Eighty-one per cent of retail executives surveyed by Deloitte expect brand loyalty to weaken as AI agents take over more of the purchasing process, according to the firm's The future of commerce: Agentic shopping in Asia Pacific report.
It is not hard to see why. A shopping agent optimising for price, delivery speed and stated preferences has little reason to favour the brand a consumer might have chosen after browsing. It finds the best match. The commercial relationship that once belonged to the retailer or brand could move to whoever controls the agent.
That is the threat beneath agentic commerce, the term retailers and technology companies are using for AI systems that go beyond search and recommendations to execute purchases, manage replenishment and authorise payments on a consumer's behalf.
Worryingly, Asia Pacific could be where that threat shows up earliest. Having spent more than a decade building superapps, social commerce habits and digital payments infrastructure, the region has created the consumer behaviour and data trails that shopping agents need to work with.
Those data trails matter because agentic commerce depends on context. Recommendation engines suggest, while AI agents act. However, an AI agent can only act well if it carries a shopper's preferences, history and intent across stores, apps and search.
That is the problem most retailers have yet to solve. "As AI agents become a gateway to the consumer, Asia Pacific retailers and brand teams should act now to ensure their products and services are discoverable and compelling to both machines and people," notes Deloitte’s report.
See also: Trump’s Anthropic crackdown sets off AI alarms for US allies
Control of the customer
In Malaysia, AEON360 has struck a multi-year agreement with Google Cloud that shows how the next phase may be built.
Before deploying AI agents, the company is creating an enterprise knowledge graph — a unified data layer that reconciles customer information across its retail stores, financial services arm and lifestyle businesses. Once that foundation is in place, AEON360 plans to deploy AI agents that can build shopping carts, execute consented transactions and carry shopper context from the mall to the mobile app to Google Search.
See also: Why governance, not adoption speed, will define the winners using AI in wealth management
The build order is important. AEON360 is trying to solve the data problem first because an agent can only be useful if it can recognise the same customer across the group's stores, services and digital channels.
"With Google Cloud, we're shifting from simple digital interactions to AI agents that surface the most relevant offerings and perform complex tasks on our customers' behalf, starting in Malaysia, with a roadmap to expand across key Southeast Asian markets that we're present in. Through our Innovation Foundry, we're upskilling our staff to be the architects of this agentic commerce future, anchored on a data foundation that adapts to evolving preferences, predicated on customer consent," says AEON360 chairman Daisuke Maeda.
The standards fight
For agent-led shopping to work beyond a single retailer or app, the systems behind it need to speak to one another. Agents have to find products, compare offers, confirm identity, process payments and complete transactions across different merchants and platforms.
Google has introduced the Universal Commerce Protocol (UCP) to address part of that problem. The standard lets merchants publish their product and service catalogues into Google's ecosystem so they can be surfaced when consumers use Google Shopping or Gemini to research a purchase.
This matters because the starting point for shopping may move further away from a retailer's own website or app. Rao Surapaneni, Google Cloud's vice president and general manager for Business Application Platform, says that a consumer asking Gemini to compare laptops already shows high purchase intent. UCP allows catalogue information to be shown in that interaction before routing the shopper to the merchant's site to complete the transaction.
The harder problem is authorisation. If an AI agent is allowed to buy something for a consumer, merchants and payment providers need to know whether the agent is acting for the user, what it has been authorised to do and whether the transaction fits that instruction.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
Google is also working on the Agentic Payment Protocol (AP2) for that layer. Surapaneni explains that the system uses cryptographically signed mandates to capture a user's intent, such as asking an agent to buy a laptop up to a certain amount. The merchant can then validate that authority before completing the transaction.
Memory is the next layer because an AI agent becomes more useful when it can remember a customer's preferences, past decisions and exceptions.
As such, Google designed its memory bank to capture what happens after an AI agent is deployed, including patterns and exceptions that were not written into the original workflow. "It is not just what the developer built for the agent, but also, once the agent is deployed in production, you're able to capture the enterprise's essence and then incorporate it back into the agent," he told DigitalEdge on the sidelines of Google Cloud Next 2026 in Las Vegas.
Auditability matters here, too. Surapaneni shares that Google's observability layer tracks what tools an agent invokes, whether it has authentication and authorisation, and the decision traces behind its actions. The platform also uses audit agents to flag decisions for human review.
He adds that Google is developing these protocols openly because AI agents "have to be interoperable… [so they need] one [common] language” to work together at scale". For example, the Agent-to-Agent protocol is now hosted by the Linux Foundation and supported by many companies, including Amazon Web Services (AWS) and Microsoft.
Still, this does not remove the platform risk. A retailer may gain access to new AI-driven shopping journeys, but it is also placing more of the customer interaction inside systems built by large technology providers. The business question is whether retailers can use those systems without giving up too much control over discovery, data and the customer relationship.
Where retailers are now
Most retailers remain some distance from the agent-led purchasing model AEON360 is building toward.
Singapore's FairPrice Group shows what that earlier stage looks like. The supermarket operator is rolling out smart shopping carts, Digital Price Cards and an AI operations app called Grocer Genie across its supermarket network by end-2026, all built on Google Cloud's Gemini platform. Smart carts and Digital Price Cards are targeted at 48 FairPrice Xtra and Finest outlets specifically, while Grocer Genie will reach all supermarket store teams.
The gains from its Punggol Coast Mall pilot are operationally meaningful. Checkout times have fallen to an average of 36 seconds. Digital Price Cards replacing printed alternatives are projected to save $138,000 and 15,000 staff hours a year. The company is also testing removable cart tablets at its Oasis Terrace outlet in Punggol, allowing shoppers to detach the screen and wheel their groceries directly to the car park.
Sarah Jane Vasquez, branch manager at the pilot store, describes the immediate benefit for staff. "New digital tools like Grocer Genie help me automate manual work like eyeballing when shelves need to be restocked and even staff rostering. This frees up time for me to lead and coach my team," she says.
While these are real productivity gains, the shopper at FairPrice is still making every purchase decision herself. The AI is helping with navigation, checkout and store operations, but it is not yet buying on the consumer's behalf.
In short, tools like smart carts, Digital Price Cards and AI operations apps can make stores more efficient and shopping less cumbersome. However, agentic commerce requires a greater leap as the AI agent needs sufficient customer context, consent, and trust to act on a customer's behalf, not merely assist with the shopping trip.
The trust barrier
Consumer trust is another constraint to agentic commerce. Deloitte's report found that 74% of Asia Pacific consumers already use AI to research products and compare prices. Yet, nearly half say they would not complete a purchase without stronger security assurances.
Trust will, therefore, have to be built into the transaction instead of being added after it. Shoppers need to know what the AI agent is allowed to do, when consent is required and how a purchase can be reviewed or reversed if something goes wrong. Google's AP2 protocol can help by verifying whether an agent is acting on a user's behalf and whether the transaction matches what the user approved. Without that assurance, agentic commerce risks remaining a research tool rather than a buying channel.
According to industry forecasts, AI agents could influence up to 25% of global e-commerce sales by 2030. Retailers should now take steps to ensure their products are visible to AI agents while giving customers enough confidence to let software act on their behalf.
