44:47Greg Isenberg · Interview
Most Valuable Skill of 2026: Managing AI Agents
Ryan Carson explains why running a one-person AI startup in 2026 means becoming a full-time manager of cloud coding agents, not a part-time coder.
July 24thThe argument in one line.
As AI makes average marketing unbelievably cheap, the valuable role shifts to the marketing engineer, someone who uses AI agents, data, code, and taste to build a marketing system that keeps learning instead of running one-off AI chats that forget everything by the next week.
Read if. Skip if.
READ IF YOU ARE…
- A marketer, growth person, or RevOps person who wants a concrete way to become the person a company can't run without.
- A solo founder who doesn't want to hire a marketing team but wants AI agents doing research, content, and outbound for them.
- Someone weighing consulting, a productized service, or software as a business built on AI marketing systems and wants the sequencing.
SKIP IF…
- You want a plug-and-play tool or template rather than a folder structure and set of agent specs you build yourself.
- You're after platform-specific ad copy or growth hacks rather than a system-level way of organizing marketing around AI agents.
The full version, fast.
Every shift in marketing technology has created a new kind of marketer: story-driven marketers in the print era, channel-and-funnel marketers once the internet made acquisition measurable, and loop-driven growth hackers once software made products track themselves. Greg Isenberg argues AI agents are creating the next one, the marketing engineer, who uses agents, data, code, and taste to build a marketing system that keeps learning rather than starting over in a new chat every week. The core move is a growth repo, a structured folder holding customer truth, founder voice, outbound targeting, and creative results, that agents read from and write corrections back into. He maps six systems built on that repo, a rough tool stack, four ways to monetize the skill, and a 30-day plan to learn it from scratch.
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 – 01:46
01 · Intro
Greg names the 'marketing engineer' role, argues it becomes a $250K-$1M+ job because every company wants more leads and faster experiments from a smaller team, and previews the episode's roadmap.
01:46 – 04:29
02 · The Evolution of Marketing
Frames four eras, Traditional (story, psychology), Digital (SEO, ads, funnels), Growth (product loops, retention), Engineering (agents, data, code, taste), arguing each shift created a new most-valuable marketer and taste now matters more because AI makes average marketing cheap.
04:29 – 07:19
03 · What is a marketing engineer
Diagnoses why company knowledge stays scattered (sales, support, product, and the founder all hold a different version of the customer), then defines the role: turning market signal into pipeline using AI agents, data, code, and taste.
07:19 – 10:18
04 · Build the Growth OS
Introduces the growth repo as the fix for AI work disappearing in random chats, laid out as customer-truth, content-engine, outbound-engine, creative-testing, and agents folders, and shows how a context-rich prompt beats a generic one.
10:18 – 13:23
05 · Marketing Engineer Tool stack
Maps tools to layers: Grokbot as the internet-connected growth radar, Claude/Codex for building the repo and internal tools, Hermes for scheduled workflows with memory and approval, creative models (FAL AI, Higgsfield) for ad assets, and local AI for sensitive data.
13:23 – 14:32
06 · Live Data Workflow
Walks an SEO example end to end: pull Google Search Console and Ahrefs/SEMrush data, check the CMS, rank opportunities by buyer intent, draft the post, and route it to an approval step, the shape every agent job should follow.
14:32 – 16:56
07 · Agent Job Description
Lays out how to write an agent spec like a hire's job description (data source, schedule, filters, output, approval, metric), with a competitor-engager example and an outbound signal-engine example, and how corrections compound into the repo.
16:56 – 18:27
08 · Example: vertical SaaS for HVAC contractors
Grounds the framework in a deliberately unglamorous example, commercial HVAC software, arguing the real question is which specific pain (missed follow-up quotes after a service call) actually gets an owner to book a demo.
18:27 – 20:20
09 · System 1: Customer Truth
Defines System 1 as a living market_truth.md file pulling sales calls, support, churn, CRM, and social data into one place, with the standard being a specific receipt-backed insight, not a vague summary.
20:20 – 23:31
10 · System 2 - 4: Founder content, Outbound signal and Creative Testing
System 2 turns one insight into five assets (post, video, landing page line, cold email, calculator); System 3 times outbound to real signals like funding or hiring instead of a cold list; System 4 treats ad creative as a constant testing loop.
23:31 – 24:19
11 · System 5: AI search visibility and the growth cockpit
System 5 reframes SEO around being the source ChatGPT and other AI assistants cite when buyers ask questions directly; introduces System 6, a weekly cockpit view of what changed and what to test next.
24:19 – 25:06
12 · System 6: Eval Loop
Describes the eval loop that keeps agents honest, checking founder voice, customer words, receipts, right buyer, approval, and whether it actually created pipeline, arguing judgment is the moat once agent output becomes commodity-cheap.
25:06 – 29:41
13 · Ways to Monetize
Lists four monetization paths, in-house hire, consulting embed, productized service around one wedge, and software, and recommends sequencing services first to find the pain that repeats before building software.
29:41 – 32:24
14 · The 30-day plan
A four-week plan to learn the role from scratch: Week 1 audit one company, Week 2 build the growth repo, Week 3 ship one working system, Week 4 measure and document the result as a case study.
32:24 – 35:18
15 · Closing Thoughts
Closes on the marketing engineer as part marketer, part product person, part RevOps, part data analyst, part creator, part engineer, with judgment as the durable moat once agents are commoditized, then a low-key like/subscribe ask.
Lines worth screenshotting.
- Every major shift in marketing technology has replaced the previous era's most valuable marketer: story-driven marketers gave way to channel-and-funnel digital marketers, who gave way to loop-driven growth hackers.
- AI is about to make average marketing unbelievably cheap, which means taste and judgment matter more now, not less.
- A marketing engineer is defined as the person who turns market signal into pipeline using AI agents, data, code, and taste.
- Most companies already have customer signal scattered across sales calls, support tickets, product usage, and the founder's memory, and nobody combines it into one version of reality.
- The first thing to build as a marketing engineer isn't a campaign, it's a growth repo, a structured folder that becomes the company's marketing memory so agents stop starting from scratch every week.
- A growth repo needs five folders at minimum: customer truth, content engine, outbound engine, creative testing, and agents.
- Prompting an agent that has read the customer-truth file, the founder-voice file, and last week's top-performing posts produces categorically better output than asking for '10 LinkedIn posts.'
- Every AI agent needs a written job spec like a hire's: the data source, when it runs, what it filters, the expected output, what needs human approval, and the metric that defines success.
- Messages sent is activity; qualified replies are signal, so an outbound agent should be judged on replies and booked meetings, not send volume.
- Agents get trained the same way new hires do: start with small tasks, correct mistakes, add the correction to memory, then expand scope.
- A sharp customer-truth memo names the specific behavior that predicts conversion, for example deals that closed all mentioned missed follow-up quotes, not a vague line like 'customers want better collaboration.'
- One customer insight can become five different distribution assets at once: a founder post, a short video, a landing page line, a cold email angle, and a calculator.
- Good outbound starts with timing signals, who raised money, who's hiring for the exact problem, who posted about the pain, not a cold list of names that merely fits an ICP.
- With a billion-plus people asking ChatGPT questions instead of Googling them, being the source an AI assistant cites is becoming its own marketing surface.
- There are four ways to monetize marketing-engineer skills: become the in-house hire ($250K-$500K), consult ($5K-$30K/month per engagement), sell a productized service around one repeatable wedge, or eventually build software.
- The recommended path is services first: build the same system for five or ten clients, notice what repeats, then turn that repeated pattern into software.
- The fastest way to prove the concept is deliberately simple: build one growth-repo folder, paste in 20 real customer notes, and ask the agent for one finding plus one test, this week.
AI agents didn't kill marketing skill, they created a new one.
WHAT TO LEARN
The valuable marketer now is the one who builds a system, a growth repo the agents read from and write corrections back into, rather than the one who runs the sharpest one-off AI prompt.
01Intro
- The marketing engineer role doesn't have a fixed name yet, forward deployed marketer and AI growth operator are used interchangeably, but the underlying job is stable: doing the work of a full marketing team with AI agents.
- The pitch for why this role pays well: every company wants more leads, faster experiments, and a marketing operation that gets smarter every week with a smaller team, and whoever can build that gets to set their price.
02The Evolution of Marketing
- Each marketing era replaced the previous one's most valuable skill set: the print-era 'Don Draper' marketer who sold through story and psychology, the digital marketer who mastered measurable channels like SEO and Facebook ads, and the growth hacker who used activation and referral loops.
- The marketing engineer still needs the old skills, customer understanding, positioning, taste, but adds building the system behind the marketing: AI, agents, data, and code, wired to keep learning.
- Taste matters more, not less, in this era, because AI is about to make average, undifferentiated marketing extremely cheap to produce.
03What is a marketing engineer
- Company knowledge is scattered by design: sales hears one version of the customer, support hears another, product sees usage data, and marketing sees clicks, so every growth meeting starts from a different version of reality.
- Formal definition: a marketing engineer is the person who turns market signal into pipeline using AI agents, data, code, and taste.
- The founder-level question to ask is who on the team is actually building the growth system, not just running campaigns.
04Build the Growth OS
- The first thing to build is a growth repo (or 'Growth OS'), a structured folder or GitHub repo that becomes a company's durable marketing memory instead of disposable one-off AI chats.
- Minimum folder structure: customer-truth, content-engine, outbound-engine, creative-testing, and agents, each holding the raw material AI needs to write with real context instead of generic output.
- A prompt referencing the customer-truth file, the founder-voice file, and last week's top-performing posts produces categorically sharper output than a bare 'write me 10 LinkedIn posts' request.
05Marketing Engineer Tool stack
- Grok (Grokbot) fits best as the 'growth radar' layer because it's wired into the X ecosystem, useful for watching competitors, customer language on X/Reddit, and creator formats worth testing.
- Claude and Codex fit the 'build' layer, generating landing pages, scripts, and small internal tools that turn one-off work into something durable and repeatable.
- Hermes-style scheduled workflows suit recurring operator work (a Monday market brief, a Friday experiment review), and local AI is reserved for sensitive data, private transcripts, regulated notes, and pricing plans a company wouldn't send to a cloud tool.
06Live Data Workflow
- A real agent job (the SEO example) pulls from Google Search Console, Ahrefs/SEMrush, and the CMS, ranks opportunities by buyer intent and search volume, and researches what's already ranking before drafting anything.
- Every agent needs inputs, a job, an approval point, and a place to write back what worked, that write-back loop is what separates a system from a one-off automation.
07Agent Job Description
- Every AI agent needs a written job spec like a new hire's: the data source, when it runs, what it filters out, the expected output, what needs human approval, and the metric that defines success.
- Example spec: a competitor-engager agent that checks 20 LinkedIn accounts every weekday morning, pulls people who commented, enriches and filters bad-fit leads, and drafts 10 messages tied to the specific post they engaged with.
- Agents get trained like new hires: start with small tasks, correct mistakes, feed the correction back into memory, and expand scope as trust builds.
08Example: vertical SaaS for HVAC contractors
- Using a deliberately unglamorous example (commercial HVAC software) to make the framework concrete: the real marketing question isn't 'we need more content,' it's which specific pain actually gets an owner to book a demo.
- A sharp, specific angle like 'stop losing replacement revenue after every service call' beats a generic promise like 'run your HVAC business better.'
09System 1: Customer Truth
- System 1 (Customer Truth) pulls sales calls, support tickets, churn notes, CRM data, and social mentions into one living file, described as market_truth.md, that updates daily or weekly.
- The bar for a good insight is a receipt, not a vague summary: 'five sales calls this week mentioned emergency dispatch, but the calls that converted all mentioned missed follow-up quotes' beats 'customers want better collaboration.'
10System 2 - 4: Founder content, Outbound signal and Creative Testing
- System 2 (Founder Content) turns one customer insight into five distribution assets at once: a founder post, a short video, a landing page line, a cold email angle, and a calculator.
- System 3 (Outbound Signal) argues good outbound starts with timing, not a spreadsheet of names, watching who raised money, who's hiring for the exact problem, or who posted publicly about the pain, then having an agent draft angles for human approval.
- System 4 (Creative Testing) treats ad creative as a compounding, always-on testing loop, not a one-time treadmill: spin up dozens of hooks and angles per offer and keep the ones that win.
11System 5: AI search visibility and the growth cockpit
- System 5 (AI Search Visibility) reframes SEO around being the source an AI assistant cites when a billion-plus ChatGPT users ask questions directly instead of googling them.
- System 6 (Growth Cockpit) is a weekly digest of what changed and what to test next: which content worked, which objection resurfaced, what percentage of tests won, and what competitors moved.
12System 6: Eval Loop
- The eval loop keeps agent output honest: check that it's in the founder's voice, cites real customer words, includes receipts, targets the right buyer, and actually creates pipeline, not just cheap volume.
- AI makes output cheap; judgment about what's worth shipping is what stays valuable, and that judgment is the actual moat as agents become commoditized.
13Ways to Monetize
- Four ways to monetize the skill: become the in-house hire ($250K-$500K), consult by embedding with a company for 30-90 days ($5K-$30K/month), sell a productized service around one repeatable wedge, or eventually build software.
- The recommended sequencing is services first: do the work by hand for five to ten companies, notice which pain repeats across all of them, and only then turn the repeated pattern into software, which is also how you avoid building something nobody wants.
- A tight wedge, like 'outbound signal engines for vertical SaaS' or 'customer truth repos for seed-stage startups,' is easier to sell, deliver, and repeat than general-purpose marketing consulting.
14The 30-day plan
- Week 1 is an audit of one real company: study the site, offer, ICP, and any existing content, then output a market map of who buys, what words they use, and where the funnel leaks.
- Week 2 is building the growth repo itself and shipping the first market_truth-style file with real receipts.
- Week 3 is shipping exactly one working system end to end, not five half-built ones, and Week 4 is proving the result: did replies improve, did meetings get booked, did conversion lift.
15Closing Thoughts
- The best marketing engineers end up part marketer, part product person, part RevOps, part data analyst, part creator, and part engineer, able to talk to a customer and also wire the automation.
- As agents become a commodity, the differentiator becomes judgment: knowing what to point the agents at in the first place.
Terms worth knowing.
- Marketing engineer
- The person who uses AI agents, data, code, and taste to turn market signal into pipeline, building the system behind the marketing rather than just running individual campaigns.
- Growth repo (Growth OS)
- A structured folder or GitHub repo holding a company's marketing memory, customer truth, founder voice, outbound targeting, and creative results, that AI agents read from and write corrections back into.
- market_truth.md
- A living markdown file, updated daily or weekly, that summarizes what customer signal (sales calls, support tickets, churn, reviews) has changed, backed with quotes and links as receipts.
- ICP
- Ideal Customer Profile, the specific type of buyer a company's outbound targeting and messaging is built around.
- Productized service
- A single, repeatable offer built around one narrow niche and problem, priced and delivered the same way for every client, instead of custom consulting.
- Forward deployed marketer / AI growth operator
- Alternate names floated for the same emerging role the video calls 'marketing engineer' — the label is unsettled, but the underlying job is described as the same.
Things they pointed at.
10:38toolGrok (Grokbot)
11:45toolClaude / Codex
12:11toolHermes
12:30toolFAL AI
12:30toolHiggsfield
13:41toolGoogle Search Console
13:41toolAhrefs / SEMrush
Lines you could clip.
00:28
“Whoever can walk in and just build that with AI agents are going to get to name their price.”
02:15
“I think about it as the Don Draper era, where marketing was about making people care through the story, through the psychology.”
15:26
“Messages sent is activity. Qualified replies is going to be your signal.”
18:17
“That's interesting just because it's really specific. So when you have something specific, it's just interesting. My bunny ears go up.”
22:00
“You're selling painkillers, not vitamins, when timing hurts.”
33:01
“The agents are going to be a commodity at some point. Your judgment about what to point them to is the moat.”
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
I think one of the most valuable people in tech over the next 18 to 24 months is going to be something called a marketing engineer. Now, some people call it a forward deployed marketer and some other people are calling it an AI growth operator. I'm saying call it whatever you want.
The name is probably going to change, but the job won't. It's the person who can do a whole marketing teamwork with AI agents. And I think there's going to be a ton of money to be made in it.
I actually think this becomes a 250K, 500K, a million dollar job because every company wants more leads. They want faster experiments. They want sharper positioning and they want to read on their customers and they want their just marketing to get smarter every week with a smaller team than a bigger team.
Whoever can walk in and just build that with AI agents are going to get to name their price. So if you're a marketer, this is how you become way more valuable. If you're a founder, you know this.
You don't just want to vibe code something. You want people using your product. So you're going to have a huge edge if you can use AI agents to do your marketing for you.
By the end of this episode, you're going to know what a marketing engineer actually does. What do they build? What the tool stack looks like?
How to use things like... Grokbot and Claude and Codex and Hermes and Creative Models, how they all play together within the context of a marketing engineer and the exact 30 -day plan I'd follow to learn from scratch if marketing engineering is interesting to you. Let's get into the episode.
I can't wait to see what you build.
So something I... can't stop thinking about is marketing keeps changing. And having been a part of multiple cycles, I've started and sold three venture -backed companies.
One was in the web era, one was in the social era, one was in the mobile era. Every time the technology changes, the most valuable kind of marketer changes with it. So think about the traditional era of marketing.
I actually think about it as like... the Don Draper era, where marketing was about making people care through the story, through the psychology, getting your product in front of people on whatever channels existed at the time, things like traditional print media and radio. The best marketers at that time understood what people wanted, what they were insecure about, who they were trying to become, and how to package a product so the market paid attention.
Obviously, that skill matters a lot. But then the internet showed up and it created the digital marketer. So it evolved from traditional to digital.
Suddenly you had websites, email, SEO, Google. In 2005, I think, 2006, you had Facebook ads, landing pages, pixels. The best marketer became the person who could acquire customers through channels you could actually measure.
And a lot of people didn't know these were new channels. So the best marketers understood funnels, targeting these new channels, analytics, things like Google Analytics, and the very practical question about what happens after someone clicks. Then software created loops.
And then growth hacking became a thing around, if I remember correctly, 2008, 9, 10, 11. The best growth hacker people were all about activation, referrals, onboarding, retention, pricing. There was a guy by the name of Dave McClure who had this, I think it was called the R framework, activation and referral.
That was the, you know, the marquee, the symbol of the time of the growth hacker era. Basically, marketing moved closer to product because the product itself could become the growth engine. Now we're walking into the marketing engineering era.
And I feel like not a lot of people have spoken about this. That's why I want this to be the de facto episode about this whole era. The marketing engineer still needs all that old stuff.
It still needs customer understanding, judgment, positioning, understanding distribution, taste. If anything, taste, people talk about this all the time, but taste matters more now than ever because AI is about to make average marketing just unbelievably cheap. The new part is that the marketing engineer also builds the system behind the marketing.
The way I think about it is, traditional marketing was about, you know, Making people care, digital marketing was acquiring customers through measurable new channels. Growth hacking was about using product and data to build these loops.
And marketing engineering is about using AI, agents, data, code, and taste to build a marketing system that keeps learning. And the last phrase is an important one, because a marketing system that keeps learning is now actually possible in the agentic era. Now, most companies already have pieces of this lying around to their credit.
So they've got tools and dashboards, calls, content calendars, CRM, some SaaS tools. The problem is the learning is pretty scattered. Sales might hear one version of the market, support hears another.
Product sees the usage and marketing sees what got clicks. And the founder remembers the one customer call that just hit him emotionally that week and just can't get that one customer call out of his or her head. I know that happens to me.
Then everyone walks into the growth meeting with a slightly different version of reality. So the marketing engineer's whole job is actually to pull in these signals into one system and turn them into growth. So the way I define the role is this.
A marketing engineer is the person who turns market signal into pipeline using AI agents, data, code, taste. And that's really the job. And I'm going to get super tactical on how you can actually do this soon.
If I were a founder right now, the question I'd be asking myself is, who on my team would be building the growth system for this company? Now, I am a founder myself, so a lot of time I'm doing this myself. And I just hope that if you're a founder listening here, either you hire someone or you do it yourself.
Because the companies that are going to win in this agentic era are going to learn the market faster than anyone else. So it's crucial to know. So if you see the customer pain earlier, you spot the winning language earlier, you're testing more angles using Facebook ads, shipping more surfaces, lead magnets, and understand what's working before the competitor even notices things, you have this unfair advantage.
So the question I get asked a lot is, okay, but what is the first thing I would build? Okay, I want to become a marketing engineer. I want to do more marketing engineering.
What do I build first? And the first thing I would build is a growth repo. Yeah, I know it sounds a little bit nerdy, but even if you're non -technical, I believe you can do it.
So you're going to want to go and create a GitHub repo, or honestly just a structured folder. You can call it something like Growth OS. And it becomes a place where the company's marketing memory is going to live.
The problem it's going to be solving is that most people use AI in these random chats. So they'll open up a chat GPT or a cloud, Gemini, they'll ask for 10 posts, and maybe they'll copy one into a doc that they like, and then the work just disappears. Next week, the AI is starting from scratch again.
When what it really needed was the performance data and the founder's voice and the objection from the sales calls and the language that actually created replies. So the growth repo is going to fix that. And inside it, what we're going to have is a customer truth folder.
And that's going to have our sales calls notes or support tickets, maybe some churn notes, interviews, even live product feedback can go in there. So you've got a content engine folder with the founder voice guide, with the winning hooks and the scripts and notes on what's performed before. You've got an outbound engine folder with the ICP, your ideal customer profile, the account research, the trigger events, maybe some improved angles could be good to have there.
Even actually band language is good to have as well. Because AI outbound gets weird fast. If you let it talk like an overexcited SDR who just discovered personalization, sometimes bad things could happen.
So you've got a creative testing folder for ad angles and things like landing page tests and hooks and offers and results. And you've got an agents folder where you define the jobs your AI workers do. And that repo, is the difference between, hey, AI helped me make a thing, and AI is helping the whole company get smarter.
That's how a growth or a marketing engineer is thinking about it. And then the prompt gets way better. So instead of, hey, write me 10 LinkedIn posts, you say, read the customer truth file, read the founder voice file, read the last five posts that drove qualified replies, and draft five new posts around Payne's buyers that were actually mentioned this week.
So it's a totally different level of output because the agent is now having real context. What tools do I need if I want to become a marketing engineer? Well, I'll tell you some of the most important ones and how to think about where all the tools fit and your tool stack.
So Grokbot is new, but it's just an incredible product. So I think of Grokbot as... the growth operating system that lives close to the internet.
So marketing is a living system, the marketing is moving, competitors are moving, culture is changing, customers are changing their language, creators are picking up new formats. Grokbot is especially useful in that world because it is connected to the X ecosystem. If I were setting this up as a founder, I'd give it a few clear lanes.
So I'd say one bot watches competitors and tells me what changed. One is going to watch customer language across X and Reddit. One watches the creators in the niche and finds formats worth testing.
And one watches ads and landing pages. Basically, wherever there's a connection to the internet, Grokbot is going to be extra good there. That doesn't mean you can't use Grokbot to do everything.
You totally can. I'm one of those people that basically pick an ecosystem that you like, that you feel comfortable with. If Grokbot feels good for you, just do everything in there as well.
The way I think about it, this is just the way I'm thinking about it. So I hope it gets the creative juices flowing. For me, I use Claude and Codex and products like that in a different part of the system.
So they're going to help me build the repo and generate the landing pages, writing scripts, building the little internal tools that I was talking about, and then turn that repeatable work into something durable. I've talked on this channel about Hermes before. Hermes -style workflows are still extremely valuable, especially when you want scheduled operations with memory and approval.
So something like every Monday morning, build me a market brief, or every Friday afternoon, review the experiments. Every time a fresh batch of sales calls land, maybe put it in a folder and then pull the objections and update the positioning file. Then you have creative models.
They're going to help you move faster on ads, thumbnails, mock -ups, and video concepts. There's a bunch of those that exist. There's FAL AI, there's Higgs field, there's a bunch of them.
And local AI matters when the data is particularly sensitive or there's private customer transcripts or regulated notes, pricing plans. Basically anything a company would feel weird sending into a cloud tool. Also things that are too expensive to do into a cloud tool.
I'm going to do a whole separate episode on local AI. So stay tuned for that over the next one or two weeks and subscribe. So that comes into your feed.
The tools are going to keep changing, but the workflow is the thing to actually learn. So where it gets really interesting is when the agent connects to live business data and the tools, obviously, to actually do the work. So take SEO content.
The beginner version is asking an AI to write a blog post about a keyword. So a marketing engineer isn't going to do that. A marketing engineer is going to check Google Search Console, pulling keyword data from Ahrefs or SEMrush, look inside this CMS to see if it already exists.
It's going to rank opportunities by volume and by buyer intent. And it's going to research what's already ranking. and hopefully adding the founder's point of view and draft the post, write the meta title, suggest internal links, and just send the whole thing for approval.
That's a pretty big jump, but the agent has a job, the job has inputs, and the inputs come from the business and the output goes somewhere useful. So every agent is going to need a real job spec. And I'd write it out almost like I was hiring a person.
Here's the data source. Here's when you run it. Here's what you filter out.
Here's the output I expect. Here's what good looks like. Here's what's going to need human approval.
Here's the metric that matters. And here's what you write the result so the system gets smarter next time. So for maybe a competitor, engager, agent, that might be every weekday morning.
Check these 20 LinkedIn accounts and pull the people who commented on new posts, enrich them, drop the bad fit leads, and draft 10 messages tied to specific posts they engaged with. Oh, and then obviously write the results to a file for approval. The metric is going to be positive replies from qualified accounts because a marketing engineer cares about business results, right?
Not activity accounts. Messages sent is activity. Qualified replies is going to be your signal.
And the whole point the marketing engineer is trying to do is to generate pipeline demand. And you train these agents the same way you train a new hire. You start with small tasks, you watch it work, you correct the mistakes, you add the correction to memory, because now we have memory, and then you expand the scope as you increase your comfort level.
If the outbound agent writes a first line that sounds... like fake, for example, you've got to add the rule to the repo. And if the content agent keeps writing these generic intros that sound like generic AI, you give it three good examples and three bad ones.
If the customer truth agent makes a claim with no evidence, we've got a problem here. You've got to add the rule to... that every insight needs a quote or a link or a source.
Every correction becomes part of this operating system, this growth marketing engineering system. And that's how this whole thing compounds. And going back to how does a marketing engineer make a million dollars a year or $500 ,000 a year or $1 .5 million a year for their own startup, it's because they're building this and it's so darn valuable.
Let's actually get into a concrete example so that this gets solidified into your head. Imagine a vertical SaaS startup selling software to commercial HVAC contractors. These are companies managing technicians and service calls and maintenance contracts and dispatch.
It's a real B2B market.
a lot of money, the workflows are messy, and the language is specific, which is why I wanted to use this example. The marketing problem for that company is usually a little bit more sharper than, hey, we need some more content. The real problem that they're facing is usually something like, which pain gets the owner to take a demo?
Maybe it's dispatch chaos or maybe it's late invoices. Maybe it's that the owner has no idea which jobs were profitable until the month is over. Or maybe it's actually that the technician finishes a service call, spots a replacement opportunity, and the follow -up quote just never gets sent.
That's interesting just because it's really specific. So when you have something specific, it's just interesting. .
Stop losing replacement revenue after every service call is obviously a much sharper angle than run your HVAC business better. So this is where your marketing engineer is going to earn their keep. It's going to start with the customer truth system.
That first system is going to be the customer truth system. And every startup says they understand the customer and you talk to five people. And they get five different results.
We talked about that. But the marketing engineer is going to pull those signals into one place. The output is a file called whatthemarketistellingus .md.
I tweeted about this idea. It went viral. I'm glad people liked it.
It's basically a markdown file which updates every morning or every week, depending on how much signal the company is going to have. And it's reading the sales calls and support tickets and churn notes. Even Stripe movement, CRM notes is a good one.
And also social data, especially if it's more consumery. And its whole job is to show what's changed. So maybe the buyers are using a different phrase than they were using a month ago.
Or maybe the trial users keep getting stuck before they invite a teammate. You're just going to get some insight. And you're going to ask the agent to show, quote, snippets.
ticket links, event counts. What you don't want is obviously a vague summary, which I've seen a lot of people do this. They just get these summaries and it's pretty vague.
In this case, you'd get something like, customers want better collaboration. You want something way more sharp than that. I want the thing that's going to make the business harder to lie to.
So for the HVAC company, A good memo might be something like, five sales calls this week mentioned emergency dispatch. But the calls that actually converted all talked about missed follow -up quotes after the tech left.
Just a lot sharper. The second system is the founder content engine. So a lot of companies have great raw material, like the founder has opinions.
You've got customer stories. But you're not really capturing all the stuff. So the marketing engineer could build the loop.
So you can record founder talking to customers. You can pull from podcasts, extract the strongest ideas, and then have the system watch what performs, which hooks people keep watching because you have this data. And then you create content out of that.
For the HVAC company, Imagine something like the loss replacement revenue insight becoming five things. A founder post about the hidden revenue leak in service businesses, a short video on why contractors lose money after the first visit, a landing page line that says every completed job should create the next quote, a cold email angle could be good, and a simple calculator that estimates the lost revenue.
So the system here is learning. The third system is the outbound signal engine. So bad outbound usually starts with a spreadsheet full of names.
But good outbound starts with timing. So who just raised money? Who's hiring for the exact problem you solve?
Who posted publicly about a pain point? And then who fits your ICP and has a real reason to care this week? These people are...
They've got the pain. You're selling , when timing hurts. Having an agent watching those signals, researching accounts, drafting specific angles, and sending to human for approvals, that's the type of thing that for the HVAC company would be awesome.
Watching for contractors, hiring dispatchers, or opening new locations, getting bad reviews, and then having the agent actually go and reach out outbound. is going to be huge. The fourth is the creative testing engine.
Taking one offer and spinning up 20 hooks, 10 ad angles, recording the results and testing them. A lot of people say, Facebook ads don't work for me.
Yeah, maybe. Or maybe you're just not testing enough creative with the right angle. A good marketing engineer could create thousands of pieces of creative based on your positioning.
It's basically like having creative becoming this learning system, not really a treadmill that you actually have to do. You're going to have it on repeat, having these agents go and create creative based on just how the world is changing and how that data is changing too. Again, another huge insight around, wow, this is a new way of doing marketing, marketing engineer.
The fifth system is AI search visibilities.
There's a billion plus people using ChatGPT asking, just ChatGPT, I'm not talking about Google AI overviews or Gemini or Perplexity or Cloud. I just saw Sam Altman said they have a billion users, it's insane. So you've got to think about whether your company is even understandable to those systems.
And then having agents actually go pull in that data and actually create content and optimize your website such that you're ranking high there. Getting cited by AI is such a huge opportunity and something that a marketing engineer is thinking about, of course. The sixth system is the growth cockpit.
This is like a weekly view that tells the team what has changed and what to do about it. What content has worked? What campaign created real conversations?
Which objection came up again? What test won? How many tests won?
What percentage of tests won? What competitors moved? Well, customer pain is getting louder and what to test next.
For the HVAC company, the cockpit might say something like, hey, this week the lost replacement revenue angle drove fewer clicks than the dispatch angle. But twice as many demo requests from owners with more than 20 tech. So that's the kind of memo that if you're an executive, you want to wake up to that.
And that's super, super valuable. So if you've gotten this far, what are some ideas on how you actually can make money with marketing engineering? And I think there's a few ways that you can do it.
The first is becoming the person inside the company. So if you're already a marketer or a RevOps person, a growth person, a creator, or just honestly a curious, marketing -minded person, this is one of the clearest ways to become way more valuable because this work sits directly next to revenue.
All the ideas that we talked about, all the systems that we talked about is things around creating pipeline, lifting conversion, cutting wasted spend is huge with... things like these marketing agents. And you've got this direct line to business value and that's how someone becomes a $500 ,000 hire.
Because they look at it and they're like, well, if I'm going to save $2 million and I'm going to increase revenue this much and I'm going to double the conversion rate, that's a huge, huge, it's a win -win situation. So why in the original I said I think that there's going to be people make a million dollars doing this.
And I actually think that's conservative. I think there will be versions of that, of the best marketing engineers making millions of dollars a year is because they're going to be... just driving insane amounts of value in the same way that forward deploy engineers are driving insane amounts of value for companies right now.
The second way is just do consulting. So you create an offer, you embed with a founder -led company, maybe it's 30, 60, 90 days. You build one growth system and then you sell the outcome.
So hey, we'll build your customer true system and turn it into a weekly. Or we'll build your founder content engine. Or we'll build your outbound signal engine.
Some of these ideas that we talked about, you just embed yourself, you build it, and you charge $5, $10, $30 ,000 a month depending on what you're actually building. The third, somewhat less talked about, is productized services. So you can pick one wedge and then repeat it.
So for example, outbound signal engines for vertical SaaS. That's what you focus on. Or founder content engine for B2B CEOs.
Customer truth repos for seed stage startups before they hire a full marketing team. So the tighter the wedge, the easier it is to sell, deliver, and repeat. And that's the only thing that you focus on.
That's why it's called productized services. Because it's not like you're doing services, custom things for everyone. There's this one thing you do for this one niche and you charge X amount of dollars for it.
The fourth is software. I think the biggest outcomes are going to come from this. But I do think that I would start with services first.
So you do the work by hand. You build the same system for five companies, ten companies. And you notice the pain that repeats.
And then that's when you turn it into software. And that's also how you avoid building something that nobody wants. The fun part is all these ideas actually stack together.
You can start by consulting to learn what actually works. You can notice the same system every client needs. You productize it.
You eventually turn it into software like a set of agents. If I were doing this tomorrow morning, I would keep the first version almost painfully simple. I would build that growth OS folder.
I'd have five of those files, customer truth, founder voice, experiments, agent jobs. And then I would paste 20 real customer notes or call summaries. And then I would ask the agent to do one job.
I'd say, tell me what's changed, show me the receipts, suggest one marketing test that could create pipeline. This week, not next week, not a month from now. And then built one thing from that output.
For that HVAC company I was talking about, maybe it's the loss replacement revenue calculator or something like that. The first goal is just to prove the system can turn this messy market data into one useful action. So if you listen to this and you're like, wow, being a marketing engineer sounds really cool.
I want to go... hone my skills in the next 30 days to become a marketing engineer, be it as an employee, as a founder, whatever it is, here's the plan that I would run. Week one, I would do an audit.
So I'd pick one real company. It could be yours, a friend's, whatever you can get access to. I would study the website, the offer, the ICP, the founder's content if there is any.
Oh, sales calls and support tickets if you can get them, obviously. And then you output a market map. Who's the customer?
What pain do they describe? What words do they use? And what would you test first?
What are they buying instead of your product or this product? Where's the funnel leak? And what would you test first?
So week one is just studying all that stuff. Week two is the growth repo. So create it, add the folders, and build your first what is the market telling us markdown file.
You can use whatever tools you like. It could be Claude, ChatGPT, Grokbot, Gemini, local models, whatever it is. The tools actually matter less than the workflow here.
The goal is basically just to turn that scattered signal into the memo with real receipts and actually just start feeling like a true marketing engineer. Week three is your first machine. So you can pick one system and actually build it.
It could be the content engine, the outbound signal engine, a landing page tester. Obviously this is going to vary depending on what the company needs and wants. But just pick one because you're going to get better outcome with one.
And one working system is going to beat five half -built ones. And then week four is just all about results. What changed?
Okay, you did this thing. Did replies improve? Did meetings get booked?
Did any conversion lift? Did the founder sound sharper? Did the founder like the post?
At the end of your month... You should have a case study that sounds something like, I audited this company's growth, we built this customer truth repo, I found it was one high intent pain that they didn't know about, and I turned it into an outbound signal engine which shipped 75 targeted messages, got nine warm replies, booked three calls, and I documented everything what I learned.
And then you're showing a real... business result, tangible value. And that's how you get hired, that's how you get clients, and that's how you become credible.
I think the best marketing engineers are going to feel like part marketer, part product person, part rev ops, part data analyst, part creator, and part engineer. So they can talk to a customer, they can build the workflow that uses that insight. They can write the positioning, they can wire the automation, and they can make the landing page, and they can read the conversion, and they can set the outbound agent, and they know when personalization sounds fake, and they can use AI to make more, and they've got the judgment and taste to know what should exist in the first place.
The agents are going to be a commodity at some point. . And that's the job of the marketing engineer, really.
And I think it's going to be one of the most valuable jobs out there. If you're a marketer, this is how you become the person your company literally cannot run without. And if you're a founder, this is how you get agents running your marketing for you.
I think there's a real edge you can have when you're actually using marketing agents to actually grow your startup ideas because people are still stuck in the old growth hacker or even worse, digital marketing era of marketing. I think this window is open right now.
I think a lot of people haven't built the machine and I wanted to give you the sauce so that you can... internalize it so you can process it, so you can get your hands dirty around building some of these agents, some of these marketing agents, because it's all about increasing your probability of success when it comes to building your own startup.
And I thought that, hey, if you can get a promotion, if you can... have more fun being an employee working within an organization, why not do this? So hope this has been helpful.
Obviously, I could have gone deeper in so many parts of this episode. There just wasn't enough time. But do let me know what you want me to go deeper in.
Is it the Grokbot part? Is it the markdown files, skills? You let me know.
I live to serve. I'm here to just give that information to you. Hopefully you enjoy it.
Hopefully it gets your creative juices flowing. And if you haven't liked, commented, and subscribed at this point, I don't know what you're doing. Hook it up.
You're hooking yourself up. You're getting more quality content in your feed, less slop. So thank you for giving me your time.
Hope it's been helpful, and I'll see you next time.
The Hook
The bait, then the rug-pull.
Greg Isenberg opens by naming a role that doesn't have a settled label yet, forward deployed marketer, AI growth operator, marketing engineer, and argues whichever name sticks, the job itself is worth a quarter to a full million dollars a year because every company wants the same thing: a smaller team whose marketing gets smarter every week.
Named ideas worth stealing.
01:59list
The Timeline (four eras of marketing)
- Traditional: story, taste, psychology
- Digital: SEO, ads, funnels
- Growth: product loops, retention
- Engineering: agents, data, code, taste
Each technology shift produced a new most-valuable marketer archetype; 'Engineering' is presented as the current, still-unclaimed era.
Steal forframing a career pivot or a positioning statement around 'which era of marketing you actually practice'
04:43model
The job is turning signal into pipeline
- 1. Listen (customer + market)
- 2. Build (assets + agents)
- 3. Ship (content + outbound)
- 4. Learn (results + memory, feeds back to Listen)
A four-step feedback loop that defines the marketing engineer's core operating cycle.
Steal forstructuring any AI-agent workflow as a closed loop instead of a one-shot task
07:31list
Growth OS folder structure
- customer-truth/
- content-engine/
- outbound-engine/
- creative-testing/
- agents/
- results/
The minimum viable folder layout for a company's marketing memory that AI agents read from and write corrections back into.
Steal forany team's first AI-agent repo, marketing or otherwise
18:27list
Six systems built on the growth repo
- 1. Customer Truth
- 2. Founder Content Engine
- 3. Outbound Signal Engine
- 4. Creative Testing Engine
- 5. AI Search Visibility
- 6. Growth Cockpit
The concrete systems a marketing engineer builds on top of the growth repo, each mapped to the HVAC example in the video.
Steal foran audit checklist when evaluating a company's current AI-marketing maturity
26:22list
Four ways this becomes money
- In-house ($250K-$500K)
- Consulting ($5K-$30K/mo)
- Productized (repeat one wedge)
- Software (turn repeated pain into SaaS)
A monetization ladder for the marketing-engineer skill set, with an explicit recommendation to start with services and productize what repeats.
Steal forsequencing any skill-to-business plan, not just marketing
30:21list
The 30-day plan
- Week 1: audit one company
- Week 2: build growth-os
- Week 3: ship one machine
- Week 4: prove result
A four-week, one-company learning plan that ends in a documented case study rather than a course completion.
Steal forany 'learn by doing on a real account' onboarding plan
How they asked for the click.
VERBAL ASK
34:48subscribe
“if you haven't liked, commented, and subscribed at this point, I don't know what you're doing. Hook it up. You're hooking yourself up.”
Casual, low-pressure subscribe ask folded into the sign-off rather than a hard pitch; no product, course, or paid offer is pushed anywhere in the video despite the topic being how to monetize a skill.
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
