There is no Gulf Ethan Mollick. I went looking for one. I did not find him.
Earlier this month I mapped the AI voices in this region. It was not a formal study. It was four separate sweeps, run at different times on different terms, because I did not trust the first result. Each one came back the same.
Three lanes, all crowded
Gulf AI content divides cleanly, and once you see the split you cannot unsee it.
Government and policy. Officials speaking in five year roadmaps. National strategies, transformation targets, adoption figures. This is important work and it tells you where the money is going. It does not tell you what happens on a Tuesday when you point the tool at a real process and the process pushes back.
The vendor and the consultancy. Frameworks, maturity models, case studies with the client's name removed. This lane is well funded and well staffed. It is also selling something. That is fine, as long as you read it knowing that.
The enterprise executive. The senior leader explaining the transformation. Usually accurate. Usually written after the fact, once the outcome is known and the false starts have been quietly edited out.
Then I looked for the fourth lane. The practitioner. The person standing inside their own operation, showing their own work.
What that lane looks like where it exists
In the US the practitioner lane is filled and crowded. People publish their actual builds. Their actual prompts. The version that did not work, and why. It reads like a lab notebook rather than a press release. Somebody tries a thing, it half works, and they publish the half.
That writing is useful in a specific way. It is not inspiring and it is not strategic. It is checkable. You can take what they did and run it yourself on Monday morning.
Here, almost nobody does that. The mandate pressure is enormous. Every organisation of size now has an AI directive. Very few people are reporting from inside one.
Why the gap exists
I have a theory and I will label it as a theory.
Showing your work costs something. It means publishing a failure with your name attached, in a business culture where being seen to be right carries weight. The safe move is to publish the framework instead of the field note. Frameworks cannot be wrong. They can only be unhelpful, and nobody gets fired for unhelpful.
There is a structural reason underneath the cultural one. Most people here with real operational depth are busy running the operation. They are not paid to write. The people who are paid to write mostly sit inside vendors and consultancies, which puts them back in lane two before they start.
So the lane stays vacant. Not for lack of value. For lack of anyone assigned to it, and because standing in it is uncomfortable.
What showing the work actually looks like
Two examples from my own systems this month, both of them failures.
The first. I keep a written memory for the AI assistant I run my business on. Every correction gets appended, because appending is safe and overwriting is not. In August I measured what that had produced. On one revised business model there were fifty seven stale statements against thirteen current ones. Four and a half to one, against the truth. Every one of those stale lines read as confidently as the day it was written, and retrieval has no sense of which is which. The memory was not wrong in any single place. It was wrong on average.
The second happened this week. I keep the source for this website on my own machine. When I went to put it under version control, the file on disk looked right, was dated correctly, and was stale. Three lines still carried an old description of my own background. Publishing from it would have quietly overwritten the corrected version that was already live, and nothing would have flagged it. I caught it by comparing against what the live site was actually serving, byte for byte, rather than trusting the file.
Neither of those is a good look. Both are the kind of detail that gets edited out of a case study, and both are the only kind that transfers.
The point
Eleven years running operations inside real businesses. A production company built from nothing in Sydney, then eight years as operations manager of an audio visual and cinema company in the Eastern Province. I have carried a payroll, which changes what you find interesting about technology.
I build AI systems now and I run them on myself first. Some hold. Some quietly stop working and take a week to notice.
That is the lane I am standing in. This is where the work gets shown, including the parts that did not work the first time.