Inside The UAE’s Two-Year Agentic AI Push: What Organisations Should Actually Do
The United Arab Emirates has given its private sector two years to adopt agentic artificial intelligence and has committed its own government to becoming fully agentic, and the organisations treating those deadlines as an invitation to rebuild how they work, rather than as a box to be ticked, are the ones positioned to have something real to show at the end of the window.
The mandate has supplied a clear sense of direction, and the path to delivering on it is already well mapped by the technology leaders working most closely with enterprises in the region, who describe a sequence that is achievable for any organisation prepared to begin now and begin small.
Government is travelling quickly because the structure of its work lends itself to speed, and that early momentum is establishing a template the wider economy can study. Jessica Constantinidis, Innovation Officer at ServiceNow, said the public sector would meet its target because so many of its services are wired to one another that a single update cascades through the rest, where a passport renewal flows automatically into residency, the Emirates ID, and a worker’s permit rather than four separate submissions.
“I think government will get there because the process allows it to be faster and the process allows it to be more, you know, push through because a lot of the things are very similar,” she said.
That similarity matters because it removes the hardest part of agentic automation, which is having to reason through fresh data, security, and control questions for every new case, and it is reinforced by the appointment of Chief AI Officers within each government department, giving the public sector a density of ownership the private sector rarely matches. The agricultural authorities in Abu Dhabi have already moved, and Constantinidis described the work with evident enthusiasm.
“They’re using these technologies and they’re using AI to figure out how can we cultivate more produce in the UAE in a desert with the optimal combination of not having water wastage,” she said, noting that the benefit the authorities were already seeing was considerable and, tellingly, that it sat on data carrying little sensitivity, which is precisely why it made a safe place to begin.
The private sector target is modest by design, and that is its strength
The private sector begins from a more complex position, and the practical question is how far an organisation can realistically travel in two years when its processes are varied rather than uniform and its data far more sensitive. Constantinidis was clear that the early expectation is intentionally measured, somewhere around 20% to 30% of process steps becoming agentic in the first phase, and that the discipline of starting matters more than the completeness of the result. “If they have done 20% of the pieces of the process or agentic, then they tick the box and they’ve done it,” she said, adding that “the reasoning behind it and the start of it is basically the more important part.”
Her recommended method is deliberately granular, and it inverts the instinct to deploy AI across an organisation in a single sweep. An organisation instead builds one workflow at a time, creating an agent that performs a single task, securing only the data that task touches, then repeating the exercise task by task, with an orchestrator above directing each agent on what to do while remaining unable to see the data any individual agent holds, so that the agents stay isolated and each piece can be governed in turn. “You find the most tedious job that you can do every single day that your employees do, find that one, do that and do the second one and do the third one and learn from how you do this,” she said, recommending that organisations stand up a cross-functional AI council bringing cybersecurity, compliance, and business teams into the same room, so that the workforce learns how the technology behaves rather than fearing it.
Governance is the foundation that lets early wins multiply instead of fragment
The discipline beneath the method is governance, and it is the area Dataiku has been pressing hardest, with real cause for optimism about how far enterprises have already travelled. Siddharth Bhatia, General Manager and Area Vice President of Sales for Dataiku META and India, said the most encouraging signal in the market is how much preparation organisations have done before the first conversation begins. “A lot of these prospects that we speak to for the first time have already done their homework. They already have identified a library of use cases,” he said, noting that the realisation governance is paramount has taken firm hold, because a single agent might perform impressively on its own while a growing population of ungoverned agents would, over time, produce chaos rather than value.
The answer Bhatia advocates is a single unified foundation rather than a scattering of disconnected pilots. “Organisations need a central framework that has data, that has models, that has business applications, that has governance, that has even oversight,” he said, describing a common control layer that sets standard policies for auditability, access, and security so that five strong use cases can scale safely towards a hundred. He identified integration as the genuine bottleneck ahead, given that CRM, ERP, and workplace vendors each build their agents differently, and argued the remedy is architectural. “It’s important to have both an open architecture plus seamless integration, and that’s the only way you can actually scale AI operations,” he said.
The most valuable preparation can begin before any technology is bought
The most empowering part of Bhatia’s argument is that the foundational work owes nothing to a purchasing decision, which means organisations waiting on vendor selection are leaving easy progress on the table. An organisation can cleanse and organise its data, define its policies, and establish a top-down, cross-functional mandate now, independent of any platform. “You don’t have to wait to buy your technology to set these standards from a governance perspective. You have the ability today, as an organisation, to define your policies for security, for compliance, for approvals,” he said.
That preparation pairs with a hard-headed test for which use case to pursue first, since the region has seen organisations buy software and hope value emerges, when the reverse sequence works far better. “It’s important to be very clear why you’re investing in AI, what is that use case, is it repeatable, is it going to be having a huge impact, and then constantly monitor the returns on that particular agentic AI use case,” Bhatia said, framing a single proven, measured use case as the soundest possible launch point before any organisation attempts to scale.
Proving value on low-risk work is what earns the confidence to go further
The case for sequencing was reinforced by Levent Ergin, Chief Industry Strategist for Agentic AI and Data Foundations at Informatica, now part of Salesforce, who set out a maturity curve that strips much of the risk out of the journey. Ergin advised beginning with low-risk, high-volume use cases that face inward, where a person stays in the loop and any error never reaches a customer, illustrating the point with a wealth manager preparing for client meetings. “That’s probably saved 19 to 20 hours of my time a week in preparing for that customer meeting, and if it goes wrong, no one externally will really hear about it because you’re still the human in the loop,” he said.
Those early, contained wins compound into the confidence to attempt more, which is what turns a deadline into a programme rather than a panic. “You’ve proven wins. You’ve gained confidence to then say, okay, now let’s take something with slightly more risk and then try that with the right guard rails,” Ergin said. The organisations best placed to move, he argued, are simply those that understand their own operations most precisely, which is encouragingly within everyone’s control. “How well companies know how they operate today are the ones that are going to be in a position to adopt agentic AI in the quickest and fastest way,” he said.
The prize is a different operating model, not a faster version of the old one
The deeper opportunity Ergin described is not speed for its own sake but a rethinking of how an organisation is shaped, and he reached for history to make the point, recalling that electricity sat inside factories exactly where the steam engine had stood for thirty years before anyone reimagined the floor around it. The agentic equivalent is to look past piecemeal automation towards what compressing a two-week home-loan decision down to ten minutes does to marketing, to sales, and to the operating model entire.
That reframing is why Ergin insists the work is human before it is technical, and why the organisations getting their people and processes ready now are the ones that will convert the mandate into advantage. “I honestly think in most of the cases it boils down to the people and the process more than the actual technology itself,” he said, locating the decisive ingredient in something every organisation already holds, namely its own people and their understanding of the work in front of them.
