Are AI shoppers coming? Here’s what retailers need to know.

Your next customer might never browse your website. Instead, they’ll be telling a machine to shop for them.

eCommerce has long followed a predictable sequence: search, browse, compare, decide, and checkout. Now, that journey is starting to change as AI agents are becoming the new buyers. This shift towards agentic commerce moves in stages.

Today, shoppers are mostly using AI as a search and discovery tool. In a recent Visa report, nearly half of US shoppers said they already use AI to enhance their shopping experience, whether to find gifts, compare prices, or discover new products. The discovery happens on AI platforms, but the purchase still happens on a website.

The next step brings the transaction closer to the AI interface itself, where payment capabilities are increasingly being built directly into LLMs and agent interfaces, on terms that merchants and platforms are still negotiating. Shoppers could search, compare, and pay without ever visiting a site.

Agentic AI takes that further, where we’re expecting agents to complete multi-step tasks on shoppers’ behalf. A shopper could simply tell AI, “Find and purchase white running shoes, size 6, under $150, that ship this week.” The agent does the heavy lifting: comparing products across retailers, checking reviews, weighing shipping options, and authenticating payment credentials, before completing the purchase. What once required dozens of clicks and half an hour of browsing can happen in seconds, with a single prompt.

Across the industry, early infrastructure for agentic commerce is emerging: agent payment protocols allowing AI agents to transact safely, commerce platforms building their own AI experiences, and payment networks adapting their rails to support agentic transactions.

So, what‌ changes when AI becomes the new buyer?

Shopping is shifting from clicks to conversations

When AI agents become the first point of interaction, every stage of the shopping journey changes with it.

Product discovery no longer hinges purely on search engine rankings. A brand would also have to be discoverable to agents that prioritise data quality, reliability, and speed. So besides a well-designed product page, a well-structured product catalogue with clean, contextual, accurate, machine-readable data is needed. If agents can't reliably parse a product catalogue, the merchant effectively doesn’t exist in that transaction. As agent preferences start to shape decision-making, businesses need to think about how to become the brand that agents prefer. That means designing for two distinct experiences: the human-facing UX that builds trust and desire, and the agent-facing infrastructure that puts a brand into consideration.

Eventually, agents may complete purchases directly. With tokenized payment credentials stored securely in agent wallets, shoppers could authorise transactions without ever touching a "buy" button.

But shoppers won’t hand over their wallet to AI overnight 

While interest around agent-led shopping is growing, consumer adoption will take time to catch up. Here’s why:

Trust is the real bottleneck

Today, many shoppers are still cautious about storing their payment details online. 69% of consumers who use digital payments worry that these methods might not be as safe as other payment options. Granting an AI agent access to those credentials introduces additional uncertainty. Shoppers would likely feel more comfortable when agentic transactions take place inside ecosystems they already trust, whether that’s familiar retailers, established marketplaces, or technology platforms with strong security reputations. That trust needs to be built gradually, through repeated positive experiences.

Trust also depends on clarity around responsibility. If an AI agent makes an incorrect or unauthorised purchase, it’s not clear yet where that liability falls. Would it fall on the consumer, the agent provider, the platform, the retailer, or the payment service provider? Governments and regulators are only beginning to address these questions. Singapore, for example, has launched the world's first agentic AI governance framework, establishing early rules around accountability, transparency, and consent.

Shoppers will only trust AI when they feel in control

Agentic commerce can't truly take off without consumer trust. Trust grows when they understand how the system works and know that they can intervene at any time when something seems wrong. Shoppers must feel in control, even when they’re delegating decisions to the AI. That means agentic commerce systems need to be designed for this sense of control, with key elements such as:

• Mandates and limits that let shoppers define exactly what agents can do, including spend limits, merchant categories, and specific purchase triggers

• Transparency around every agent-initiated transaction, with clear labelling and full audit trails

• Easy pause-or-cancel mechanisms to stop an agent's activity quickly

• Dispute resolution for contesting agent-initiated purchases

• Returns and refunds for items that weren't as described or simply unwanted, regardless of how the purchase was initiated

• Access to human-led customer support, even in an automated environment

Agentic commerce will complement the human shopping experience

AI won’t take over shopping entirely, because the shopping experience isn't one-dimensional. Think of the shopping experience as a spectrum. On one end, there are purchases that are purely functional and repetitive, like restocking household essentials and renewing subscriptions. These are ideal for full automation, where shoppers could set their preferences once and let agents handle purchases from then on. For example, autoship functions already exist on some marketplaces, but agents can take this further by managing recurring deliveries across multiple platforms simultaneously.

On the other end of the spectrum are purchases that are deeply personal: choosing a gift for someone special, exploring a new style, or discovering a brand for the first time. These experiences carry an emotional weight that AI agents can't replicate, like the joy of browsing and discovering something unexpectedly. Part of these experiences come from the emotional connections that retailers have built with customers over time, creating brand affinity ‌through storytelling and memorable interactions. Many shoppers still want to own these experiences directly. Agents may play a role in these situations, but more as assistants than decision-makers. For example, Google's AI Mode in shopping supports a fully-assisted commerce experience: finding products, tracking prices, enabling virtual try-ons, and completing purchases.

But even as agents take on a supporting role, they still sit between brands and their customers, changing how a brand gets experienced. Today, perception is shaped through design, storytelling, and on-site interactions. In agent-led shopping, those touchpoints may be bypassed entirely. The upsell prompts, the “you may also like” nudges, the new-drop notifications, the well-timed discount codes – these tools have driven eCommerce revenue for years. But agents don’t have moments of hesitation or desire. They simply filter for data points that meet a brief. This raises an interesting question: how much of the customer experience do businesses want to own?

Some businesses will rely on third-party platforms and compete within those ecosystems. Others will build their own AI-powered experiences through LLM integrations, proprietary apps, or platform-specific tools to maintain their voice, values, and customer relationships even as purchasing becomes more automated. Amazon’s Buy for Me feature, for example, lets shoppers purchase from external retailers without leaving the app, keeping shoppers within its ecosystem and retaining the brand experience.

From eCommerce to aCommerce

We’re moving into a new phase of commerce, where AI doesn’t assist transactions, but starts to run them. McKinsey estimates that AI agents could orchestrate up to $3–5 trillion in global consumer spend by 2030. If this holds true, agentic commerce won’t be a marginal shift, it will reshape how transactions happen, much like the move from brick-and-mortar to eCommerce did.

You can already see the early version of this with one-click checkout. Returning customers skip the friction and complete a purchase in seconds. The next step is zero-click checkouts. Shoppers set their preferences once – payment methods, spend limits, and preferred retailers – and agents take care of the rest. Card details are tokenized and stored securely in an agent wallet, removing the need to re-enter details. Routine purchases happen automatically, with users notified and given control to step in, adjust, or cancel when needed.

Agents will interact with each other to run more complex workflows: negotiating prices, routing purchases across regions, and optimising for delivery times, FX, and shipping costs. When commerce starts to operate as a network of interacting agents across platforms, marketplaces, and payment networks, the underlying infrastructure needs to keep up.

Infrastructure will become the new storefront

Your brand still matters. Brand design, storytelling, and product pages – these are what get shoppers through the door. But agents shop differently. They evaluate your infrastructure instead, assessing whether it can support fast, secure, and reliable transactions globally. They’ll prioritise retailers whose platforms provide rich, contextual product data, clear pricing, efficient checkout flows, and trusted payment systems.

The key is making it easy for agents to understand and transact with your platform. That requires secure, programmable infrastructure that can:

• Communicate with agents and accept agent-initiated payments

• Authenticate legitimate agents while protecting against malicious traffic

• Make product catalogues easily discoverable by agents

At Airwallex, we’re building towards this future. Airwallex powers the agent-ready payment layer and finance stack that helps merchants be part of agentic commerce, without having to build their own infrastructure.