Science fiction to strategy: Agentic AI for world-class government in the GCC

This November at MWC25 Doha, I will speak about a shift already reshaping public services across the Gulf: the rise of Agentic AI. For some time now, AI in government has meant copilots and point tools that help people complete a single step faster, but the next chapter will be markedly different. Agentic systems can plan, reason and act across whole workflows – drawing on multiple systems, applying policy rules, triggering follow-up actions, and recording every step for audit.

That may sound like science fiction, but in practice it is the pragmatic path to public services that are faster, clearer and more transparent. And it is fitting that this discussion takes place in Doha. MWC25 Doha, the first MWC in the MENA region, signals how central this region has become to digital innovation. As a sponsor of MWC Doha, Oliver Wyman will show how to move from pilots to platforms responsibly, and how Quotient — AI by Oliver Wyman helps governments measure, govern and scale agentic systems with confidence.

What Agentic AI really is – and is not

Whereas today’s AI copilots wait for instructions, Agentic AI goes much further. Agents can collect data from trusted systems, reason through policy steps, act – such as sending a notice or scheduling an inspection – and record every action for human review. If a copilot accelerates one step, an agent completes several and hands you the result.

This shift, from passive assistance to active collaboration, transforms AI from a feature into a strategic capability. Yet multi-step government workflows demand high reliability. Small error rates compound quickly. Oliver Wyman’s A Pragmatic View Of AI Agents In The Enterprise highlights common failure modes – memory limits, error cascades and planning gaps – and recommends disciplined scoping, testing and governance. The goal is human-centered autonomy: agents handle repetitive steps while people retain oversight, judgement, and accountability.

Why the GCC is ready now

The Gulf has built the foundations that agentic systems require: sovereign cloud, digital infrastructure and coherent AI strategies. In Qatar, Microsoft launched a local cloud datacenter region in 2022, and in 2025, the Ministry of Communications and Information Technology (MCIT) announced a partnership to bring Azure OpenAI Services to government entities under the country’s Digital Agenda 2030.

Across the region, UAE’s National Strategy for Artificial Intelligence 2031 and Saudi Arabia’s National Strategy for Data & AI set ambitious long-term goals for AI leadership. Readiness indices such as the UN E-Government Survey and World Bank GovTech Maturity Index consistently rank GCC nations among the world’s most advanced in digital government. With these foundations in place, the region is poised to turn Agentic AI from concept into capability.

From forms and follow-ups to guided journeys

With the infrastructure now in place, governments across the Gulf can apply Agentic AI where it matters most – in the everyday interactions that define citizen experience.

The clearest opportunity lies in service delivery. Residents often navigate licensing, permits and benefits by submitting documents, chasing updates and waiting for officials to reconcile data. Agents can now pre-validate information, check compliance, and complete steps autonomously while escalating exceptions to humans. The result is guided, transparent interactions instead of fragmented bureaucracy.

Within the back office, agentic workflows can automate repetitive tasks – procurement, recruitment or inspections – while keeping human oversight for final decisions. In policy support, agents can synthesize data across ministries, generate scenarios, and present decision packs that make trade-offs clear.

The telecom industry provides a useful precedent. Digital agents in contact centers have reduced average handling times by up to 2× and cut post-call work by half. Network-operations agents have lowered costs by 20–40 percent and improved ROI by 10–15 percent. The same architecture – orchestration, guardrails, multi-model design – can now modernize public services.

Early use cases to prioritize

These early transformations point to a broader set of practical applications that can be rolled out today.

• Licensing and permits: Agents can pre-validate documents, check compliance and draft decisions for officer approval – cutting turnaround times.
• Procurement:Agents can draft specifications and requirements, compile RFPs, support bid evaluations, compile tender packs, verify policy compliance and manage supplier Q&A, improving transparency.
• Social programs: Agents automate eligibility checks and proactive citizen engagement, while humans handle exceptions.
• Human resources: Automated intake, scheduling and screening accelerate recruitment and reduce administrative load.
• Inspections and fieldwork: Mobile agents guide officers through digital checklists, capture data consistently and trigger follow-up actions.

Beyond these, multi-agent ecosystems will soon support transport and logistics, utilities, healthcare, finance and education – sectors where data and workflow automation already exist. Each builds on the same design rule: narrow scope, strong oversight, measurable results.

A Pragmatic and Responsible Path

While the potential is clear, scaling Agentic AI also requires structure and safeguards. Across our client work and Quotient assessments, we see a clear roadmap for doing this safely – and the guardrails that must accompany it.

The journey begins with measurement. Before deployment, agencies should evaluate agent performance against expert human output, defining what “good” looks like and setting clear reliability thresholds. Next, start where speed matters more than perfection. Use agents for “discovery tasks” such as drafting, research or summarization, where human review is built in and rapid iteration drives learning.

The third step is to constrain trust-critical actions. For citizen-facing or regulatory tasks, narrow the scope, embed human-in-the-loop checkpoints, and ensure every decision is logged for audit. Finally, build hybrid systems. Combine deterministic code, domain-specific models and fine-tuned language models, and only extend autonomy when evidence proves reliability.

This disciplined approach – outlined in Oliver Wyman’s A Pragmatic View Of AI Agents In The Enterprise – keeps experimentation safe while capturing real productivity gains.

Rigorous governance underpins every step. Multi-step workflows magnify small errors; privacy risks increase when agents access sensitive data; and transparency is essential because citizens must understand how AI contributes to decisions. Reliable systems require step-wise validation, fallback paths, human checkpoints, and live dashboards to track performance and drift.

International experience reinforces this. Anthropic’s Project Vend showed how an autonomous AI, asked to run a small business, made plausible but costly mistakes. Salesforce’s CRMArena-Pro benchmark revealed similar fragility – agents achieved 58 percent success on single-turn tasks, dropping to 35 percent for multi-turn tasks, with little built-in confidentiality awareness. The conclusion is clear: autonomy without governance erodes trust. Governments that instrument their agents and keep humans clearly accountable will earn public confidence – and sustainable value.

In my experience, the most successful transformations balance vision with discipline – starting small, proving value and scaling with governance built in from day one.

Laying the right foundations

Technology alone is not enough to prevent errors or ensure adoption. Success depends on five organizational foundations:

• Trusted data. Shared, interoperable data models with quality controls and secure APIs.
• System access. Agents must integrate with core platforms – case management, identity, payments – without disrupting operations.
• Agnostic by design. Supports interoperability across multiple model providers, cloud and infrastructure layers ensuring flexibility, vendor independence and future readiness.
• Agent-to-agent interoperability. Enables seamless communication between agents across workflows and entities, enabling shared learning and coordinated action.
• Governance. Human-in-the-loop policies, auditability, cybersecurity and ethical oversight aligned with national frameworks.
• Cross-functional teams. Policy experts working alongside engineers and data scientists. Key roles include agent orchestrators to design workflows, evaluators to test reliability, and product owners who focus on measurable outcomes.
• Culture. The hardest piece: a shift from one-off pilots to continuous learning, where officials see AI as an enabler rather than a threat.

These conditions create the scaffolding for scalable, accountable innovation. Once they are in place, governments can begin small and build momentum through targeted pilots.

A 90-day plan to start now

Governments can start within a single quarter:

1. Select one workflow with visible citizen impact – for example, procurement for public services or goods, or HR recruitment of government employees.
2. Map and classify tasks as discovery or trust, adding checkpoints accordingly.
3. Build a small pilot integrating one or two agentic steps plus human review.
4. Measure results against baseline metrics for time, accuracy, and satisfaction.
5. Document and share the pattern for reuse across departments.

This disciplined approach converts experimentation into evidence – and evidence into momentum.

Economics and operating models

When implemented correctly, Agentic AI is not just about efficiency; it is about rethinking how work gets done. In our experience, impact comes less from technology itself and more from operating-model redesign.

In telecom and IT operations, our research has showed that embedding agents into engineered workflows has achieved 30-40 percent faster code development, 20-30 percent quicker refactoring, and 14-35 percent IT-spend optimization when scaled. Similar efficiencies can be realized in government once common platforms and evaluation methods are in place.

These shifts also reshape sourcing. Agencies should insource processes where data sensitivity and differentiation matter, partner for speed or specialized expertise, and consider build-operate-transfer models to develop capability before localizing ownership. The goal, as outlined in Oliver Wyman’s Location Strategy In The Age Of AI And Digital Agents, is not cheaper labor but smarter allocation of talent and accountability.

More broadly, Quotient provides the frameworks, tools and governance models that make responsible AI practical. It helps organizations define what “good” looks like, benchmark agents against human expertise, establish audit standards and track adoption over time. 

The GCC’s moment

Across the GCC, national strategies, digital platforms and governance frameworks are aligning. UAE AI 2031, Saudi NSDAI and Qatar’s cloud investments create both ambition and capability. Readiness indices such as the UN EGDI and World Bank GTMI show sustained progress.

Agentic AI will not replace civil servants; it will empower them. By embedding intelligence into processes, governments can operate with the speed of technology while maintaining the accountability of public service. The leaders who act now – pairing governance with scale – will set the benchmark for the world’s next generation of digital government.

That is how the region turns science fiction into strategy.