I spent most of my career being judged on rankings, click-through rate, and conversion percentage. Nobody asked whether I understood tokens, context windows, or agent orchestration. Then the job postings changed, and the question became whether a growth marketer could also build the system, not just write the content that goes in it.
I built a Filipino tech community from a bare domain in my twenties, then spent a decade-plus running affiliate sites and marketing for a crypto-fintech startup. These days the title on my site says AI growth engineer. This is the honest version of that transition, what carried over without friction, what I had to learn from zero, and what I'd tell a recruiter trying to evaluate someone making the same jump.
Why I made the jump
Affiliate marketing taught me to treat every page as a system: traffic in, intent matched, action out. For years that system ran on keyword research, backlinks, and content calendars. Then the automation layer got good enough to run half of that system without me typing each step by hand.
I didn't wake up one day and decide to become an engineer. I noticed I was spending more hours building the workflow that produced the content than writing the content itself. I wrote about that shift in more detail here, the short version is that encoding my own judgment into a repeatable system felt like a better use of a decade of marketing instinct than doing the same manual pass for the thousandth time.
So I started calling the thing I was actually doing by its real name. Not SEO. Not content marketing. Growth engineering, with AI doing the execution layer I used to do by hand.
What transferred directly, no relearning required
Some of this surprised me. I expected to start from zero on the technical side and keep my marketing instincts as the one asset I brought with me. Instead, a chunk of the marketing instinct turned out to be the exact skill the AI side needed, just pointed at a different output.
| Affiliate/SEO skill | AI growth engineering equivalent |
|---|---|
| Reading search intent behind a query | Reading the real intent behind a prompt before writing it |
| Structuring content for crawlers (schema, headings, internal links) | Structuring data and context for a model to parse reliably |
| A/B testing headlines and CTAs | Testing prompt variants against the same output criteria |
| Tracking funnel drop-off in GA4 | Tracking where a workflow fails or produces bad output |
| Cold outreach to build a community from nothing | Pitching AI workflow work to clients who've never bought it before |
The outreach line is personal. The instinct to send a cold email with no guarantee of a reply, the one I used to book the first community event back in 2009, is the same instinct I use now pitching a workflow automation project to someone who isn't sure what they're buying yet. The channel changed. The nerve required didn't.
Analytics discipline carried over almost unchanged too. A marketer who's spent years staring at GA4 funnels already knows how to find where a process is leaking value. Pointed at an AI workflow instead of a landing page, that's the same question: where does this break, and how do I know it broke.
What I had to learn from zero
The gap wasn't strategy. It was execution mechanics I'd never needed before. Marketing rewards you for knowing what should happen. Building the system that makes it happen rewards you for knowing exactly how, down to the API call.
- Actually building workflows, not just briefing them. Chaining calls in n8n, handling errors when an API times out, deciding what happens when a step fails halfway through a run. There's no creative workaround for a workflow that silently breaks at 3am.
- Version control. I never touched git as a marketer. Now every workflow and every prompt I rely on lives in a repo, because I need to see what changed when something that worked yesterday stops working today.
- Treating model output as unreliable by default. A model will answer confidently and wrong. Marketing copy that's slightly off still reads as copy. A workflow that's slightly off produces bad data downstream, and nothing flags it unless you built the check yourself.
- Reading and writing structured data. JSON, schema, API payloads. None of it is hard once you've done it a dozen times, but a dozen times is a real cost when you're starting from a content background with no engineering reps behind it.
Not the tools, the trust calibration. I spent months either checking every single AI-generated output by hand (too slow to be worth automating) or trusting it too much and shipping something wrong (worse than doing it manually). The real skill, the one no course taught me, was figuring out exactly which steps need a human check and which ones genuinely don't.
The skills map, visually
Laid out, the transition looks less like starting over and more like a shift in what the same underlying judgment points at.
What this means if you're a recruiter looking at a resume like mine
If you're screening someone making this same jump, the resume will look mismatched. The job titles say marketing. The portfolio has workflow diagrams and code. That mismatch is the thing to look past, not the thing to flag.
Ask what they built, not what they were titled. A marketer who automated their own reporting, built a content pipeline with an AI step in it, or ran a growth system end to end without a dev team backing them up already has the habit a pure engineering background sometimes lacks: starting from the outcome and working backward to the system, instead of starting from the tool and looking for a use for it.
The weakest version of this transition is someone who learned the vocabulary, agent, prompt, workflow, without ever shipping something that broke in production and having to fix it. Ask for one specific example of something that failed and what they changed because of it. That answer tells you more than any certificate on the resume.
Key takeaways
- Marketing instinct and AI workflow building share more surface area than the job titles suggest, intent reading and analytics discipline transferred almost unchanged.
- The real gap was execution mechanics: building workflows that don't silently fail, using version control, and treating model output as unreliable until proven otherwise.
- Trust calibration, knowing which steps need a human check, was the hardest skill to learn and the one no course covers.
- Cold outreach nerve from building a community years ago is the same nerve needed to sell AI workflow work nobody's bought from you before.
- When evaluating someone on this path, ask what they shipped and what broke, not what words are on their resume.
What I'll do next
I'm writing down the specific trust-calibration rules I actually use, which steps in a workflow get an automatic pass and which ones always get a human check, as an actual checklist instead of a gut feeling. Right now that judgment lives in my head, and it's the one part of this transition I haven't productized yet.
FAQ
Can an affiliate marketer or SEO really move into an AI growth role without a CS degree? +
What should go on a resume for someone making this transition?
Do I need to learn to code to make this jump?
What's the biggest red flag in someone claiming this transition?
How long did this transition actually take?
Want this built for your team?
I design AI agents and growth automation that run without babysitting. If that sounds useful, let's talk.
Get in touch →