UAE Publishes Design Guide for Agentic AI-Powered Public Services

Announcing a target is one thing. Defining how the system behind that target should actually treat the people using it is another, and considerably harder to get right. The UAE's National Committee for the Agentic AI Project has published a unified design guide setting out core principles for how agentic AI systems should be built and how they should interact with citizens across government services, a document that moves the country's agentic AI transformation from stated ambition into specified behaviour.

The guide was released during a workshop attended by more than 100 government officials, part of the same national push to convert 50 percent of government operations, services and tasks to agentic AI-driven models within two years, a mandate that has turned agentic, governed and human-supervised AI into a matter of national competitiveness across the wider Gulf, not just a UAE-specific push. Mohammed Rashid bin Talia, Assistant Minister of Cabinet Affairs for Government Knowledge Exchange and head of the government services track within the project, described the guide as marking a shift in how administration for public services is delivered, moving the burden of navigating bureaucracy from the citizen to an AI assistant. Under the model he described, a resident would simply state what they need, and the AI assistant would coordinate directly with the relevant authorities to complete the transaction on their behalf.

That framing is worth taking seriously as a genuine change in interaction design, not just a technical upgrade. Most government digitisation to date, in the UAE and globally, has focused on making existing processes faster: fewer forms, quicker uploads, shorter queues. What bin Talia describes is a different model entirely, one where the citizen no longer navigates the process at all, an agent does it on their behalf, treating the underlying bureaucracy as something to be abstracted away rather than merely streamlined.

The guide itself sets out specific, binding design principles rather than general aspirations. Two stand out for what they reveal about the underlying philosophy. First, an agentic AI system should never ask a user to provide information they have already submitted previously, a principle that sounds simple but requires genuine cross-entity data interoperability to actually deliver, information submitted to one government body needs to be accessible, with appropriate authorisation, to any other agent handling a related request. Second, users must be kept informed at every step of their service journey, a transparency requirement that pushes back against the risk of an autonomous system operating as an opaque black box between request and outcome.

Privacy and security protections are built into the guide as core requirements rather than appended considerations, though the public reporting on the guide's release does not specify the technical mechanisms involved, an area regional analysts have already flagged as a growing priority as agentic systems scale across UAE, Saudi and Qatari government and enterprise infrastructure alike. What the guide does specify clearly is how the system should handle situations requiring human involvement. The guide anticipates instances where a citizen may need to interact with an actual person, built around the same "human leads, AI enables" principle that has anchored the UAE's broader agentic AI messaging since the project's launch. Codifying that principle into a design guide, rather than leaving it as a general reassurance, gives it more practical weight: it means system designers building agentic services are working from a specified requirement for human handoff points, not simply a stated intention.

This guide does not exist in isolation from the UAE's broader AI governance architecture, and understanding that context matters for assessing how seriously to take it. In June, the government merged its Office of Artificial Intelligence, Digital Economy and Remote Work Applications with the UAE Data Office to create a single national authority consolidating public data, AI and digital government capabilities, building on a separate structure announced for January 2026, in which a National AI System sits as an advisory member of the Cabinet specifically to oversee ethical deployment. That consolidation gives the new design guide an institutional home with genuine authority over both the data infrastructure the "never ask twice" principle depends on and the AI systems the guide is meant to govern, rather than the guide sitting as a standalone document with no clear enforcement mechanism behind it.

The UAE's approach to AI transparency requirements is not new territory either. Last year, the country implemented what it described as a world-first policy requiring political candidates in Federal National Council elections to declare and register any AI tools used in their campaigns, a measure Minister of State for AI Omar Sultan Al Olama framed as necessary because AI could be "misused to influence voters unfairly" even as it helped candidates reach broader audiences. That willingness to codify specific, enforceable AI behaviour rules rather than relying on general principles is the same instinct visible in this new design guide, translating high-level commitments into concrete, checkable requirements.

For government technology teams and vendors building agentic AI systems for the UAE public sector, this guide functions as a genuine specification document, not just a values statement. Vendors and integrators bidding on government agentic AI projects now have a defined bar to design against: no redundant information requests, continuous status transparency, and specified human handoff points. For citizens and businesses interacting with UAE government services as this transformation proceeds, the guide is the clearest public signal yet of what standard those interactions are meant to meet, and, by extension, a benchmark against which any rollout falling short of "never ask twice" or "always inform" can reasonably be measured.