Cody Schneider (Graphed): What It Takes to Build AI Marketing Agents That Work transcript

You have like a board meeting with your agents. You're like, how's that product going? It's very very good. Listen, I think you need to cut down on cost. I just be like, agent, run Facebook ads. It's like, holy you know? Like just like a human that that has never done it before. And it's also just like it's very rare you can find somebody that can make good creative, that can like actually run manage the ad account, that can analyze the data, that like can do all those pieces.

[music] So it's Like vibe coding from zero to 80% is pretty cool. 80 to production [music] and live is like a chasm and unless you know what you're doing, basically crashed out. I got banned from Etsy for mining their They weren't stoked about that. I still have a ban, which is hilarious. Oh, man. Oh, my goodness. One of those days, man. >> It's you. I know. I know. You know, totally. Yeah. I am I don't I don't know how you are these days, but I am in the mode where uh 4:30 a.m. my brain goes, oh, you should fire up some Clyde code tabs. >> Literally I'm like in the same I woke up at 2:30 this morning and I was like, oh, here I could go and have this graft agent that like does this thing. >> totally.

Yeah. And then my you know, my wife is like, what time did you get up? I was up at 4:30 and here's what I did. Here's what I've already had three coffees. Like >> [laughter] >> It's ridiculous. Oh. We're talking to this We're talking to We're in this with this private equity company and they're like, hey, we need to like source com you know, business owners that are wanting to sell their businesses. And we're like, okay, cool.

How can we do that? And this is literally like this morning. I was like, oh, I bet you I could go and look at the site map changes of like business sites, like sell my business sites. And then from that public information, like have Exa basically be like this company is at this location that is described like this. And then find the LinkedIn profile of the person and then cold email them and it was like, yes, was the answer.

And so literally like built that agent this morning for them, which was ridiculous, but I don't think people understand what's about to happen with all this marketing agents. That is crazy, man. But don't you don't you in your brain a lot of times end up becoming the GIF of like All right, wake up, Jarvis. Let's go. Exactly. >> Like, you know, and you just start like moving stuff around. So good to see you again, sir.

What would you like to build today? Exactly. [laughter] Exactly. Oh, um You know what's funny is when I was looking back, so I want to get into Graft. Yep. And and and the birth of of Graft and what I love about it and where you guys are headed. But funny enough, I was looking at your background sort of, you know, understanding your your trajectory. Um I emailed you two years ago at Swell AI and was like, I love it. It's great.

[laughter] And funny enough, I had found you off We just bought it back and we're like, we're we're resurrecting it. We sold it to an operator and it just kind of went to a terrible place and we're like, all right, we just our names were still sort of connected to it and we're like, pretty quick we're like, all right, we need to get it back and like refurbish it. So That's what has been happening for like the last two months.

It's like it was one of the biggest pains of, you know, of podcasting is trying trying to figure all that out. I I love the product. And funny and then you and I were talking about Danny Grant today and in the email that I sent you two years ago, I said if I found you off the exchange with Danny on X. >> Oh, nice. Small world, man. Small world. We yeah, I want to hear like so, you know, typically one of the things we dive into that I love, you know, I'm I mean I love Darius CEO and Patrol and all this, but like but what I really love to hear about is that like the week you you have an aha and you're like, oh, man, is this going to be the thing?

Like am I going to buy some domain names and now that now this is like this is what I'm doing now. Like I'm putting everything into the, you know, into this basket. You know, and then the >> bought trackable QR code.com cuz I have no compulsion [laughter] control cuz I'm like, all right, I'm just I'm literally going to have an agent build it like I'm going to have a graph agent like go and like build the actual product and then market the product and just see if we can run it autonomously, which is hilarious. just be like, how's it going?

You have like a board meeting with your agents. >> You're like, how's that product going? It's very very good. Listen, I think you need to cut down on cost. Yeah. Totally. But you know, more honestly, when I'm I'm thinking about companies, it's like either like this is a problem that I'm facing that like, you know, and there's a certain amount of people that have that same profile that this like are, you know, they're also trying to solve the solution for.

Or it's just like based off of search intent. Like for example, that that domain name, like the trackable QR code, it's like getting I think it's like 1,200 searches per month and like it's growing really rapidly. So I'm like, oh, this is like a problem that people are trying to solve. When you look at the solutions that exist that that are before them, it's like Bitly, right? Is an example. Um they're not really focused on that and then it's like, oh, who's buying this?

It's like restaurants and like, you know, companies that are sending um uh like I didn't realize this, but um a lot of um like consumer packaged goods companies are starting to put QR codes on their products as a way to like they have some offer that like takes the user further etc. Yep. And so they need a way to like a Ralph Lauren, you know, that sort of has like this is exactly what it is and and that's so great because you think being able to just actually shoot all the stuff in your closet.

Totally. Totally. I love it. Yeah. So anyway, stuff like that is kind of my origin of like when I'm like, okay, is this a a brand or a company? And I think there's like a bunch of different components of the you know, different levels of this of like you know, there's a fun side quest thing and then there's like, oh, this is like got the biology for like a large company. And then I I think that's just like a pulse thing.

Like I said, you know, call myself an entrepreneur. I just build businesses and like but at its core it's like you you as you do this more and more, you start to get a understanding of the market and like what actually has you know, again, the biology to be a large organization or what's going to be like this can be a cool side project that makes whatever eight grand a month and It's got legs legs enough to get to the eighth floor, you know, or whatever, which could be a good flip, but then yeah, that's cool.

Yeah, so I don't know if that answers your question directly, but >> [laughter] >> I and so funny enough going back, you know, I want to go back to to sort of like upbringing and you know, and and and are there parts of your DNA and your upbringing that sort of, you know, is is built into this. One other small world thing is I interviewed Tara. Oh, nice. Cool. on the podcast. So Max, my co-founder and I, like we we met at Ruphae Health.

I was like employee nine or 10 and he was right after me and so we basically our my who who recruited both of us is named Kobe Conrad. He's running his company now called Sunflower Sober Now and um but anyways, he like you needed to meet Max. I think you guys would get along and like immediately basically like we were both into techno and skiing, so we just kind of became friends. I think that I immediately like while we were still working at Ruphae, Wait, are we best friends now?

Yeah, basically. Yeah, I think we are. I mean, he probably wants to kill me some days. He's the better half, man, where like, you know, I'm out here just kind of running around on the marketing side and he's the he's the Oh, my gosh. the slow and steady on that he's the he's the That's Martina. That's Martina on on on that side. I'm the one like she'll show me something that she just built and it's amazing. And I'm like, oh, you know what phase seven of this could be?

And she's like, you know what? Can we just Can we stop? Yeah. Can we just hold on for a second? Exactly, man. Exactly. So like wait like, you know, where where'd you grow up and and was there anything in in your, you know, family dynamics or or just in the way you were built that was like, oh, I love figuring stuff out and, you know, solving problems or like where do you get some of those like Puzzles and trend analysis was always the things I was interested in.

Like give me a bunch of data, what does all this mean and like how do I synthesize it? Um I studied economics in college. I I think I asked like I was wanting to do engineering and then like finance, right? And then I was like I was like basically that the thought was, how do I make money and how do I have a skill to sell? Um And really like how does money work and how does how do I have a skill to sell? And then I ended up getting an economics degree, which makes zero sense, but Anyway, yeah, I mean background wise, I grew up in a small town in North Idaho called Coeur d'Alene.

It's now become super crazy popular since the Kardashians went there during COVID and it's basically like turned into this, you know, suburb of LA and San Francisco, which is kind of funny. But yeah, it's like an old mining and timber town. My parents were like My from LA originally and then basically we're like, we want to get out into nature. They were both kind of dirt bags as I would describe them, a little hippy.

Um and uh yeah, so we did that. My both of them are I build companies. That's all I really know. I've like had maybe I've started more of the companies than I've had jobs. >> [laughter] >> Yeah. Yeah, I really got >> actually shutting them down and like, you know, doing all the like all the paper trying to back out of Google Workspaces. >> [laughter] >> We just rolled a bunch of credit cards like recently for a bunch of stuff cuz we're like, we don't even know where any of this is.

And so we're just like we're like, cool, we're just going to roll all of these and like just see what happens. Whatever breaks, we'll go and fix and then, you know, restart. That's good. Um it's kind of funny. That was literally like we did that like two months ago. I I had to update a credit card this morning that was like needed it for you know, one of the one of the companies. But Anyways, um yeah, so started that.

Always was just kind of like buying and selling stuff. Um first uh you know, in high school I learned how to do like graphic design. Kind of taught myself Photoshop and Illustrator and then figured out how to do screen printing and then got a direct to garment printer. And then I was like, how do I sell things on the internet? And then I ended up uh figuring out how to mine Etsy for sales data. At the time it was public, so you could see the best selling products and I was basically doing a script to do extraction for that.

And then I was pulling that uh and we would remix the best sellers uh based off of the target keywords that they were going after. And um then uh from there I was like, "Cool." I was literally like running this out of my basement and like we were print you know we were like 20 and making 10 grand a month, you feel like God, right? Um So that was happening and then I was like, "All right, cool. Like how do I not do the operations?" And so a couple It's really early on print on demand and basically rode that wave when that happened.

Um basically crashed out. I got banned from Etsy for mining their data. They weren't stoked about that. Um I still have a ban um which is hilarious. Uh I like to every year I'll just try to like sign up and see what happens, but all my stuff is fine. Um But uh yeah. >> Lifetime Etsy ban. I don't you know, if you need to call me after we can talk about it. Like you know, if >> [laughter] >> If you um But yeah, so from that um uh But really honestly it was the best thing that could have happened cuz learned that like brand was such an important thing that you need to be building along with selling the product.

Um I you know, I was basically just selling commodities which is chill, but you know, you make money doing that, but it's if you want to build like something that has value that kind of compounds uh brand comes from that. Ended up working uh uh I was just doing like honestly I went into the fine dining uh industry for a minute and was like working at this uh hotel and was like I think I want to do this. Like it was like so intense and like very you know It's very intense and it was the type of I I kind of thrive in environments like that.

So there's this like moment where I was like, "All right, I'm going to go. This is like what I'm going to pursue." And I could you know, you could go and travel on that and that's what I was interested in, but Anyway, yeah. Um ended up meeting uh uh this boss though. His name's Jeff Reynolds um and he'd gone through Y Combinator and you know, done a startup um uh like basically went out into the valley, you know, made God came back and was selling it to Fortune 500 companies.

Here's how you do like digital strategy and like adoption, right? And so Nice. Um I basically went and uh did some contract work for him and he's like, "Cool, here's a test." I remember it so vividly. It was like a it was like a bathtub manufacturer. Like so unsexy when you think about it as a product and he's like I was like, "Yeah, I think Pinterest will work for it. Like I think you can like" they were trying to launch this new like shower base that was like a smaller model.

And I was like, "I think we could make like a e-book that's basically like here's how to do like a shower, you know, in your van or your RV." Right? Cuz I was like this like >> Nice. way that they were trying to do it. I was like, "I think Pinterest would work for this." And he's like, "Awesome, here's the test budget." And it was like just a stupid amount of money and cuz it's like this massive I mean they they sell to like Holiday Inn, right?

Like that's like you know, the scale like the scale that we're talking about where they're like, "Cool, we need like you know, >> [laughter] >> a couple thousand shower bases." It's like you know, or a hundred thousand shower bases or whatever the the size of these end up being. So anyways, yeah, I through that um like built this massively list on like from Pinterest. He's basically like, "What are you doing right now?" And I'm like, "I don't know, man.

I'm just like kind of around." He's like, "If you want a job, like come work for me." Um I was like, "Cool." And so anyways uh That's really like where I cut my teeth and learned B2B marketing. Like how do you sell over long sales cycles? Like how do you do account based marketing? Like where are all these levers that you can pull? I mean how do you like service, you know, large organizations, right? With these types of services.

Um so then like four years with him uh doing that and uh Yeah, just basically like my job was here are these budgets. Like these are the outcomes they want. Like go and figure out like how do we do distribution and you know, it's really just like a I was I lived in a studio apartment and I just worked. >> [laughter] >> Um Yeah, so that's the that's the hacker apartment lifestyle. >> Yeah, exactly. Exactly. And uh Anyways, yeah, just did that uh and then I I uh left this company.

I was just doing consulting. I was chasing snow for the winter. I was actually living out of a van at the time and uh with my my friend and we were just like chasing the storms on the West Coast. And uh my friend Kobe called me uh which was uh he was working at Rupa Health and um he's like, "Hey, what are you doing now?" And I'm like, "Nothing, man. I'm dirt bagging." And he's like, "Do you want to like come help me do this?

Like uh you know, at Rupa." Um and I was like, "No, I don't >> [laughter] >> I don't want a job, right? Like I'm I don't want to do anything. I'm just I'm you know, I'll talk to I'll talk to you later, right? Basically. He's like, "Let me send you the data and you can see like you know, what the the composition of the company looks like from a growth standpoint." So he sent me the data and it was just ridiculous, right?

Like it was like a curve that just looked stupid. And I mean it's the the the the best curve that you can have where it's like like slow incremental month over month growth with like 3 to 4% compounding. And then like you know, this massive curve and I'm like, "What did you do?" And he's like, "I turned on ads, Cody." I was like, "Holy shit." And so I literally I was like he sent me that on a Friday. Um on Sunday I was like, "I'm in.

I'm in now." And uh yeah, I was like again still living like in the van at the at the time of all this. Um And yeah, he was uh he basically um was like, "Awesome, I'll set you up with a call with Tara on Wednesday." I had the call, got the offer letter Thursday, uh bought a ticket Friday, moved to San Francisco that Monday. Um slept on a friend's couch for two weeks and then basically we just went and ran up Rupa and did that.

So was there, learned so much. >> those things though. If you think about you being in a van and sort of getting that data sent over. There's that like I get the feeling you're so tapped in also to just your you know, that you listen and these signals and and you know when to sort of activate stuff, but it really sounds like one of those where you saw it and you went, "Okay, here we go." Like there's there you know what I mean?

Like uh and those are so important from an entrepreneurial perspective like to not like sit and debate all the things uh in your mind, right? But just when you when you hear and feel something, you're like, "All right, done. Chapter as soon as you have signal, I think that's the like most important thing is to act on it, right? Like and it's the hardest thing as well. Like signal and noise especially right now. It's so frothy.

Like I you know, what is the right thing to be building is the hardest thing in the world cuz we can build anything. Um I mean we literally had a product meeting like two weeks ago where it was like, "Fuck. Like what do we what do we do? Do we go you know," cuz it's just there's you can build anything. >> 4/7 is out now with auto approve. So who can >> [laughter] >> Exactly. It's like it's like where does this turn into?

But I I think more and more like I'm realizing it's just like a business at its core is always the simplest thing. Like it's it's basically like what does the market want to buy and like how you know, how do I sell it? Like how do I get in front of them basically? Um And if you just like look at that as that first principle, then you know, you have all the signal that you need for like what what what I should be working on.

And so like we're finding this right now. Like we're like we got we just added this agent feature on top of us. Um we were data pipeline data warehouse and we were building this like BI tool. People were interested. We you know, we're growing, but we kept having people like, "Cool, I have this insight now. I want to go act on it." And so it's like, "Yeah, like that's what everybody wants to do, right?" And like we were seeing this within our like even our distribution or go-to-market where they're like, "Hey, like I want the full system.

Like I want the you know, I want the GTM engineering basically." And so we started like you know, in the last literally like two weeks offering that as like a hey, we'll forward deploy. We'll build out these systems and processes for you and build these AI, you know, agents for marketing and it's market's just been like everybody wants it, right? Like cuz it it it's you're telling me I can like basically hire you guys and I get a team you know, an infinite marketing team that scales.

Like that's based off of the actual outcomes that we're looking for and we I have that feedback loop. And I think the thing that's changed with all of this is like for the first time like there's all those AIS uh uh like SDRs that came out in the first like wave of all this. And uh people are like, "Oh, like you can't have this because it it fails." Like they like they they and it's they they basically like look at that as like the example of like why this won't work in this like system that isn't um like finite.

And um it's really funny cuz I when anybody that has that opinion, I'm immediately like, "Oh, you're a rookie. Like you don't realize that like growth is just like I'm just doing engineering, right? Like at its core." It's like I have an outcome that I'm trying to get. I have limited resources of time, money, and mental energy. And then like how am I basically like distributing, you know, those resources most effectively to create that outcome?

So that's just that's just a system. If it's a system, then I can you know, turn it into a process into into software, etc. So anyway, on the to go back to the AI SDR thing uh what I find really interesting is when you do a like a postmortem of what actually happened, the agent was performing unbelievably well. Like it was getting leads. The leads were just terrible quality. So basically it it self-optimized into like you know, it's it's a great example of this is like if you run Google if you've ever ran Google Ads, um you can have this happen where it basically will self-optimize the ML will self optimize into like a a a a subset of your audience that that it's bidding on that is like creating a sign-up event at a cheaper cost, but it's not it's it's like the wrong person to get in basically.

So it basically ends up getting >> cuz it thinks you want lowest CAC ever, you know, exactly. And so it's like if you have the wrong signal going back to it, then that's like it's not the agent's fault. It's just it's doing its job, right? It's just like a human. It's just basically it's optimizing for the wrong metric. And so what we we've kind of realized now is like if you have this whole data like solution where it's like all my data's here and I can provide that to the agent and I can be like, "Agent, I'm trying to make revenue.

Here's all of the resources that you have available. Here's the systems that you got." So to give an example, like I on this live stream like two days ago I built this AI for SEO agent. Uh or sorry, this agent for SEO. So it basically like does keyword research. It researches what's ranking on page one of Google. It then I gave it like a transcript of me talking about this specific topic category. Um like how AI is changing Facebook Ads is this is exact example just to be explicit.

And then based on that it writes the content. It publishes the content. And then it like calls the web indexing API to index the content, and then we have another agent that's then looking at the live data, optimizing that article based off of the feedback loop that we're getting from Google Search Console. And then from that, on top of that, there's another agent that's basically looking at, "Okay, what pages are actually making sign-ups?" It's looking at the post-hoc data.

What pages are actually turning into revenue? Let's have that go influence the next round of creative. So, like it creates this feedback that when you when you look at any like good go-to-market like system, that it's that exact process. It's like, I try things, I look at what works, I do less of what doesn't and more of what does, and I just like repeat that growth loop. And then to make a company grow really fast, you're basically just layering on this growth curve, right?

So, I do that one channel, and then I layer on channel two, and I layer on channel three. You never stop what you're doing, you're just layering on these next elements. So, um Anyway, yeah, I I I think that it's just like that has been the realization that like we have come to, and I think everybody has come to. And then the infrastructure around that ends up being like the hardest part, right? Where it's like, "Okay, how do I actually give my agents the things that they need so that they can go and do, you know, these tasks?" And also, it's just like nothing is built for this.

Like if you were if we had this conversation like 6 months ago and somebody was like, "Oh, should I pick HubSpot or Salesforce?" I'm like, "Okay, it depends on where your company size is at." But like, you know, HubSpot hit historically, like I would pick it cuz the UI is more like user-friendly, and like I you know, I can do more within there like as a as a normie that's not technical. Um now I'd just be like, "Absolutely not.

All you're looking for is like what has the most APN like API endpoints, and like what can control the product at like the largest like scale via the API?" And and per like because agents are going >> And we were talking about Life Field. Like that's one of those I was going to look at CRMs, and I was just meeting with the co-founder of Life Field. And basically what you're getting is a cloud code plus an MCP that happens to be a CRM.

And you know what I mean? And just being that brain instead of the you know, all the drop-downs and the navigation >> in too like at this point? Like we're like Yeah, yeah. I mean, I haven't touched HubSpot, right? But I've been using it. Like we're I'm in sales cycles right now. I'm literally living out a graph that has an API key to HubSpot, and I'm like that is my whole interaction with it. And then I'm like, "Agent, like keep tabs on my deal flow.

Like tell me what I should be focusing on, like who hasn't followed up, like you know, like I can build out these systems on top of this, right?" And so >> And this idea like 2 years ago when you you know, when when ChatGPT really sort of emerged into the into the zeitgeist, this idea that like, you know, website pages, the way that we know them now are going to go away, but just this mythical thing. But what's interesting, it doesn't happen the way that you think it's going to happen.

There are no websites. It just so happens that the thing that you used to go in and do all the drop-downs for, you just never touch it anymore. Yeah, exactly. You know, I think the other thing that's like we're realizing too, and we're getting this feedback from users is they're like, "Hey, I want I want the software to evolve and mold to be like what I need it to be." Um so, like we built this dashboarding solution initially, and then we have people that are like, "Hey, I want to I want to add this like, you know, basically like a drop-down filtering mechanism." And we're over here and we're like, "We're getting in the way of what the user like the user can just ask the agent to add that to the dashboard." And then it just like does that, right?

Like we just give the agent a skill so it knows how to like dashboard like effectively. But then that those those final, you know, mile like solutions that they're trying that are really, you know, custom and specific to them, you have to almost make like dynamically generating UIs to do this. And so like how we're solving this right now is like, "Okay, say I want to do a dashboard." So, we built a dashboarding product initially that's like this like very specific like, you know, it's like components and modules, and it's basically just like you can vibe code a dashboard with the idea, right?

Um But now uh now what like where we're going with this is we're like, "Okay, we're just going to give the agent like here is an endpoint where you can create an artifact, and like you can basically go and design whatever dashboard like the user wants, but you can host it here, and like this is in the safe, you know, space. So, it's like it's not being like publicly created or added to like a website, but that can be shared if you want to.

Like you can have like a shareable link if you want to like give that to a to somebody external. Um but what that quickly does is then you realize like, "Oh, it can make any document type." Right? So, like I'm talking about a dashboard that's on top of your live data. So, say you're trying to do reporting on your paid ads, right? And you're like, "Cool, I built a custom reporting paid ads dashboard." But I can then be like, "Agent, build me a slide deck that has these charts like as PNGs in this slide deck, and gives a report, and like, you know, I want you to create that on every Monday and send that to my client." And so because it has this dynamic UI generation, it's like now it's like any asset that you need that's like a snapshot in time or a deliverable, it has the ability to go and create that for you.

And it just turns into this like really, you know, I I also the surface area gets really scary as like a as a company and a founder I would predict this experience with that when like the UI is dynamically generating, and then it's like dependent on the user to have like the right way that it's describing the output. But like again, what we're finding more and more as companies are like, "Hey, like we want this done for us.

Just like, can you just air drop in and build out these solutions and then maintain and run them for us? And like instead of us going and hiring 10 people and having to pay those salaries, you know, we're going to pay you like 1/10th of the cost basically to do that." Yeah, yeah. So. And I think that, you know, and that's where that's where you and I sort of re-intersected was when I was right at the place where Martin and I spent, you know, June through January through March sort of heads down rebuilding for enterprise and teams cuz the the one-to-one accounts weren't working.

And then it was ready, and we started selling to businesses, and businesses started taking off. So, we thought, "Okay, we really need to build the go-to-market, and it needs to be not just me talking to people." And so and so I started looking around, and I I hired a few teams. And the few teams that came in were as I told you, it was like it was like the older school like, "Let's take a list and put it into GoHighLevel, and then let's email those people, and let's see if there's any movement." So, like You know what I mean?

Or whatever. And you sort of spend like 2 weeks doing that or whatever it is. And it's, you know, and and so you think about those old cycles of standing up SDRs and and AEs, and then teaching them and training them. They got to come in in the culture. They got to get in a notion. They got to do email box right or whatever. And how far away that is from being able to spin something up and get signal this afternoon. Uh and and start iterating on it is >> has changed that too.

Like we we didn't even have this like, you know, done-for-you service or this forward-deployed service. And I was like basically just like, "Do people want this?" Like let me just like literally put it on Twitter and LinkedIn. And I'm going to throw up a tally form, and we'll see like what happens. And like we got like over 100 people to inbound. I'm like, "Oh, there's an appetite for this." Cuz they're like everybody wants I think this is the thing too with all this AI agent stuff is like everybody wants this.

Um but they like they conceptually understand it, but the actual like build-out and implementation of this, the gap between that is is so large. They just want the outcome, and they just want to pay for the outcome. Like can I just buy the outcome? >> like vibe coding, right? Like everybody said you even had investors being like being like, "Can't someone just vibe code this?" And and and not understanding the idea that like vibe coding from 0 to 80% is pretty cool. 80 to production and live is is like a chasm.

And unless you know what you're doing, How do I do agent evals so that it hits brand style guidelines? Like how do I do agent evals so that like the image that's coming out of Nano Banana from, you know, these two different API endpoints that we're hitting, like there's a feedback loop where it's like, "Oh, you're not We have an agent that's observing, and it's like you're not hitting these brand style guidelines. Like you have to go back and like do these modifications." Cool, now it passed that eval, goes to step two where that's uploaded.

Like all of those pieces that you just when you get down into like the, you know, the weeds with it. Like cuz you talk about like an AI agent for like SEO as an example. And people are like, "That's easy. Like I give it a keyword, and it writes it." And then they do this, and they're like it doesn't rank. It doesn't, you know, there's I I'm not getting anything to happen from it. And then you like when you look under the hood of like what is good content, there's actually so much that's happening to actually make that, right?

Or like what is good content? >> between an agent conceptually, which feels like a robot that does instructions, which what you're talking about is actually a a special like um a special team of agents with instructions and orchestration that are working in concert, which is totally different, which is not what what people understand with the idea of just agents. For sure. We think about it as like agents forms where they're like, and this is like again, like we're very early.

I'm barely sleeping. We're figuring out like how do we do all this, right? Um But like what this what we're realizing is like, "Okay, for if you want if I just be like, 'Agent, run Facebook ads.'" It's like, "Holy you know?" Like just like a human that that has never done it before. And it's also just like it's very it's very rare you can find somebody that can make good creative, that can like actually run manage the ad account, that can analyze the data, that like can do all those pieces.

So, it's like And you wouldn't know that for at least two, three months until you really got into working with someone. Totally. >> Yeah. And so what we're finding though is like if you give a scope of work to the agent where it's like, "Your job is research, right?" So, like you are researching the pain points of our target customers on social media and the outcomes that they want on social media. That's all your job is.

You just basically like scour the internet, and like ingest all, you know, you build a corpus of information. So, that gets handed off to uh you know, the script writing agent. And their job is to write scripts based off of that. And uh they They like, you know, have access to the HeyGen API so they can make the raw video file and then, you know, there's like a loop where it's like make the HeyGen, you know, make the the AI UGC ad, remove the the silence, like add captions, okay?

That then gets like uploaded into, you know, the Facebook ads account, totally separate agent. Managing the ads account, here's these rules that you have, like if the CPM hits 50, that goes off. If it's like, you know, you see a a CPA of X price, you know, up basically promote that to its own ad set, delegate budget to that and then, you know, if you start to see a CPMR that's going like like higher like at the end of the ads life cycle after 30 days, then I want you to automatically turn that off because it's basically we're seeing ad fatigue happen, right?

If I try to get an agent to like think about all of that and do all of that, impossible. If I try to get a human to do that, impossible. It's just very hard. And then the other layer on top of that is like, okay, now I have an agent that's like needs to analyze what's actually working and give that feedback loop to the rest of them, so they're like ads like this work. >> [laughter] >> Based on what you're seeing, like you know, how do you do that loop?

And so, how we how we're starting to approach this is like imagine you have like just like you would have a team, I have like a team of agents and they're like all listed in their own little sandbox and they're working on, you know, their jobs and tasks and then I have these like ICs and then I have a manager and then above that you probably have like, okay, like this whole other, you know, analytics system where it's like that's looking at Facebook ads, it's looking at Google ads, it's looking at SEO and looking at the composition of the entire organization.

I think these are all the things that are important. >> It's not a mental model that we actually are used to, which is why I think it's it's so interesting that you're you're on the like the cusp of this wave of figuring out how how this is all coming together. It's not a mental model that we're used to to think, "Hey, this afternoon over coffee, I'm going to stand up and a micro company of 12 >> [laughter] >> and have them start doing all the work and to know what to do." Also the speed Yeah.

The speed of this is crazy too cuz like we had we, you know, talking to a startup this morning on a sales call and they're like, "Yo, we need GTM like tomorrow. Like how do we, you know, how do we get to that?" He's like, "We got to go raise our, you know, our A like we we have PMF but we we have to go raise our A. We need it we need to get velocity going." And he's like, "What's the He's like, "What's the onboarding look like?" And like I could just tell he's like thinking like, you know, 30 days or like, you know, to get this to scale.

And I'm like I'm like, "Yo, we've we can deploy an agent in like 40 hours. Like as fast as you want to move, like we can do that. It's like it's it's really just at your discretion." And what I'm not talking about like one channel, I'm talking about like all the channels. It's just the same playbooks like ported over. Like if it works for this SaaS company, it's going to work for that SaaS company, right? And then then the problem becomes like, okay, this is what's happening.

And everybody's doing this. Like, you know, we fast forward 12 months or whatever, 24 months. How do you create differentiation? Like what is that? I don't know. I don't I keep getting asked this question and so I'm just like addressing it kind of like head-on. I have no idea where this goes. I don't know where this ends up. I just know right now that like this is working. You can like you can build you can finally do this where like I plug in and a company is marketing itself.

Like all of the technology exists for that. Now it's just building the systems and the operations around that and like the the standards for the agents so that they have like a caliber of work that they have to do too. So, anyway. Yeah. Oh my gosh. Like this where you see with with I know it's What's so funny too is you think about it with the the life cycles have shortened. Right now if you said, "Where where would you see the company in 5 years?" It almost seems seems strange because it's like, "Well, I think I'll have flying cars and and all stuff like you know, like how how fast stuff goes." But really like where where would you see even if you jump 6 months based on what you've sort of learned over the past 6 months?

Like where where do you see Photographed headed? Yeah, I think we're a bubble. Um I think we're first in a bubble. Like everybody I know that's doing this type of work is like they're like, "This is on the horizon." And like I think it this is a a shift that's going to take 20 years, right? And then like by that time there's going to be another thing that's the next thing. And, you know, like hopefully I don't have to compete in that next cycle. >> [laughter] >> Yeah.

Uh but it's going to be gnarlier than this one, right? Yeah. >> But but I think I think with that though, like how I see this evolving, um and just like based off of like what people are asking for and like what the market and and what I know mainly is like again, I know like go-to-market agents, right? Like that's like what we're doing. So, um but I I think that these companies it's going to be just really hard to hire this talent internally.

So, you're going to have all of these organizations and again, they don't want they don't want to buy like we're seeing this more and more. They they just don't want to buy a self-service function. They want to buy like an outcome and then you're just basically building virtual employees for them. And I think that that is more and more what's going to happen is like Also I think as the the shift of understanding and learning is you to your point about not hiring internally, unless you have all neurodivergence that stay up all night all the time, you know, it's one of those that the company can't shift fast enough to be able to, "All right, everyone, we're going to start training on this like process, uh you know, and so you have to buy outcomes because the only people who understand how to do it, you know, it's going to 3 months from now it's going to shift when Opus 5 comes out and it's just going to be like 100% different." Totally.

And we already see this like, you know, we we ship a lot of what we're doing with like the mini max 2.5. Super capable for the cost. It's unbelievable. Like the the cost, you know, if you just make that assumption too, like the cost for intelligence is going to like approach zero. Like we're going to have We've already seen, you know, two cycles of this where it's like frontier model gets released, like open-source model gets trained off of derivative data and they're, you know, fast follow 3 months later and they're at 1/20th the cost, right?

And like that's basically like you're going to continue to see that same, you know, like thing occur. Um Yeah. I think that like I mean, that's the bet that we're making as well. And every time that the models upgrade, like we just see the outcomes that we can get them like things that we can get them to do are just that. Like it's a step function each time, right? And with auto approve, you can even see start to see the future a little bit of >> Totally.

I'm going to give you everything you need. Well, I've got agents that will end up doing the plan. The plan has been tested. The the plan is bulletproof. I'm going to do a one shot and this thing's going to go at it for the next 2 days and then, you know, >> Totally. uh I built a directory site like 2 weeks ago just to see like for fun. So, this is it was when I was initially testing like Hermes. So, I was like, "All right, can we just like have like go scrape every business in this category and just like pull out everyone you can find in the US that's in like a medical space." Um went and did that.

And it's like, "All right, like let's create an action plan to build out like a directory website that has like, you know, this structure and on the homepage you can like filter by location and like that those pieces. Needs to have a login and then I wanted like add it where it's like, you know, they pay to have their listing show up at the top, you know, so that like in SEO geography. Whatever, right? Um and gave it that plan like, you know, dangerous permissions on cloud code and I'm like, "Go run." And it's like I was like, "Cool, I'll probably, you know, end of day it'll probably be done." It was like 45 minutes later it's like this thing is completed, right?

The agents that like collect the data like took time but like the actual one to get from an output it was like nothing to build the actual product. And I I think that that is like, you know, is it it's a great great fortune-telling for like what is what is what is coming down the road. And like I I don't see this again, if if you have I'm really obsessed with like can a company market itself? Like like can I Like all of it exists.

All the data exists. Like the intelligence exists. That's all you need. >> Yep. And then it's just like, how do I create that feedback loop based off of the action? Then the hard part becomes what is the action that I want to happen? So, it's like it always comes back to human problems of like, "Okay, what am I optimizing for?" Um cuz then it's like, you know, you're pointing like a nuclear weapon at whatever [laughter] this thing is.

And so and so it's like, okay, if you choose wrong on like what it's optimizing for, then it's that you're just the outcomes aren't going to be what you expected. And so, uh anyway, yeah, I I I think for us right now like week by week it's changing and companies like there's no way they they can move at the velocity and speed of this right now. So, I think there's just an a huge for company like for startups to like go and build these types of I don't know what you want to call them like agents as a service or like it's it's there's there's all these different terms.

They're like service businesses are the new like >> But if you're not diving in and breaking a bunch of stuff and learning, then it's going to be real it's going to be tough to catch up. You know what I mean? >> going to be a rude awakening. We're we're just onboarding this company right now. Um like this old boomer entrepreneur um owns like 10 like 10 businesses in his portfolio. Um the guy that, you know, is running point, younger guy is like technical.

And um like the you can just tell he's like, "I know this is coming. I have no idea where to start." And like that, if you just like take that and look at the entire market and be like, "Okay, like that is every person's feeling." Um how do I go and serve them, right? Like there's just so much to to to do within within that. So, I I think also like on the personal like there's things that are just like fun. Like you can have agents that are like I mean, we we've been throwing this around uh like my fiance and I have like, can we get agents to like just start cold emailing like people in geographies asking them about like things that are happening in their community as like a a way to do polling, right?

So, like polling I used to like call and like do thing. But like imagine like I've been in an email chain with this thing for like whatever, 20 messages. And then it asked me a question about like, you know, some bill that's about to go through or whatever that is it being. And like I have this relationship that's been cultivated at scale like with this, right? Like what what happens in that that situation? Like it like there's these things that you can do now that are just like incredible, right?

I mean we we've been experimenting like with that. We've been experimenting with like can I have like autonomous like news, right? Like where it's like We we started with like podcasts where it's like a basically like we have an agent like research like a podcast category sorry like I say for example it's like e-commerce and it's like cool go find me like weird e-commerce companies that nobody's like heard of but that are like growing quickly.

Like one was like this like random mushroom company functional mushrooms from like Canada or something. It's crazy. But they're like growing really fast. And I'm like okay, go find me everything you can about them. And then I want you to write out like a monologue script and then 11 Labs voice put that in and then we do an email newsletter on the back of it of like, you know, basically promoting the podcasts and and you know, then we you can put ads in the email newsletter, right?

Totally autonomous running on its own. I mean like list is at like 20,000 or something and it's just like I I don't know what any of this looks like. I think I think what's going to happen I see is you're going to do this a few times and then you're going to flip over on the VC investment side, but you're going to end up having these agents that sort of like and you go yeah, invest 2%. Yep, [laughter] invest 5%. Like you just start like collecting all these pieces of companies just because of all the signal that you're getting.

Totally. I think now too it's like you can just touch the data of these companies and instantly know the health of them. Like I it's wild man. Like if you unify all of the data and this used to be a massive lift for organizations, right? Like if you unify all the data and like >> was it Omni? No, what am I thinking of? Like Yeah, yeah, yeah. Totally. Even the BIs that like for a startup to get into is like well, I talked to one it was like all right, it's about 30 grand a year to like get started or whatever and you're like wait, what?

But you still need to bring in the consultants who end up standing it up and they try and sort of, you know, sanitize everything and I mean it's just ridiculous amount of money. Totally. And I this is going to be a become a large problem too is like the agents are assuming about the underlying data. And so like unless you're teaching them the ontology of the underlying data, it becomes a massive problem. So like a great example of this is there's like 96 tables in Facebook ads alone and there's like hundreds of columns of like you know, in each of those tables.

Um if you ask it like, you know, show me link clicks or show me show me CPC over time, right? There's like all link clicks like ad link clicks Like what are the rates and notions for every block? Exactly. And then then maybe like adds them together. Like you it doesn't know like, you know, it's just it's it's it's it's guessing off of its own training and like what it thinks that column title means in relationship to like what you asked.

And so what you have to do is like you basically have to like teach it like okay, when somebody says this they mean this and then you can actually run like an eval program over the top of that where you're giving the agent more context on like what the underlying data set means so that in the future like maybe it and you know to just give an example it's like okay, maybe it like it it tries five times to write this SQL query to pull in, you know, the data that you're looking for.

Um and like that's like a failure, right? Like it's like that's not quick. It's not doing you know, it's being very inefficient. But if I can show it like hey, here's how to do that and then basically have it test, you know, basically runs its eval program to like improve when, you know, user asks for something that's in the shape of this this is what they mean. Here's how to write the SQL code and then it one shots the first time.

So you can get it they can get more performant under the hood. And I think that's this the this whole other thing is the agent evals of like both the data and like the decisions it's making on which is going to be a huge problem because the worst thing that you can do is make decisions on bad data. Like that's why everybody is like again the AIBDR thing or like I see so many people right now that are like I just hooked up Cloud Code to like the Facebook ads API and they're like look at everything I can do.

And I'm like I guarantee you the data that you're getting out of it. Like if you're trying to do data analytics out of the Facebook ads API I and you're running a legitimate account where you're like you're spending real money. I guarantee you that that data isn't correct. Like you're going to be >> You drink from a firehose and and you don't even know what to to analyze. >> the the like you're you're it's just and they don't know like they're not technical, right?

Like it's like a marketer that's doing this and so they just like they're assuming that this is right and it's like that is not representative whatsoever of the actual like in it like base data. So the information that you're turning that data into is entirely wrong because like you're running into rate limits, but it's not telling you that. It's just like yes, here's all the data or you're having like truncation errors when you're using some MCP where it's like again telling you it's all of the data, but it's not.

It's just like page one because it has trouble with page nation or you're like okay, uh there's like endless, you know, challenges that happen from this. One we see all the time is they're basically like just stuffing all the data into the context window of these agents or like of these of these models and they're like okay, analyze this. And it's like what? Like this is tabular Read war and peace and if you can tell me, you know, like you know what's funny is with 47 coming out with Opus 47 coming out today, I saw it do something really unique which I was actually working on something when the switch happened.

I was working I have a Webflow agent that goes into Webflow MCP and a few of our clients that we have helped sort of level up or build their websites for and I connected the MC the Webflow MCP in Opus 46 and it had that problem where it would try and go and and take every little piece of every little page and figure out what was what. I [snorts] did it today on 47 and it said, you know what this is taking too long and it fire crawled the whole thing on the outside and then said okay, here's what we do.

Like it basically knew that it was just taking a dumb route and then it went out on the outside and it read the pages from the outside made markdown files from them and then solved the problem from this which is crazy. It's crazy. I I Uh It's like those those are the things where it's like it's escape patching itself in a way that is so like unexpected and exciting. Um I the challenge with that though is like how do you create safety in it?

And so like this is something that like we're figuring out like in real time, right? Cuz it's like there's only so much you can do. But The repo wasn't organized well, I deleted it. I'll make a new one here. Like yeah, yeah, yeah. Exactly. [laughter] Exactly. Exactly. What are the limitations? So like we see this cuz we interact with a lot of API keys, right? Like it's like oh, you want to publish something to your CMS like we need to do that through the API key.

And um so how we're handling that is we basically like the agent is actually never touching an API key. We're basically storing it in like a secrets, you know, file that's local. We give the agent like a, you know, a ghost key it thinks is the Webflow key. And so like when it's interacting and writing, you know, the code that it means it's basically doing that like using that key. And then right as the the call happens we're basically proxying in the real key so that like the agent is never like interacting with it.

Um and what that creates is it's just like at least some So it's like you don't have like who knows where this inference is going. Like is that API key getting served up? You know, what like what is happening on the outside of that? And then the other things that we're doing is like we actually like a remove the agent's ability to like use the web. We just like gave it like fire crawl and then it you have to approve like its ability to go out and do specific things.

Um so that it's like again it can't get prompt injected from a random Reddit post that it reads and suddenly your, you know, Mercury bank account information is now in some exactly. Anyways, but I think again these are the these are the problems that it's like we're all facing we're all trying to figure out and like we're all trying to solve that are going to just be like trial and error. And I I think that's also like the hard part for these companies is like they don't even know what the surface area from a safety standpoint for this stuff looks like.

And we all don't, you know, we're all just building like the best version that we can of this. But with anything where it's like if you're accessing the internet, like you are a vector, right? Like a human is too. And so it's just thinking about it in that in that mindset as you approach this like with agents, you know, working on the behalf of businesses. Where I think it gets really exciting though is like what we're starting to see with these like these services that are like for agents.

Like our friend is building this company that's basically like it's an it's it's data for agents, right? Like oh, you need like >> Email inboxes Totally. Exactly. Exactly. >> Like so cool. Like you got you have a Gmail box for your agent. And so then you end up running like rock reports and those come in via email and then the agent can pick up that email and go okay, I'm going to take all these things which is amazing. >> Exactly.

Exactly. And so I think that will be really interesting to see what happens. And like we're also like we want to experiment with this of like can we like give an agent a credit card and like it has like a $50 limit and like we're like go do this thing. And then it comes back and And all of a sudden you're like steak dinner, $250. Exactly. Exactly. Exactly. And it was like but it was on sale. It's like [laughter] all right.

Yeah, exactly. I saved you so much money. Totally. Totally. >> Yeah, I it's it's a it's a fun period. What would you say to someone like So as a like typically I I ask you know, guess who were on like if you if you were going to have sort of tell yourself when you were first coming up or entrepreneurs. But I think what's interesting now with what you're pushing into is I would love to hear what your advice would be for let's say an entrepreneur that is either early in their journey or or has has an established company.

They're sort of going through pre-seed or seed and but they are they're a founder but they wouldn't call themselves technical and they they maybe they use Cloud Code or ChatGPT, but like in the desktop or the web app and that's about as far as they've gone. Like what would your like okay, this is Cody's homework for the next month is you should go do what? Yeah, I I think that the most valuable I mean everybody's saying this right now, but like I'm just going to parrot it cuz like I'm seeing it.

Like I just had a friend who got hired at a company. He's getting paid like 650k a year and like literally it's because he's a go-to-market engineer and he's like you hire me. It's like I got 20 like work clothes and agents behind me. It's like you're hiring a team, right? And you know, if you're a young person and you're like you know, I know I want to start a company someday. Um I I want to try to get a job at one of these organizations that I can go and and and you know, provide value to.

Um it the the pool of people that know how to do marketing and know how to use like this tooling to actually like do the middle work for you is like Yeah. I mean I it's a handful, right? I can call them on my phone. Like that's >> [laughter] >> Yeah. It's like it's like it's like it's essentially it's like there's a hundred brain surgeons that know how to do this one very specific thing. Uh >> And and and with that like I think that it's going to become like more you know, adopted.

I think it's just going to become the standard as it as it as it starts to go through this ecosystem, but right now the arbitrage is that like if you become become like a marketing engineer or go-to-market engineer, the the GTM engineer has kind of evolved. They're Originally that was just like you were doing cold outbound, but now it's like okay, you're doing everything, right? And like the marketing engineer is like okay, you were just doing like paid ads management, but now you're doing cold outbound and like it's all it's kind of evolving into the same, you know, role.

But I think it's there's something in this space >> design and development, right? You end up having you know, my what Microsoft just said that they basically are taking people who do content and design and development and they're hiring that one person and just fusing those things together. >> And and and when you look back at like the the people that always made impact, they they typically had a combination of skill sets.

They weren't like deep T in one thing. They were like deep T and like two things and then that like you know, an example of this is like the founder of Polaroid, right? Where it's like deep T's in a couple things and like that's what makes this like unbelievable company, right? Um and I I think that that is like what's happening here. Like if you can get pretty competent and like you can get so far now with Claude code.

Like it's like Again, I it's just looking back like at my I was like scrapping to get YouTube videos that was actually showing how to do stuff and now they just like exist for literally everything, right? Like every tutorial that you like in a weekend you could learn how to do paid ads cold email SEO and also you know, probably like email nurture campaigns. Like that there, if you have that and you go to a company and you're like yo, I can do this, but I can do it out of scale because I'm like you know, using agents for this.

Like you are now like top 1% basically. Right? Because it's there's just so few people that understand these pieces. Like it's and they don't know how to deploy they don't know how to use this new tooling that exists. So Yeah. Well, and I'm going to get I'm going to give a little graph plug because I love I love it so much, but I also So like this is a great example of and I'm not an engineer by trade. Like I I poke around with everything and I I want to understand everything.

But >> [snorts] >> uh when I when I signed up for graph um there was a Saturday uh not too long ago that I uh sat down with a cup of coffee in the morning and I hooked up uh Apollo um our Stripe two Stripes, QuickBooks, Brex and PostHog Analytics and um by about 2:00 p.m. had a full dashboard command center of basically all of our cash, our burn scenario modeling all of the product analytics, but like all in one thing that only my co-founder Martin and I could log into and I could I could click a button and download the markdown file for all of that scenario.

And then drop it into Claude and go what if? And I could what if forever. And these are things that normally as a as a startup founder would have been taking a bunch of different things and getting freelancers, getting specialists, getting you know, and so and now essentially with Claude code and graph I have a command center that I probably would be at least paying, you know, 1,500 bucks a month for whatever plus having specialists and analysts doing it.

So I'd probably be spending a couple thousand dollars a month. But that literally now it is this is running our business, which is crazy. It's so like I you know, I'm building this product, so I'm biased obviously, but like the stuff I had to do in it like would take me literally hours or be impossible like previously. Like I you know, I taught myself Looker Studio off like YouTube. It's like I have to click 20 buttons to get this like filtering mechanism.

Like you just you just >> anymore. I like graph and PostHog by the way. Oh my god, it's it's crazy. And like I think this is also something that the PostHog thing I just you made me think of this, so I just want to piggyback off of it. The setup now is going to happen by these agents. Like we just I I needed to do like get uh we switched from Amplitude to PostHog. I was like I don't want to set this up. And so I literally was like here's the API key, just like do it.

And it went and built out like what we needed. It had access to our codebase and you were needed to like put the like >> And PostHog is the best example I think of we actually went through and I hated like I went through the Google Analytics, then I went to Mixpanel and I had two consultants in two different periods coming and saying we're going to take a while and analyze how all this is working or whatever and I couldn't stand I spent so much money on that.

And then when I dropped PostHog in and just started watching the data over the next little bit, it's so fascinating to see how PostHog really is just a big SQL lake. Totally. >> can just ask it all the questions. And I think that's where and and that's very similar to graph, right? That you're basically like great. Now I have everything in one place and we can ask all the questions. I just love that we're getting out of all these like expensive you know, overhead layers that that were unnecessary.

Um and now we can just get to the data to make the decisions. >> Totally. And it's just the time to insight. Like we found this when we were just running the analytics product. It's like how do we get the first insight to be as fast as possible? Like that was like our like metric of success for you know, activation. And when you think about it like it every person at a company your company right now, if you're listening to this, every person at your company right now has a a data question that they wish they could ask that they like can't get to the answer to currently because either there's not enough bandwidth on the data team or like if they do put a request in, it takes three to five business days and if they have a follow-up, it's another three to five business days.

And uh if also the data team is always prioritizing whatever finance and the CEO needs. So it's like they're getting jumped in the line. If you unlock them to where it's like okay, I can now interrogate the data that I live on top of to understand this >> my Claude code desktop is Reddit still our number two referrer? And it goes hold on a second and it checks graph and it goes yep. It's amazing, right? It's just like >> That's it.

Which is which is crazy. Uh Totally. Amazing, man. Thank you again for hosting me and sorry again about this morning and I I No. No. No. deal with the dog and then it was just like I was behind on calls and the combination of all of that and stuff. Anyway, >> Like the the I think the life we've got is this shell game of being like a little bit over here, a little bit over there, a little bit over here, you know? And like and everything everything bumps into place.

Thanks for the time and I I appreciate it.