Beyond Chatbots: Why Agentic AI Is the Strategic Frontier

The public conversation about AI in Saudi Arabia — as elsewhere — has been dominated by large language models, chatbots, and content generation. These are real applications with genuine value, but they represent the first wave of AI deployment, not the strategic frontier. Agentic AI — systems that take sequences of actions to accomplish multi-step goals, interact with external systems and environments, and operate with degrees of autonomy that go beyond single-turn response generation — is the second wave, and it is already arriving in Saudi Arabia’s most ambitious infrastructure programs.

The distinction matters enormously for investors and policymakers. A chatbot that answers customer service questions in Arabic creates efficiency; an AI agent that autonomously manages supply chain procurement across thousands of vendors, negotiating prices and updating inventory systems, creates structural competitive advantage. The compute requirements for agentic AI are also substantially different — multi-step reasoning chains, tool use, environmental feedback loops, and persistent memory require far more sustained compute than single-turn inference. Saudi Arabia’s $77 billion AI compute buildout is therefore better understood as agentic AI infrastructure investment than as chatbot infrastructure investment.

SDAIA’s NCAI: The Research Foundation

The National Center for AI (NCAI), SDAIA’s applied research arm, has published research on agentic AI frameworks with a specific focus on Arabic-language agentic systems. The research agenda includes multi-agent collaboration (multiple AI agents working in parallel on complex tasks), agentic systems for government service automation, and Arabic-language tool-use models that can interact with Saudi government digital infrastructure.

NCAI’s work is particularly important because it defines the technical standards and evaluation frameworks for Saudi agentic AI deployment. When government entities procure agentic AI systems, they will increasingly reference NCAI benchmarks for safety, accuracy, and Arabic language capability. Companies that engage with NCAI’s research agenda — through research partnerships, benchmark contributions, or joint publications — are building the relationships and technical credibility that translate to procurement advantages.

The NCAI’s agentic AI research is also feeding into Saudi participation in international AI safety discussions. Saudi Arabia’s position in the global AI governance conversation — hosting AI safety discussions at FII, engaging with the UK AI Safety Institute’s framework — is partly informed by NCAI’s technical work on agentic system risks and evaluation.

Humain’s Agentic AI Product Roadmap

Humain’s positioning as a “full-stack AI infrastructure provider” implies an agentic AI layer — not just compute and model hosting, but AI agents capable of operating Saudi-specific workflows. The Humain-NVIDIA partnership (one of the most significant in the $77B buildout) is particularly relevant here: NVIDIA’s Agent Intelligence initiative and NIM (NVIDIA Inference Microservices) platform are specifically designed for deploying production agentic AI applications.

Humain’s product roadmap, inferred from public announcements and partner statements, suggests several agentic AI priorities: government service automation agents (integrating with the Absher national identity and services platform), enterprise workflow automation for Saudi corporations, and sector-specific agents for energy, healthcare, and financial services. The stc-Humain JV creates a telecommunications-AI integration that could enable agentic AI applications across stc’s 22 million Saudi subscribers.

The key question for Humain’s agentic AI ambitions is the availability of high-quality Arabic-language foundation models capable of reliable tool use. Current Arabic LLMs, including Allam, have demonstrated strong language understanding but have not yet matched the tool-use reliability of leading English-language models. Closing this gap is a prerequisite for deployable Saudi agentic AI at the standards that enterprise and government customers require.

Allam as Backbone for Saudi AI Agents

SDAIA’s Allam model (34B parameters, 8 PB Arabic training data, trained on 5,000 Blackwell GPUs at the SDAIA Hexagon Data Center) is designed explicitly as a platform for downstream applications, not just a standalone chatbot. The model’s architecture and training data selection were made with agentic use cases in mind: strong Arabic instruction following, tool call formatting, and structured output generation — all prerequisites for agentic AI deployment.

For Saudi-specific agentic applications, Allam provides capabilities that GPT-4o or Claude cannot replicate: deep familiarity with Saudi government service terminology, Islamic legal and financial concepts, Saudi dialect and formal Arabic fluency, and an understanding of Saudi cultural and institutional context that emerges from large-scale Arabic training data. An AI agent helping a Saudi citizen navigate the Absher platform, applying for permits, or accessing government services needs precisely this contextual grounding.

The Humain-SDAIA collaboration on Allam’s agentic extensions is ongoing, with expected releases of Allam-Agent variants through 2025–2026. These will compete with Arabic-tuned versions of international models (GPT-4o with Arabic system prompts, Claude Sonnet with Arabic fine-tunes) for enterprise agentic deployment in Saudi Arabia.

NVIDIA Physical AI and Omniverse: Robotics and Industrial Agents

The NVIDIA-Humain partnership extends well beyond language model inference. NVIDIA’s physical AI platform — combining the Omniverse simulation environment, Isaac Robotics framework, and Metropolis vision AI — represents the foundation for agentic AI in physical systems: manufacturing robots, autonomous vehicles, smart building management, and industrial process control.

Saudi Arabia has specific strategic interest in physical AI. The Vision 2030 industrial diversification agenda (through NEOM’s Oxagon industrial zone, King Salman Energy Park, and the NIDLP industrial cities) involves building modern manufacturing capacity where AI-driven automation is built in from the start rather than retrofitted onto legacy systems. NEOM’s Oxagon is designed as an “AI-first industrial city” — a testbed for physical AI applications at scale.

NVIDIA’s Omniverse platform at NEOM enables “digital twin” simulations of industrial processes — training physical AI agents in simulation before deploying them in physical systems, reducing the safety risks and data collection costs of in-situ AI training. The compute requirements for large-scale Omniverse simulation are substantial (among the most GPU-intensive workloads in enterprise AI), which connects directly back to the Humain/NEOM compute infrastructure investment rationale.

Lucid and Ceer: Autonomous Vehicle Agentic AI

Saudi Arabia’s ambitions in autonomous vehicles represent one of the most concrete agentic AI programs in the Kingdom. Two programs merit specific attention:

Lucid Motors (majority-owned by PIF following a SPAC transaction) is developing next-generation electric vehicles where the autonomous driving stack is a core product differentiator. Lucid’s AI-driven driving assistance and autonomous capability development is proceeding in partnership with NVIDIA’s DRIVE platform (another NVIDIA-Saudi connection). The Saudi market context — high temperature operation, desert driving conditions, specific road infrastructure — requires localized AI training that Lucid’s Saudi ownership facilitates.

Ceer is Saudi Arabia’s first domestic electric vehicle brand, a joint venture between PIF and Hon Hai Precision (Foxconn). Ceer is developing connected and autonomous vehicle features explicitly for the Saudi market. Ceer’s strategic position is distinct from Lucid: while Lucid is a premium global brand, Ceer is designed as a volume Saudi (and GCC) brand where affordability and Saudi-specific features — including Arabic voice control, Saudi navigation integration, and desert operating performance — are the differentiators. The AI driving and autonomy stack for Ceer is being developed with regional focus.

The Saudi autonomous vehicle ambitions connect to agentic AI investment in a specific way: both programs require substantial simulation, testing, and validation compute — exactly the kind of workload that justifies Saudi Arabia’s Blackwell GPU cluster investments in ways that go beyond language model inference.

NEOM’s Tonomus: Smart City Agentic AI

Tonomus, NEOM’s technology and digital infrastructure company, is arguably Saudi Arabia’s most ambitious agentic AI deployment — operating AI agents at city scale. The concept is that The Line, MIRRORED, and Oxagon will be managed by AI systems that continuously optimize energy distribution, transportation flows, predictive maintenance of infrastructure, and public safety monitoring.

The agentic AI architecture for a smart city of NEOM’s scale requires thousands of specialized agents operating in coordinated multi-agent systems: an energy management agent, transportation orchestration agents, predictive maintenance agents for individual systems, security monitoring agents, and orchestration layers that coordinate them. This is not a single AI product but an AI operating system for urban infrastructure.

Tonomus has partnerships with Ericsson (smart city connectivity), Cisco (networking), and multiple AI vendors. The primary AI infrastructure runs on NEOM’s dedicated compute capacity, which is being connected to the broader Saudi AI infrastructure through data center interconnects. The Tonomus agentic AI program is one of the few places in the world where the “agentic AI at city scale” concept is being built rather than merely planned.

Agentic AI for Saudi Government Services

The most scalable near-term agentic AI opportunity in Saudi Arabia is government service automation. Saudi Arabia has approximately 3,000 distinct government services available digitally through Absher, Etimad, and related platforms. The Vision 2030 government efficiency agenda targets dramatic reductions in service completion times and human touchpoints — mandates that point directly to agentic AI deployment.

An AI agent that can, on behalf of a Saudi citizen: gather required documents from connected government databases, verify identity through the national ID system, complete form submissions with contextually appropriate responses, check application status, and proactively notify the citizen of requirements or decisions — this replaces multiple human processing steps and creates genuinely transformative government service efficiency. The Arabic language requirement and deep knowledge of Saudi government processes makes this a domain where Allam-based agents have structural advantages over international alternatives.

SDAIA has piloted government service AI agents through its national e-government programs, with early deployments at the Ministry of Commerce and the Ministry of Human Resources. The scale deployment — integrating agentic AI across all major government services — is expected to be a significant program announcement at LEAP 2026.

Compute Requirements: Why Agentic AI Justifies the $77B

The compute requirements for agentic AI are substantially higher than for chatbot or content generation applications. A single agentic workflow might involve dozens of LLM inference calls (planning, reflection, tool use, verification), code execution, database queries, and environmental feedback loops. At enterprise or government scale — millions of concurrent agentic workflows serving Saudi citizens and businesses — the compute demand can easily be 10–50x higher than an equivalent conversational AI deployment.

This is the under-discussed justification for Saudi Arabia’s $77 billion AI compute commitment. The frequently asked question — “does Saudi Arabia need $77B of AI compute just for Arabic chatbots?” — misunderstands the target workload. The Humain infrastructure is being built for the agentic AI generation, where Saudi Arabia’s position as a regional AI hub requires compute capacity at a scale that can support both Saudi-originating agentic workloads and neighboring country inference demand routed through Saudi infrastructure.

Agentic AI Safety and Saudi Governance

Agentic AI introduces safety challenges that go beyond the bias, accuracy, and privacy concerns of static LLM deployments. An AI agent that can take external actions — send emails, modify database records, execute financial transactions, submit government forms — has failure modes with real-world consequences that a text-generating chatbot does not. A misunderstood instruction, a reasoning error in a multi-step chain, or a prompt injection attack on an agent’s tool-use pathway can produce irreversible actions.

SDAIA’s AI ethics framework and NCAI’s safety research have begun addressing agentic AI-specific risks, but the governance frameworks are still early. Saudi Arabia’s position as a “Year of AI” 2026 country creates pressure to deploy agentic AI at scale before governance frameworks fully mature — a tension that international observers in AI safety should monitor.

The Saudi approach to agentic AI safety is likely to be pragmatic rather than precautionary: deploy in lower-risk domains first (scheduling, information retrieval, document drafting), build human-in-the-loop requirements for higher-stakes decisions (financial approvals, medical recommendations, legal filings), and develop post-deployment monitoring requirements rather than pre-deployment certification. This approach prioritizes deployment speed while building safety evidence — a reasonable balance given both the governance capacity constraints and the competitive pressure from faster-moving AI deployments in peer countries.

International Agentic AI Competition: The Saudi Context

Saudi Arabia’s agentic AI ambitions compete with parallel programs in the UAE (Abu Dhabi’s Technology Innovation Institute has an active agents research program), Singapore (IMDA’s national AI program includes agentic systems for government service delivery), and South Korea (NIPA’s AI R&D program includes agentic AI for e-government). The competition is for both technical primacy and for the role of regional hub — which country’s agentic AI infrastructure becomes the platform on which Gulf and broader Middle East applications are built.

Humain’s infrastructure scale ($77B commitment, multi-GW compute target) gives Saudi Arabia a structural compute advantage over UAE and Singapore competitors. The question is whether compute advantage translates to agentic AI deployment leadership, or whether smaller, more agile jurisdictions build working agentic systems faster on less compute. Singapore’s pragmatic deployment model — smaller scale, faster iteration, rapid feedback from government service operations — may produce learnable agentic AI systems faster than Saudi Arabia’s infrastructure-first approach.

The resolution to this competition will significantly shape the regional AI ecosystem for the 2030s. If Saudi Arabia’s Humain infrastructure becomes the platform on which Gulf agentic AI is deployed, it locks in both revenue and strategic influence. If the UAE’s G42-MBZUAI-ADQ ecosystem or Singapore’s IMDA programs move faster to deployment, they capture the relationship and data advantages that compound over time. This race is genuinely undecided in 2025, which is precisely why the Humain launch represents an urgency play by PIF — investing at speed and scale to foreclose competitive options.

Workforce Implications: The Automation Paradox

Saudi Arabia faces a specific tension around agentic AI: the Vision 2030 Saudization agenda aims to increase Saudi employment across all sectors, while agentic AI systems automate precisely the administrative, processing, and routine cognitive tasks that absorb much of the current Saudi workforce in government and semi-government entities. The automation paradox is that the Kingdom’s AI ambitions, if fully realized, may structurally reduce demand for the kinds of roles that Saudization targets.

The resolution to this tension — partial, contested, ongoing — is the bet that AI-augmented Saudi workers will be more productive than replaced by AI, and that new AI-native roles (AI trainers, agents supervisors, AI product managers, AI safety evaluators) will absorb some of the displaced workforce. Whether this bet pays off at the scale and speed Saudi Arabia requires is one of the most consequential questions in the Kingdom’s economic transformation.

The structurally important observation is that Saudi Arabia has a narrow window — roughly 2025–2030 — during which the oil revenue base is still sufficient to fund the transition costs. Agentic AI automation that displaces government workforce without a functional alternative employment sector creates social stability risk that the Vision 2030 economic diversification program is designed to prevent. The pace at which agentic AI deployment is scaled in Saudi government services will be calibrated, at least partly, against the pace at which the private sector is absorbing Saudi workers in AI-native roles. This makes Saudi agentic AI deployment more politically managed than commercially optimized — a nuance that pure technology analysts often miss.