39:43Greg Isenberg · Interview
Making $$$ with Loop Engineering
Elie Steinbock walks Greg Isenberg through the build-verify-learn loop he's using to run SEO, ads, and product feedback on autopilot.
July 13thThe argument in one line.
A real marketing agent is defined by three things — unified pipeline data, decisions made on a cadence, and a feedback loop that learns from results — and building one today comes down to a data pipeline, a warehouse, and cloud-hosted code that reads that warehouse and writes decisions back into ad platform APIs.
Read if. Skip if.
READ IF YOU ARE…
- A solo founder or small team with a built (or vibe-coded) product that is stuck on the 'how do I get customers' problem.
- Someone semi-technical who can run a coding agent like Claude Code and wants real marketing infrastructure instead of hiring an agency.
- A marketer or agency operator trying to figure out how much of ad management AI can realistically take over.
- Anyone evaluating WordPress-adjacent SaaS or plugin niches as a startup idea.
SKIP IF…
- You want a plug-and-play no-code tool — this describes custom infrastructure (pipelines, warehouses, API integrations), not an off-the-shelf product.
- You have no budget for paid ad spend — the whole system assumes you're running and iterating on live Facebook ad campaigns.
The full version, fast.
A marketing agent only earns the name if it unifies your pipeline data, makes decisions on a running cadence, and learns from the results it reads back — a linear Zapier or n8n workflow doesn't qualify. Cody Schneider walks through building one for a hypothetical AI-for-WordPress product: mine Reddit for real pain points, generate static and video ad creative seeded from competitor ads, and wire Airbyte (pipeline) and ClickHouse (warehouse) so an agent can read the full path from ad to revenue and write publish/pause/promote actions back through the Facebook Marketing API. The system runs an elimination bracket on its own ad sets every few days, and stays fresh by pulling new material from competitor ad libraries, YouTube, and podcast transcripts. The actionable conclusion: paid ads let you test a thousand angles and get a market verdict within 48 hours, and marketing agents now make that loop cheap enough to run continuously instead of hiring an agency.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Who's talking.
00:00guestCody Schneider
Where the time goes.
00:00 – 01:54
01 · Intro
Cold open framing the episode: getting customers, not building the product, is the hard part now.
01:54 – 04:00
02 · Defining a Marketing Agent
Cody sets a three-part bar for what counts as an agent versus hype.
04:00 – 07:27
03 · Startup Idea: AI for WordPress
The pitch: a vibe-coding, token-metered layer bundling WordPress plugins into one AI-run product.
07:27 – 09:55
04 · AI-First Plugin Ideas: Yoast, WPForms, WooCommerce, Akismet
Rebuilding proven, already-monetized WordPress plugins as agent-run versions.
09:55 – 12:48
05 · The current state of Meta Ads
How Facebook's Andromeda algorithm shifted targeting from audiences to creative and landing pages.
12:48 – 15:23
06 · Bundling the Stack and Choosing Channels
Why Facebook and Google Ads are the first channels and how conversion events should be set up.
15:23 – 17:25
07 · Two creative pipelines: static and video
Parallel static-image and AI-avatar-video pipelines sharing the same research inputs.
17:25 – 24:11
08 · The Data pipeline and warehouse
Airbyte plus ClickHouse as the minimum infrastructure so an agent can read data and write ad actions.
24:11 – 25:51
09 · Ad strategy
A live example: daily ad-set publishing, a hold period, then automated culling into a winners' pool.
25:51 – 28:01
10 · Solving for Entropy
Why agent creative goes stale and the fresh-input sources that fix it.
28:01 – 34:26
11 · Let the market pick the winner
The case for wide creative testing and why marketing has to run continuously now.
34:26 – 37:47
12 · Closing Thoughts
Cody has to jump off; both ask for comments on what to cover in a follow-up series.
Lines worth screenshotting.
- WordPress powers roughly 43% of all Google-indexed websites, yet almost nobody is building AI-first products for that installed base.
- A system only counts as a marketing agent if it unifies pipeline data, makes decisions on a set cadence, and updates itself from the results it reads back — a scripted workflow like Zapier or n8n doesn't qualify.
- Facebook's Andromeda algorithm reads the ad creative and the landing page itself to decide who sees an ad, so the ad copy now carries the targeting instead of interest-based audience selection.
- Facebook Ads accounts get banned for violating terms of service by pulling hundreds of millions of rows through the API, not because an agent touched the account — safe usage limits the Marketing API to writes: publish, pause, promote.
- Reddit threads get scraped for pain-point research because they read as unfiltered complaints, even as that same scraping is part of what's eroding Reddit as an authentic human data source.
- A minimal agent infrastructure stack is three pieces: a pipeline (Airbyte) to move data, a warehouse (ClickHouse) to unify it, and cloud hosting to run the agent code against it.
- Static ad creative gets seeded from one competitor ad or a tangentially related industry's ad, then a vision model checks every output against brand fonts, colors, and text readability before it ships.
- AI avatar video tools currently cap individual clips around nine seconds, so any longer ad has to be assembled by stitching multiple generated clips together.
- One production example runs two ad sets a day, five ads per set, lets them run untouched for two to three days, then an agent cuts the worst performers and promotes survivors into a winners' pool that competes for budget.
- Ad creative goes stale when the agent gets stuck generating variations of the same idea — the fix is feeding it fresh source material: competitor ad libraries, YouTube transcripts, and podcast transcripts.
- Content and marketing trends now cycle roughly ten times faster than in the early 2000s, which is why a set-it-and-forget-it campaign no longer works.
- Running paid ads lets a founder test a thousand creative angles and get a clear market read within 48 hours instead of guessing at positioning up front.
- The typical failure mode with paid ads is quitting after two or three variations — the same core ad idea often needs ten to twenty different positioning angles before one works.
- A repeatable startup play is finding plugins with already-validated demand — Yoast SEO, WPForms, WooCommerce, Akismet — and rebuilding each as an AI-first agent instead of a manual tool.
Marketing agents replace ad management, not marketing judgment.
AGENT INFRASTRUCTURE
A real marketing agent is a data pipeline, a warehouse, and code that reads results and writes decisions back into ad platforms — not a linear workflow tool, and not a replacement for having something worth marketing.
02Defining a Marketing Agent
- A marketing agent needs three things to earn the name: unified data across the whole pipeline, decisions made on a set cadence, and a feedback loop that learns from its own results.
- A linear workflow tool like Zapier or n8n does not qualify as an agent, even if marketed that way — it lacks the thinking loop and the read-back of results.
- The useful version isn't a fully autonomous AGI-style system — it's a monitored process running on a cadence that you steer using the data it produces.
03Startup Idea: AI for WordPress
- WordPress runs roughly 43% of all Google-indexed websites, and almost no one is building AI-first products for that installed base — a wide-open niche.
- The pitch is a vibe-coding layer for WordPress: bundle the plugins people already stitch together (forms, CRM, security) into one package sold as a token-metered subscription starting around $29/month.
- Bundling ten separately-paid plugins into one AI-run package at a lower total price is the wedge — customers save money and manage one product instead of ten.
04AI-First Plugin Ideas: Yoast, WPForms, WooCommerce, Akismet
- The repeatable play is finding plugins with already-validated demand and shipping the AI-first version: Yoast SEO's manual fix-it-yourself flow becomes an agent that writes the meta and internal links itself.
- A static form becomes a conversational agent that qualifies the lead and answers questions instead of just collecting fields.
- A WooCommerce store becomes an AI storekeeper that writes product descriptions and manages abandoned-cart flows automatically.
- The hard part was never the idea — it's building the product well and then finding a distribution channel to reach people already paying for the old version.
05The current state of Meta Ads
- Facebook's Andromeda algorithm reads the ad creative itself — image, text, video script — plus the landing page, and uses that to decide who sees the ad, so interest-based targeting is no longer necessary.
- The ad copy now carries the targeting: write creative that speaks to a narrow pain point, and Andromeda finds the small set of people who actually have that problem.
- This shift has made Facebook an unusually strong B2B ads channel — narrow, specific pain-point ads can find the handful of relevant buyers in a large population.
06Bundling the Stack and Choosing Channels
- For an AI-for-WordPress product, Facebook and Google Ads are the first channels to test because that's where people actively searching for a fix already are.
- Set up a conversion event for the actual signup or payment action deep in the funnel, not just clicks, so the ad platform optimizes toward real revenue.
07Two creative pipelines: static and video
- Static ad images are generated by seeding an image model with one competitor ad or a tangentially related industry's ad, then checking every output against brand fonts, colors, and text readability with a vision model.
- AI avatar video ads currently run on tools that cap individual clips around nine seconds — anything longer requires stitching multiple clips together, which is the current bottleneck.
- Both the static and video pipelines pull from the same pain-point research — only the rendering engine changes between formats.
08The Data pipeline and warehouse
- The minimum viable agent infrastructure is three pieces: a data pipeline (Airbyte) to move data, a warehouse (ClickHouse) to unify it, and cloud hosting to run the agent's code against that warehouse.
- A data warehouse's job is to put every data source — ad platform, analytics, CRM, payments — in one place so an agent can reason about the full path from ad spend to actual revenue.
- Facebook Ads bans blamed on agents are almost always a terms-of-service violation from pulling excessive data through the API, not a ban on agent usage — the safe pattern is read from the warehouse, write to the ad platform only for publish, pause, and promote actions.
- Once the pipeline and warehouse exist, a business owner can ask plain-language questions about their own data through a coding agent instead of building custom dashboards by hand.
09Ad strategy
- One production example: two ad sets published per day, five ads per set, left untouched for two to three days to gather signal, then the agent cuts the worst performers and promotes survivors into a winners' pool.
- The system isn't picking individual winning ads — it's running a recurring elimination process every few days, continuously replacing the weakest performers.
10Solving for Entropy
- The recurring failure mode for agent-generated creative is entropy: the agent gets stuck generating variations of the same idea and quality plateaus.
- Two fixes for stale creative: pull ad examples straight from competitors' public Facebook ad libraries, and mine YouTube and podcast transcripts from channels in the same niche for fresh angles.
- Trend cycles now turn over roughly ten times faster than in the early 2000s, largely driven by short-form video, so a creative system has to keep ingesting new source material.
11Let the market pick the winner
- Running paid ads lets you test a thousand creative angles and get a clear signal on what the market wants within about 48 hours, rather than guessing at positioning before launch.
- The most common founder mistake is testing only two or three ad variations and quitting — the same core offer often needs ten to twenty different angles before one lands.
- Marketing no longer works as a start-and-stop campaign; because feedback loops are fast, it has to run continuously with ongoing adjustment, closer to a habit than a project with an end date.
Terms worth knowing.
- Marketing agent
- Software that owns unified data across an entire marketing pipeline, makes decisions on a running cadence, and updates its own output based on the results it reads back — as opposed to a fixed, linear automation workflow.
- Andromeda
- Facebook's ad-ranking algorithm, which reads an ad's creative and landing page directly to decide who sees it, replacing older interest-based audience targeting.
- Data warehouse
- A single location that unifies every data source — ad platform, analytics, CRM, payments — so software can reason about a metric like revenue in the context of everything that led to it.
- Airbyte
- An open-source, self-hostable data pipeline tool with prebuilt connectors for moving data from apps like Facebook Ads or Stripe into a warehouse.
- ClickHouse
- An open-source data warehouse used to store and query the unified data a marketing agent reads from.
- Creative entropy
- The failure mode where an AI creative system gets stuck generating variations of the same idea instead of producing genuinely new angles.
- Winners pool
- A holding group for ad sets that survived an initial performance cutoff, where they continue competing against each other for ad budget.
- Facebook Marketing API
- Facebook's programmatic interface for managing ad accounts; safe agent usage is limited to write actions like publishing, pausing, and promoting ads, not bulk-reading account data.
Things they pointed at.
07:27toolYoast SEO
07:50toolWPForms
08:05toolWooCommerce
08:20toolAkismet
09:20toolPerplexity
16:10toolKai AI
16:15toolGoogle Nano Banana
16:40toolHeyGen
17:05toolSeedance
17:40toolClickHouse
31:40toolVirlo
Lines you could clip.
02:20
“An agent is just code under the hood. It's just a decision tree.”
10:30
“Facebook has turned into the best B2B ads channel that exists right now.”
16:55
“The agent is not the reason it got banned — that's a TOS violation, not an agent problem.”
22:40
“Your job now becomes, like, I'm an agent jockey.”
29:40
“I can test a thousand different ad creatives and know what's winning within forty eight hours.”
Where the conversation goes.
01:54 – 04:00denseDefining what a marketing agent actually is
04:00 – 09:55denseAI-for-WordPress startup pitch and plugin rebuilds
09:55 – 12:48denseHow Facebook's Andromeda algorithm changed targeting
12:48 – 17:25steadyCreative generation pipelines (static + video)
17:25 – 24:11denseData pipeline and warehouse infrastructure
24:11 – 25:51steadyRunning the ad elimination loop
25:51 – 28:01steadySolving creative entropy
28:01 – 34:26steadyWhy continuous testing beats campaigns
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogystory
Let's be real. The hardest part about building a startup today is getting customers to the startup. You can vibe code stuff now super easily, we all know that, but getting customers to drive revenue is hard.
So how can you actually set up marketing agents that help you do this? I feel like a bunch of people have talked about this, but no one's really gone in-depth and clearly explained what a marketing agent is and how you can use it to go and get you customers.
In today's episode, it's two parts. We give you a startup idea that I really love and then we also explain what marketing agents are, how you can set them up, what are the tools you need to understand if you actually wanna be a marketing agent manager, and Cody Schneider just does an incredible job on this episode of explaining everything end to end.
So enjoy the episode and I'll see you at the end.
Thank the good lord that Cody Schneider is back on the podcast.
Cody, by the end of the episode, what are people gonna get out of this? Yeah. I'm gonna show you and, like, share everything that you need to actually deploy real marketing agents.
And I'm not talking about some, like, Zapier automation workflow that's linear. I'm talking about, like, code in the cloud that's making decisions off of your live business data. I'm And also gonna give you three businesses, um, that are concrete so you can basically feel like, okay.
What do these agents actually look like for a specific company type? So you're gonna walk away with three business ideas and also the agents that you should potentially deploy for them, how to actually deploy them for them, uh, so you can grow the company autonomously.
As you say, may the MRR gods be with us.
Blessed be thy name. Let's go.
Awesome, man. Cool. So I think just to start out, like, what is a marketing agent?
Right? Like, I see this thrown around so much, and it drives me nuts. And I've talked to so many people and that I, you know, talk huge game on the Internet, and then you would, like, actually get down to the brass tacks with them on a conversation.
You're like, you're so full of shit just to be blunt. So to me, like, what is it what is a marketing agent? It has to be three things.
It's like, you have to solve the data problem to, like, give the agent unified data clarity of, like, the whole pipeline. And then you have to be able to have the agent, autonomously making decisions, you know, on some type of cadence, like, with, like, a thinking loop in it.
Right? So in practice, like, the infrastructure that's necessary for that, you gotta build a data pipeline, a data warehouse, and then have, like, the ability to host these agents in the cloud. And when I say an agent, like, it doesn't have to be this, like, AGI Hermes thing.
Right? It like, or it can think on its own. I actually don't want that.
Like, I want something that is, like, doing a process for me, and I'm looking at the data and then basically, like, improving based off the data that's getting back from it. So I just wanna, like, define that really quickly for the audience because I think that there's so much out there that's like, this is a totally autonomous agent that's like running your your, like, entire marketing stack.
And it's like, no. It's not. Like, this n eight n workflow, I guarantee you, is not entirely running your marketing stack.
But in contrast, like, I can, and we will like, we'll talk about this today. Like, you can make an agent entirely run Facebook ads now. And I'm talking about, like, it is researching the pain points.
It is researching the outcomes of your target customer. It's making creative that's on brand both statics and also, like, AI avatar UGC.
It's publishing that content into your Facebook Ads account. It's turning off the losers. It's promoting the winners, and then it's going in a feedback loop where it's like, okay.
What what is actually working? And how do I go and basically make more content like the best performers? And I'll talk through all of that today and try to show you as much as I can.
It's not not really easy to show because it's just like code. But, anyway, we'll go through that. So yeah.
And then just to jump into it, I the first business that I wanna talk about to use is kinda like a just a a sandbox for, like, how would you actually build and deploy these agents for this, like, this company type Is AI for WordPress?
I've talked about this before online, but, like, I I I've actually pitched, like, three startup founders this that are like, we got into YC. We don't know what to pivot into. And I just they always just, like, are like, no.
I don't wanna do this. So I just wanted to hit on this idea. So let me let me just screen share real quick and kinda show you the data for this because I think it's just fascinating.
And I think it's such a huge category that, like, nobody is talking about right now. Yeah.
Like, nobody's really executing on. And before people's, like, eyes glaze over and they're and they tune this out, WordPress is bigger than you think. So Oh, WordPress powers, like, 50% plus of the Internet.
Right? 40 Like, it's I just checked 43%
of the Internet websites that are, like
that are indexed by Google or WordPress websites. No one's building stuff for WordPress. It's insane, man.
It's insane. And it's, like, it's such an easy space to compete in. It's kinda for this.
And all I'm gonna pitch to you today is, like, literally, it's just lovable or anything.com but for WordPress. Right?
So you're just all all you're doing is you're making it so that people can, like, vibe code their website, but, like, on top of the WordPress platform.
And then the the business structure is super simple. It's like, hey. You get all this stack.
Like, you get forms solved. You get, like, you know, maybe a CRM. You like, all of these plugins that people would go and, like, hodgepodge together, you just bundle that up into a single package and then you sell tokens.
So it's like whatever. You know, base tier is $29 a month that gets you x amount of tokens and then that makes you so that you can go and basically, like, you know, build out design the site with this, like, pre built in and like, you know, get forms, these these pre built features.
So you got me thinking that I actually think that the there's a bigger opportunity here than just the lovable for WordPress because and you don't know this about me, but back when I was 19 years old, 2000 and I don't know.
I'm bad with math. He was in, like, 11 or something. I was a partner in an agency, and the agency's job was basically to get people off their existing CMS and move it to WordPress.
And we were officially partnered with WordPress on doing that. So if you called WordPress and you're like, time.com, Time Magazine, you're on some custom CMS, they would they would be like, work with Greg and team to move and do the migration to WordPress.
WordPress is happy because they want more people. Right? And and time.com's happy because they basically need someone to handhold through the process.
Totally. So it was a great business. We did, like, time.com, techcrunch.com, a bunch of big websites.
The amount of money I saw that in the in the WordPress automatic ecosystem and plug ins was absolutely absurd. So what you got me thinking about is, you know, the the play is basically how do you find plug ins that have validated demand that don't have an AI component and building the AI first version of each?
Because you already know that, you know, there's proven demand, people are paying for it, and then you just make them 10 x better with agents. So I'll give you a couple of examples.
The Yoast SEO tool. Do you know about that one?
No. Yes. I've used it forever ago.
Forever ago. Right? Yeah.
I think it was I think it's a guy named Yoast. I I I could be wrong, but I think it's a guy named Yoast. There is no way that they, you know, that they have an AI first version of that.
So, you know, instead of showing you red and green dots and telling you to fix things yourself, just have an agent do it. It writes the meta.
It restructures the content. It adds the internal links. Like, make it AI first.
And I'll give you a couple more. WP forms. That was a huge plugin back in the day.
Make an AI form that talks back. So instead of doing, like, a static form, you do, like, a conversational agent that qualifies a lead and answers questions. WooCommerce, one of the most popular, you know, WordPress plugins.
You know, do like an AI storekeeper so it automatically writes product descriptions, abandon cart flows.
You know, it it can go on and on. Another another one is Akismet, which is the spam plug in.
Like, there's just so many. So all you have to do WorkVent security. Like, there's Yeah.
It's just endless. Right? It's endless.
So many of these. There's so many, and people are paying for it.
43% of the Internet is is is powered by it.
The hard part is is obviously building the product, you know, making it good, making it work, and then finding, you know, a distribution channel that you can actually get customers.
Totally. And I think that, again, the the bundling of all this is where the opportunity is. It's like, hey.
I'm gonna give you all, like, all the plug ins that you would normally be paying for instead of the aggregate of those 10 that are necessary. Now you pay a tenth of the cost, you know, or a fifth of the cost, and I'm gonna give you everything, and you're gonna also be able to, like, vibe code to the design of this. Right?
Like, that is the business opportunity here. And then to your point, like, how do you market this? Right?
And this is where this gets so fun.
Like, I'm having the most fun of my entire career right now with what we're doing, and so I can't wait to, like, share this with you guys. But, anyway, so the op like, how when I think look at this company, like, where do you meet the people that are, like, trying to learn? Like, how do I do this or how do I, you know, do this work with my WordPress site?
So, like, Facebook Ads and Google Ads are gonna be the first two places that you can go to. I'm gonna talk through a bunch of different things, different opportunities. Um, but this is just like, we're gonna just run down the list of all the things that you could do, and here's how to deploy agents for them.
So on the Facebook Ads side, so for for targeting on Facebook anymore, um, after Andromeda, you don't do interest based targeting. Historically, how I would probably do this is I would actually go and I would target people who have a WordPress interest on Facebook ads. Um, now you don't need to do that.
You just basically have to make ad creative that is talking to the pain points or the outcomes of the person that you're trying to sell to and then send them to a landing page that, like, describes what the product is. So what is Andromeda?
What how does it work? And, like, what what did what do you need to know about this? So Andromeda is, uh, Facebook's new, like, ad algorithm.
Right? It basically it decides what ads get shown to what people. And how it functions is the the AI is basically looking at the ad creative, and it's understanding, okay.
Like, what is in this static ad? What is the what is the image? What is the text?
What is the video, what's in the video, what's the script of the video, and then it also understands the landing page that it's sending the people to. And then from that, it basically uses that information to decide what ad gets shown to what person on the platform.
So I you know, jokingly on Twitter, I've been saying this a lot, and I I mean this, though, from, like, the bottom of my heart. Facebook has turned into the best b to b ads channel, and that exists right now. And the reason for this is, like, we'll make these ads that are so obscure.
Like, it it's like maybe there's, like, 10 people in the in The US, right, that, like, would would would be having this problem right now. Um, and Facebook will find those people and show that ad to it, and we'll get that lead, like, a qualified lead for a b to b business. It's ridiculous.
So, anyway, to take a step back to the the WordPress side, like this AI for WordPress thing, what you end up doing and, like, I can already see what the creative looks like. Like, the script end up ends up being like something like this. Like, oh, uh, this is kinda crazy.
I used to be paying my agency a thousand dollars a month to maintain my WordPress site, and they were not responsive and they never, uh, like, actually could do the designs that I wanted. But I just found this, uh, this, uh, this plugin for WordPress that makes it so that I can, uh, chat with AI to change my website, and it concludes all these plugins that I used to have to pay extra money for.
It's just a part of it for free. Click the link to learn more. If you run an ad like that, Facebook is gonna automatically identify, okay.
These people are WordPress website owners. They have this problem. It's gonna show that to them, and then you basically set up a conversion event for the sign up action, for the payment action that's deeper within the funnel.
So how do you actually do all of that? Like, I'm just I'm talking through all this. Let's just, like, go into this deep and, you know, break down, like, what this actually looks like from an AIMogen perspective.
So okay. So the first thing is researching the pain points of your target customer.
I'm just gonna use perplexity for this because I've just found that it's easy to do live, like, sharing this. Um, actually, let me stop this, and I'll screen share a different way just so that you can just see the window. And for this, we're just gonna do, like, scrape Reddit, pain points, and outcomes, uh, that people that run WordPress websites, uh, have for like, why would they buy an AI, uh, WordPress tool that basically, like, solves all the plug in problems and makes it so that they can vibe code the design of the website?
The reason you're doing Reddit, by the way, is because those are real people just complaining,
basically. Exactly. That's exactly why it is.
It's basically dialogue. And, again, a lot like, Reddit is being eroded right now by people manipulating it for, like, LLMs to rank on them.
But this is the way like you. By people like me that are basically creating the dead Internet. Let's be real.
It's already happening. I am the problem. But with this, though, what it enables is you basically can identify, like, what are the problems that they're facing.
And then from this, this crew, uh, basically seeds the creative that I want that I that I go and make. Right?
So I found these pain points. Right? Let's, like, pull out, you know, uh, one of these that makes sense.
So content workflows, customization requires development skills. Right?
Like, this is one that you could talk to. What I would then go do is, like, once I found this information, I'd be like, rank stack based off of, uh, the most referenced.
Just to give us, like, what are the top three? And then once I've identified that, so the plug and layer, most reference, This is the biggest pain point for people. Awesome.
That is what ads are gonna be about. Um, performance and speed. Cool.
Can we make this super lightweight? Can we use something like ASTRO, um, to actually, like, serve it so the content's really fast?
Security and maintenance. Okay. How do I make it so they never have to think about plug in security again so that their site doesn't, you know, randomly get gambling ads on it and then get de indexed by Google for their pest control business.
Right? So with this, what I then go do is I'm going to use, uh, Kai AI is what, uh, we're using right now for all of our image generation.
Basically, just all of the image models, uh, and video models in the same place. And so with this, I can then go and bulk generate static creative.
And the way that we found to do this, uh, is you basically give it an example of an ad that is already being ran by either a competitor or by somebody, like, you know, somebody in a tangentially related industry. And we're just using Google Nano Banana for this.
Um, so the agent then, again, is generating this creative using the static. You can make this as advanced as you want.
This is what we're doing a lot now so that it always hits brand style guides. And what do I mean by brand style guides? That's like, what are the fonts that need to be used?
What are the colors that need to be used? All of these specifics about the company, about the brand. Right?
So with this, you basically can put a vision model over the outputs and be like, hey.
Does this match brand style guides? So what I'm getting back from Google Nano Banana, I can be I can, you know, basically qualify the like, is is the text all readable?
You know? Does are the fonts that are being used actually on brand, etcetera?
And then from that, you basically have this way to mix statics. And then on the video side, what we are still using and still seeing great results with, which is crazy to me because it's not even the best, is HeyGen.
But we're starting to experience more and more with Seedance, and this is, like, where this is all going to go. It's just hard to do the stitching across, uh, like, Seed Dance, like, for longer, you know, thirty second videos. You can only do up to believe it's nine seconds.
Don't quote me on that, but it's it's less than that. So you basically have to, like, stitch these frames together is the challenge, but this is where this will end up going. And then it's just like AI avatar UGC that's talking about the pain points or outcomes that you just extracted from what people are saying online.
And everything that I'm describing here, this is just an agent that's done this work.
Right? It is an agent that's done this research. It's an agent that is calling doing the API calls to actually build out the creative.
And then from there, how do we actually get it into Facebook ads? Right? So to get it into Facebook Ads, then what we're going and doing is we need to use what's called the Facebook marketing API.
And you probably if you're online at all and not living under a rock, you've probably seen tons of people where it's like, I got my Facebook Ads account banned, right, because I was using an agent with it. The agent is not the reason it got banned. Alright?
The reason it got banned is because they violated TOS, and they spammed the API because they were trying to pull in, like, you know, hundreds of millions of rows of data from their Facebook Ads account, and that is not the way that you you can't use the API in that capacity.
Like, it's literally like that that that's against Facebook's COS. That is the problem. So how you actually use the Facebook Ads API is you use the API only for rights, like, to publish content to the Facebook Ads account or to turn off or to basically, like, promote ads, and I'll talk about that more in a second.
And this I'm realizing just based on time, this whole video is just gonna be, like, how do you build, like, an actual, like, marketing agent. Like, we're gonna only probably get through one of these. If you want more content like this, this is all I talk about on the Internet, so just, like, go follow me across channels, and this is what I share.
But, anyway, with this, though, the to solve the data analytics problem, you have to build a data pipeline and a warehouse.
So the open source way to do this is using a tool called AirByte. Um, you can host this, like, you know, anywhere. You can literally have Clog code go and do this for you.
Right? Set this up. And then you can use a data warehouse called ClickHouse.
These are both open source solutions that you can self host, and I'm just giving you those as the that's, like, you know, the the the way that that is totally accessible to anybody. And and, Cody, can you explain, like, what is a data warehouse?
Like, what are you saying? Absolutely. So a data warehouse is a way to unify all of your data sources into a single location so that your agent has the ability to understand all of the data in context with all, like, all channels in context or all data sources in context with all the other data sources.
So I'll give you specifics. That was a very, like, you know, broad way to explain it. So say, for example, I'm running Facebook ads.
How do I actually run effective Facebook ads? I need to do probably four things. I need to have my Facebook ads data.
I need to have my Google Analytics data. I need to have probably my post hoc data, maybe my my CRM data, so, like, my HubSpot data. And then I also need to probably have, like, my Stripe data to understand the entire stack of, like, I ran ads and did customers come out the other side.
Right? And then how do I actually tie the literal specific ad that is driving revenue for the company?
Right? This is how you do that. It's by having this data warehouse that has all of the data in a single place.
Now to get all of that data into a single place, you have to build a a data pipeline. And so AirBite is a data pipeline solution. It's self hosted, um, and it enables you basically, they have all these prebuilt connectors that enable you to pipe all of your data into a data warehouse.
So when you think about, like, you know, visually how this is happening, we can just, like, TL draw this real quick to actually show you. Basically, what I am doing is I am giving, uh, my agents so say I have, you know, these, like, five different data sources, um, that I'm pulling from.
We'll just do three for the sake of time, and I'm putting them all into a data warehouse. And that data warehouse, I'm then giving my agents access to it.
So imagine I have this agent here that can read this information.
It's back and forth. Right? And this is the Facebook Ads agent.
And this Facebook Ads agent, um, how it works is it basically is looking at that data stream, and then it is making updates within Facebook ads.
Right? And then that Facebook ads data is going back to, like, where we're pulling from, which goes back in basically, back into the data warehouse.
Actually, this is probably more like this. Like, back into the data warehouse. Right?
And so this is how I'm creating this, like, loop that is basically learning. Right? Um, I'm giving the agent access to the live business data, access to the livestream.
Also, what this enables for you as a business owner is you can also interact with this within your Claude code, within your codex, whatever that is, right, and whatever your daily driver is.
And the reason that you want this is you can have conversational analytics and also build, like, custom dashboards with this data.
Right? So you can literally ask like, okay. Like, we're, like, we're having trouble like, say you're, again, pest control control company.
Or or we'll say, I don't know, something like your commercial HVAC. Like, we're having trouble hitting payroll this month.
Like, what is going wrong? It's like, oh, your accounts receivables are, like, off. You can literally have those conversations with your marketing data, with your business data.
This is the way that you can enable both yourself and your team to be able to do this. Right? So how do you actually so, um, how do you actually set this up?
So we've got the data pipeline in the warehouse solution. Now for hosting the agents, everybody's doing this already, like, locally with, like, their cloud code, and they, like I I mean, I've seen these crazy setups where it's like, yeah.
I'm deploying, uh, these solutions onto, you know, my Mac mini that I bought for OpenClaw, but now it's actually just, like, working as a server. Like, you don't have to use a Mac mini locally as a server to do this. Alright?
You can just deploy this into the any cloud. It can be into a Roku. It can be into Railway.
It can be into, like, any of these solutions. And, again, I'm just giving options that are available. But, basically, an agent is just code under the hood.
Right? It's just a decision tree that's like, cool. Here's a live data stream.
I have some, like, decision loop that's happening where an LLM is thinking, and then I'm basically having it optimized for an outcome, which for most companies is like, how do I make the business more revenue? Right?
So, anyway, that is kind of the the the infrastructure that's necessary to actually make these agents function. Once you've got this whole thing set up, what is possible, just to give you, like, a scope of this, and we're doing this for companies right now, is you can actually, like, create so for example, like, for one company, we're doing two different ad sets per day, five ads per ad set.
They automatically get uploaded into Facebook ads based off of all the research that we've done and also the data that exists within the data warehouse, like, of what has worked. Right? And we are based so there's, like, a learning loop that's happening on top of this.
The agent is looking at, okay. I published these 10 ads.
It lets it run for, a two to three day window for, like it just get initial signal. It turns off the worst performing ads. And, again, how is it getting that information?
It's pulling that information from the data warehouse. Right? It's pulling, okay, what is the worst performing ads?
Let's turn those off. The winners live on. And then the winning ad sets, they go into a winner's pool, and they're competing against each other for the ad budget that's available.
And then over the top of that, you can have a vision model or, honestly, like, how we're doing this more and more is we build a database of all the ads that we have created, and it's literally like the JSON prompts that we're sending to Nano Banana or the ad scripts that we're using and we're sending to HeyGen or SeedNance.
And that is what we are letting the agent analyze of, okay. What is actually working so that it can it can go and, you know, basically become more effective at making creative that creates outcomes.
So the problem that happens with this year and, again, I'm gonna get into the weeds and talk through a lot of this, you know, in in detail just to share the the issues that come up, is you have to solve for entropy. So, like, what is entropy? It's like, basically, the agent gets stuck thinking in the same way.
So how do you solve for that? The two ways that we found that we can solve for it is you can use the Facebook ads library to pull ads from competitors' libraries, and that basically, like, puts new DNA into the system.
So you're basically feeding that source text that you did that research on. And then the other thing that we found to work is taking YouTube transcripts and also podcast transcripts and basically pulling out the insights from those.
There's, like, hundreds of YouTube channels that are just designated to, like, you know, for example, WordPress. You can pull out the insights from those and then run ads basically off of off of those insights that you aggregate. So this is a way to solve that that that that, you know, kind of, like, entry problem.
But what this ends up looking like in the in the, you know, the final form is, like it's like a , Greg, is the best way to describe it. This virtual employee, all it's doing is focusing on Facebook ads as the channel and how it can actually, like, operate and function is because it has all the necessary data it needs to think through what is actually driving, like, outcomes that we are like, that that drive business revenue.
Right? And this is where all this is going. Like, it's so funny.
I talk to, like, so many marketers on a weekly basis, and they're all like, is my job secured? I'm like, I don't know. Like, can you do this?
Like, this is what this turns into. It's like, your job now becomes, like, I'm an .
You're taking your domain knowledge that you have, and you're basically, like, building this into code. And, like, I'm not technical. Right?
Like, I can set this up now. Like, the models are smart enough now. We're, like, a semitechnical person or just if you have access to Clog code and literally just give it the transcript that I just, like you know, we just created on this call.
Be like, okay. I wanna set this up. Walk me through this process.
Like, you can go and build out this solution. Like, that is the the possibility here. So, anyway but, yeah, happy to answer questions about that.
That was one agent. I'll I can go down into a deeper one as well. So Okay.
Yeah. And I think
taking a step back, it's like, well, why are I I know it's crazy to say this, but why are Facebook ads important? Well, okay. If you're gonna steal an idea from this podcast or come in come up with an idea on your own, you're gonna vibe code something that you think has value in the world.
The biggest question I get is, how do I actually get customers to this thing? And the truth is Facebook is the most cost efficient way to do it If you don't wanna go and create content which takes time to actually, you know, build an audience, like, I I I still think that people should create content, but at the same time, be thinking about how can I create an, you know, an ad set that is gonna perform?
Now what people the mistake that people make is they try a few ads. It doesn't work, so they give up.
And they don't realize that you actually should be taking the same ad and just, like, changing the positioning a little bit because and and and changing it a little bit, like, 10 times or 15 times or 20 times.
Right? And what happens is, like, I'm shocked sometimes with some of the Facebook ads that I do, and I'm like, wow.
I didn't think that this ad set was actually gonna work. This positioning was gonna work. Like, for example, you know, let's say we wanted to do that WordPress plugin idea.
And let's say we wanted to compete with Yoast, the SEO plugin. Like, maybe the positioning is Yoast is bad. And I you know, it's just like an anti Yoast,
you know, and and you never would have thought that. Do work rather than doing work for you. Right?
Right. There's a great there's so many angles that you can pull out of that. To your point too, I just wanna piggyback off of this.
Like, I think a lot of people come into marketing and advertising, and they're like, I know what the market wants. I'm gonna impose this idea on the market. That is not how this works.
Right? We're not Don Draper. We are, like, putting creative into the mark.
Like, why is why is paid ads the most, like, why is Facebook ads? Why is Google why are they the most powerful assets? Like, why the most powerful marketing channels that have ever existed.
I can go and test a thousand different ad creatives, and, like, I'm gonna have clear delineation on, like, what is winning and what is being the most what is the market most receptive to.
Right? And you're gonna know that within forty eight hours. And I see this all the time where, like, oh, they're like, oh, I don't wanna go run ads.
I don't wanna like, I only wanna do organic. Right? Or, like, how do you grow?
It's only by word-of-mouth. I'm like, this is such a that's such a rookie way to look at the market. Like, you should be doing all of it, number one.
Number two, like, you need to think about how does this like, this is the only system where I put a dollar in and, like, $5 can potentially come out the other side. And I can just keep where it's like, just keep putting money in, and that same, you know, volume will like, that same multiplier will come out the other end.
So, anyway, just a And and so what's changed
like, why I now I'm into Facebook ads, honestly, is because of the systems that you're describing. So, like, in the past, if you were a founder and you wanted to do what you're talking about, you had to hire an agency that would cost you tens of thousands of dollars a month to go and run this at scale.
Now you just need to set up the system that you talked about.
And, of course, it needs to be fine tuned. Of course, there needs to be it it needs to work.
Right? Of course, you need to solve for entropy, which I'm happy you spoke about because no one speaks about that. People are like, oh, yeah.
You set it up, and then this is good. It's good. Like, no, dude.
That's
lit. Though the Through the opposite.
Yeah. Have have the people that say that have never set up anything.
Right? Because as soon as you it might feel good, like, on day one, but on day two, it gets a little worse.
On day three and four and five until you get this point where you're like lifestyle choice, man. I think that's the thing that people, like, have the hardest. Like, when you think about marketing, like, people think about it in campaigns.
Like, I start a thing and then I stop a thing. And that is not how this works anymore. Like, how marketing actually functions now is, like, you're continuously doing this.
And because the feedback loops are so fast, like mean, when you look a great example of this is if you just look at, like, the fashion industry, like, how quick the the the trends occur now. It's like 10 x faster than what it was twenty years ago. Right?
And what it was in, like, the early two thousands. And so when you think about that and apply that to everything, like how quick the market moves, you have to basically have, like, you know, some type of observability of what's happening so that you are adjusting your, you know, ads, your positioning, your all of the strategy that you have across your your your go to market based off of those trends that are that are are basically, like, evolving.
And where does all this happening? Right? It's all on short form social.
Like, that is where all this kind of originates from, and it trickles down in everything else. So you can actually do this. Right?
There's another way to solve this entry problem. You can use tools like Viralow. They have an API.
Basically, it's like it scrapes, um, like, all of TikTok, and I think they're doing Instagram reels now. Don't quote me on that.
Um, but, basically, I can be like, hey. Show me all of the most viral, like, posts that are in the beauty category from the last week. Cool.
What are there are there trends that are happening here? Are there, like, certain content formats that are happening here? And can I use that information to basically influence the creative output that I'm doing?
Yeah. 100% you can do that. This is, again, like, how you could how it's this push and pull with the market that you're doing constantly.
Right? And, like, agents for the first time have can enable you to do that. Like, what I just if I wanted to go and create a 100 ads, right, and get that into Facebook ads, like, that is probably, like, two weeks of work, right, historically.
And now you, as a founder, like, literally in the next hour and a half, you can have this whole system that I just described set up and be able to actually go and publish this content, like, at scale with the understanding of what's actually working. And maybe there's a human in the loop to begin with, but, like, what this gets to is, like, you're gonna understand, like, this is what's actually functioning.
Cool. I now want an agent to go and run this. I deploy an agent into the cloud with the necessary data stream that it needs so that it can actually make these decisions.
So Cody, what else did you wanna cover? Dude, I have so much.
I have to jump in two minutes, though. So I'm like, it's I'm kinda bummed.
So I I think So, folks, comment if you want more marketing I'll come back and just do a whole series on this. Like, here is literal, like, agent by agent, like, descriptions, like, showing the actual, like, structures, talking through the actual campaign structures. If that is what people want, like, this is, again, all I'm thinking about right now.
This is all I do from when I wake up to when I go to bed at the moment because I'm obsessed with it. That is uncomfortable how like, it's a compulsion, and this is not really a choice.
So anyways, yeah. Greg, didn't mean to cut you off. No.
No. No. No.
Let us know. I think, like,
it you know, people like I said, people wanna figure out ways how they can use AI agents to get them customers.
So I think marrying, like, here's a couple ideas with here's some here's some agents is gonna be helpful for folks. Let me rattle off some, and then Yeah. You comment what you want, and we'll come back and we'll do this.
Right? So I can we can do Google Ads agents that are entirely running your Google Ads accounts. We can do influencer out outreach agents that are researching influencers within your category, cold emailing them, and then negotiating the pricing with them and hand raising for when, you know, they think it's a good person.
We can do cold email of, like, how do I go and find all of the customers, you know, that I'm trying to sell to, find their email addresses, like, cold email them, and then have an agent managing that inbox, like, to them to book new meetings.
We can do TikTok reel farms where we're basically, like, making 10 TikTok accounts in the cloud and then at scale making slideshows that are related to your product, scheduling those and posting and aggregating all of the views across that. We can do, uh, what are some of other things?
We can do SEO agents that are doing keyword research, researching the article, writing the article, having the article be on brand for your voice. Like, how do you actually give it a unique perspective or point of view so that it has that, like, you know, basically personality to it so that it actually ranks well and it's, like, good to read?
We can talk about AI search and how do you build citations? How do you actually do cold outreach to get your brand listed on the citations that are already being shown in the category that you're in? Uh, we can talk about social media management and, like, how do you get an agent to run your LinkedIn account, to run Twitter accounts?
We can talk about how do you get agents to build email newsletters that are related to podcasts and actually research the podcast, write the podcast script, use Elevenlabs' voices to read the podcast script, and then build lead magnets to build an email list on the top of that podcast for an industry like ecommerce.
So if you comment what you wanna learn, I'll come back. I'll share all these things that we're seeing and all these things that we're deploying. I'd I'd love to do that.
I think it'd be awesome. So God bless you, Cody Schneider.
You know? I appreciate you. It looks so weird.
We'll see you next time. And,
I'm curious to see or hear what people think. Yeah. If people have questions, let me know too.
I'm active on Twitter and LinkedIn. Those are the biggest places.
And companies graph.com, all this work that I just talked about, that's where we're doing this. So thanks for having me, G.
Later.
The Hook
The bait, then the rug-pull.
Vibe-coding a product is trivial now; getting it customers is not. Greg Isenberg presses Cody Schneider for a working definition of a marketing agent, and Cody spends the next half hour proving there is one: unified data, a decision cadence, and a feedback loop, built out live as a Facebook ads system for a fictional AI-for-WordPress company.
How they asked for the click.
MENTIONED ON CAMERA
