Agentic AI is set to outpace GenAI in growth. Learn the 3 architectural essentials every organization must adopt to stay ahead of the AI curve.

In the United Arab Emirates (UAE), AI is now an everyday tool. We use it as individuals, and we use it as professionals. Among the businesses that use it, the more successful implementers follow a Universal AI adoption path that changes the corporate culture from within and infuses the workforce with AI literacy. Stemming from this enthusiasm, analysts foresee AI in the UAE as a multibillion-dollar segment, with generative AI (GenAI) alone taking about US$383 million in 2025 and more than US$2.5 billion in 2031, a CAGR of nearly 37%.

But AI itself has changed. As businesses have come to understand the limitations of GenAI and the importance of taking an operationally centric approach to tool procurement, decision makers have begun to explore the idea of having AI agents with modular autonomy take over from other forms of AI. The UAE’s agentic AI market garnered revenues of around US$34 million in 2024. By 2030 it is expected to be worth more than 10 times this figure, some US$352 million. At a CAGR of almost 48%, agentic AI, in the UAE at least, will be adopted at a faster rate than GenAI.

As with GenAI, or any AI, or indeed any technology, procurement of agentic AI is no guarantee of success. We must be diligent about how we build our architecture, the ideal example of which, I believe, has three basic characteristics.

1. Flexible

AI waits for nobody. At its current speed of evolution, modern business IT environments find it difficult to keep pace. To stand a chance, CIOs must look at how easy or difficult it is to maintain their tech stacks. If a new version of the GPT core model arrives on the market, will it be easy to adopt, or will it require weeks of overtime work from the DevOps team and others?

To streamline adoption, enterprises should ensure that the underlying framework, in which AI agents will operate, is flexible enough to support plug-and-play models. Modular architecture is crucial to the success of almost any modern technology; but if the regularity of recent versions of GPT is anything to go by, then the journey organizations will take with agentic AI is likely to be marked by particularly frequent upgrades.

Architectures should be crafted around four layers: the generative model layer, the feedback layer (which implements learning loops across multiple models), the deployment layer, and the monitoring layer.

2. Matches models to jobs

To apply the FOMO principle to AI procurement is to invite disaster. The individual or team that oversees the organization’s Universal AI journey should be laser-focused on business issues first and AI only as the means to overcome challenges. Organizations should be fully cognizant of what issues are being addressed by AI. Is it an exercise in optimization? Is it the addition of a completely new business capability or a new product or a new service?

Whatever is being added, it should come with a net-positive value. The AI procurement team should work with targeted beneficiaries to ensure everyone knows how to measure success and what constitutes a risk. For example, giving GenAI-powered virtual assistants to sales or customer-service employees may lift their productivity, conversion rates, and even profitability ratings. But these benefits may be neutralized if employees share sensitive data with a cloud-native model.

Thankfully, formal metrics like answer correctness and B-score allow analysis of models for their suitability in a use case. “LLM as judge”, where AI models are used to monitor the effectiveness of other AI models is also viable.

3. Backed by strong governance

The AI journey is fraught with risk. Today, we see many organizations prioritizing speed over security, and we see AI budget growth outpacing that of IT budgets. The introduction of AI must align with the compliance obligations and financial limitations of the enterprise. The only way to achieve this is through appropriate governance.

Governance has a broad remit. On the security side, it can mandate content-filtering to ensure customers are never exposed to output that would be damaging to the organization’s brand. On the financial side, it can prescribe dashboards that monitor costs and categorize them by project and user.

So critical is governance to AI success that some modern AI platforms include it as part of the suite, signaling that solutions vendors now consider it as important as the building of ML models. Even when AI was in its infancy, some industry leaders were calling for “responsible AI” that cracked open the black box and presented models’ innermost workings for scrutiny. Guardrails and trust go hand in hand. Security builds trust with customers. Cost-effectiveness builds trust with the C-suite.

The path to Universal AI

We can have the AI future we want, but only if we apply due diligence. By ensuring we take the right steps towards security and cost-effectiveness we can introduce agentic AI in ways that produce the right results. It may be the talk of the town right now, but we must adopt agentic AI strategically if we are to prosper from its merits.

By Sid Bhatia, Area VP & General Manager – Middle East, Turkey & Africa, Dataiku

Related post

View all