Agentic AI in Dubai: How UAE Businesses Can Prepare for 2028

Dubai’s two-year agentic AI programme began in May 2026. For businesses planning around that announcement, 2028 is a useful horizon—but the preparation period is already underway. The first job is to choose a workflow, define what an agent may do and establish how its work will be judged. Dubai’s programme launch.

There is a concrete reason to act now. On 1 September, Dubai Chambers launched training aimed at more than 14,000 member companies of the Business Groups and Business Councils operating under Dubai Chamber of Commerce. That figure describes the intended audience; it does not mean 14,000 companies have completed training or deployed agents. September training announcement.

For a UAE business, preparation should produce a decision: which part of the operation could benefit from an agent, under whose authority, and with what evidence? Buying a tool does not answer those questions.

What Dubai’s agentic AI programme actually covers

Three announcements explain the direction and the present opportunity:

• 4 May 2026: Dubai launched a two-year private-sector transformation programme, including training tracks for business councils affiliated with Dubai Chamber of Commerce. Programme announcement.
• 11 June 2026: The executive plan set out ambitions to empower 295,000 companies, develop and deliver 100 specialised AI assistants over the next two years, and support the establishment of 50 agentic AI companies. These are programme targets, not counts of businesses already transformed. Executive plan.
• 1 September 2026: Dubai Chambers announced the training launch through Dubai Chambers Academy, its new e-learning platform. Training announcement.

Keep the federal initiative separate. The UAE Cabinet’s April announcement targeted agentic AI across 50% of government sectors, services and operations within two years, with phased implementation across ministries and federal entities. The 50% figure belongs to that government programme. Federal framework announcement.

This article uses 2028 as a planning horizon inferred from the Dubai programme’s May 2026 launch. It does not establish an individual company’s legal obligations or a company-specific deadline. Businesses elsewhere in the UAE can use the preparation method below while checking which local programmes apply to them.

What changes when AI can take action?

An AI assistant might summarise a customer request. An agentic workflow can use tools to retrieve information and carry out actions beyond a conversation. Microsoft’s architecture guidance describes that extension from conversation into external information and services. Hassan’s perspective on agentic AI and decision rights focuses on how much authority a workflow should have.

Consider a service enquiry. A proposed agent could read the request, consult approved service information, prepare a reply and suggest the next owner. Sending a price commitment or changing a customer record introduces a different decision. Each permission needs an owner and a clear boundary.

Use the simplest workable method. If fixed fields and rules can route a request reliably, test that option first. An agent becomes worth evaluating when the work needs interpretation across varying inputs. Neither approach earns wider access merely because the demonstration looks convincing.

Choose a workflow with a visible finish line

Start with work whose outcome you can inspect. The examples below are proposed pilot designs, not claims about deployed systems or guaranteed benefits.

New enquiry intake

• Initial scope: Summarise the request and propose a route using approved service information.
• Human decision: Review ambiguous cases, price commitments and final routing during the pilot.
• Useful completion: The reviewer accepts the route and can trace the supporting information.

Customer service follow-up

• Initial scope: Assemble the case history and draft a response.
• Human decision: Approve sending the response and resolving exceptions.
• Useful completion: The response addresses the request without unsupported promises.

Internal purchasing preparation

• Initial scope: Organise supplied requirements and draft a comparison of approved options.
• Human decision: Retain supplier choice, purchase approval and payment.
• Useful completion: The comparison is complete, traceable and useful to the buyer.

Pick one workflow and name its business owner. Record today’s handling time, correction work and completion quality before testing a change. Include awkward cases—missing information, contradictory records and requests outside the usual pattern—so the pilot has more to demonstrate than success on tidy examples.

For the first evaluation, use cases and information the team is authorised to use. Keep proposed actions reviewable. Increase scope only when the evidence supports the specific change being considered.

Agree on the evidence before expanding access

Write a short evidence record for the pilot. Its purpose is to make “this works” an assessable statement. These five questions can structure it:

1. What counts as finished? Define the business result, its owner and the cases excluded from the pilot.
2. What information may the agent use? Name approved sources, identify who maintains them and decide what happens when information is missing or contradictory.
3. What exact action is being approved? Show the proposed recipient, message or record change. Give the reviewer enough information to judge the action itself.
4. Where do exceptions go? Assign a person or queue, state when the workflow must stop and record why a case was handed over.
5. What would justify expansion? Agree on acceptable completion quality, review effort, recurring cost and reasons to pause. Set these before seeing the pilot results.

This is consistent with Microsoft’s responsible-AI guidance: establish information and access boundaries during design, decide which actions require approval, and retain human judgement for consequential actions. The guidance also calls for ongoing monitoring as agents and their operating conditions change. It provides design principles, not a guarantee of safe or profitable operation. Microsoft Learn: apply responsible AI.

At the end of the pilot, review accepted work, corrected work and escalated work together. A lower average handling time is useful only alongside an acceptable result. If the reviewer must repeatedly reconstruct the source information, include that effort in the business case.

Count review and rework in the business case

Here is a hypothetical example for screening a pilot. Every input is illustrative; these are not customer results, salary benchmarks or a vendor quote.

Assume 400 cases per month currently take eight minutes each. After introducing the proposed workflow, each case still needs three minutes of review and an average of two minutes of correction or exception handling. The net time released would be:

400 × (8 − 3 − 2) ÷ 60 = 20 staff hours per month.

If the business assigns an illustrative value of AED 100 to an hour of staff capacity, those hours represent AED 2,000 of capacity value. Subtract assumed recurring tool and support costs of AED 1,200, and the example leaves AED 800 per month before implementation costs.

That is not automatically a cash saving. The business still needs to show how the released capacity will be used, or which spending it can actually avoid. The example also excludes implementation effort, potential losses from errors and taxes; it is not a complete investment case.

Replace every assumption with observations from your own pilot. Count time spent checking, correcting, chasing missing information and maintaining the workflow. If that leaves little value, narrow the scope, simplify the process or stop the experiment.

A preparation plan from now to 2028

Treat the following as suggested decision points, not a guaranteed rollout schedule.

During the rest of 2026, establish the baseline. Assign an owner, choose one candidate workflow and document its current performance. Relevant member companies can investigate the Dubai Chambers training route. The announcement alone does not establish your organisation’s eligibility, course schedule or price; confirm those with the programme provider.

During 2027, build evidence from a bounded pilot. Compare completed work with the baseline. Review the mistakes and exceptions, as well as the time released. Keep permissions narrow while the team learns where interpretation helps and where fixed rules or human decisions work better.

Approaching 2028, decide what deserves wider use. Expand a successful workflow only when its owner can explain the result, recurring costs, unresolved failure cases and oversight required. If the evidence is weak, improve or retire the pilot. A calendar milestone does not make an unreliable workflow ready.

The useful output is a record of what your business has learned: what works, what remains uncertain and what it can responsibly do next.

Frequently asked questions

When did Dubai’s two-year agentic AI programme begin?

The private-sector programme was announced on 4 May 2026 and described as spanning two years. That supports a 2028 planning horizon; it does not give a business starting in September 2026 a fresh two-year countdown or establish an individual company’s deadline. May announcement.

Does every business workflow need an AI agent?

No. Evaluate fixed rules for stable, predictable steps and consider an agent where interpretation adds useful value. The objective is a dependable outcome with appropriate oversight. Hassan’s workflow and autonomy framework explains that decision in more detail.

Which actions should keep human approval?

For an initial pilot, retain approval for external commitments, spending and changes that are difficult to reverse. Define the exact action a person approves and the exceptions that must stop the workflow. Microsoft’s guidance supports human approval for consequential actions; the appropriate design still depends on the work. Responsible-AI guidance.

Where can a Dubai business look for training?

The September announcement identifies Dubai Chambers Academy and targets member companies of the Business Groups and Business Councils operating under Dubai Chamber of Commerce. Use that as a starting point to confirm access and course details directly. Training announcement.

If you are working through the connection between AI, operations and growth, get in touch with Hassan. Bring a specific workflow and the decision you are trying to make.