAC company or you own a software company or you own a whatever, like an insurance business, right? Like there is, I guarantee things that you do, whether it's writing, whether it's like, you know, data analysis, whether it's like lead list cleaning, whatever that is that you do on a daily basis that you can basically delegate to these models. And it costs money, but it's like out of, you know, a tenth of the cost or a hundred of the cost of having an employee do that. And it's also going to be like, you know, at a caliber or a level that's way like better, especially if you provide it more and more context. Like for example, how I work with, I use Cloud Code as like my daily driver. And I've just built up this like, you know, whole collection of information about my business. Like the things that we're trying to accomplish, like here's our brand style guides. Like all of these things are basically saved like within its context. And so as you're working with it, it gets like smarter and smarter about your business. It almost turns into like a co-founder, right? And again, I think this is like the best way to have the adoption of this. Don't try to jump initially to like, I'm going to build an agent that entirely, you know, automates my whole business. It's, I think that that's a mistake. And it's kind of a tar pit that people fall into accidentally a lot of the times. What ends up being way more effective is it's like, okay, what am I doing today? Can I automate a portion of this with like, again, something like Codex or Cloud Code or whatever that is, even ChatGPT. Honestly, they're building all these into them now where it's like the local desktop apps like have Codex in it. They have Cloud Code in it, et cetera. So it's basically, what am I working on? Can I automate what I'm working on right now currently? And then, you know, basically from there, now I can start to think about, okay, what is this workflow that I can do? Can I use something like a Zapier or like an N8N or any of these other workflow building tools to actually go and automate this? Or can I, you know, set up Hermes? Like there's tons of these services now that just like are, it's like a Hermes agent in the cloud where you pay like whatever, $9 a month to have, you know, Hermes in the cloud. And you can literally like message it on, you know, Telegram or iMessage, right? Like you can send voice memos to it and like have it do work for you. It's pretty wild. But that progression is the best way I've seen from an adoption standpoint. And I assume you need a separate computer for this that you run off of. Yeah. When I say a computer, like I do this like on my MacBook, right? Like what I was describing when I'm co-working with it. When I say like a separate computer, like a Hermes agent, you can technically run a Hermes agent on your local machine. Like you could 100% do that. The problem is that like if you like turn your computer off or like you, like another piece of this is the, like how much access do you want it to give, right? Like if it's running on your local machine, it literally has access to like everything, right? And so you kind of have to think about, okay, like how am I sandboxing this? How am I, you know, in the best way that we've seen is using, you know, some type, like you can do this on Hostinger as an example. It's really cheap. I think it's, I want to say it's like $15 a month and they basically have like a one click Hermes instance that you can spin up. And then like that Hermes instance is like on a computer in the cloud. Like whenever you hear the cloud, just think like it's just literally racks of computers in some data center, you know, somewhere that's using a ridiculous amount of water to cool it. And that agent is basically like running on top of that, that computer that's running in the cloud. And then how you interact with it, it just, again, it can be through text message. It can be through like you connect it to your Slack. And if you're like, okay, how do I do any of this? How do I actually, you can literally ask it. Like I'm trying to do this. How do I accomplish that? And it's going to walk you through that process and that stuff. And again, it's super early. There aren't perfect, like it's not perfect yet, but every month it feels like a year right now in my industry where it's like, even just looking at our company and like what we do, like it is an entirely different shape than what it was like in January. And we're only halfway through the year. So anyway, yeah. Yeah, I'm sure it moves so fast. And I was going to ask you this. You kind of touched on it. If someone's listening, they're like, I'd love to learn how to do this a little bit more hands-on. You mentioned like a webinar or a class. That is something you do as well? Yeah, yeah, yeah. You can go to gtmengineeringcourse.com. It's just literally, I just use it to collect emails. And then we do, it's every two weeks now. I'm trying to get to weekly where we're like some type of class where I'm talking through like an agent build. But yeah, that and like honestly, YouTube is the best resource for this. Like if you just search like how to build X agents or whatever, like for example, a voice agent say that you wanted to like answer your phone. There is so much information out there that like people are sharing about. Don't pay for a course. Like the URL I just told you is like it has literally a course in the name of it. It's free. It's sponsored by my company. Like we never will charge for it. Everybody that's selling a course is selling old information. Everybody that's like on the front lines of this and actually doing it. They're all just giving it away for free because they just find it super interesting. And it's like, it's also just like so exciting, right? Like you're telling me I've been running businesses, like online companies for the last whatever, 15 years. And I mean, we're at the point now where it's like not unreasonable to think like, hey, I can have this small, you know, website. Like something I'm experimenting with right now that I'm like having a lot of fun with. That's totally a side quest outside of my, you know, our business. But it's basically building like directory websites for like random, you know, things like, for example, like, you know, best dog parks with shade. You know, just very obscure, like long tail things. But like for the first time now, you can go and build these sites at scale that are really like they can be autonomously run by software, like by these agents. Right. Which again, it's just software under the hood with some type of inference, like it using tokens to think. But yeah, it's a pretty exciting time. And again, everybody that's legitimate, they're going to be sharing exactly this, you know, like on whatever it is, their YouTube or like, I mean, I hate to say it, like Twitter is probably the like most, like as people are discovering things, that's where it initially like it's like shared. And then from that, it goes to YouTube. And then like two weeks later, LinkedIn picks it up. And then like a week after that, it's Instagram, basically. But it all originates typically from like Twitter and YouTube are kind of the two origins of all this new information. Okay. That's really helpful to know. And at what point do business owners or whoever your client typically is come to you? Is it after they've been trying it and they just can't get it to work or they realize how much time they need to invest? What makes someone come and work with your company versus doing this on their own? I think like we have two different customer types. So like one is like, they're a repeat founder that's like done multiple companies. And they're basically like, can we partner with you instead of having to hire a marketing team or like hire a sales org or whatever that looks like. Where they're like, I don't want to have to go and hire 10 people to be able to run these five channels. Can we work with you guys? That's kind of like one customer type. The other is totally where they've like, they've tried to do this implementation. They've gotten 90% of the way there. And then the last 10% is just like some hard technical problem that they can't get over the hurdle. And they're basically like, hey, I just want to pay you for the outcomes, right? Like just, I just want to pay you to have this actually doing the work. And so that's like where we like come in and how we function. So we forward deploy engineers to the businesses to basically like solve these problems, right? So we have this menu item of agents that you can deploy. And then like we have software engineers that like literally embed with your company and like build these systems and these solutions out, right? And then we, again, we're the data pipeline, the warehouse and the agent infrastructure. And so it's just like the whole, like it's full service in the sense of it's everything that you need to actually implement these within your companies. We're seeing this a lot with like a business, you know, businesses where they're like, okay, we tried to implement this. Like we've tried three different, you know, organizations. We failed three times and we're just like, please just get us to this like final yard line. Because again, just what we've seen in this space is like there's just a lot of snake oil right now because it's like, you know, for a lot of companies, like this is the dream, right? You're telling me I can just like plug into this thing and my business starts growing. And like I just like all I have to think about is like providing a good product. And like at its core, it's like, you know, when you look at a company, you know, from the fundamentals, like what is a business? It's like, you know, you have something that people want to buy and you have the ability to sell it, right? Like that's all of businesses. And like for a lot of these, you know, business owners, like there's great product people that have terrible distribution skills. And so they're like, okay, if I can just focus on my service and providing that and this is handling, you know, all of my basically growth and new customer acquisition. You know, that's again, the dream of everybody that's starting this and honestly where I started as like a founder as well too, like in whatever, 10 years ago. So anyways, yeah, but that's kind of what we see it. A lot of the times we're working with companies where they're like, hey, we want to do these five different channels simultaneously. We have some companies where they're like, we just want to do Google ads or we just want to do Facebook, like individual agents. And we service that. But a lot of the times, like the marginal cost of doing everything now is so small that it's like, why wouldn't we do everything? Like why wouldn't we try to be literally everywhere that we can? Because we know that's going to like make our business grow faster and be more sustainable. So what would someone expect as far as like the timeline to get an agent running? So they're say they're doing one. Yeah, we get an agent out the door in like five business days typically. So if like the Facebook ads agent, again, just give you an example, like a concrete one, like using that landscape design firm. It's like the first ads were going live in like three days. Right. And then like, so we took their average cost per lead. We took it from $17 on average before we started. We brought it down to $1.70 in three weeks, just like by the agent running the process. And it's not anything like tricky or like special. It's just like, what are the top 1% of like Facebook ads marketers do right now? And when you look at it, it's like they test a lot of creative. They have a conversion action that they're trying to optimize for, which is like typically a form submission or like a qualified lead, you know, whatever that is. That gets sent back to Facebook. And then it's like, I'm just trying to look for like, what is the cheapest creative to create, you know, that action. And then I'm just testing. So 10 new pieces of creative go into testing every day. Losers get turned off. Winners get promoted to a winner's campaign. The winners are competing for budget for like, you know, who's the best performer. They're all optimizing for that same conversion action. And then we just have an agent that's in the loop. That's basically looking at the data. And then also looking at the winners, we have it like attached to a vision model where it's basically like, okay, what do they have in common? What are the themes? And then lets that influence the creative agent. And then you just go, you know, in this loop, right? Where it's like, yeah, it's just what a human would do. It's like I research my market. I make creative that speaks to the pain and like to the outcomes that my market wants. I test all these different formats. I look for the winners. I double down on the winners. I turn off the losers. I repeat that process. And that's all that's happening under the hood with any of these. So for custom builds, it takes longer. Like we're doing this right now for a company where we're like building. This is going to be really esoteric. But like we're called TikTok Real Farms. So it's basically imagine like 100 TikTok accounts that are in the cloud that like, you know, they basically like they give us an API endpoint and we can go and bulk upload like TikToks too. So for them, they're like a mobile application. That's like an AI photo editing app. And so they have this format that they found that works organically on mobile or sorry, on TikTok where it's like basically like three different options of what they're basically like a photo studio app. Like you take a selfie of yourself and then you can like put yourself in different settings. Right. Is that the idea. But for them, it's like, okay, we're going to go and like create this whole organic content engine. It's just what a human was doing previously. We're just taking that same process and remixing it. But for custom builds, it just takes more time to actually like build out the solution. But for these like prebuilt solutions that we've already made for other companies, it's just like on our menu where it's like, cool, I want that one. Like, you know, give me that one. Let me go implement it. And then we basically as we build out these other agents for other companies, they just get added to our menu as well. So it's, you know, we'll build this like TikTok cloud agent and or sorry, TikTok Real Farming agent. And then we'll basically like have this playbook that we can go and apply to these other organizations. And I think this is like what is most exciting for me with as all this goes is it's like, as you start to see like, hey, companies that have this type of shape, these are the things that are working. Let's like, you know, you should focus on these five things to begin with because we've seen, you know, similar companies have success with that. And then it just kind of cascades with all the cross learnings and everything that happens. And again, it's just moving so fast. Like the impact we can get is like typically in like the first month, like, well, so in the first two weeks, we typically get two to three agents out the door. By the end of like 90 days, we're in the range of like 10 agents that are live. And like, you know, three of those end up being custom, right? Like specific to the company. And again, it's just like you're telling me that I come from a business owner. Like I've done this before, right? But you're telling me that like I don't have to go and hire. First off, hire the staff, train the staff, implement the staff or hire an agency, have them like, you know, maybe do well for six weeks. And then they absolutely just crumble because they delegate it to some junior staff member that is managing 10 different accounts simultaneously. Or it's like, you know, somebody in like the Philippines or India that's like now managing my Google ads and doesn't care. Right. It's just the you can just create like at its scale and an impact that previously was just I think this is where all this is going. Like you're a business owner. Like what ends up happening right over the next couple of years? Like I think anybody can go and make, you know, a thousand Facebook ads right now. Actually knowing what's working is going to be the hardest part with all this. And this is like what we're obsessed about is like what is the closest thing to revenue that we can track for and like give to the agent so that all of its focus is like the closest thing to revenue. Right. Again, if that's like a qualified lead or like the actual payment that occurs, like how can we like send that back to it? Because without that, it's just going to turn into like and we're seeing this. There's tons of these companies that are just like AI slop generators. Right. And like to be blunt, I built AI slop generators like before I built this company was like kind of the first iteration of this was like, OK, can I go have it write a thousand blog posts? And it's like, yes. And can I can that make leads? Yes. But then it's like it gets to that point where it's like, OK, well, like of all make anything now. It's how and how we're thinking about it more and more is like marketing used to be like the cost, you know, your ad spend costs. Right. And more and more, it's like, what is the cost per token or what is the token cost to get a lead is basically like our measurement of success. It's like, how do we reduce the token cost to basically grow the band or grow the brand, whatever that outcome is, like for some of the companies that we're working with are huge. And they're like, it's just brand authority. That's what they're trying to build. It's like notoriety within their category. Totally different outcomes than like. But for a lot of people listening, it's probably like I just need more customers, like get me more customers. Right. Anyway, I have two follow up questions. So one, is there any type of business or stage of business where AI agents just wouldn't be a fit for them? Yeah, I think early stage companies that don't like have an understanding of who their target customer is or like what they're selling, like if it's amorphous and it's like, okay, I'm still just figuring out like what is the market positioning that I'm going after. We just have like, you can't build a system for a company that doesn't have a system. Right. And so I think that's one piece. Also, like we see this with a lot of organizations. Like you look at like, you know, anybody in real estate to be totally transparent. It's like they have some crazy paper trail that they're running. That's just like super, super messy. And it's just like hard to like, there's some human that's worked there for 12 years. And, you know, Stacy understands the whole system, but that's the only person in the whole company that actually understands it. And there's all this nuance. So I think for like a lot of these organizations, and this is where you're seeing a lot of like consulting starting to happen now, especially for like SMBs, like small, medium sized businesses is basically like somebody comes in. They look at your whole process, like structure, and then they're like, okay, cool. Like how do we like, you know, automate this or make it more streamlined? Those are the types of companies where we're seeing this be a problem. Just where it's hard on the agent deployments, because it's like, again, how I would just think about it is like, if I could take a task and I could delegate it to somebody on my team and they could operate, you know, they could run that task, right? On a cadence with some type of measurement of success. If I could do that, that's like a great sign at this point that that's probably an agent that can be deployed that actually does this or accomplishes this. Yeah, that makes sense. And the second question I had in any of the agent workflows, any that you work with or with your clients, I'm sure the answer is yes. Is there a point where humans have to like input some sort of content? So for example, say it's a client who, you know, is doing YouTube videos or social media and they do actually have a human in the videos. Is that ever something that is part of the workflows or is the goal to have AI create all the content instead of the human? Does that make sense? I think the human is the most important part. So we for sure have built, we call it like human in the loop, right? Where it's like, say they're in a regulated industry and they need a human to check off of, you know, on something, right? So they're in like finance. And so it's like, cool, like the agent does the work. It sends a Slack message and it's like, Todd, is this all right? It's like, no, you know, you need to change this section. It changes, makes the change. It sends it back, right? And it can like be like a coworker almost in that way. So I think that that is like a way that we're seeing it used. The other side of it too is like, if I just go and ask an agent to like write me a good piece of content about my industry, it's like going to be the most terrible, most like unoriginal, most like the least novel thing that you could imagine, right? Because it's just writing to the average of the industry. But in contrast, if you're like, hey, here is, you know, 30 minutes of me talking about some specific thing that's happening within my category, you give it that context. So it's like, here's the transcript of that conversation. And now you're like, agent, okay, now go write, you know, a thought leadership piece of content based off of the source material that I provided. The output quality that you're going to get out of that is going to be like top 1%, right? And I think that's the thing, you know, for anybody that's listening, like how do you take that domain knowledge that you have or the expertise that you have and basically capture it in some capacity and then provide it to, again, whether you're using cloud code or codex or it's an agent that's in the cloud, like give it to that employee to like basically have access to. And the impact you're going to get, like, and a lot of the conversations we have, it's like, that's what we hear. It's like, oh yeah, we tried to do this and the ad quality that we got was terrible, right? Like from Nano Banana. Like it looks, it's not on brand. It looks like, you know, slop, like somebody that did it in, you know, Microsoft Paint, right? And we're like, okay, like, did you give it brand style guides? And they're like, no. Like, did you give it the color palette that it has to be in? It's like, no. Did you give it like the language that it's okayed and like language that's not okay? It's like, no. And so I think more and more about this is like, I'm trying to like basically give AI like a boundary of like no's. Yeah. And then this is where you can live in. Because like when it's unbounded, it will just go and do, you know, basically unhinged things. But if I, instead I'm like, hey, here's the line in the sand. Like, here's the box that I need you to live in. And like, this is like, and it's easier to say that. Like what I found too is when you're like, this is what good is, it kind of gets distracted. And when you're like, instead like, hey, this is what bad is. Yeah. You should like do whatever isn't bad. It's going to create just a better outcome for you. So anyway, yeah. Just learnings again that we found that to be impactful for us internally. Yeah. That makes sense. So I have two final questions. Then I'll, I'll ask you where people can connect with you. So next question would be if a business owner, marketing professional, someone who's interested in this is listening. What is one thing they can do that can make a significant impact in their business if they want to get started with this? Yeah. I think that the biggest thing, so it depends on the organization side. So if you're just a founder and you're like, okay, how do I just adopt this myself? Again, I would get something like a cloud code or a codex and whatever your work is, like just co-work with it. Like literally have it in another tab and just go back and forth. As you're doing your job, try to figure out, can I delegate some of this work to it? Okay. So that's the first step. The second step is basically going and then incentivizing, like if you have an organization, like how do you get your entire staff to like adopt this? And this is the hardest part because like everybody's also fearful. Like how do I, I don't want this to replace my job. Like what is my, you know, what is my role going to be as all this gets adopted? I don't have a really strong opinion or answer about this. I think there's going to like probably be like some, you know, I think about it as like a snow globe. Like the snow globe is going to get shaken and then it's going to like tilts and like go into a new position. But like, it's still the same amount of snow in the snow globe. It's not like that's changing, right? It's just the jobs are going to be in like a different place. But if you're a business owner, like incentivize, the best way I've seen to incentivize people is to have like a weekly lunch and learn where somebody on your team shows what they did with like this AI tooling. And it's like, hey, I automated this job function, like a part of my day to day. And this is like how I did it, like everything I learned. And you're just going to immediately be like, you'll, you'll see people just become like, they're like, oh, like you did this. Well, this relates to what I'm doing. Like, let me go try and do that. And it also creates this feedback loop where you're trying to make like a hero out of this person that is using the tooling. Because if you set requirements, like companies right now, they're setting token requirements where it's like you need to use X amount of tokens per week. Like I have a friend that's at a huge organization that this is a requirement at his job. Oh, wow. Like I have, it's a payroll software you would know the name of. And the, like, he's like, Cody, my job, I have this requirement that I'm supposed to hit. All I'm doing is separating columns. Like I'm separating them and putting them back together. And that's how I'm hitting my token limit on a weekly basis. It's like, it's not actually providing any value. He's just like doing what's required of him. Right. I think that if you're actually trying to get this adoption, it's like, you have to create that like lunch and learn setting and then create that hero where it's like, you know, here is again, like, like Stephanie. And she basically like, it's going to share in almost like a setting where like praise occurs as well, where it's like, look at what she built. Isn't this amazing? Right. And like, it's like that spotlight is focused on that person, which in it, like it creates this like feedback loop where it's, you know, of adoption. And you can, you can start to incentivize stuff where it's like, I don't know, like tie it to revenue and those pieces. But I find it's always very hard to actually quantify, like, what is the real value that's coming out of this? What I've seen is like, when you're like, listen, there's parts of your job you hate. This is going to make you hate those parts. Like, it's going to make that part of your job way easier. Right. Yeah. And if you can facilitate that, that understanding and actually get them to start playing with it, you're just going to get like, so the speed that you'll be like of adoption that you'll be able to get to happen at your organization will be incredibly fast. So, yeah, that's great. And last question for you is, what do you think is coming with the future of AI agents? Do you think that all businesses will either need to adopt this kind of process or they're not going to make it? Or what do you see coming in the future, like 10, 15, 20 years down the road with AI agents? Yeah, I have no idea on that. Like, I didn't think agents, like in the capacity that we're even using them on a day to day now, we're going to be possible like this year and like they have been, which is crazy. I think the thing, you know, if you're a business owner, I think there's a lot of hype in this category and people think like, oh, I'm way behind. You're really not way behind. I think the biggest thing right now to think about, and again, there's going to be people that are trying to sell, like we can totally automate your entire business. That is not reality. Right. But we can probably like realistically, like maybe 40% of it. And it's like all the stuff that creates employee burnout. And you also hate to do individually. Right. That sounds like a pretty good, you know, outcome. So I think it's, it's also just like have bite-sized pieces is what always works. Like pick one workflow, try to automate that. Don't try to like restructure the entire organization in a weekend or, you know, something. It's just, it's just, I've never seen it be successful. It's always like small, like eat the elephant, like, you know, piece by piece. Right. And that's what always ends up being like the best, you know, outcomes. And then again, just like, I think if you just know, like just, even if you just learn about the industry, right. You don't even know they're implementing the things, but just kind of have a pulse on like what is happening. And like listen to conversations like this, where it's like, okay, here's possible. Right. At least you're going to have like an understanding and not get fleeced by some, you know, kid who's 20, who's in college. And he's like, I can do this whole thing that I, you know, and it's in reality, it's like, yeah, you can probably do like one tenth of that. And it will make incredible like impact for the business. Right. Right. Yeah. But the, the whole overhaul of the entire organization, like it's just going to take time. I mean, we're seeing this where it's just adoption is slow. It's hard. There's like, there's human parts of this too, right. Where it's like political internal conversations. Like we just talked to this large company and like they have 50 employees and it's like, they are entirely against the adoption of this. Cause they see the writing on the wall. Cause like, oh, my job is, is, as you know, potentially like can be taken away. Right. And so you're like having to fight that internally as well. And so I think the bigger thing is like, again, just the small, like, like it's boring really to be blunt. But it's doing the boring thing. That's always what actually makes companies work. Right. It's like, what is the most boring thing I can do consistently? And that's what actually makes growth happen. So anyway, we know that provides a lot of value. But can you share with us where people can find you, learn more about your work and connect with you? Yeah, absolutely. So it's graph.com. Like I have data and I graphed it past tense. My Twitter handle is Cody Schneider, personal websites, Cody Schneider. And then LinkedIn is, you just Google Cody Schneider. It should come up. If I don't come up, I'm doing my job terribly well or not doing my job well. But yeah. And then personal email, if you have questions, I'm more than happy to have a conversation with you. It's codyatgraph.com. You can feel free to reach out there. Amazing. Incredible. Well, thank you so much. This was such a good conversation. I know I learned a lot. So I'm sure that our audience did too. We really appreciate you being here. Absolutely. Thank you for hosting me again. I really appreciate it. Yes, of course. And to our audience, I hope you found this as valuable and eye-opening as I did. If you want to get started, you have everything that you need to kind of take this and run with it. So make sure whatever platform you're listening on that you subscribe, you follow, and you continue to stay posted for upcoming episodes. Thanks again for listening to the Growth Boss Podcast. Thank you.