18:02Kev Builds Apps · Tutorial
I Built a Self-Improving SEO Agent With Claude Code
A looping Claude Code harness that writes blog posts to win AI search and rewrites your checkout page based on what the data says is converting.
October 4thThe argument in one line.
A coding agent can run real marketing operations end to end, but only if it reads from a unified data warehouse instead of hitting ad platform APIs directly, because direct API calls hit rate limits, blow the context window, and cause hallucinated joins.
Read if. Skip if.
READ IF YOU ARE…
- A marketer or growth operator who already uses Claude Code or a similar coding agent and wants to automate ad creative, uploads, and reporting.
- Someone evaluating whether to build their own data pipeline and warehouse stack versus paying for a bundled tool.
- A solo operator or small team running ads on Facebook, Google, or LinkedIn who wants one chat interface instead of five dashboards.
- Someone curious what the emerging 'marketing engineering' job title actually involves day to day.
SKIP IF…
- You have no marketing budget or ad accounts to connect, this is infrastructure for running live campaigns, not a general AI tutorial.
- You're looking for prompt-engineering tips rather than a systems/infrastructure explanation.
The full version, fast.
Marketing engineering means handing the manual middle-work of marketing to a coding agent: writing copy, making creative, uploading ads, researching keywords, finding leads, and publishing content. The video argues the agent needs ten infrastructure pieces to do this reliably: a data pipeline, a data warehouse, a cloud server, media storage, a database, cron jobs, artifact hosting, authentication, a git manager, and an API gateway, because calling ad platform APIs directly causes rate limits, context blowouts, and hallucinated data joins. The agent should read from the warehouse and only write back to ad platforms via their APIs. Cody then runs a live Claude Code session using the Graphed CLI to generate a Facebook ad image, get a Facebook Marketing API token, upload and toggle the ad, query Facebook/GA4/Search Console data into a shareable dashboard, scrape LinkedIn post engagers into enriched emails, research SEO keyword gaps, and auto-publish a blog post, all from chat prompts.
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 →Where the time goes.
00:00 – 01:36
01 · Intro
Defines marketing engineering as having a coding agent do manual marketing middle-work (copy, creative, uploads, research) and previews the demo.
01:36 – 02:39
02 · Why marketing engineering is exploding
Shows a Google Trends chart for growing search interest and rising job postings, claims salaries above $150K, compares it to the GTM engineering trend.
02:39 – 07:35
03 · What your coding agent needs: data pipeline + warehouse
Explains why an agent should never call ad platform APIs directly (rate limits, ToS bans, hallucinated joins, context limits) and why a unified data pipeline feeding a SQL data warehouse solves it.
07:35 – 10:26
04 · Cloud server, storage, databases, cron jobs
Lists the remaining infrastructure pieces: always-on cloud server, S3-style media storage, Postgres databases, recurring cron tasks, artifact hosting, authentication, and a git manager for multiplayer work.
10:26 – 11:47
05 · The full marketing engineering stack
Recaps all ten pieces and gives open-source/self-hosted equivalents (Airbyte, ClickHouse, Netlify, S3, Railway, GitHub) before introducing Graphed as the bundled alternative.
11:47 – 12:37
06 · Install the Graphed CLI
Introduces the Graphed CLI as an NPM package that bundles the data pipeline, warehouse, and tool access, installed directly into Claude Code.
12:37 – 13:47
07 · Make a static ad with AI
Prompts Claude Code to generate a 1080x1080 ad image via the ChatGPT Image 2.5 API routed through the Graphed CLI.
13:47 – 16:08
08 · Get a Facebook Ads API token
Walks through creating a Facebook developer app, selecting the marketing API use case, and generating a system-user token that never expires, then stores it in the CLI's secrets manager.
16:08 – 17:18
09 · Graphed tools: the unified API
Lists the ~250 endpoints available through the unified API gateway: ad scripting across platforms, email finding and validation, lead lookups, keyword research, image and video generation.
17:18 – 19:14
10 · Upload the ad to Facebook Ads
Pastes the generated image's URL into Claude Code and prompts it to upload the ad to an existing Facebook ad set using the stored API token.
19:14 – 19:47
11 · Turn ads off from Claude Code
Demonstrates toggling a specific ad off by name directly from chat and verifying the change in Facebook Ads Manager.
19:47 – 20:52
12 · Ask questions across Facebook, GA4, Search Console
Runs a single prompt asking for Facebook signups, pixel data, GA4 referral traffic from Facebook, and branded search volume from Search Console, all pulled from the warehouse at once.
20:52 – 21:33
13 · LinkedIn post engagers to emails
Pastes a LinkedIn post URL and prompts the agent to extract everyone who engaged with it, then look up their emails via a leads endpoint.
21:33 – 22:18
14 · Keyword research for versus content
Prompts a DataForSEO query for 'versus' content keywords relevant to the product category to seed future blog posts.
22:18 – 24:08
15 · Find page-two keywords
Queries Search Console data via the warehouse for keywords ranking in positions 11-20 with meaningful impressions, then filters out duplicate versus-phrasing, surfacing 398 candidate keywords.
24:08 – 24:53
16 · Write and publish a blog post with the SEO plugin
Uses an installed Graphed SEO plugin to research the top-ranking page for a target keyword, write a post with Claude Sonnet, and auto-publish it to the CMS.
24:53 – 25:57
17 · Waterfall email enrichment
Runs a waterfall enrichment across providers like Lead Magic and Apollo for engagers with no email found, then guesses and verifies remaining addresses with Million Verifier.
25:57 – 26:52
18 · UGC ad scripts from Reddit pain points
Prompts the agent to write ten UGC ad scripts using pain points previously scraped from Reddit, with specified hook styles.
26:52 – 28:06
19 · Edit the dashboard
Returns to the earlier cross-platform dashboard artifact, requesting a chart-type change and a date-range picker added, both completed by further chat prompts.
28:06 – 29:08
20 · Wrap-up
Points to graphed.com/platform to get started for as little as $10, plugs the weekly channel and the free GTM engineering course with a signup form.
Lines worth screenshotting.
- Marketing engineering job listings and search volume are both rising fast, with salaries that can exceed $150,000 for people who pair marketing knowledge with coding-agent fluency.
- Calling platform APIs like Facebook Ads directly from a coding agent risks getting the ad account banned for exceeding rate limits or violating terms of service.
- Joining large datasets, like 50 million Facebook ads rows against 75 million Google Analytics rows, inside the agent's own context window causes hallucination and blows the token budget.
- The correct architecture separates reads from writes: the agent reads from a unified data warehouse for analysis and only uses platform APIs to write actions back out, like uploading media or toggling ads.
- A Facebook Marketing API system-user token never expires, unlike a personal access token, which is why it's the recommended way to give an agent lasting ad account access.
- Storing an API token directly in a coding agent's plain context is a security risk; routing it through a secrets manager keeps the token referenced but not exposed.
- A single natural-language prompt can pull data from Facebook Ads, Google Analytics, and Search Console simultaneously because all three already sync into one warehouse on an hourly cadence.
- Scraping LinkedIn post engagers and running a waterfall email enrichment (trying progressively more providers) only recovered emails for about half the list in this live run, the same success rate the presenter says they typically see.
- When an email can't be found through enrichment providers, guessing likely address patterns and verifying them with a service like Million Verifier can be cheaper than paying for additional enrichment lookups.
- SEO content ideas were generated by querying Search Console data directly for keywords ranking on page two (positions 11-20) with meaningful impressions, rather than using a generic keyword tool.
- A dashboard built as a shareable artifact from the agent's own output lets a team view and even modify visualizations, like changing a bar chart to a line chart, through further chat prompts rather than a BI tool.
- The agent can generate ad scripts for UGC creators directly from pain points scraped off Reddit, skipping manual customer research.
Agents need a warehouse, not direct API access, to run marketing
WHAT TO LEARN
Reliable AI-run marketing depends more on data infrastructure, a pipeline into a central warehouse that the agent reads from, than on which coding agent or model you use.
- Calling an ad platform's API directly and repeatedly from an agent risks rate limits and can get the ad account banned for violating the platform's terms of service.
- Centralize data from every channel, ads, analytics, search console, CRM, payments, into one warehouse so the agent queries one source instead of joining raw API responses in its own context.
- Keep a hard split between reads and writes: pull analysis data only from the warehouse, and reserve direct platform API calls for write actions like uploading media or toggling a campaign.
- Use a system-user access token instead of a personal token for platform APIs like Facebook Marketing, since system tokens don't expire and survive personnel changes.
- Route API tokens through a secrets manager rather than pasting them into the agent's chat context, since raw tokens sitting in context are a real security exposure.
- A single natural-language prompt can answer cross-platform questions, like whether branded search is rising, once the underlying data already lives in one warehouse synced on a regular cadence.
- Waterfall enrichment, trying progressively more providers for the same lookup, is standard practice for filling gaps in lead data, and typically only recovers emails for about half a list.
- When enrichment services run out of real matches, guessing likely email address patterns and verifying them is often cheaper than paying for another enrichment provider.
- Search Console data can directly drive content strategy by surfacing keywords already ranking on page two with real impression volume, a more targeted approach than generic keyword tools.
- Dashboards and reports generated as agent output can stay living documents, since further chat prompts can modify chart types or add interactive filters without touching a BI tool.
Terms worth knowing.
- Marketing engineering
- Using a coding agent to do the manual middle-work of marketing operations, such as writing copy, making creative, uploading content, researching keywords, and finding leads, that a human marketer used to do by hand.
- Data pipeline
- A system that collects data from multiple sources, such as ad platforms and analytics tools, and funnels it into one unified location the agent can query.
- Data warehouse
- A database built to store large volumes of data and query it quickly, typically with SQL, serving as the single source of truth the agent reads from.
- System user token
- A type of Facebook Marketing API access token tied to a business account rather than a person, which does not expire the way a personal token does.
- Secrets manager
- A service that stores API keys and tokens securely and hands them to an agent on request, instead of pasting raw credentials into the agent's context.
- Waterfall enrichment
- Trying multiple data providers in sequence, typically from cheapest to most expensive, to find a missing piece of information like an email address.
- Unified API gateway
- A single API layer that routes an agent's requests out to many underlying third-party services, like image generation or ad platforms, so the agent only has to learn one interface.
- GTM engineering
- Go-to-market engineering, a closely related discipline to marketing engineering that applies the same coding-agent-driven automation to broader go-to-market operations.
Things they pointed at.
10:50toolAirbyte
10:51toolClickHouse
13:02toolChatGPT Image 2.5 API
19:10toolSuper Whistler (transcription tool)
25:00toolApollo
25:00toolLead Magic
25:13toolMillion Verifier
21:43toolDataForSEO API
Lines you could clip.
01:30
“Marketing engineering is having your coding agent do all of the middle work that you used to have to do by hand as a marketer.”
03:40
“You can easily make $150,000 plus a year if you understand how to do this.”
17:08
“Your agent is now the smartest marketer in the world, full stop.”
27:15
“Every Monday I used to spend like three hours pulling scenes... all just to answer one question. Did we actually make money last week?”
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.
metaphoranalogy
In the next 10 -ish minutes, I'm going to teach you how to use your cloud code or your cursor or your Hermes agent or your Grokbot. It doesn't matter what harness that you're using, but I'm going to teach you how to do marketing engineering using these tools. In marketing engineering, what that is, is having your coding agent do all of the middle work that you used to have to do by hand as a marketer.
That could be writing copy, that could be making creative, doing graphic design, uploading the actual blog post to your website, making the landing page, researching the keywords for Google ads, researching your target customer, finding their emails and adding them to a cold emailing software, building an email newsletter so that you can send it out to your audience for a webinar that you're hosting.
monitoring your competitors' ad libraries on Facebook, Google, and LinkedIn so you can understand how they're positioning themselves. Everything that I just described that used to be manual work, a human having to do that work, can now be completely augmented with these AI coding agents. And I'm going to teach you what you need to do to get started with this.
I'm going to explain marketing engineering to you. I'm going to explain why it's so important for you to learn as a marketer so that you have a job two years from now. And I'm going to give you real world examples of the outcomes that come companies are making by using this tooling.
With that said, I'm Cody Schneider. I've been doing digital marketing for the last 15 years of my life. And for the last four years, all I've been doing is figuring out how to get AI to do marketing operations for us.
And quickly, before we jump in, this YouTube video is brought to you by graph .com, the marketing engineering cloud. Click the link in the description to get started below for free. Okay, so first off, why should we even be caring about this?
Why should we even be having this conversation? I just want to show you this chart. So this is marketing engineering searches over time.
This is a very quickly growing category and companies are trying to hire for this role. If you look at the... total number of job listings that are happening for this role, it is skyrocketing.
And the salaries for this are insane. You can easily make $150 ,000 plus a year if you understand how to do this. If you take your marketing knowledge and you learn how to do this engineering component by using these coding agents.
And this marketing engineering trajectory has a really similar trajectory to what you're seeing with GTM engineering. Also an extremely fast growing category of job. Companies want this more than anything.
It's basically somebody who's technical enough to build out these go -to -market motion operations, whether that's in marketing and in the go -to -market. And on this channel, we actually talk about both of these. We talk about GTM engineering and marketing engineering.
So subscribe if you want content like this. But for today, we're just going to be doing this marketing engineering crash course. So let's get into it.
What do you need to even do marketing engineering? This is... everything.
This is everything that we have found that is necessary to be able to do this function, to actually get Claude code, to do the work for you, to actually get your coding agent, like a grok bot, to actually be able to do this marketing, to actually be able to do this marketing operations. So first you need a data pipeline.
And you need a data warehouse. What are those? So a data pipeline is how we collect all of the data from all of our different sources into a single location.
So a data pipeline enables us to basically pull data from Facebook ads, from Google ads, from Google analytics, from search console, from our CRM, like a TO or HubSpot, from our payment processor like Stripe. And it puts it all into a location that the agent has access to. Why do you need to do this?
Why can't I just use the APIs to call? to these data providers. There's a couple of things that will go wrong.
One, you'll run into rate limits with those API providers. If you spam those APIs too aggressively, for example, with Facebook ads, it can be against their TOS or their terms of service. So they'll ban that Facebook ads account and also doing joins across all of this.
joining data across these sources in the agent is extremely hard. And you're going to find that it hallucinates. You're also going to run into these challenges of eating up all of the context.
If you pull in 50 million rows from Facebook ads and 75 million rows from Google Analytics and are trying to join that data to... understand the historical context and provide that to your agent, you're going to very quickly run into the token input limits that that data provider has or that coding agent has. All right, so we have to get the data from all of these different locations and then we have to put it into a place, into a single location that's unified so that our coding agent could then go and access that data whenever it needs it.
And why do we even care about this data? Because this is the source of truth that the agent is making decisions on. It has to have this to be able to actually do the work.
And if I only have a single source, if I only pull from Facebook ads, that can misrepresent what's actually happening. I have to be able to go all the way down into what is actually driving revenue, not just that top. initial data source.
And I see this mistake happening all the time with people when they're doing this work, they don't realize that like, I need to be able to tie the ads all the way into what is actually producing outcomes. So once I've unified that data, where am I unifying it? I'm unifying it in a data warehouse and a data warehouse is a very specific type of database.
It's built so that it can hold a lot of data and query that data really quickly using a coding language called SQL. And your cloud code or your cursor, they're incredible at writing SQL. All of these modern models are really good at writing it.
And this is actually what we give the coding agent. So it has that source of truth. The data is all going to the data pipeline and the data warehouse.
We give the coding agent access to the data warehouse. That's where it's pulling its data to do its analysis. And then with that analysis that it does, it then can go and do actions.
So when we think about the structure of this, I'm pulling the data into the data warehouse. The data warehouse is going to something like cloud code. So let's just type that here, cloud code.
And then I only use the API keys, for example, for Facebook ads, for Google ads, whatever other channels that I'm running. I'm only using the APIs to write to those platforms. And what do I mean by write?
So read is what's happening here. I'm reading from this data warehouse and then I'm writing to, for example, Facebook ads. via the API to upload media, to turn ads on and off, to move ads inside of ad sets.
And again, to reiterate, why are we doing this? Why are we having this separation? It's because we will run into rate limits if we're trying to query all the data that's in Facebook ads and not having this kind of circular loop that's occurring.
And in reality, what this actually looks like is something more like this. So basically we are writing. back into Facebook ads.
The data is going back into the data warehouse and it's the circular loop that's kind of happening continuously. And this is the proper way to do this. And I say this is the proper way to do this because you can probably run a small account on the back of these APIs, a single channel.
But as soon as you get the scale, if you're a real business, you're going to be making so much data that you have to structure it in this way to be able to actually understand and give the coding agent what it needs to be successful without constantly having to solve for the API. Limits the context we know that you're burning all these things that you'll run into and I promise you you will run into these We've been doing this for two years now.
We always see the same thing. We always get companies that come to us and they're like, oh yeah, I did this locally. And then I've realized that I ran into these problems.
And that's great. I'm happy that they're discovering this, right? Because it's so game changing when you think about the amount of work that a person can do.
But structurally, this is the correct way to do this. All marketing, everything that we're doing is now code. We're trying to give the coding agent what it needs to be able to program and do marketing with code successfully.
All right, so coming back to this. Next item is a cloud server. So a server, when you hear that word, just think this is a computer that's in the cloud that I can go and deploy software to that can run on.
any time frame, and that computer is always on. It's always accessible. And I need this if I want to be able to run one of these scripts or one of these pieces of software, one of these agents, without having to have my own personal laptop on constantly.
So whenever I hear a server, just think about that. The other thing you need is media storage. The other thing you need is media storage.
Where do I store images and videos? For example, we make a bunch of media and marketing, whether that's social posts or ads. or AI avatars, you need a place to store those.
The best place to store those is some type of S3 bucket. Super cheap to do this. And then the other thing that you also need to provide is databases.
So Postgres databases is a database type that enables you to store information so you can access later. How we see this used a lot and how we use this a lot is by giving the agents a database for it to store, for example, information that it extracted for its total addressable market mapping for cold email. a way to organize the images and videos with a specific skew, and the JSON that was actually sent to, for example, ChatGPT Image 2 .5 or Seed Dance to actually make that piece of media.
And these databases are flexible. You need to be able to create these on a whim as you need them. The next thing that you need to do is have recurring tasks or what are called cron jobs.
So these are things that can run on some type of cadence. It could be hourly, it could be at a minute, it could be daily, but it's basically a way to run a piece of software, you know, in some type of timeframe. You also need artifact hosting.
So this is like if you need to share a dashboard or some type of webpage that maybe has all of the ads that you made so that you can get approval, some way to share that. And as soon as you have that artifact hosting that's at an accessible URL, this is like a share link. You have this problem if you have to think about authentication.
So how do I give people the appropriate authentication so that they can go and access it, but other people can't. And that kind of relates to the shareable links thing. And then the final piece is you want to make this multiplayer.
You want to make this so multiple people can work on these systems simultaneously. So you have to have what's called the Git manager. So a Git manager is basically a way for you to store code and have multiple people work on it without having conflicts of code being overwritten.
Imagine two people working on a document or three. people working on a document together and they're writing over each other and they don't know the other changes that people are making. This is a way to basically solve for that.
And then the final piece is an API gateway to all of the APIs that you need to actually do this work. APIs to make images, APIs to get emails, APIs to make videos, APIs to scrape the web, scrape social media accounts, etc. And once you have all these tools, , full stop.
It is the best marketer that exists on planet Earth. But you have to basically remove these barriers so that it can actually go and do this work. Give it the right resources for it to be able to do these actions.
And everything that I just described, you can just go and build some type of amalgamation of all this. A data pipeline that's open source is called Airbyte. A data warehouse that's open source is called ClickHouse.
Cloud server, you could run on something like Netlify. Media storage could be an S3 bucket. Databases could be stored on something like a railway.
Recurring tasks, just a cron job function that's running on that server. Artifact hosting, potentially you're shipping that to Vercel. Application authentication, you make a login database, shareable links.
Again, related to that login database and a Git manager, you could use something like a GitHub. And then for the tailored API gateways, you could go and get subscriptions. to the 40 different APIs that are necessary.
But I'm not going to do that today. Instead, I'm going to use the marketing engineering cloud from graph .com. It's literally free to get started.
It's pay as you go, put $10 in and you can go and extract a thousand emails or something in that range. It's ridiculous. And the reason I'm using this is because everything that I just described is being necessary to do this work.
It's all bundled up into that single cloud provider. And all of that becomes accessible through the graph CLI. And the graph CLI is just an NPM package.
This is just a code package that you give to your clog code. You literally give it this URL. And you say something like cloud code, install this.
And what I just did there is I'm using a transcription software from a friend's company called Super Whistler. Shout out Neil. We love the work they're doing there.
And this NPM package, it has all of the knowledge and everything is bundled up into that CLI so that you can do all of this work. The data pipeline, the warehouse, et cetera. It's all captured within there.
Just to show you what I mean by this, you can go. and connect any data source, whether it's Go high level or Google Analytics or Google Search Console or Instantly or Facebook ads, Post -its, Twilio. There's 800 plus data sources that you can connect into this.
And then all of these other tools, they're now accessible to you within Cloud Code. So now let's actually go do some work together. So I've installed this CLI already.
And the first thing that I want to do is I want to make an ad. And I want to upload it to Facebook ads. And that's going to be the first thing that we do today.
So I'm going to start a new chat. I just did command new. And again, I'm just working within cloud code.
And I'm going to say, use the graph CLI. I want to make a static ad that's a 1080 by 1080 pixel ratio using the chat GPT image 2 .5 API endpoint. And that ad, I wanted to say something like marketing engineering crash course and go to graph .com to see what the design style needs to be for this ad.
All right, I'm just going to hit enter. And now what's going to happen with that graph CLI installed, it knows that it needs to go and pull through the unified API. So this here, whatever's necessary to call the chat GPT image 2 .5 URL.
And it's going to go and make an ad using that endpoint. Now we're just going to make a single ad, but I can make as many variations. of these that I want.
And this is how I can make creative at scale, statics at scale. And I can put other requirements into them. I can have it be specific fonts.
I can have it be specific colors. I can have it be specific language that I can and can't use if I'm in a regulated industry. And while that's working in the background, I want to now talk about how do you get the API key for Facebook ads.
So we're going to use the Facebook ads marketing API. And the marketing API is basically our bridge to access Facebook ads. It enables us to go and manage an account.
And this is the entire process to actually go and set up the Facebook ads token that's necessary. So first, you're going to go and create a developer app by going to business settings. Click add in the top right.
I'll include this Notion document in the description below as well so that you can go and access this. And then you name it whatever you want. For example, graphed ads.
And then you select the use cases that you want to create and manage ads with the marketing API and create and manage ads with the meta ads manager. Connect the business portfolio to its own account on requirements. Click next.
confirm the overview and create the app. And the last thing you need to do is add a privacy policy. I won't bore you with this.
It takes about 10 minutes to get this set up. But once you have it set up, you'll have an API key that never expires. All you have to do is add a system user and that system user enables that API token.
And then you just assign the assets that you want it to have access to. And that token gets generated. And then you just provide that to your coding harness.
given the graph CLI access and set that up. And you can store this token locally, but I would not suggest doing that largely just because it's a security vulnerability. The best way to do this is you.
do it in like a one pass and then it only can use it. With that, like one pass setup, it's just complicated. The way that we do it is we basically do a secrets manager and that's what's actually happening under the hood is we're basically doing secrets management for you automatically.
So once you have that token, you paste it into cloud code. I would paste it right here and I would say, add this Facebook ads API token to my secrets. And then I can say something like list the secrets in the graph CLI.
And you can see all the secrets that are available here. I've got my Facebook ads, my Google ads, my instantly API, my strappy, which is my blog post and all the other API keys that I've got stored in that internal account. But I can add anything that I need to this, but it's largely just these ones that are necessary for the ads management or for my CMS.
Everything else is already built into this. And so let's take a look. It made this image just.
Basically decided this is the right way to build it. And how did it actually do that image? So we can read it here.
It basically used the chat GPT image 2 .5 through the graph tools to send that image to a model called Kai AI. And then that's where it generated that image and brought it back. So what are graph tools?
List all of the graph tools. So graph tools are the unified API that I was talking about previously. It's like 250 endpoints that we can call to have it do work for us.
You can see all these different tools. and new tools are constantly being added based off of different workflows that we identify and we aggregate. But here's these tools that are available.
We can script the ads from TikTok, from Google, from Meta. We can find emails. We can validate emails with million verifiers and get leads.
We can look up people. We can look up their phones. We can find funding signals, news search.
We can research keywords using... the data for SEO API. We can create images.
We can create videos with Seed Dance, with HeyGen. We can do LinkedIn profile extractions. There's tons of these and all of these are on a pay -per -use basis.
So it's just usage -based. All the data that's syncing is just usage -based and all of the API tokens or credits that you need are just usage -based as well. So we've made this ad and now let's actually get it into Facebook ads.
So let me show you how easy this is to do this now. So I've got this ad set, this cloud remix. go and take that and I'm going to just give this URL the cloud code.
I'm literally just going to paste it in. And I'm going to say, upload the marketing engineering crash course ad to this ad set. And I'm just going to add a piece at the front, use the graph CLI and upload this.
So it now knows that it has that API token that we generated previously stored within its secrets. And when I say upload this, I want it to upload it into the platform. And this is how I can get it to do all of this middle work that used to take me hours, but I can do this at scale.
I can write a hundred different, I can make a hundred different ads. I can write a hundred different titles, a hundred different descriptions based off of the pain points and the desired outcomes that I go and extract from Reddit by using the XAI API. And we can go do that while that's working in the background.
Let's go do this. Extract the pain points and desired outcomes by somebody would buy a data pipeline data warehouse solution tool and then give it access, you know, their clog code access instead of having to use Looker Studio or Power BI or any of these other tools. Extract from Reddit.
Those desired outcomes, pain points that they're trying to solve. We're going to use this for ad creative. And again, just adding, use the graphs to see a lie for this.
All right. So that's running in the background. And again, this is how I actually work.
This is how you should actually work. It's having multiple things going on in the background. Once I build the system, once I build this process, I then go and deploy this into the cloud so that it can run on its own cadence without me having to jockey or manage it.
All right. So it looks like it just completed that upload. Let's go check to see if it's there.
I'm just going to refresh this page. And that marketing engineering crash course ad is now in that ad set. That's how easy this is to do this process.
I can do this for videos. I can do this for statics. Anything that you can imagine, you can upload into this.
And I know I personally never want to touch a Facebook ads library again. This is a way that you don't have to do this. Now, say I wanted to turn off this G ads questions.
I want to turn that ad off within that ad set. Again, I can just go back to clog code and say, in that ad set, turn the ad off that has this. title all right it just turned that ad off let's go check it all right you're gonna see that G ads questions is now off so this is how you can manage the entire ad account you can upload ads you can turn off ads you can move ads around everything is just you telling the agent what to do from your cloud code or your codex.
And this is what marketing engineering looks like. So this is still scraping the pain points and desired outcomes. Well, that's working in the background.
Let's do a couple other things. Now say I want to analyze this data. I want to analyze my Facebook ads data and my Google analytics data and maybe my search console data.
Say I wanted to see if branded searches are increasing. So I'm going to just say that. Look at my Facebook ads data, how many signups have happened from Facebook ads, what Facebook ads pixel is recording.
Also tell me how much traffic is coming from Google analytics for with referral as Facebook. And tell me how many branded searches are happening. Like, are we seeing an increase in branded searches?
Use the data source, Google search console for this. So again, all I'm doing is getting it to go and aggregate this data for me to answer these questions. And because I have all this data unified into a data warehouse, it can pull all.
that data from all those locations extremely quickly and respond back and this enables me to do conversational analytics with all of my live data stream that's coming from my data pipeline data warehouse it syncs on an hourly cadence so this is in real time you're getting up to speed and because this is happening in real time you can build agents on top of this that are making it decisions autonomously.
You can build dashboards on top of this that you need to share with your team. And while it's pulling that data, let's do some traditional go -to -market engineering. I'm going to find all the people that engaged with the post on LinkedIn, and then I'm going to find their emails so that we can call the outbound to them.
So let's do that now. So I'm going to LinkedIn. I'm just going to find a post that I think anybody who engaged with would be my target customer.
Cool. Claude just booked these meetings. Awesome.
Let's take this one. So I'm literally just going to take this post URL and I'm going to go back to Claude and I'm going to paste it in. And then I'm going to say, Use the graph CLI, extract all of the engagers from this post, and then use the get leads endpoint to find their emails.
And I'm going to hit go, and that's going to start running in the background. And what we're going to get back is all of the emails that we can find for all the people that engaged with that post. And while that's happening, let's do some keyword research.
So I'm going to say, use the data for SEO API. I want you to find all of the versus content for customer data platforms. I'm thinking like high touch versus X.
click house versus X, anything that would be in this stack. We're going to write some blog posts about this, but just find me all those keywords. And again, use the graph CLI for this.
All right. So it's going to do that keyword research for me. And again, because it has the data for SEO API built into that unified API, it's just going to go and use the credits that it needs to basically do these polls.
And I can interject or stop at any point. Let's go back to the static ads. Here we go.
It's now pulled that data from the data warehouse. We can see the signup data from Facebook. We can see the Facebook.
pixel records for landing page views. We can see the branded searches and search console. We can see the Google analytics for traffic.
And then I can say something like, all right, build a dashboard for me, an artifact that has all this data that we just pulled using the graph CLI. And now we're back at that LinkedIn post engagers. It found those 69 reactions and there's 316 comments that exist.
So it looks like it's probably duping those is what I'm guessing is happening. And that research that we had at using XAI for it is pulling all. of the relevant threads, the pain points, the desired outcomes, the objections, and it's building clusters around that for it to basically categorize for our ads.
Now it's making this Facebook ads funnel artifact for us. It's going to give us a URL in a second for us to go check that out. Let's do some analysis on what are all the keywords that are ranking on page two of Google that we're ranking for organically.
So let's go do that. Find all the keywords that we're ranking for organically on page two of the search results for Google. Use the Google search console data source and the graph CLI data warehouse to source this data and this information for us.
I'm basically looking for like high impact keywords or keywords that. we could potentially rank for really easily if we just write some content or designate some content to it. Have that start doing that research for me.
That CDP versus content came back. We've got results of 765. Isolate the ones that are only the same or unique search intent.
Like if they're like X versus Y or Y versus X, I don't want them. I just want the unique search intent ones. Again, getting cloud code to do that work for me.
And then once I have these, I can actually go and write the articles and publish them. You can see now that it found 398 keywords that sit on page two, positions 11 to 20 that have a thousand or a hundred plus impressions in the last 90 days. And we have that analysis.
Now we can go and say, okay. Okay, should we make some new content for any of these? Rank stack the top 10 that we should based off of where they're positioned currently.
We've got that X versus Y keywords. Let's just post one of these just for the fun of it. So what I'm about to do is I'm going to use a plugin that is already created on Graft.
It's called the SEO plugin. So if you wanted to install this plugin, all you'd have to say is install the Graft SEO plugin. Help me set it up to my CMS.
But I've already done this. So I can instead say. Use the Graph CLI, use the SEO plugin that we have installed.
And I want you to write a blog post and publish it to the Graph blog for this target keyword. So it's going to go now. It's going to research the article that it should write for this target keyword phrase by going to the front page of Google and finding what's ranking currently.
It's then going to go and write the article using Sonnet or Cloud Sonnet and then upload it to the CMS. And it will give me that URL in a second. So we're going to let that run in the back.
Again, all I did was wire this together by using the Graft SEO plugin. You can modify this plugin however you want. This is just what we have found to work and that's available to you.
All right, so we're coming back to those LinkedIn engagers extractions. We have now found all of the emails that we can. Found 44 people that don't have any emails, five emails found.
This is kind of a terrible one to be totally honest with you. We typically see about half, but what I can do is a waterfall enrichment. So I can say, hey, those 44, let's do a waterfall email enrichment now with what the cheapest provider is work down that for these other 44.
And then also let's just guess the emails of the other 89 and use mill and verifier to see if they're actually real. So what's about to happen is it's going to do a waterfall enrichment through the other service providers that are on the platform. Something like lead magic, Apollo, you're only going to pay for the credit that you use.
And then it's just going to guess the emails of the other ones. So only 32 different or 32 variations of what an email can be. And then checking the emails is so cheap that sometimes it's just cheaper to guess the person's email at this point, basically, than anything else.
All right. So pull back those data pipeline pain points. And I'm going to say something like write some scripts for AI Avatar UGC for UGC.
Have the hooks be something like, OK, this is kind of crazy or I wish I found this sooner. You know, basically something like that. and then talk to one of the pain points and the outcomes.
And again, this is an ad script that a UGC creator is going to read. Write 10 of these for me based off of the most impactful ideas. All right, so it's gonna write those scripts and I'm gonna be able to go and make an AI avatar ad or do the script read myself if that's something that I want.
All right, coming back to that dashboard, we now have made that dashboard with that data. So this artifact is now created. I have these visualizations.
I can modify this in any way that I want. Like for example, say I wanted to change this bar chart, change this to a line chart, the spend by month instead of a bar chart.
And then we'll just go and do that vibe code modification of this. This is a shareable URL your team has access to, and it's not public. So nobody else can go to it, but your team can go to it.
You can also build interactive software here. So add a date range picker to the dashboard. It's going to go and create that for me or add that date range selection back at the select LinkedIn profiles and finding those emails.
Wrote those scripts for me as well. Okay, this is kind of crazy. Every Monday I used to spend like three hours pulling scenes because I have a Shopify meta, Google ads, managing a bad spending, updating the pivot table, all just to answer one question.
Did we actually make money last week? And there you have this ad. You can do all these ad reads.
You can send these to the AI avatar to do the ad reads. Let's check back on that SEO. And these page two organics has come back.
Okay, Google ads for Facebook ads for X industry. Fascinating. Hearing aids, that's a great actually angle for us to go after.
We'll probably do that. I can add those to the queue. So I'm going to say that now.
Find all the other keywords that are similar to this one. Use the data for SEO API and the graph CLI. Cool.
I'm still adding that date picker to the artifact that we created. Found all the keyword variations that matched what we were going for. It's pulling those leads from Apollo.
And that URL post got published based off of what was ranking currently and the research that it did. And I'm going to end it there, guys. If you want to get started with graph .com, go to...
graph .com forward slash platform. Just click the link in the description wherever you're watching this. You can get started for free for as small as 10 bucks.
You can do a ton of work. Almost everything that I just showed you today, you can do. And if you want more content like this, subscribe to my YouTube kit.
If you want more content like this, subscribe to my YouTube channel, Cody Schneider. We'll be dropping marketing engineering content like this every week. If I don't drop a video, my team gets paid $200 for the all hands off site.
I've been slacking, so hold me accountable. And I talk about everything that we are learning in real time on both Twitter and on LinkedIn. Go follow me there to learn all of this as we learn it.
And if you want to be notified when every video like this drops, go to gtmengineeringcourse .com. Every video that I make, I'm building into a free course that is sponsored by Grafted. It is entirely free.
I teach you everything that I know about marketing, about GTM engineering, about marketing engineering. Go sign up for free. All right, that's it.
We'll see you next time.
The Hook
The bait, then the rug-pull.
The video opens with a direct promise: in under ten minutes, any coding agent can be taught to take over the manual middle-work of marketing, and the next twenty-nine minutes are spent proving it live.
Named ideas worth stealing.
02:42list
The ten-piece marketing engineering stack
- Data pipeline
- Data warehouse
- Cloud server
- Media storage
- Databases for agents
- Recurring tasks (cron jobs)
- Artifact hosting
- Application authentication
- Git manager
- Tailored API gateway
The full list of infrastructure pieces the presenter says a coding agent needs before it can reliably run marketing operations, shown on a whiteboard slide and repeated throughout the first half of the video.
Steal forscoping what a self-built AI marketing-ops stack actually requires before promising it to a client or team
05:44concept
Read from warehouse, write to platform
The agent should only read analysis data from the unified data warehouse, and only use direct platform API calls to write actions back out (uploading media, toggling ads), never to pull bulk data, to avoid rate limits and hallucinated joins.
Steal forany agent integration that touches a rate-limited third-party API
How they asked for the click.
VERBAL ASK
28:10product
“If you want to get started with graph.com, go to graph.com forward slash platform... you can get started for free for as small as 10 bucks.”
Soft product pitch woven in at the top as a sponsor-style mention, then repeated at the close alongside a free email-gated course signup and a channel subscribe ask.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
