On 4 May 2026, Sheikh Hamdan bin Mohammed bin Rashid Al Maktoum, Crown Prince of Dubai, announced a two-year initiative to transition Dubai’s private sector to Agentic AI. The directive came under His Highness Sheikh Mohammed bin Rashid Al Maktoum, with three concrete instruments behind it: specialised training tracks for every business council under the Dubai Chamber of Commerce, dedicated incubators for Agentic AI companies, and dedicated funds to back the shift.
The stated goal: for Dubai to become the world’s leading city in adopting these technologies economically and commercially.
This is not aspirational positioning. This is a Crown Prince directive with budget, infrastructure, and a Chamber of Commerce execution arm. Over the next 24 months, every Dubai business owner will be asked some version of the same question: what are you doing about Agentic AI?
This article is a builder’s view on that question. Not a policy commentary. Not an opinion piece. A practitioner’s reading of what Agentic AI actually involves, what good implementation looks like, and what businesses should ask before adopting it.
What Agentic AI actually is
The official definition in the announcement is precise: self-executing and self-leading artificial intelligence. That phrasing is doing real work. It distinguishes Agentic AI from the kind of AI most businesses already use, like chatbots and image generators.
The simplest way to understand the difference:
- Generative AI answers questions. You ask, it responds. The interaction starts and ends with you in the loop.
- Agentic AI completes workflows. You give it an outcome. It plans the steps, calls the tools it needs, makes decisions along the way, and produces a result. The interaction is goal-driven, not turn-driven.
An accountant asking ChatGPT to summarise an invoice is using generative AI. An accountant telling a system “process this month’s invoices, flag anything over the approval threshold, draft the supplier emails, and queue payments for review” is using agentic AI. The difference is not in the model. It is in the architecture around the model.
The difference between generative AI and agentic AI is not in the model. It is in the architecture around the model.
Why this matters for Dubai businesses now
Three things shift the moment a directive like this lands.
The category got nationalised. Until last week, Agentic AI was a technical term that needed explaining to most clients. Now the Crown Prince has defined it for them. Every Dubai business owner has Agentic AI on their radar. Many will spend the next 24 months trying to figure out what it means for their business. The conversation has changed from “what is this” to “what should we do about this.”
Adoption became visible. When a directive like this exists, not adopting becomes a choice that has to be defended. Boards will ask. Investors will ask. Government tenders may begin to ask. The cost of inaction shifts from invisible to visible.
The market is going to fragment fast. By the end of this year, hundreds of agencies and consultants will be calling themselves Agentic AI providers. The vast majority will be repackaging chatbot work with new vocabulary. The technical reality of what makes a system genuinely agentic is not understood by most of them — and the gap will only become visible after a botched implementation. The risk is not in moving slowly. The risk is moving with the wrong partner.
What businesses should ask before adopting it
Two years from now, when the dust settles, the businesses that adopted Agentic AI well will have asked the same questions early. The businesses that adopted it badly will have asked them too late.
The questions worth asking — of yourself, and of any partner proposing to build it for you:
- What is the actual workflow we want the agent to take over? Not a vague capability. A specific business process with measurable inputs, defined steps, and a clear definition of success. If you cannot describe the workflow in plain language, the agent will fail to automate it.
- What does the agent need to read? Internal policies. Contracts. Past correspondence. Product manuals. Pricing sheets. RAG (retrieval-augmented generation) is what grounds the agent in your reality. Without it, you have a hallucination machine. With it, you have a domain expert.
- What systems does the agent need to act on? Email, calendar, CRM, ERP, document management, internal database, WhatsApp. Each system is an integration. Each integration has a complexity cost. The number and quality of integrations defines half the project.
- Where does the human stay in the loop? A good agentic system has explicit checkpoints where humans review, approve, or override. Without them you have an autonomous system you cannot defend in audit, in court, or to a regulator. Define the checkpoints before you build the system.
- Where does the data live? Cloud, hybrid, on-premise, air-gapped. Each tier has cost, capability, and compliance implications. For a small business serving private clients, cloud is fine. For a regulated entity processing personal data, it is not. Decide before you select tools.
- Who owns the system after delivery? Most agency models end at launch. Agentic systems require ongoing operation: monitoring drift, refining prompts, updating knowledge bases, integrating new data sources. Define stewardship up front, not as an afterthought.
- What does failure look like? Every agent will produce a wrong output sometimes. The right question is not “can we prevent this” but “what happens when it does, who detects it, and how is it corrected.” This is the audit trail conversation, and it is non-negotiable for any production deployment.
If a partner cannot answer these questions clearly, they are not ready to build a production agentic system for your business. They are selling a demo.
What real implementation costs and looks like
The most common question we are asked is: how much does it cost to actually do this.
The honest answer depends on the tier. A small business automating a single workflow is not the same engagement as a federal regulator deploying a sovereign agentic platform. Below is the realistic cost and timeline ladder for production-grade Agentic AI implementation, derived from systems Hashinclude has actually deployed:
| Tier | Description | Timeline | Investment |
|---|---|---|---|
| Compact | One agentic workflow, one integration, RAG over your existing documents. Cloud-hosted. Suitable for SMEs automating a real business process. | 6 weeks | AED 60K–90K |
| Standard | Multi-workflow, multi-integration. Mid-market enterprise. Includes basic governance, monitoring, and a stewardship retainer for the first year. | 8–12 weeks | AED 150K–350K |
| Platform | Enterprise-grade. Custom architecture, full integration surface, audit trail, and ongoing operation. Production reliability for thousands of users. | 12–20 weeks | AED 500K–1M+ |
| Sovereign | Government and regulated. Air-gapped, Arabic-first deployment with custom OCR, full compliance documentation, and on-premise infrastructure. | 16–24 weeks | AED 1.5M+ |
Anything below the Compact floor is consulting, not delivery. A real agentic system requires real engineering: chunking, embeddings, retrieval, orchestration, prompt design, integration testing, deployment, and observability. Below that floor, you are either buying vapor or losing your supplier money on every engagement. Both destroy trust.
Practitioner note — the compact tier is what most Dubai SMEs should look at first. A six-week, AED 60K–90K engagement that automates one real workflow and produces a measurable outcome is the right entry point for a small or mid-sized business. It is also the right scope to validate Agentic AI in your context before committing to anything larger. Start narrow. Prove value. Expand.
What we have already deployed
Hashinclude has been building agentic systems in production for some time. The work pre-dates the announcement, which is the only reason this article can be written from a builder’s perspective rather than a forecaster’s.
A short list of relevant deployments:
- UAE Department of Finance — AI Compliance Platform. Agentic compliance monitoring deployed into a federal regulatory environment. Every AI decision traceable to source documentation. Live in 2024.
- PoliSync — Enterprise Insurance Quoting Engine. 127 production API endpoints. A hybrid agentic system combining intelligent document processing with deterministic rule engines. Three weeks from first commit to production v1.
- Anas AI — Conversational Agent on hashinclu.de. A live agentic system trained on 12+ years of Hashinclude project knowledge. Speaks to clients, qualifies leads, generates proposals, and emails them — autonomously.
- Ekamati — Government Insurance Platform, Ministry of Interior UAQ. Government-grade document intelligence platform processing insurance workflows. Live since 2023.
None of these systems were built last week in response to the announcement. They are the basis of the perspective in this article.
What the next 24 months will look like
Three predictions, offered with the usual caveats about prediction.
The market will split into builders and packagers. Within months, the term “Agentic AI” will be everywhere. The technical reality of what makes a system genuinely agentic — planning, tool use, memory, decision-making with explicit checkpoints — will be understood by very few of the providers using the term. Buyers will need to learn the difference quickly.
The Compact tier will become the dominant entry point. Most Dubai SMEs will not start with a platform overhaul. They will start with one workflow. Document processing, customer triage, internal knowledge query, lead qualification. The providers who can deliver a real outcome at the Compact tier in six weeks will be the ones who get repeat business.
Stewardship will become the differentiator. The first wave of agentic deployments will work. The second wave will degrade silently. The difference between systems that keep working and systems that quietly fail is ongoing operational discipline: monitoring, drift detection, prompt refinement, knowledge base updates, integration maintenance. The providers who treat agentic systems as software you operate, not software you ship, will be the ones whose deployments are still working in 2028.
The honest closing
Agentic AI is real. It works. It produces measurable business value. It is also fundamentally an engineering discipline, not a product category, and that distinction will matter more as adoption accelerates.
The Crown Prince’s directive is going to compress two years of mainstream adoption into the next 24 months across Dubai. That is good for businesses willing to think clearly about what they are adopting. It is a real risk for businesses that are not.
If you are a Dubai-based business trying to figure out what to do about Agentic AI, the answer starts with the seven questions above. If you can answer them, you are ready. If you cannot, the honest first step is not buying a system. It is mapping the workflow you actually want to automate and the systems you actually want to integrate. Until that exists, no amount of technology will help.
If you would like to talk to someone who has actually built and operated these systems in production, we are here. We listen first. If we are not the right fit for what you are trying to do, we will tell you so.