‘AI-driven’ vs a real AI agent: questions to ask a Dubai agency

Open ten agency websites in Dubai and nine will say ‘AI-powered’, ‘AI-driven’ or ‘AI-first’ somewhere above the fold. Very few can show you a system behind the words. The phrase has become a design element — a badge, not a build. For a buyer, that is a problem: the label tells you nothing about what is actually automated, what it connects to, or whether it works in the languages your customers use.

The good news is that the gap between decorative AI and a real AI agent is easy to expose. You do not need to be technical. You need seven questions and the confidence to keep asking until you get a specific answer. Here they are, with what a good answer sounds like — and what a red flag sounds like.

Why this matters more in 2026

From 15 January 2026, Meta restricts the WhatsApp Business Platform to task-specific automation — booking, lead qualification, order status and similar defined jobs — rather than open-ended general chatbots. That single rule change is a useful filter. It means the compliant, durable way to run AI on WhatsApp is exactly the way a real agent should already be built: narrow, purposeful, tied to an outcome. Given that WhatsApp reaches roughly 87% of UAE internet users, an agency that still pitches a general ‘chat with our AI about anything’ bot is selling you something that is both less effective and, on the channel that matters most here, no longer permitted.

The seven questions

1. Can I message a live demo bot right now — not a screenshot?

This is the single most revealing question, so ask it first. A real agent exists as a running system, which means the agency can hand you a number or a link and let you talk to it this minute.

A good answer: ‘Here is a WhatsApp number — message it and watch it qualify you, then offer a booking slot.’ You interact, it responds, it does something. A red-flag answer: a screenshot, a Loom recording, a slide that says ‘AI-driven pipeline’, or ‘we can build you a demo once you sign’. If the only proof is an image, the AI is a claim, not a capability.

2. What exactly is automated?

‘AI’ is not a task. Push for the specific jobs the system performs so you know where the human effort actually goes.

• Lead response — does it reply to a new enquiry within seconds, on the channel it arrived on?
• Qualification — does it ask the right questions and score or route the lead?
• Follow-up — does it chase a quiet lead on a schedule without someone remembering to?
• Reporting — does it log outcomes and surface them, or does that stay manual?

A good answer names these steps and tells you which are automated today versus on the roadmap. A red-flag answer stays abstract — ‘it handles your marketing with AI’ — and cannot draw the line between what the machine does and what a person still does by hand.

3. Which systems does it integrate with, by name?

An agent that cannot read or write to your tools is a novelty. Real automation lives or dies on integration, and integration is specific.

A good answer names them: ‘It writes qualified leads into HubSpot, books into Google Calendar, and posts a daily summary to your Slack.’ A red-flag answer waves at ‘your CRM’ without naming one, or promises integrations that turn out to be a spreadsheet someone updates by hand. If nothing is named, nothing is connected.

4. Is it task-specific or a general chatbot?

Tie this directly to Meta’s 2026 rule. A task-specific agent is designed to complete defined jobs — qualify, book, fetch an order status — and hand off cleanly when it hits the edge of its job. A general chatbot tries to answer anything, drifts off-topic, and, on WhatsApp, is now non-compliant.

A good answer: ‘Each agent does one job well, and there is a specific one for lead-qual and another for support.’ A red-flag answer: ‘It can talk about anything’ — which sounds impressive and is precisely the pattern that both underperforms and falls foul of the platform rules.

5. What languages does it operate in natively?

In the UAE this is not a nice-to-have. Your customers message in Arabic, English and Hindi, often switching mid-conversation. ‘Native’ means the agent understands and replies correctly in each — not that it runs everything through a translation layer that mangles tone and detail.

A good answer: ‘It operates in Arabic, English and Hindi, detects the language of the incoming message and replies in the same one.’ A red-flag answer: ‘English, and we can add translation later.’ In this market, later means lost enquiries.

6. Who is the human in the loop, and when do they take over?

The best automation is honest about its limits. You want to know exactly where the machine stops and a named person steps in — because that boundary is where deals are won or dropped.

A good answer defines it: ‘The agent handles first reply, qualification and booking. The moment a lead is high-value or asks something outside its job, it flags a human and hands over the full conversation.’ A red-flag answer is either extreme — ‘it does everything, no humans needed’, or ‘a person checks all of it eventually’, which is just manual work with an AI sticker on it.

The question is never ‘is it fully automated or fully human’. It is ‘does the machine do the repetitive work fast, and does the right person own the moment that closes the deal’.— Awais Tahir Khan, founder, Agile Services

7. How do you measure and report the automation’s impact on revenue?

If an agency cannot connect the agent to money, they are selling activity, not outcomes. Automation should be measured against the same numbers your sales team is: response time, qualified leads, booked meetings, conversion, revenue influenced.

A good answer: ‘You get a report showing leads handled, how many were qualified, how many booked, and what closed — so you can see the pipeline the agent produced.’ A red-flag answer counts messages sent or ‘engagement’ with no line back to revenue. Volume is not value.

What a real answer looks like, put together

Ask all seven and a pattern emerges quickly. A decorative ‘AI-driven’ agency gives you adjectives, screenshots and roadmaps. A real one hands you a number to message, names the systems it writes to, shows you the Arabic and English replies, tells you who takes over and when, and reports the result in leads and revenue.

Real automation is task-specific, integrated, multilingual and human-in-the-loop. Those four are not marketing positions — they are the properties of a system that actually runs. And an agency that has built one is glad to show it, because the demonstration is the strongest pitch it has.

So make the demo the first meeting, not the last. If ‘AI-driven’ is real, it will be sitting in your chat window within minutes.