Introduction

Most AI education and tooling currently in use across the GCC is translated, not native. The interface is Arabic. The prompts are Arabic. The examples are from American small businesses. The regulations are European. The case studies are from companies that have never operated under GCC procurement rules. The audience reads Arabic and recognises that they are reading a foreign document in their own language. This pattern is not a niche complaint. It is a strategic vulnerability. Translation is not enough. The Gulf is large enough, fast-moving enough, and culturally specific enough to require its own institutional architecture for AI — including the curriculum, the regulatory references, the case studies, and ultimately the underlying models themselves. Building that architecture is the work of the next ten years. It will not happen by accident. It will not happen by translation. It will happen because someone made a deliberate choice to design Arabic-first.

What 'first' actually means

Arabic-first does not mean Arabic-only. The Gulf has a deeply multilingual operating environment, and an Arabic-first system that cannot communicate in English will fail in practice. Arabic-first means the design starts in Arabic. The vocabulary is selected in Arabic. The examples are written in Arabic about Arabic-speaking situations. The regulatory references are to GCC regulators. The case studies are from companies the audience has actually heard of. The English version comes second, derived from the Arabic, not the other way around. The reason this matters is that translation flattens. A document translated from English into Arabic carries the assumptions of the English original even when the words are Arabic. The assumptions are foreign to the reader, even when the language is not. The Arabic reader recognises the foreignness immediately and discounts the content accordingly. The Arabic-first reader does not. The first cost is paid in design. The second cost is paid in audience trust, which is harder to recover.

The translation gap

  • Vocabulary mismatch — translated technical terms often have multiple equivalents in Arabic, none of which match the working vocabulary of the local industry.
  • Example mismatch — translated training materials reference Silicon Valley start-ups, US regulatory regimes, and small-business workflows that are not the Gulf operating environment.
  • Regulatory mismatch — compliance examples reference GDPR, HIPAA, and SOX. The Gulf operates under different frameworks, with different agencies, with different enforcement patterns.
  • Cultural mismatch — the unstated assumptions about business relationships, hierarchy, and decision-making baked into translated content are American or European. They do not describe how the audience's organisation actually works.
  • Pacing mismatch — translated content is typically written for fluent English speakers consuming long-form material. Arabic-first content respects different reading rhythms and shorter average reading sessions.

An AI course, run twice

I ran an AI adoption course in 2025 for a public-sector audience in the Gulf. The first cohort received the standard, English-translated curriculum that the consulting world had produced for the region. The feedback was polite but quiet. The exercises were completed but not engaged with. The post-course surveys produced satisfaction scores in the high seventies — respectable, not transformative. We rewrote the course Arabic-first. We started with the vocabulary the audience used in their actual work. We replaced the American small-business examples with Gulf-specific case studies — a UAE federal authority deploying AI for permit processing, a Saudi holding group automating accounts payable, a Qatari hospital triaging patient communications. The compliance references shifted to the regional regulators the audience already engaged with. The post-course surveys for the second cohort produced scores in the low nineties. More importantly, the participants quoted the material back to me in their own meetings. The difference was not the technology. The technology was identical. The difference was that the second cohort recognised themselves in the material. Recognition does most of the work that translation cannot.

On vocabulary and audience

"You cannot teach a discipline in a vocabulary the audience experiences as foreign."

— a CIO who reviewed the rewrite and asked me why we had not started this way

Why this matters in 2026 specifically

The window for this work is now. Three forces are converging. First, AI capability is becoming a national priority across the GCC, with the largest Gulf states making infrastructure investments at scale and recruiting talent globally. Second, the open-weight model ecosystem has matured to the point that Arabic-first fine-tuning is technically feasible at reasonable cost. Third, the institutions that absorb AI in the next five years will compound their position relative to the institutions that wait. The choice between translated AI and Arabic-first AI is therefore not a matter of taste or branding. It is a question of which institutions are positioned to lead the region's AI capability ten years from now, and which are positioned to be customers of capability built elsewhere. Both are valid positions. Only one of them produces national capability. The other produces a procurement bill. The decision tends to be made by default, which is why most regions end up customers.

Where Arabic-first design lands in practice

  • Training and education — the single highest-leverage application. Most of the GCC's AI workforce will learn from materials produced this decade. The materials should not feel foreign.
  • Public-sector AI products — government services in the GCC are delivered in Arabic. AI products embedded in those services have to work natively in Arabic, not just translate output at the edge.
  • Customer-facing AI in regulated industries — finance, healthcare, hospitality all operate primarily in Arabic with their customer base. Translated AI here is a service failure.
  • Internal enterprise AI — even where the technical team works in English, policies, audit trails, and compliance documents have to land in Arabic for the institution to defend them.
  • AI ethics and governance frameworks — imported ethics frameworks carry imported assumptions. The Gulf will need its own.

Common mistakes

  • Translating English content first and then 'reviewing for Arabic accuracy'. The structural assumptions remain English.
  • Hiring translators to produce Arabic content instead of hiring Arabic-native writers and subject-matter experts.
  • Treating Arabic-first as a marketing exercise rather than a design discipline.
  • Building Arabic interfaces around English-trained models without evaluating the Arabic-language performance of those models.
  • Imagining that bilingual users will tolerate weak Arabic. They will switch to English, and the Arabic product will atrophy.

Closing

The Gulf has a window. The next decade will not produce another moment in which AI capability is this fluid, this fast-moving, and this open to regional differentiation. The institutions that design Arabic-first now will compound. The institutions that translate will fall behind, slowly at first and then suddenly. This is not a language question. It is a capability question. The language is the most visible part of a deeper choice about who the AI is built for and who it makes powerful. My practice exists in part to make this case to organisations that have the resources to act on it. I am betting twenty-five years of credibility on the belief that this will be the most consequential structural choice of the decade. It is also one of the few choices that is still genuinely available. The institutions that take it will compound their way into a different position. The institutions that do not will pay a procurement bill the size of an industry.