The Signal & The Noise · Vol. 08
Abu Dhabi will be the world's first AI-native government next year, Dubai wants half its services delivered by AI agents, and people's lives run straight across the lines between them. What has to lie underneath, shown for the first time in public.
This week the region gathers for Ai Everything in Abu Dhabi, at ADNEC, and I am walking the floor. The ambitions are big, and they should be. According to GovDigital, Abu Dhabi will be the world's first AI-native government next year, a government built on AI from the ground up, with more than forty entities in a single ecosystem. Dubai wants half of its government services delivered by AI agents within two years, and in June the federal government announced its own authority for AI and data. In this country, AI gets built rather than talked about.
Many of you who read this newsletter are not in the Emirates. You are in Europe, in Africa, in Asia or further away, in countries with the same ambitions and, often, less room to move this fast. For you, this is worth a closer look. The Emirates are good at something that is rarer than it seems: setting an ambition so large that everyone has to help build it, and then actually building. That is the example this country sets, and it extends to what such an ambition asks of the foundation underneath.
Each emirate is building its own ecosystem, and that is exactly right. People's lives, however, run straight across those lines. Someone who lives in Sharjah and works in Dubai, or moves from Abu Dhabi to Dubai, keeps arriving at the point where two systems have to understand each other. The UAE National Strategy for AI 2031 has a name for the layer where that happens: its bottom layer is called, quite literally, the Foundation. The strategy says what has to be there and leaves the how to the people who build it, because that is craft, not policy.
Samture builds that layer. Today I am showing in public, for the first time, what it looks like, and I do so with pride, because it works.
At an event like this, the conversation is about the model: which model, how many use cases, which agent goes live first. I understand why, because that is where the visible progress is. But an assistant that reads each office on its own simply inherits the confusion between those offices, politely and with confidence. The difference between a service that works and a service that looks as if it works rarely sits in the model itself.
Take a family moving from Dubai to Sharjah. They deal with a landlord, a rent registry, a municipality, two utilities, an identity authority and two education bodies. Seven offices across two emirates, and each one knows only its own form. The same tenancy is an Ejari certificate in one system, an attested lease in the next and a tenancy contract in a third. Which step comes first, and what depends on what, is written down nowhere; it lives in the citizen's head. When a school in Sharjah asks for an Ejari, the family hands over the only one they have, the Dubai tenancy they are in the middle of closing, and no system notices.
On a shared foundation, the same question becomes one plan, in the right order, across all seven offices, with the source of every document shown at every step. The foundation even catches the wrong document before it is handed in. No office has to replace its own systems; the meaning is captured once, and every office uses it. You can see it in a short film.
The same foundation carries two more things we are showing this week. A register of eight AI systems, measured against the principles of the UAE Charter for the Development and Use of AI, that shows on its own where a system falls short, such as a decision on a benefit taken without human review, and puts the evidence next to it. And a test bed in which a young company tries its model on data from three government offices and receives figures for groups, never a row about a person, computed inside the foundation. Revoke the key, and the next question is refused. The data never leaves the house.
Three very different questions, and the same work underneath each time. We capture, once, what something means, what depends on what, and where it comes from. Who supplies the data and who uses it, a municipality or a school, a citizen or a regulator, makes no difference to that. This is what I am proud of: you lay the foundation once, and every next use is faster to build and easier to check than the one before. All three were recorded from a working demonstration on fictional data. They sit side by side at samture.com/solutions, where you can also try the demonstration yourself.
That is why I take this layer, with the same confidence, into a very different world: a bank. Picture an experienced anti-money-laundering analyst who leaves for a competitor. The risk rating of every client she handled is on file. The judgement behind it is not: why this client and not that one, which signals she always weighs, which clients she knows belong together. That walks out with her. Meanwhile, since February the Central Bank of the UAE has expected banks to be able to explain how their AI reaches a decision, and has made clear that the responsibility stays with the bank, even when the work is outsourced. And in this field the responsibility is personal: this year the regulator fined not only a bank but its head of compliance, 300,000 dirhams, for failures in exactly this area.
Look at it the same way, and it is the family's question again. What the analyst knows is largely about what things mean, what depends on what and where it comes from. Once that is captured in a foundation, she takes her experience with her and the reasoning stays in the building, and the bank can answer the regulator with something that will still hold tomorrow. A family and a bank seem to live in different worlds, and for us it is the same craft. I will come back to it in more depth in a later edition; a first outline is already at samture.com/banking.
You do not have to take my word for it. Pick one event that sends someone in your organisation past more than one department or office: a move, a new client, a permit, an employee who leaves. Write down what each department calls the key document, and who knows the order in which the steps belong. If you end up with more than two names for the same thing, or only one person knows the order, you know where your foundation begins.
I am at ADNEC this week. If you would like to see what this could look like for your own processes, the three films and the demonstration are at samture.com/solutions, and I would be glad to spend half an hour on it, on the floor or afterwards.
And if you are reading this from outside the region: look over my shoulder, and take whatever is useful. Some of you solved part of this long ago, with shared registries that every office draws on. The AI version of that question is being answered here now, at a speed worth watching, and the lesson about the foundation is the same wherever the data lives.
Three demonstrations on one shared foundation: connected public services, an AI register that checks itself against the UAE AI Charter, and a secure test bed. Recorded from a working demonstration on illustrative data, and open for you to try.
See the solutionsThe family, the offices and the AI systems in the demonstrations are illustrative, built on fictional data and modelled on publicly described procedures. No real agency, person or bank is depicted.
The Signal & The Noise
A fortnightly read on what is actually changing in AI across the UAE and the Gulf, written personally. No content calendar, no generated filler.