SDAIA sets an AI-first design rule for Saudi public services

Saudi Arabia is moving toward an approach in which artificial intelligence is considered from the outset when public services and products are designed, rather than added to them afterwards. Abdullah Alkhodairi, chief strategy officer at the Saudi Data and Artificial Intelligence Authority, described the shift on 2 September as he opened the third day of DeepFest, the AI event running alongside LEAP 2026 in Riyadh. He framed the principle as putting value before technology: applying AI where it offers a genuine advantage rather than applying it uniformly. And he set the case for it against an adoption figure, citing Stanford's latest AI Index Report finding that more than 80% of surveyed employees in Saudi Arabia use AI regularly in their daily work. On that evidence, he argued, the challenge has moved from encouraging people to experiment toward converting widespread individual use into institutional value. The distinction is more consequential than it may sound, because the two problems have different solutions. High individual adoption means employees are using AI tools on their own initiative for their own tasks. Institutional value means the organisation has changed a process, retired a step, or measurably shortened a cycle. The first can be achieved by licensing a product. The second requires redesigning how work moves through an agency, which is a considerably slower undertaking. Design-time AI is the mechanism for closing that distance. A service that has AI attached to it after launch inherits whatever data model, approval chain and interface it already had; the AI can only improve steps that were designed for a human doing them manually. A service designed with AI assumed from the start can be architected differently — different data capture, different points where a decision is made, different places where a person intervenes. It also fits the shape of the governance apparatus SDAIA already has in place. Its AI Adoption Framework, released in November 2025, sets a mandatory baseline for public sector entities across five pillars: data governance, model accountability, transparency, human oversight and risk management. Those five are considerably easier to satisfy at design time than retrospectively. Documented data lineage, a named model owner and a defined human intervention point are all straightforward to build into a new service and expensive to retrofit into a live one. For vendors, the implication runs through procurement. Saudi Arabia's public sector is the dominant technology buyer in the Kingdom, and a framework binding on government entities reaches suppliers through contract terms rather than through statute. A design-time expectation extends that further up the process: it applies before a tender is written, at the point where an agency decides what it is buying at all. A supplier arriving with a finished product to bolt onto an existing service is answering a different question from the one being asked. The adoption figure is worth holding carefully. Stanford's AI Index draws on surveys, and a self-reported measure of regular AI use captures a wide range of behaviours — from occasional drafting assistance to genuine workflow dependence. What it establishes reliably is direction and scale: individual familiarity with these tools inside Saudi organisations is high and is not the constraint. That is a materially different starting position from most public sectors, and it explains why the emphasis has moved to the institutional layer. DeepFest ran alongside LEAP across the same four days at the Riyadh Exhibition and Convention Centre in Malham, and its programme has been weighted toward demonstration rather than announcement. The design-rule framing is the more durable output: an announcement commits a budget, while a design principle shapes every service specified after it.