The entire architecture of digital commerce rests on one assumption: that a person initiates a transaction – a consumer browses, selects, confirms and pays. Every layer of security, authentication and fraud prevention is calibrated to that sequence.

Agentic AI, systems that can reason through a complex instruction and plan what needs to happen and act autonomously with minimal human input, disrupts this model at its foundation.

This is not a hypothetical shift, but one already well on its way to impacting commerce. In the UAE, 70% of consumers already use AI tools when shopping – a 44% increase on 2024 figures, according to Adyen’s 2025 Retail Report. In travel, 68% of UAE consumers used AI to book holidays in 2025 – a 57% year-on-year rise.

“What makes agentic AI different from the AI tools we’ve seen so far is that it doesn’t just respond or recommend,” says Daumantas Grigaravicius (pictured, right), head of Middle East at Adyen, speaking to MEED. “It can take a complex instruction, reason through it, plan what needs to happen and act autonomously on a user’s behalf.”

In retail, he says, that means AI agents handling the entire customer journey – discovering products across multiple platforms, comparing prices, applying discounts and completing the purchase – based on a single instruction.

In hospitality, an agent could plan and book a trip end-to-end, adjusting plans if flight schedules change. In financial services, it could monitor accounts and time international transfers to secure better exchange rates.

From browsing to delegating

When AI agents take over the discovery process, the consumer will shift from navigating individual apps and websites to setting preferences that inform how an AI agent acts.

“The customer journey becomes less about navigating touchpoints and more about setting preferences and letting AI handle execution,” Grigaravicius says. For UAE consumers who already value convenience and efficiency, this is a natural evolution.”

AI will select products based on data: price, quality metrics, delivery times and sustainability scores – replacing the current advertising, social media and consumer algorithms.

“This puts pressure on merchants to compete on substance rather than just marketing appeal,” notes Grigaravicius, though there will remain a distinction between the routine and the personal.

“Consumers will still want to be involved in choices that carry emotional weight,” he says. “What changes is that the mundane, repetitive aspects get automated, which makes the whole process feel far less cluttered and more streamlined.”

The merchant’s dilemma

For service providers, the challenge is clear: their offering needs to be easy for AI agents to find; their systems have to connect smoothly; and their value proposition needs to deliver.

The risk is that if the entire customer journey is contained within a chat interface, merchants could find themselves cut off from the relationship they have spent years building.

“There’s a real concern that hard-won brands could be reduced to commodities, perhaps just a featureless API endpoint in a bot’s decision-making logic,” says Grigaravicius.

The industry has confronted versions of this anxiety before. The leap from desktop e-commerce to mobile prompted similar fears of disintermediation.

“Mobile didn’t replace digital storefronts; it added a powerful, specialised channel for high-intent customers,” he says. “Agentic AI is likely to follow a similar path.”

One defence is tokenisation. “When an AI agent completes a purchase, the merchant can still recognise the customer through their secure tokenised credentials,” says Grigaravicius.

“This allows them to apply loyalty benefits, personalise offers and maintain a cohesive relationship across channels.”

Rethinking identity and fraud

If AI agents are executing transactions at scale, the security apparatus designed around human behaviour also needs to adapt.

The traditional fraud-prevention toolkit assumes that personal data alone is sufficient proof of identity, but this assumption weakens when the entity initiating the transaction is an AI agent.

“The old way of proving identity no longer holds,” says Grigaravicius. The counter is dynamic identification based on patterns of real commercial behaviour – looking at how customers and businesses actually transact, rather than relying on one-off checks that can be faked.

In principle, AI agents could reduce overall fraud by detecting behavioural anomalies across millions of data points, validating transactions in real time and flagging suspicious patterns before a transaction completes.

“AI agents don’t fall for phishing emails, don’t share passwords and can’t be socially engineered in the traditional sense,” says Grigaravicius. “So the net effect, if designed correctly, should be a reduction in overall fraud.”

Liability and standards

Where a compromised AI agent executes a fraudulent transaction, the chain of responsibility nevertheless needs to be resolved. Grigaravicius argues for a shared model between the AI platform provider, the merchant, the payment processor and the consumer.

“Where it gets complex is in cases where an AI agent is manipulated through no clear fault of any single party,” he says. “These scenarios require pre-agreed frameworks for liability allocation, which is why industry collaboration on standards is so important.”

Adyen is a partner of the Google-led Agent Payments Protocol initiative, which includes more than 60 tech and payment firms, and has also joined the Agentic AI Foundation, which aims to bring together companies to shape how autonomous systems interact.

Two-year horizon

The next phase – the transition from experimental, single-task agents to collaborative, multi-agent systems managing complex end-to-end processes – is likely to mature within two years, according to Grigaravicius.

The barriers are structural, with the sector needing robust authentication processes and interoperability across merchant systems, as well as consumer trust.

For now, the technical talent pool also remains thin. “The demand for people who understand both the commercial and technical dimensions of agentic AI far exceeds what is currently available,” Grigaravicius notes.

For the Gulf’s service economy, the opportunity is to serve as a proving ground. E-commerce penetration is high, regulatory appetite for fintech innovation is strong and consumer willingness to adopt runs well ahead of global averages.

The foundational questions – who verifies identity, who bears liability and whether merchants retain autonomy over their own customer relationships – need to be settled before adoption outpaces the infrastructure designed to support it.

“The rise of agentic AI is not a zero-sum game,” says Grigaravicius. “For agentic AI to become sustainable and profitable, we must build infrastructure that delivers genuine trust, transparency and merchant autonomy – because only that way will we achieve outcomes that benefit all.”