Operational test at LEAP: why deal numbers and attendance crowds are no longer enough to gauge AI maturity in the region

With the fifth edition of the LEAP conference taking place in Riyadh at the end of August, the battle to demonstrate the ability to attract capital and rally international attention has become completely settled. The record figures previously logged, attendance exceeding 200,000 participants, the announcement of AI investments worth $14.9 billion on the opening day of the last edition, and the cumulative total of announced tech-infrastructure investments in Saudi Arabia surpassing $42.4 billion since the conference began in 2022, have closed the debate about the region’s appetite for investment and growth.

But the real challenge in 2026 is no longer the size of the announcements, but the metric of conversion and actual operation. The cumulative figures announced in major technology markets suffer from a known structural flaw: they always rise and never fall, because there is no public record that excludes projects that have been quietly restructured or frozen after years of being announced. What regional market credibility needs today is an accounting audit that shows the computing capacity actually running in data centres versus what has been announced, and the technical functions that have already been deployed compared with the stated targets.

The genuine shift to agency-based artificial intelligence starts with government tender specifications, not with demo presentations at conference platforms.In this trajectory, the United Arab Emirates has taken advanced steps toward governance of agency systems, with the National AI System, an advisory body, participating in the Cabinet, the Ministerial Development Council and the boards of federal agencies since January, and with the AI and Data Authority established in June to develop a unified digital government framework based on intelligent agents. Advising governments fundamentally requires settling mechanisms for exchanging sensitive data between ministries, so the real test of these technologies’ effectiveness in events such as Deep Fest, the side event of the conference, lies in the appearance of binding conditions within procurement tenders that define service levels, escalation paths and audit logs for the processes executed by agents.

This transformation also reaches the dilemma of developing authentic Arabic software. Throughout the past decade, Arabic has been treated as a later localisation step added to English interfaces, even though a language spoken daily by more than 400 million people cannot be classified as a marginal market. Building Arabic software and models from the ground up, capable of understanding local dialects in specialised fields, entails costs that exceed simple translation, and this will not be achieved through theoretical calls but by assigning explicit weighting factors to Arabic linguistic capabilities in government procurement and large-company evaluation tables; only then will global firms race to build them within a few months.

On the talent-building front, Saudi Arabia’s National Data and AI Strategy aims to train 20,000 specialists by 2030; the Sadaia training programmes have reached 14,495 specialists and experts, the Samay initiative has trained more than one million citizens, and the UAE has awarded 100,000 golden residencies to programmers. In contrast, estimates from IDC project that more than 90 percent of organisations worldwide will face a technical-skill shortage by 2026, causing losses of about $5.5 trillion, putting the Gulf states in direct global competition for qualified engineers.The most important metric that has not yet been published for building technological sovereignty is not the number of trainees, but the retention rate of national technical personnel within the ecosystem after three years.

What is changing in practice today for organisations and entrepreneurs in the Gulf, Egypt and the Levant is a redefinition of skills and production costs. Notably, accelerated data from Wi-Computer for the winter-2025 cohort showed that roughly a quarter of companies relied on AI to generate 95 percent of their code, and the RepLit platform reported that 75 percent of its users do not write code; the specialised engineer is no longer a prerequisite for building a prototype. This shift enables a doctor in Dubai, a logistics supervisor in Jeddah or an entrepreneur in Cairo to build a complete application model and test it before requesting massive engineering budgets, moving competition from the size of funding to speed of deployment and operational efficiency.

The evidence that will demonstrate the stage’s maturity does not require new promotional statements; it lies in publishing clear operational logs for government services that rely entirely on intelligent agents with error rates disclosed, announcing procurement matrices that support Arabic dialects, providing realistic figures for professional retention rates, and ensuring companies are transparent about tasks they have eliminated manually thanks to automation and intelligent operation.