From licenses to rebuilding work, why IBM puts governance before AI agents in Saudi Arabia

The question Saudi organisations are asking today is not which AI tool they will buy, but what needs to change for that tool to actually work within their operations. This is the line drawn by Mohammed Ali, senior vice-president and head of IBM Consulting (the consulting arm of IBM), in his interview with Middle East on the sidelines of LEAP 2026 in Riyadh: moving to an enterprise that operates AI at scale starts with processes, data and governance, not with distributing new licenses to employees.

Data is not a technical detailAccording to Ali, data readiness remains one of the biggest barriers to scaling, because many organisations have historically not been built to feed their data directly into AI systems. What is needed, in his description, is to extract the data and make it available in a format that systems can use, then re-engineer the process itself so that humans and digital capabilities operate in a single operating model. This is an important point because buying a model or a chat interface does not automatically change decision-making or workflow within the organisation.

The piece distinguishes two paths that should not be conflated. Riyadh Air, the Saudi airline, started from a new environment, away from decades of legacy systems, and used IBM Consulting Advantage (a solution from IBM Consulting) and AI agents to build software and address new requirements. The interview notes that AI is also used there for customer service, field operations and financial planning. The existing organisation, however, faces a different problem: it does not start from a blank page, but from applications and data accumulated over years.

Updating what lies beneath the toolFor this, Ali cited IBM’s work on updating hundreds of applications for stc (Saudi Telecom Company), re-architecting them to expose APIs and make the data needed for AI applications available. He also discussed the use of AI-for-operations, or AIOps, to analyse operational information in the network infrastructure. The message here is practical: if the interfaces, data and processes remain unprepared, adding a smart layer on top will not resolve the core bottleneck.

The issue becomes even more sensitive when agents shift from processing documents to working near industrial facilities and infrastructure. Ali places human intervention in the design and says that some decisions will not be left to an automated process that runs on its own. He does not present this as a hesitation to use AI, but as an intentional delineation of tasks the system can perform and those that require human review.

The vision shifts to cost and risk controlIn the IBM Consulting (the consulting arm of IBM) experiment Ali described, about 6,000 AI agents are managed through a unified layer, with 35 agents operating simultaneously at the time of the interview. He explains that governance gives management the ability to see what the agents are doing, and to monitor usage, risk and cost. This puts governance before scaling: it is not a compliance step added to the project, but a prerequisite for knowing whether resources are being consumed on valuable work.

For the region, and especially for teams that build or operate enterprise systems in Saudi Arabia, this proposition adds a clearer benchmark for the required skill set. It is not enough for an employee to know how to use a ready-made application; they also need to work with data, develop systems and redesign processes. In this context, Ali said IBM is aiming to expand its skill programmes from roughly 500,000 Saudis to one million beneficiaries in technology and AI, with a centre created in partnership with the University of Al Baha that includes local students and talent working on projects linked to the Saudi economy.

The takeaway is not that every organisation should follow the Riyadh Air model; the interview itself confirms that building a new company and updating an old one are different challenges. However, it offers a useful test before deploying any agent: are the data usable, has the process been redesigned, and can management see what the system is doing and control its risk and cost? Without these answers, the experiment remains an add-on tool, not an operational transformation.