YT YouTube2Text

YouTube2Text → Transcripts → Marketing Agents Are Too Good Now

Marketing Agents Are Too Good Now — Transcript

by Greg Isenberg · 7,447 words · 1,127 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Let's be real, the hardest part about
0:02building a startup today is getting
0:04customers to the startup. You can vibe
0:07code stuff now super easily, we all know
0:09that, but getting customers to drive
0:11revenue is hard. So, how can you
0:13actually set up marketing agents that
0:16help you do this? I feel like a bunch of
0:18people have talked about this, but no
0:20one's really gone in depth and clearly
0:22explained what a marketing agent is and
0:24how you can use it to go [music] and get
0:26you customers. In today's episode, it's
0:29two parts. We give you a startup idea
0:31that I really love, and then we also
0:34explain what marketing agents are, how
0:36you can set them up, what are the tools
0:38you need to understand if you actually
0:41want to be a marketing agent manager,
0:43>> [music]
0:43>> and Cody Schneider just does an
0:45incredible job on this episode of
0:47explaining everything end to end.
0:49>> [music]
0:50>> Um so, enjoy the episode and I'll see
0:52you at the end.
0:55>> [music]
1:00>> Thank the good lord that Cody Schneider
1:03is back on the podcast. Uh Cody, by the
1:06end of the episode, what are people
1:08going to get out of this?
1:09>> Yeah, I'm going to show you and tell
1:12like share everything that you need to
1:13actually deploy real marketing agents,
1:16and I'm not talking about some like
1:17Zapier automation workflow that's
1:19linear. I'm talking about like code in
1:22the cloud that's making decisions off of
1:24your live business data. And then I'm
1:25also going to give you three businesses
1:28um that are concrete so you can
1:30basically feel like, okay, what these
1:32agents actually look like for a specific
1:34company type. So, you're going to walk
1:35away with three business ideas and also
1:38the agents that you should potentially
1:41deploy for them, how to actually deploy
1:43them for them, uh so you can grow the
1:44company autonomously.
1:46>> As you say, may the MRR gods be with us.
1:50>> Blessed be thy name.
1:52>> [laughter]
1:52>> Let's go.
1:53>> Awesome, man. Cool. So, I think just to
1:56start out, like what is a marketing
1:58agent, right? Like I see this thrown
2:00around so much and it drives me nuts and
2:02I've talked to so many people and that I
2:05you know, talk huge game on the internet
2:07and then you would like actually get
2:08down to the brass tacks with them on a
2:09conversation. You're like, you're so
2:12full of [ __ ] just to be blunt. So, to me
2:15like what is it what is a marketing
2:16agent? It has to be three things. It's
2:18like you have to solve the data problem
2:20to like give the agent unified data
2:23clarity of like the whole pipeline. Um
2:25and then you have to be able to have the
2:27agent
2:28like autonomously making decisions
2:31um you know, on some type of cadence
2:33like with like a thinking loop in it,
2:34right? Um so, in practice like the
2:36infrastructure that's necessary for
2:37that. You got to build a data pipeline,
2:39a data warehouse and then have like the
2:40ability to host these agents in the
2:42cloud. And when I say an agent, like it
2:43doesn't have to be this like AGI Hermes
2:47thing, right? It like where it can think
2:49on its own. I actually don't want that.
2:50Like I want something that is like doing
2:52a process for me and then looking at the
2:54data and then basically like improving
2:56based off of the data that's getting
2:57back from it. So, I just want to like
2:59define that really quickly um for the
3:02audience because I think that there's so
3:04much out there that's like this is a
3:06totally autonomous agent that's like
3:07running your your like entire marketing
3:10stack and it's like no, it's not.
3:11>> [laughter]
3:12>> Like this end-to-end workflow I
3:13guarantee you is not entirely running
3:16your marketing stack, but in contrast
3:18like I can and we like we'll talk about
3:20this today. Like you can make an agent
3:22entirely run Facebook ads now. And I'm
3:24talking about like it is researching the
3:26pain points. It is researching the
3:28outcomes of your target customer. It's
3:31making creative that's on brand both
3:33statics and also like AI avatar UGC.
3:37It's publishing that content into your
3:39Facebook ads account. It's turning off
3:41the losers. It's promoting the winners.
3:44And then it's going in a feedback loop
3:45where it's like, okay, what what is it
3:47actually working?
3:49And how do I go and basically make more
3:51content like the best performers? And
3:52I'll talk through all of that today and
3:53try to show you as much as I can. It's
3:55not not really easy to show cuz it's
3:57just like code.
3:58But anyway, we'll go through that. So,
4:00yeah, and then just to jump into that
4:02the first business that I want to talk
4:03about to use is kind of like a just a a
4:06sandbox for like how would you actually
4:08build and deploy these agents for this
4:10like this company type
4:12is AI for WordPress. I've talked about
4:14this before online, but like I have
4:17actually pitched like three startup
4:18founders this that are like we got into
4:20YC, we don't know what to pivot into.
4:23And I just they always just like are
4:25like no, I don't want to do this. So, I
4:26just want to hit on this idea. So, let
4:28me let me just screen share real quick
4:29and kind of show you the data for this
4:30cuz I think it's just fascinating and I
4:32think it's such a huge category that
4:34like nobody is talking about right now
4:37or that like nobody's really executing
4:38on.
4:39>> And before people's like eyes glaze over
4:41and they and they tune this out,
4:44WordPress is bigger than you think. So,
4:46>> powers like 50% plus of the internet,
4:49right?
4:49>> Like it's I just checked 43% of the
4:52internet
4:53>> that are like
4:54>> All websites.
4:55>> that are indexed by Google are WordPress
4:57websites.
4:57>> No one's building stuff for WordPress.
4:59>> It's insane, man. It's insane. And it's
5:01like it's such an easy space to compete
5:03in. It's kind of blue ocean for this.
5:04And all I'm going to pitch to you today
5:05is like literally it's just lovable or
5:07anything.com
5:09but for WordPress, right? So, you're
5:11just
5:12>> [laughter]
5:12>> all you're doing is you're making it so
5:15that people can like vibe code their
5:17website, but like on top of the
5:19WordPress platform. And then the
5:22business structure is super simple. It's
5:23like hey, you get all this stack. Like
5:25you get forms solved. You get like, you
5:28know, maybe a CRM. You like all of these
5:30plugins that people would go and like
5:32hodgepodge together.
5:34You just bundle that up into a single
5:36package.
5:37And then you sell tokens.
5:40So, it's like whatever, you know, base
5:41tier is $29 a month. That gets you x
5:43amount of tokens. And then that makes
5:44you so that you can go and basically
5:46like, you know, build out design the
5:48site with this like pre-built in like
5:51you know, again forms these these
5:52pre-built features.
5:54>> So,
5:55you got me thinking that I actually
5:58think that the there's a bigger
5:59opportunity here than just the Lovable
6:02for WordPress.
6:04Because and you don't know this about
6:06me, but back when I was 19 years old,
6:092000 and I don't know, I'm bad with
6:12math, he was like 11 or something.
6:14Um I was a partner in an agency and the
6:17agency's job
6:19was basically to get people
6:22off
6:23their existing CMS and move it to
6:25WordPress.
6:27And we were officially partnered with
6:29WordPress on doing that. So, if you
6:30called WordPress and you're like
6:32time.com Time magazine, you're on some
6:34custom CMS,
6:36they would they would be like work with
6:38Greg and team to move and do the
6:41migration to WordPress. WordPress is
6:42happy cuz they want more people, right?
6:45Um and and time.com is happy cuz they
6:50basically need someone to handhold
6:52uh
6:53through the process.
6:55So,
6:56it was a great business. We did like
6:58time.com, techcrunch.com, a bunch of big
7:00websites.
7:01Um the amount of money I saw that I in
7:04the in the
7:06WordPress automatic ecosystem and
7:08plugins was absolutely absurd.
7:10So, what you got me thinking about is
7:13you know,
7:14the the play is basically how do you
7:16find plugins that have validated demand
7:21that don't have an AI component and
7:23building the AI first version of each.
7:26Cuz
7:27you already know that, you know, there's
7:28proven demand, people are paying for it,
7:31and then you just make them 10x better
7:33with Agent. So, I'll give you uh a
7:35couple of examples.
7:37The Yoast SEO tool. Do you know about
7:39that one?
7:40>> No, yes. I've used it forever ago.
7:43>> Forever ago, right? I think it was I
7:45think it's a guy named Joost. I I I
7:47could be wrong, but I think it's a guy
7:49named Joost. There is no way that they
7:51have you know, that they have an AI
7:53first version of that. So, you know,
7:55instead of showing you red and green
7:57dots and telling you to fix things
7:58yourself, just have an agent do it. It
8:00writes the meta,
8:02it restructures the content, it adds the
8:04internal links, like make it AI first.
8:06And I'll give you a couple more.
8:08WPForms.
8:11That was a huge plugin back in the day.
8:13Make an AI form that talks back. So,
8:15instead of doing like a static form, you
8:17do like a conversational agent that
8:19qualifies a lead and answers questions.
8:21Um WooCommerce, one of the most popular
8:25you know, WordPress plugins.
8:27Um you know, do like an AI storekeeper.
8:30So, it automatically writes product
8:32descriptions, abandoned cart flows,
8:35you know, it it can go on and on. Um
8:38another another one is Akismet, which is
8:40the spam plugin. Like there's just so
8:42many. So, all you have to do
8:44>> Wordfence security, like there's it's
8:46just endless, right? There are so many
8:48of these.
8:49>> There's so many
8:51and people are paying for it. 43% of the
8:54internet is
8:55is uh is powered by it. The hard part is
9:01is obviously building the product, you
9:03know, making it good, making it work.
9:06Um and then finding, you know,
9:09a distribution channel
9:12that you can actually get customers.
9:15>> Totally. And I think that again, the the
9:17bundling of all this is where the
9:18opportunity is. It's like, okay, I'm
9:20going to give you all like all the
9:21plugins that you would normally be
9:22paying for.
9:23Instead of the aggregate of those 10
9:25that are necessary,
9:27now you pay a tenth of the cost, you
9:29know, or a fifth of the cost, and I'm
9:31going to give you everything, and you're
9:32going to also be able to like vibe code
9:34to the design of this, right? Like that
9:35is the business opportunity here.
9:38And then to your point, like how do you
9:39market this, right? And
9:41this is where
9:43this gets so fun.
9:44Like I'm having the most fun in my
9:46entire career right now with what we're
9:48doing. And so I can't wait to like share
9:50this with you guys, but anyway, so the
9:52like how when I think of look at this
9:53company, like
9:55where do you meet the people that are
9:57like trying to learn like how do I do
9:58this or how do I, you know,
10:00do this work with my WordPress site? Um
10:02so like Facebook ads and Google ads are
10:04going to be the first two places that
10:05you can go to. And I'm going to talk
10:06through a bunch of different things,
10:07different opportunities,
10:09um but this is just like we're going to
10:10just run down the list of all the things
10:12that you could do and here's how to
10:13deploy agents for them. So
10:15on the Facebook ad side, so for for
10:17targeting on Facebook anymore, um after
10:19Andromeda, you don't do interest-based
10:21targeting. Historically how I would
10:22probably do this is I would actually go
10:24and I would target people who have a
10:25WordPress interest on Facebook ads.
10:28Um now you don't need to do that. You
10:30just basically have to make ad creative
10:33that is talking to the pain points or
10:35the outcomes of the person that you're
10:36trying to sell to and then send them to
10:38a landing page that like describes what
10:40the product is. So what is Andromeda?
10:42What how does it work? And like what
10:44what
10:45what do you need to know about this? So
10:46Andromeda is uh Facebook's new like ad
10:50algorithm, right? It basically it
10:51decides what ads get shown to what
10:53people
10:54and how it functions is the uh
10:57the AI is basically looking at the ad
10:59creative
11:00and it's understanding, okay, like what
11:02is in the static ad? What is the what is
11:04the image? What is the text? What is the
11:06video? What's in the video? What's the
11:08script of the video? And then it also
11:09understands the landing page that it's
11:11sending the people to. And then from
11:14that, it basically uses that information
11:17to decide what ad gets shown to what
11:19person on the platform. So I, you know,
11:22jokingly on Twitter I've been saying
11:23this a lot and I mean I mean this though
11:25from like the bottom of my heart.
11:26Facebook has turned into the best B2B
11:29ads channel and that exists right now.
11:31And the reason for this is like we'll
11:33make these ads that are so obscure. Like
11:36it it's like maybe there's like 10
11:38people in the in the US, right? That
11:40like would would would be having this
11:42problem right now.
11:44Um and Facebook will find those people
11:46and show that ad to it and we'll get
11:47that lead like a qualified lead for a
11:49B2B business. It's ridiculous. So,
11:51anyway, to take a step back to the the
11:53WordPress side like this AI for
11:54WordPress thing. What you end up doing
11:57and like I can already see what the
11:58creative looks like. Like the script
12:00then does ends up being like something
12:01like this. Like, "Oh this is kind of
12:03crazy. I used to be paying my agency a
12:05thousand dollars a month to maintain my
12:07WordPress site and they were not
12:09responsive and they never like actually
12:12could do the designs that I wanted. But
12:14I just found this
12:16this this plugin for WordPress that
12:18makes it so that I can chat with AI to
12:21change my website and it concludes all
12:24these plugins that I used to have to pay
12:25extra money for. It's just a part of it
12:27for free. Click the link to learn more."
12:29If you run an ad like that,
12:32Facebook is going to automatically
12:33identify, okay, these people are
12:35WordPress website owners.
12:38They have this problem. It's going to
12:39show that to them and then you basically
12:41set up a conversion event for the sign
12:44up action, for the payment action that's
12:47deeper within the funnel. So, how do you
12:49actually do all of that? Like I'm just
12:50I'm talking through all this.
12:52Let's just like go into this deep and
12:55you know break down like what this
12:56actually looks like from an imaging
12:57perspective. So,
12:59>> [sighs and gasps]
12:59>> Okay, so the first thing is researching
13:02the pain points of your target customer.
13:05I'm just going to use Perplexity for
13:06this cuz I've just found that it's it's
13:08easy to do live like sharing this.
13:11Actually, let me stop this and I'll
13:13screen share a different way just so
13:14that you can just see the window.
13:16Um
13:17and
13:19for this we're just going to do like
13:20scrape Reddit.
13:23Uh pain points
13:25and outcomes uh,
13:28people that run WordPress websites, uh,
13:30have, um,
13:31for like why would they buy an AI, uh,
13:34WordPress, uh, tool that basically,
13:36like, solves all the plugin problems and
13:38makes it so that they can vibe code the
13:40design of the website.
13:41>> The reason you're doing Reddit, by the
13:43way, is because there's a real people
13:47just complaining, basically.
13:49>> Exactly. That's exactly why it is. It's
13:51basically dialogue and again, a lot like
13:54Reddit is being eroded right now
13:56by people manipulating it for like LLMs
13:58to rank on them. Um, but this is a way
14:01>> like you.
14:02>> By people like me that are basically
14:04creating the dead internet. Let's be
14:05real. It's already happening. I am the
14:07problem.
14:08Um, but with this though, what it
14:10enables is you basically can identify
14:12like what are the problems that they're
14:15facing and then from this, this crew,
14:18uh, basically seeds the creative that I
14:20want that I that I go and make, right?
14:23So, I found these pain points, right?
14:25Let's like pull out, you know,
14:27one of these that makes sense. Um,
14:31so content workflows,
14:33customization requires development
14:35skills, right? Like this is one that you
14:37could talk to. Um,
14:39what I would then go do, is like once I
14:41found this information, I'd be like rank
14:43stack, um, based off of, uh,
14:46the most referenced.
14:51Just to give us like what are the top
14:52three? And then once I've identified
14:54that, so the plugin layer, most
14:56referenced. This is the biggest pain
14:58point for people. Awesome. That is what
15:00ads are going to be about. Um,
15:02performance and speed. Cool. Can we make
15:04this super lightweight? Can we use
15:05something like Astro, um, to actually
15:08like serve it so the content's really
15:09fast? Security and maintenance. Okay,
15:12how do I make it so they never have to
15:13think about plugin security again, so
15:15that their site doesn't, you know,
15:16randomly get gambling ads on it and then
15:19get de-indexed by Google for their pest
15:21control business, right? So, with this,
15:24what I then go do is I'm going to use uh
15:27Kai AI
15:28um is what uh uh
15:31uh we're using right now for all of our
15:33image generation. Basically, just all of
15:34the image models uh and video models in
15:37the same place.
15:38And so, with this, I can then go and
15:40bulk generate static creative. And the
15:43way that we found to do this uh is you
15:45basically give it an example of an ad
15:48that is already being ran by either a
15:50competitor by somebody like, you know,
15:52somebody in a
15:53a tangentially related industry.
15:56And we're just using Google Nano Banana
15:57for this. Um so,
16:00the agent then, again, is generating
16:02this creative using the static. You can
16:05make this as advanced as you want. This
16:07is what we're doing a lot now so that it
16:08always hits brand style guides. And what
16:10do I mean by brand style guides? That's
16:12like, what are the fonts that need to be
16:13used? What are the colors that need to
16:15be used? All of these specifics about
16:17the company, about the brand, right?
16:19Um so, with this, uh you basically can
16:23put a vision model over the outputs
16:26and be like, "Hey, does this match brand
16:28style guides?" So, what I'm getting back
16:30from Google Nano Banana, I can be I can,
16:32you know, basically qualify the like, is
16:35the is the text all readable?
16:37Um you know, does are the fonts that are
16:39being used actually on brand, etc. Um
16:41and then from that, uh you basically
16:44have this way to make statics. And then
16:45on the video side, what we're still
16:47using and still seeing great results
16:49with, which is crazy to me cuz it's not
16:51even the best, is HeyGen.
16:53Um
16:54but we're starting to experience more
16:56and more with Seedance, and this is like
16:58where this is all going to go. It's just
16:59hard to do the stitching across uh like
17:02Seedance like for longer, you know,
17:0430-second videos. You can only do up to
17:06I I believe it's 9 seconds, don't quote
17:08me on that, but it's it's less than
17:09that. So, you basically have to like
17:10stitch these frames together is the
17:12challenge. But, this is where this will
17:14end up going. And then it's just like AI
17:16avatar UGC that's talking about the pain
17:19points or outcomes that you just
17:22extracted from what people are saying
17:24online. And
17:25everything that I'm describing here,
17:28this is just an agent that's done this
17:30work, right? It is an agent that's done
17:32this research. It's an agent that is
17:34calling doing the API calls to actually
17:36build out the creative. Um and then from
17:39there, how do we actually get it into
17:41Facebook ads, right? So, to get it into
17:44Facebook ads, then what we're going and
17:45doing is we need to use what's called
17:47the Facebook marketing API.
17:50And you probably, if you're online at
17:52all and not living under a rock, you've
17:53probably seen tons of people where it's
17:54like, I got my Facebook ads account
17:57banned, right? Because I was using an
17:59agent with it.
18:00The agent is not the reason it got
18:02banned, all right? The reason it got
18:04banned is because they violated TOS and
18:07they spammed the API because they were
18:09trying to pull in like, you know,
18:11hundreds of millions of rows of data
18:13from [laughter] their Facebook ads
18:14account. And that is not the way that
18:17you can't use the API in that capacity.
18:19Like it's literally like that that
18:21that's against Facebook TOS. That is the
18:23problem. So, how you actually use the
18:26Facebook ads API is you use the API only
18:31for rights, like only to publish content
18:34to the Facebook ads account or to turn
18:38off or to um
18:40uh
18:40uh basically like promote ads and I'll
18:42talk about that more in a second. And
18:44this I'm realizing and just based on
18:46time, this whole video is just going to
18:47be like, how do you build uh like an
18:50actual like marketing agent? Like we're
18:51going to only probably get through one
18:53of these.
18:54Um if you want more content like this,
18:55this is all I talk about on the
18:57internet, so just like go follow me
18:58across channels and this is what I
19:00share, but um anyway, with this so, the
19:04to solve the data analytics problem,
19:07you have to build a data pipeline and a
19:09warehouse. So, the open-source way to do
19:11this is using a tool called Airbyte.
19:13You can host this like you know
19:15anywhere. You can literally have Claude
19:17code go and do this for you, right? Set
19:19this up. And then you can use a data
19:21warehouse called ClickHouse. These are
19:22both open source solutions that you can
19:24self-host. And I'm just giving you those
19:25as the as like you know the the the way
19:29that that is totally accessible to
19:31anybody.
19:32>> And And Cody, can you explain like
19:34what is a data warehouse? Like what are
19:36these things?
19:37>> absolutely. So a data warehouse is a way
19:40to unify all of your data sources
19:43into a single location so that your
19:46agent has the ability to understand all
19:49of the data in context with all like all
19:52channels in context or all data sources
19:54in context with all the other data
19:56sources. So I'll give you specifics.
19:57That was a very like you know broad way
19:59to explain it. So say for example, I'm
20:01running Facebook ads. How do I actually
20:03run effective Facebook ads? I need to do
20:04probably four things. I need to have my
20:06Facebook ads data. I need to have my
20:08Google Analytics data. I need to have
20:10probably my Posthog data, maybe my my
20:13CRM data, so like my HubSpot data. And
20:15then I also need to probably have like
20:17my Stripe data to understand the entire
20:19stack of like I ran ads and did
20:21customers come out the other side,
20:23right? And then how do I actually tie
20:26the literal specific ad that is driving
20:28revenue for the company, right? This is
20:30how you do that. It's by having this
20:33data warehouse that has all of the data
20:35in a single place. Now, to get all of
20:37that data into a single place, you have
20:39to build a data pipeline. And so Airbyte
20:42is a data pipeline solution. It's
20:44self-hosted
20:45and it enables you basically they have
20:47all these pre-built connectors
20:50that enable you to pipe all of your data
20:53into a data warehouse. So when you think
20:54about like you know visually how this is
20:56happening, we can just like TL draw this
20:58real quick to actually show you. It
21:00basically what I am doing is I am giving
21:05my agents. So say I have, you know,
21:08these like five
21:10different data sources
21:12um that I'm pulling from. We'll just do
21:14three for the sake of time. And I'm
21:17putting them all into a data warehouse.
21:20And that data warehouse, I'm then giving
21:23my agents access to it. So, imagine I
21:27have this agent here that can read this
21:31information. It's back and forth, right?
21:34And this is the Facebook Ads agent.
21:39And this Facebook Ads agent,
21:42um how it works is it basically is
21:44looking at that data stream.
21:47And then it is making updates within
21:52Facebook Ads,
21:53right?
21:55And then that Facebook Ads data is going
21:58back to
22:01like
22:02where we're pulling from, which goes
22:03back and basically back into the data
22:05warehouse. Actually, this is probably
22:06more like this. Like back into the data
22:08warehouse, right? And so, this is how
22:10I'm creating this like loop that is
22:12basically learning, right? Um I'm giving
22:15the agent access to the live business
22:17data, access to the live stream. Also,
22:19what this enables for you as a business
22:21owner is you can also interact with this
22:25within your Cloud Code, within your
22:27Codex, whatever that is, right? And
22:31whatever your daily driver is. And the
22:33reason that you want this is you can
22:35have conversational analytics and also
22:37build like custom dashboards with this
22:40data, right? So, you can literally ask
22:42like, "Okay, like we're like we're
22:44having trouble like say you're again
22:46pest control company
22:48uh or or uh we'll say um
22:51I don't know, something like your
22:52commercial HVAC. Like we're having
22:53trouble hitting payroll this month. Like
22:55what is going wrong?"
22:56>> [laughter]
22:56>> It's like, "Oh, your accounts
22:57receivables are like off." You can
22:59literally have those conversations with
23:01your marketing data, with your business
23:02data. This is the way that you can
23:04enable both yourself and your team to be
23:06able to do this, right?
23:08So, how do you actually
23:10how do you actually set this up? So,
23:12we've got the data pipeline in the
23:13warehouse solution. Now, for hosting the
23:15agents,
23:16everybody's doing this already like
23:18locally with like their cloud code and
23:20they like I I mean I've seen these crazy
23:21setups where it's like, yeah, I'm
23:23deploying these solutions onto you know,
23:26my Mac mini that I bought for open claw,
23:28but now it's actually just like working
23:30as a server. Like you don't have to use
23:32a Mac mini locally as a server to do
23:34this, all right? You can just deploy
23:36this into the any cloud and be into
23:38Heroku, it can be into Railway, it can
23:40be into like any of these solutions. And
23:42again, I'm just giving options that are
23:44available. But basically, an agent is
23:46just code under the hood, right? It's
23:48just a decision tree that's like, cool,
23:50here's a live data stream. I have some
23:52like decision loop that's happening
23:54where an LLM is thinking and then I'm
23:56basically having an optimized for an
23:58outcome, which for most companies is
24:00like, how do I make the business more
24:01revenue, right?
24:03So, anyway, that is kind of the the
24:07the infrastructure that's necessary to
24:09actually make these agents function.
24:11Once you've got this whole thing set up,
24:14what is possible, just to give you like
24:16a scope of this, and we're doing this
24:17for companies right now, is you can
24:20actually like create So, for example,
24:22like for one company we're doing two
24:25different ad sets per day, five ads per
24:28ad set.
24:30They automatically get uploaded into
24:31Facebook ads based off of all the
24:33research that we've done
24:36and also the data that exists within the
24:38data warehouse like of what has worked,
24:41right?
24:42And we
24:44are base So, there's like a learning
24:45loop that's happening on top of this.
24:47The
24:49agent is looking at, okay, I published
24:51these 10 ads and lets it run for like a
24:532 to 3-day window for like it's just get
24:55initial signal. It turns off the worst
24:58performing ads.
24:59And again, how is it getting that
25:01information? It's pulling that
25:02information from the data warehouse,
25:03right?
25:05It's pulling Okay, what is the worst
25:07performing ads?
25:09Let's turn those off. The winners live
25:11on.
25:12And then the winning ad sets, they go
25:15into a winners pool, and they're
25:16competing against each other for the ad
25:18budget that's available.
25:21And then over the top of that, you can
25:22have a vision model or honestly, like
25:24how we're doing this is we build a
25:26database
25:27of all the ads that we have created.
25:30And it's literally like the JSON prompts
25:32that we're sending to Nano Banana or the
25:35ad scripts that we're using and we're
25:37sending to H and or Seed Dance.
25:39And that is what we are letting the
25:42agent analyze of like, okay, what is
25:44actually working so that it can it can
25:46go and, you know, basically become more
25:48effective at making creative that
25:50creates outcomes.
25:51So, the problem that happens with this
25:53here and again, I'm going to get into
25:54the weeds talk through a lot of this,
25:56you know,
25:57in in detail just to
25:58share the the issues that come up. Is
26:00you have to solve for entropy. So, like
26:01what is entropy? It's like basically the
26:03agent gets stuck thinking in the same
26:04way. So, how do you solve for that? The
26:07two ways that we found that we can solve
26:08for it is you can use the Facebook Ads
26:10Library
26:12to pull
26:13ads from competitors' libraries.
26:16And that basically like puts new DNA
26:19into the system. So, you're basically
26:20feeding that source text that you did
26:22that research on. And then the other
26:24thing that we found to work is taking
26:26YouTube transcripts and also podcast
26:29transcripts and basically pulling out
26:33the insights from those. There's like
26:35hundreds of YouTube channels that are
26:36just designated to like, you know, for
26:38example, WordPress. You can pull out the
26:41insights from those and then run ads
26:43basically off of off of those insights
26:45that you aggregate. So, this is a way to
26:46solve that that that that, you know,
26:48kind of like entry problem.
26:50But what this ends up looking like in
26:52the in the you know, the final form is
26:54like it's like a virtual employee, Greg,
26:56is the best way to describe it. This
26:58virtual employee, all it's doing is
27:00focusing on Facebook ads as the channel
27:02and how it can actually like operate and
27:04function is because it has all the
27:06necessary data it needs to think through
27:09what is actually driving like outcomes
27:11that we are like that are that are drive
27:14business revenue, right? And this is
27:16where all this is going. Like I it's so
27:17funny. I talked to like so many
27:19marketers on a weekly basis and they're
27:20all like, "Is my job secure?" I'm like,
27:22"I don't know. Like, can you do this?"
27:24Like, this is what this turns into is
27:25like, your job now becomes like I'm an
27:29agent jockey. You're taking your domain
27:31knowledge that you have, and you're
27:33basically like building this into code.
27:35And like, I'm not technical, right?
27:37Like, I can set this up now. Like, the
27:39models are smart enough now where like a
27:41semi-technical person or just if you
27:43have access to Claude code, and
27:45literally just give it the transcript
27:46that I just like, you know, we just
27:47created on this call, and be like,
27:49"Okay, I want to set this up. Walk me
27:50through this process." Like, you can go
27:52and build out this solution. Like, that
27:54is the the possibility here. So, anyway,
27:56but yeah, happy to answer questions
27:58about that. That was one agent. I I can
27:59go down into a deeper one as well. So.
28:01>> Okay. Yeah, and I think um
28:04taking a step back, it's like, well, why
28:06are I I know it's crazy to say this, but
28:08why are Facebook ads important? Well,
28:10okay, if you're going to steal an idea
28:12from this podcast or come in come up
28:13with an idea on your own, you're going
28:14to vibe code something
28:16um that you think has value in the
28:18world. The biggest question I get is,
28:20how do I actually get customers to this
28:23thing?
28:24Um and the truth is,
28:28Facebook is the most cost-efficient way
28:30to do it. Um if you don't want to go and
28:32create content, which takes time
28:35to actually,
28:37you know, build an audience. Like, I I
28:40still I still think that people should
28:41create content, but at the same time,
28:44be thinking about, how can I create an
28:46you know, an ad set that is going to
28:48perform? Now, what people the mistake
28:51that people make is they try a few ads,
28:54it doesn't work, so they give up.
28:57And they don't realize that
28:59uh
29:00you actually should be taking the same
29:02ad and just like changing the
29:06positioning a little bit.
29:08Um because
29:11and and and changing it a little bit
29:13like 10 times or 15 times or 20 times,
29:15right? And what happens is like I'm
29:19shocked sometimes with some of the
29:20Facebook ads that I do and I'm like,
29:22wow, I didn't think that this ad set was
29:24actually going to work. This positioning
29:25was going to work. Like for example,
29:28you know, let's say we wanted to do that
29:30WordPress uh plugin idea. And let's say
29:33we wanted to compete with Yoast, the SEO
29:35plugin. Like maybe the positioning is
29:38Yoast is bad. And I, you know, it's just
29:40like an anti-Yoast.
29:42You know, and and you never would have
29:44thought that.
29:44>> do work rather than doing work for you,
29:46right? Like there's a great there's so
29:48many angles that you can pull out of
29:50that. To your point, too, I just want to
29:51piggyback off of this. Like I think a
29:53lot of people come into marketing and
29:55advertising and they're like, I know
29:57what the market wants. I'm going to
29:58impose this idea on the market. That is
30:01not how this works, right? We're not Don
30:03Draper.
30:04>> [laughter]
30:04>> We are like putting
30:06creative into the market. Like why is
30:08why is paid ads the most Like why is
30:11Facebook ads, why is Google Why are they
30:13the most powerful assets? Like why the
30:14most powerful marketing channels that
30:16have ever existed? I can go and test a
30:18thousand different ad creatives and like
30:21I'm going to have clear delineation on
30:23like what is winning and what is being
30:25the most What is the market most
30:27receptive to, right? And you're going to
30:29know that within 48 hours. And I see
30:31this all the time where like oh they're
30:32like oh I don't want to go run ads. I
30:34don't want to like I only want to do
30:35organic, right? Or like how do you grow?
30:37It's only by word of mouth. I'm like
30:39this is such a that's such a rookie way
30:41to look at the market. Like you should
30:43be doing all of it, number one. Number
30:45two, like you need to think about how
30:47does this like this is the only system
30:50where I put a dollar in and like $5 can
30:53potentially come out the other side and
30:56I can just keep feeding that ATM where
30:58it's like I just keep putting money in
31:00and that same, you know, volume will
31:02come like that same multiplier will come
31:04out the other end. So anyway, just a
31:06>> And and so what's changed, like why I
31:09now am into Facebook ads honestly is
31:12because of the systems that you're
31:13describing. So like in the past, if you
31:16were a founder and you wanted to do what
31:19you're talking about, you had to hire an
31:21agency that would cost you tens of
31:22thousands of dollars a month to go and
31:24run this at scale. Now, you just need to
31:27set up the system that you talked about
31:31and of course it needs to be fine-tuned,
31:33of course there needs to be it needs to
31:35work, right? Of course you need to solve
31:37for entropy which I'm happy you spoke
31:39about cuz no one speaks about that.
31:41People are like, "Oh yeah, you set it up
31:42and then it's just good it's good."
31:43Like, "No, dude, that's lit
31:45The the people that say that have never
31:47set up
31:47>> [laughter]
31:47>> So they're the opposite.
31:49>> Have you have The people that say that
31:50have never set up
31:52anything, right? Cuz
31:55uh
31:56as soon as you it might feel good like
31:58on day one but on day two it gets a
32:00little worse and day three and four and
32:01five until you get this point where
32:02you're like
32:03>> choice, man. I think that's the thing
32:04that people like have the hardest like
32:06when you think about marketing, like
32:08people think about it in campaigns. Like
32:10I start a thing and then I stop a thing.
32:12That is not how this works anymore.
32:14Like how marketing actually functions
32:15now is like you're continuously doing
32:18this and because the feedback loops are
32:20so fast like I mean when you look a
32:22great example of this is if you just
32:24look at like the fashion industry, like
32:25how quick the the the trends occur now.
32:28It's like 10x faster than what it was 20
32:31years ago, right?
32:32>> [laughter]
32:32>> Than what it was in like the early
32:332000s. And so when you think about that
32:36and apply that to everything, like how
32:39quick the market moves,
32:41you have to basically have like,
32:44you know, some type of observability of
32:46what's happening.
32:47So, that you are adjusting your, you
32:50know, ads, your positioning, your all of
32:52the strategy that you have across your
32:54your your go-to-market based off of
32:56those trends that are that are are
32:57basically like evolving. And where does
32:59all this happening, right? It's all on
33:00short-form social. Like that is where
33:02all this kind of originates from and
33:03then trickles down and everything else.
33:05So, you can actually do this, right?
33:06There's another way to solve this entry
33:08problem. You can use tools like Viral
33:10Loop. They have an API. Basically, it's
33:12like it scrapes, uh,
33:14like all of TikTok and I think they're
33:16doing Instagram Reels now. Don't quote
33:17me on that. Um, but basically, I can be
33:19like, "Hey, show me all of the most
33:21viral like posts that are in the beauty
33:23category from the last week." Cool. What
33:25are there
33:26Are there trends that are happening
33:27here? Are there like certain concept
33:30formats that are happening here? And can
33:31I use that information to basically
33:34influence the creative output that I'm
33:36doing? Yeah, 100% you can do that. This
33:37is again like how you can how it's this
33:40push and pull with the market that
33:41you're doing constantly, right? And like
33:43agents for the first time have can
33:44enable you to do that. Like what I just
33:46If I wanted to go and create 100 ads,
33:48right, and get that into Facebook Ads,
33:51like that is probably like 2 weeks of
33:53work, right, historically. And now you,
33:56as a founder, like literally in the next
33:58hour and a half, you can have this whole
34:01system that I just described set up and
34:04be able to actually go and publish this
34:06content like at scale with the
34:08understanding of what's actually
34:10working. And maybe there's a human in
34:11the loop to begin with, but like what
34:13this gets to is like you're going to
34:14understand like this is what's actually
34:16functioning. Cool, I now want an agent
34:18to go and run this. I deploy an agent
34:19into the cloud with the necessary data
34:21stream that it needs so that it can
34:23actually make these decisions. So,
34:26Cody, what else did you want to cover?
34:28Dude, I have so much. I have to jump in
34:292 minutes though, so
34:31>> [laughter]
34:31>> I'm like,
34:32it's um, kind of bummed. Um
34:34>> I think so, folks comment if you want
34:37more marketing and
34:38>> back and just do a whole series on this.
34:40Like, here is literal like agent by
34:42agent like descriptions, like showing
34:44the actual like structures, talking
34:46through the actual campaign structures.
34:48Um if that is what people want, like
34:51this is again all I'm thinking about
34:53right now. This is all I do from when I
34:55wake up to when I go to bed at the
34:57moment cuz
34:58I'm obsessed with it and it's
34:59uncomfortable how like it's a compulsion
35:01and this is not really a choice. So,
35:03anyways, yeah. Greg, I didn't mean to
35:05cut you off.
35:06>> No, no, no.
35:08Uh let us know. I think like
35:10it, you know, people like I said, people
35:13want to figure out ways how they can use
35:15AI agents to get them customers. Um so,
35:19I think marrying like here's a couple
35:20ideas with here's some here's some
35:22agents is going to be helpful for folks.
35:24>> Let me rattle off some and then you
35:26comment what you want and we'll come
35:29back and we'll do this. Right. So, I can
35:30we can do Google Ads agents that are
35:32entirely running your Google Ads
35:33accounts. We can do influencer outreach
35:36agents that are researching influencers
35:38within your category, cold emailing
35:40them, and then negotiating the pricing
35:43with them, and hand raising for when,
35:44you know, they think it's a good person.
35:46We can do cold email of like how do I go
35:48and find all of the customers, you know,
35:51that I'm trying to sell to, find their
35:52email addresses, how like cold email
35:55them, and then have an agent managing
35:57that inbox, like responding to them to
35:59book meetings.
36:00We can do TikTok real farms where we're
36:02basically like making 10 TikTok accounts
36:05in the cloud and then at scale making
36:07slideshows that are related to your
36:09product, scheduling those and posting
36:11and aggregating all of the views across
36:13that. We can do um
36:16what are some of the other things? We
36:17can do SEO agents that are doing keyword
36:20research, researching the article,
36:22writing the article, having the article
36:24be on brand for your voice, like how do
36:26you actually give it a unique
36:28perspective or point of view so that it
36:30has that like, you know, basically
36:32personality to it so that it actually
36:34ranks well and it's like good to read.
36:36We can talk about AI search and how do
36:38you build citations? How do you actually
36:41do cold outreach to get your brand
36:43listed on the citations that are already
36:45being shown in the category that you're
36:46in?
36:47>> [snorts]
36:47>> We can talk about social media
36:48management and like how do you get an
36:50agent to run your LinkedIn account, to
36:52run Twitter accounts? We can talk about
36:54how do you get agents to build email
36:56newsletters
36:57that are related to podcasts and
36:59actually research the podcast, write the
37:01podcast script, use 11 Labs voices to
37:05read the podcast script and then build
37:07lead magnets to build an email list on
37:10the top of that podcast for an industry
37:12like e-commerce. So, you comment what
37:14you want to learn, I'll come back, I'll
37:15share all the things that we're seeing
37:17and all the things that we're deploying.
37:19I'd love to do that. I think it'd be
37:20awesome. So.
37:21>> God bless you, Cody Schneider. You know,
37:23I appreciate you.
37:24>> It was so weird.
37:27>> We'll see you next time and uh
37:30yeah, I'm curious to see or hear what
37:31people think.
37:32>> Yeah, if people have questions, let me
37:34know too. I'm
37:36active on Twitter and LinkedIn. Those
37:38are the biggest places. And
37:40companiesgraph.com, all this work that I
37:42just talked about, that's where we're
37:43doing this. So, thanks for having me, G.
37:45>> Later.

About this transcript

This page contains the full transcript of Marketing Agents Are Too Good Now by Greg Isenberg, generated from the public captions YouTube serves with the video. The transcript has 7,447 words across 1,127 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.