Human-centric AI: Inside the UAE's strategy for safe government automation

The UAE is moving beyond using AI as a tool for answering questions or automating individual tasks. It is now attempting to redesign how government itself works around Agentic AI—systems capable of planning, coordinating and executing multi-step tasks with limited human intervention.

At the heart of this transformation is an emphasis on keeping humans at the centre of AI-enabled government. The UAE’s National Agentic AI Project is an ambitious effort to transition 50% of federal government services and operations to Agentic AI models within two years. If achieved, the UAE says it would become the first government in the world to undertake such a transformation at this scale.

While earlier government deployments largely focused on chatbots, predictive analytics or other narrow applications, agentic systems can execute end-to-end workflows across government entities. They can synthesise data, coordinate between agencies, execute routine processes and manage complex, multi-stage tasks.

But the UAE’s model does not envisage simply handing government decision-making to machines. Its approach places considerable emphasis on human capability, governance and accountability. AI agents can take on greater autonomy in execution and coordination, while human officials remain responsible for setting priorities, establishing governance frameworks and exercising judgment over consequential matters. The government has also created a governance framework defining the roles and responsibilities of ministries and federal entities in implementing the project.

Greater autonomy, however, creates new risks, including erroneous decisions, unchecked system behaviour and the difficulty of determining who is accountable when an AI agent acts across multiple systems.

The UAE is therefore attempting to build a governance system with a framework that uses powerful tools of AI but assigns responsibility to individual ministries and federal entities, with dedicated teams overseeing implementation and performance indicators used to measure efficiency and productivity. The initiative is consequently being treated not simply as a technology upgrade, but as an organisational transformation involving operations, services, workforce capabilities and governance.

The transformation has already moved beyond strategy. More than 50 federal entities have participated in implementation workshops, while the government has launched an initial cohort of AI agents covering areas such as procurement, tax auditing, customer service and technical support. The UAE is also pursuing large-scale training, with a national programme targeting 80,000 federal employees, from ministers and senior executives to recruits.

The UAE experience offers an interesting question for India. The two countries operate at vastly different scales: the UAE is a centralised, high-income federation, while India is a 1.4-billion-plus democracy in which governance is distributed across the Union, states and local bodies.

India has a mature Digital Public Infrastructure (DPI)ecosystem built around platforms such as Aadhaar, UPI and DigiLocker. The challenge is whether this infrastructure can evolve from enabling digital transactions and access to services into an agentic layer capable of executing complex public-service workflows.

That could mean moving beyond citizens or officials navigating multiple portals themselves. Verified AI agents could potentially coordinate processes spanning departments—for example, welfare delivery, licence renewals, grievance redressal or other services that currently require multiple human and institutional hand-offs.

But the UAE model also highlights the importance of retaining human control over consequential decisions. In India, where digital systems already play a major role in welfare delivery, errors in identification or authentication can have serious consequences. An agentic system that automatically cancels a benefit, denies a pension or triggers an adverse administrative action could magnify those risks.

The broader principle could therefore have particular relevance in India: AI agents could identify cases, synthesise evidence and recommend action, while consequential decisions remain subject to human review and accountability.

Agentic AI is transforming the UAE's government by enabling systems to plan, coordinate, and execute multi-step tasks with minimal human input, marking a significant shift from previous AI applications. This ambitious National Agentic AI Project aims to transition 50% of federal government services to Agentic AI models within two years, potentially making the UAE the first nation to achieve this scale of transformation.  While agentic systems can execute end-to-end workflows and synthesise data across entities, the UAE's approach prioritises human oversight and accountability, with officials remaining responsible for setting priorities and making consequential judgments.  The project is treated as an organisational transformation, encompassing operations, services, workforce capabilities, and governance, with over 50 federal entities involved in implementation and a national training program for 80,000 employees.  The UAE's experience raises questions for India about evolving its Digital Public Infrastructure to support agentic workflows, while emphasising the importance of human control over critical decisions to mitigate risks. The core principle for India could be using AI agents for identification and recommendation, with final decisions subject to human review.