Updated August 2026. Originally published June 2024.
In 2024 I wrote that one prompt engineer with marketing coordinator experience could absorb five marketing roles: market research, SEO, content, social, and data analysis.
I was describing myself, and I did not say so.
At the time I was acting as a one man fractional CMO for the company I had just joined. There was no team. The claim was not a prediction about the labour market, it was a description of what I was doing that week.
What I was actually doing in 2024
I did not hire to fill those five roles. I did the work, and used deep research to cover the parts that would otherwise have needed people.
What that looked like in practice:
- Competitor content strategies, pulled apart and fed back in as examples rather than as a question
- My own SEO material: the silo structure, the processes, the frameworks I had already built and tested
- Tool selection, for instance a proper comparison of social publishing schedulers instead of picking the one I had heard of
- Conversion rate optimization and value proposition methodology studies I had run myself
Every one of those started with something I already knew. That is the part I did not make explicit in the original article, and it is the part that turned out to matter most.
I said prompt engineer. I meant domain expert.
Looking back, the term was wrong.
You may never have used AI once in your life. If you are exceptional at what you do, your prompt will still be better than anyone else’s, because you know what to ask for and you know when the answer is wrong.
AI is a tool, like a knife. You can do remarkable things with it or you can do something stupid with it. Which one happens depends entirely on who is holding it.
The industry spent 2023 and 2024 hiring for the tool. The skill that actually compounds is the domain knowledge underneath it. I wrote the right idea under the wrong name.
What I hire for now
I hire AI native people. That means one of two things, and I will take either:
- Already very familiar with working alongside AI, or
- Extremely eager to learn it, without needing to be convinced first
I do not have a marketing coordinator. I have a GTM engineer who works with GitHub, works in Cowork, maintains dozens of Python scripts, and makes all of it hold together as one workflow rather than a pile of tools.
That role did not exist on my 2024 org chart, and it is the single most useful seat on the team.
Seven people against a hundred and ninety seven
The clearest number I have is appointments set per month.
In a previous role I ran marketing inside an organisation of 197 people. Our best month was 28 appointments.
Today the team is seven, and we book 123 a month.
I want to be careful with that comparison, because it would be easy to overstate. Neither number describes seven people doing one job:
- The 197 included a database enrichment team, a QA team, a cold emailing team, an SEO team of more than ten, a conversion rate optimization team, and a social media team.
- The seven include an SDR team of three, and the rest of the function around them.
So it is not a like for like headcount comparison, and I would not present it as one. What is like for like is the output, measured the same way on both sides.
What the seven actually have
The difference is not that the three SDRs work harder than the cold email team did. It is what arrives at their desk before they start.
- A morning brief pushed into the team chat automatically, no request needed
- A popup the moment a positive reply lands
- Booked appointment notes, including the email thread, sent to the sales manager for that region without anyone assembling it
- All data living in one database that is the source of truth, pushed to the CRM every hour
- Site visitor identification and behaviour scoring, so a call has a reason behind it
There is more than that, but the pattern is the same throughout: the connective work between the parts that matter has been removed, so the parts that matter get all the time. I wrote about where that started in adapting a marketing team to AI.
Why this could not have been done in 2024
Transitioning 197 people onto what we had in 2024 would have been a disaster, and I say that as someone who was enthusiastic about it at the time.
It was barely assistive. It answered from what it had read, in the shape of a conversation, and it was equally confident whether or not it knew. There was nothing underneath it that belonged to us.
What changed is that hard facts and real skills can now live in folders and in version control. Our pricing, our hardware, our documentation, our history of what closed and what did not, all of it sits where the system can reach it.
That is the difference between a tool that sounds like it knows your business and one that does.
The framework I used, and where it held up
The original version of this article leaned on Lex Sisney’s PSIU model, which sorts the forces that drive people into Producer, Stabilizer, Innovator and Unifier. I still find it useful, so it stays.
What I would revise is which part AI actually changes.
My read now is that AI strengthens the Producer and Innovator drives, and it does most of its work for people whose stabilizer drive is weak. Reading long documents, formatting, chasing details, holding process discipline: that is stabilizer work, and it is the work being absorbed fastest.
I know this from the inside. I always saw myself as a producer and an innovator, and I lost a lot of good ideas over the years because the execution required a kind of patience I do not have. That constraint is gone. I can now go from an idea to a documented, fact checked, actionable plan in a few hours.
Are stabilizers still needed?
Yes, but the job changed. Less producing the document, more judging it.
Here is the example I would give.
Our sales managers used to wait weeks for a statement of work. Someone had to write it, someone else had to review it, and it sat in a queue behind everything else.
Now it takes seconds. Not minutes, seconds. It comes out of a section of our system that acts as a solutions manager and sales enabler, connected to a folder that knows our company, our pricing, and our hardware.
The sales manager reviews it themselves, and that is close to instant too, because the output is formatted consistently enough that they know exactly where to look. The review did not disappear. It stopped being a bottleneck.
The prediction I made, two years later
In 2024 I predicted that AI would replace websites, that the exchange would become purely inbound, and that marketing touches as we know them would stop existing.
I still stand behind it, and I think the mechanism is clearer now than it was when I wrote it.
The models consumed the web. They weighted their trust toward high credibility, non commercial sources: universities, government, large community sites. Independent publishers lost enormous amounts of traffic and the revenue that came with it.
The consequence is the part people are not saying out loud. Website publishers are no longer meaningfully incentivized to publish. If the traffic does not come and the revenue does not follow, the supply of independent, first hand material on the open web thins out.
So why am I still writing this
It is a fair question to put to me, given what I just said.
I am not publishing for traffic.
My view is that a blog stops being a collection of HTML pages and becomes an identity that the models refer to. Not articles to be visited, but a body of positions, decisions and evidence that gives me a voice in the market and inside the models.
That is why what goes on it has to be mine, and has to be checkable. I wrote about what that means for the writing itself in AI content and E-E-A-T.
If you are running a marketing organisation and wondering where to start with any of this, it is not with the org chart. It is with documenting what your people actually do all day. The structure follows from that, and not the other way around.
This article was substantially rewritten in August 2026. The original June 2024 version argued that a prompt engineer could absorb five marketing roles, and listed fifteen marketing job titles as context. The core idea survives under a better name, domain expertise, and the list has been removed. The team numbers, the statement of work example, and the two year review of my original prediction are new.
