The United Arab Emirates has begun the most consequential part of its plan to deploy agentic AI across federal government: deciding which tasks machines may complete autonomously and which decisions must remain under human authority.
More than 100 federal officials attended a workshop this month to establish priorities, timelines and classification frameworks for the national programme. The government’s stated target is to convert half of federal operations, services and tasks to agentic AI models within two years.
Classification turns ambition into accountability
The UAE has spent years building digital identity, sovereign cloud and data-sharing infrastructure. It also operates an AI-supported performance system that processes more than 150 million data points each month. Those foundations can connect an agent to government information, but they do not answer whether it should be allowed to act on that information.
The planned frameworks are intended to separate tasks suitable for autonomous execution from those where AI should only make a recommendation. The source says no government has yet published a comparable method at this scale. That makes the eventual classification document more important than another catalogue of technical capabilities.
Officials describe the implementation principle as “human leads, AI enables.” That sits beside earlier language from Sheikh Mohammed bin Rashid Al Maktoum about systems managing operations and completing sequences of actions without human intervention. The two positions are not necessarily contradictory: different tasks can receive different levels of autonomy.
However, the materials released so far do not explain how citizens will challenge an incorrect automated outcome or where liability will rest when an agent makes a consequential mistake. Training programmes, indicators and committees establish governance machinery; they do not yet define individual redress.
Seven pillars organise a two-year programme
The initiative spans strategy and projects, foresight and strategic intelligence, policies, organisational structures and governance, government performance, global competitiveness, and innovation in government work. Workshop participants also discussed performance indicators and coordination intended to prevent agencies from duplicating deployments.
Oversight rests with Sheikh Mansour bin Zayed Al Nahyan, while a task force is chaired by Cabinet Affairs Minister Mohammad Al Gergawi. Deputy Minister Huda Al Hashimi described the latest workshop as the operational start of deploying the models across government activity.
The programme is already moving beyond a small specialist group. In May, the Cabinet approved an initial package of services for agentic AI alongside training intended to reach 80,000 federal employees, from ministers to new recruits. A June event in Dubai brought together more than 300 participants from 50 federal bodies with a 90-day goal of identifying suitable services and procedures. Another workshop trained 600 staff.
The same May session approved a national AI healthcare policy and frameworks making digital records the official source of core government information. Agencies are expected to collect data once and share it securely, which could reduce duplication but also increases the importance of access controls and data quality.
Measurable targets will expose measurable failures
Many governments have published AI strategies without defining how much work will change or by when. The UAE’s 50% target and two-year deadline create a clearer test. They should eventually produce evidence about service speed, error rates, staffing effects and the kinds of decisions agents can manage safely.
The target is a government commitment, not proof that half of federal work can be automated responsibly. “Operations, services and tasks” is also broad enough that the final result will depend heavily on how units are counted. Automating many small administrative steps would look different from delegating a smaller number of high-impact decisions.
Other governments will be able to study the outcome, but the UAE must confront the unresolved questions first. Its task framework will need to connect technical autonomy with legal responsibility, appeals and human oversight. Without those elements, classification risks becoming an efficiency exercise rather than a defensible boundary on state power.
The programme’s significance lies in making that boundary explicit. Deploying agents is a procurement and engineering challenge; deciding what government may allow them to decide is a governance choice.
