UAE wants 50% of government running on AI agents

The largest declared AI agent rollout on the planet moved a step forward today, and the interesting thing about it is not the model vendor. It is that nobody has named one.

What actually happened

The UAE launched the strategic track of its national agentic AI project with a workshop for more than 100 federal officials, organised by the National Committee for the Agentic AI Project. MIT Sloan Management Review Middle East reported that the session covered timeline priorities, task classification frameworks, and mechanisms for cross-entity integration that eliminate duplicated work. Huda Al Hashimi, Deputy Minister of Cabinet Affairs for Strategic Affairs, described it as the operational start of deployment.

The target itself came from Prime Minister Sheikh Mohammed bin Rashid Al Maktoum: convert 50% of government operations, services and tasks to agentic AI within two years. The program spans seven pillars — strategy and projects, foresight, policies, structures and governance, government performance, global competitiveness, and innovation in government work — under the UAE Government 4.0 framework. Its stated operating principle is "human leads, AI enables." An earlier phase in June ran with 300-plus officials from 50 federal entities.

Treat the 50% as direction, not measurement. No public definition of the denominator exists, so there is no way to audit the claim from outside. What is auditable is the sequence.

Why it matters for your business

Notice the order of operations. Before any of this becomes a procurement, the work was: classify the tasks, name the pillars, decide where humans stay in the loop, and find the duplication. The model choice is downstream of all of it — and conspicuously absent from the announcement.

That is the opposite of how most small operations approach agents. The usual sequence is: buy the seat, wire it into the CRM, then discover three months later that it automated a task two people were already doing manually in a different system.

Run the boring version. Take a week of real work and split it four ways. Deterministic tasks — invoice matching, order routing, data sync — do not need an agent at all; they need code, and code is cheaper and does not hallucinate. Judgment tasks with a checkable output — drafting quotes, triaging tickets, summarising calls — are where agents actually earn their cost, because a human verifies in seconds what took minutes to produce. Judgment tasks with no cheap check stay human until you build the check. Duplicated tasks get deleted before anything touches them.

A government with unlimited procurement budget is doing task classification first. You have a much better reason to.

Sources: MIT Sloan Management Review Middle East, Gulf Today.