On 4 May 2026, Dubai's Crown Prince Sheikh Hamdan bin Mohammed announced what is being described as the most aggressive private-sector AI mandate any government has issued. Every private-sector organisation operating in Dubai is expected to transition to agentic AI within two years. Specialised training programmes will run through Dubai Chamber of Commerce business councils. Government-funded incubators and investment vehicles are being stood up. The goal, stated plainly, is for Dubai to become the global leader in agentic AI adoption.
This did not arrive without context. Less than two weeks earlier, the UAE Cabinet had mandated that fifty per cent of all federal government services would be delivered by autonomous AI agents by the same 2028 deadline. The private-sector announcement is the complementary half of the same strategic picture: the state is automating from the inside, and it expects business to follow.
There is a problem with this picture — not with the ambition, which is coherent and well-resourced — but with the state of readiness in the organisations expected to execute it. Current data puts the share of companies actively planning agentic AI deployment at seventy-four per cent. The share with a governance model mature enough to support it sits at twenty-one per cent. That fifty-three point gap is not a technology problem. It is a programme delivery problem. And two years is not long.
What agentic AI actually requires — and why it is different
Most organisations that think they are ready for agentic AI are actually ready for a better chatbot. The distinction matters enormously.
Conversational AI responds. It answers queries, suggests options, routes to a human when it runs out of road. The human remains in the loop for any consequential decision. Agentic AI acts. It executes multi-step tasks, integrates with live systems, makes decisions autonomously within defined parameters, and handles edge cases without a human looking over its shoulder at every step. That shift from response to action changes the risk profile completely — and with it, the governance, integration, and operational design requirements that any sensible deployment demands.
An AI agent that can update a customer record, process a refund, modify a contract, or submit a procurement request is not a productivity tool. It is a decision-maker operating at scale. The same properties that make it valuable — speed, consistency, 24/7 availability — make it dangerous if the underlying process design, data architecture, and accountability framework are not built to match. When it goes wrong, it goes wrong at volume, and it does so faster than any human team can catch.
This is the gap the twenty-one per cent figure captures. Governance maturity in this context means having clear answers to questions most organisations have not yet asked: which decisions can an agent make autonomously, and which require escalation? What audit trail does every agent action create? What happens when an agent encounters a situation outside its training? Who is accountable when an automated decision causes customer harm? How does the organisation detect and respond to agent failure at scale?
These are not technology questions. They are process design and programme management questions. And the training programmes Dubai is offering — structured around customer service, procurement, logistics, compliance, and decision support — confirm that the government understands this. The mandate is not to buy a platform. It is to transform how organisations operate.
Customer service: the highest-stakes deployment area
Customer service is specifically named as one of the five training tracks under the Chamber of Commerce programme, and for obvious reasons. It is where agentic AI has the most immediate commercial application and where deployment risk is highest.
The current state of AI in customer service — across markets globally and in the GCC specifically — is characterised by a gap between what organisations measure and what they actually deliver. Containment rates look impressive. Resolution rates, when honestly measured, are often catastrophic. AI handles the query; the customer's problem remains unsolved. The customer phones in. The human agent has no context from the digital interaction. The experience is worse than if the AI had not been involved at all.
Agentic AI does not fix this by default. It amplifies it, in both directions. An agentic customer service deployment that is well-designed — with access to live customer data, integrated into order management, billing, and case management systems, with escalation logic that actually works and a feedback loop that improves its judgement over time — can transform the customer experience. One that is poorly designed creates the same resolution failure at higher speed, with no human in the loop to catch it, and an audit trail that makes accountability difficult to establish.
The difference between those two outcomes is not the AI model chosen. It is the quality of the process design upstream and the integration work that connects the agent to the systems that hold the data it needs to act correctly. That work is unglamorous, time-consuming, and where most organisations cut corners. The mandate will not change that tendency. The deadline will.
The two-year problem
Two years sounds like a long time. For AI programme delivery, it is not.
A credible agentic AI deployment in a complex enterprise — one that genuinely automates meaningful work rather than wrapping a conversational layer around an existing process — takes twelve to eighteen months to design, procure, integrate, and roll out to the point where it is reliably operational. That timeline assumes clear scope, executive sponsorship, available data, a coherent vendor landscape, and a programme management function capable of holding it together. Most organisations in scope for this mandate do not currently have all four.
The organisations that will meet the 2028 deadline are the ones that start the programme design work now — not the technology selection, not the vendor conversations, but the harder questions: which processes are genuinely candidates for agentic automation, what does the data infrastructure look like against what it needs to be, where are the governance gaps, and what does the transformation programme actually require to close them? Those questions take time to answer honestly. They take longer to act on.
The organisations that will not meet the deadline are the ones that treat the mandate as a procurement event — select a platform, run a pilot, report upward that agentic AI has been adopted. That approach is not new. It is the pattern that has produced a ninety-five per cent failure rate in enterprise AI globally. A government mandate does not change the underlying failure mode. It adds a deadline to it.
What good programme design looks like in this context
The training structures announced under the mandate — business council tracks, incubators, investment support — are the enabling infrastructure. They will help organisations understand what agentic AI is and what it can do. What they will not do is substitute for the programme management rigour required to deliver transformation at scale.
In our experience working across complex AI and CX programmes in the region, the organisations that deliver real outcomes share a consistent set of characteristics. They select use cases based on value and data readiness, not visibility. They treat integration as the primary technical challenge, not a post-selection afterthought. They build governance frameworks before they build agents, not after the first incident. They measure outcomes — resolution rates, cost per interaction, decision accuracy — rather than activity metrics. And they resource change management as a core programme workstream, not as a communications campaign that runs in parallel.
The mandate does not change any of that. What it does is create a genuine forcing function — an external deadline with government visibility — that makes it harder for organisations to keep treating AI transformation as something that will happen eventually.
Eventually is now 2028. For most organisations in scope, that means the programme starts this quarter or it does not start in time.