The rise of AI-powered agents is beginning to offer that same level of convenience – not just to the wealthy, but to anyone with access to a smartphone. AI-powered agents are quickly shifting from helpful search tools to autonomous shopping intermediaries. No longer just guiding product discovery, they’re now completing purchases – entirely on the consumer’s behalf. According to the Global Benchmark Report 2025, ChatGPT may only handle around 1% of Google’s query volume, but even that sliver represents a meaningful slice of the global search market, with significant revenue implications. What began as a digital concierge model is being remodelled into a new layer of commerce - one where the shopper has to get out their wallet.

Major payments players are taking note and moving fast. Visa recently announced a Digital Credential Innovation Hub, aimed at exploring new identity frameworks for agent-driven transactions. Stripe has confirmed development of secure infrastructure for AI-to-merchant payments. And Google has revealed plans for an “agentic” checkout flow – signalling that what once felt speculative is now becoming the standard. The groundwork is being laid for an entirely new class of consumer interaction.

But behind the convenience lies a new layer of complexity. What does it mean when your AI agent makes a purchase decision without you? What are the failure modes – and who’s accountable when things go wrong? Behind the optimistic headlines lies a more complicated set of questions.

Buying without browsing – behind the picture

There’s no universal standard for how agent-led shopping works – at least, not yet. The process is still in flux, with each platform experimenting with its own approach. But in early implementations, a few consistent patterns are becoming visible.

At the core, the shopper experience looks simple – but behind the curtain, it’s anything but. To begin, users store their full payment details with the AI platform: card number, CVV, expiry date, billing address, and delivery address. With that information on file, the agent is equipped to transact autonomously.

Consumers are less likely to turn to an AI agent for trivial purchases – a pizza, for example or for high-stakes buys like a car. Instead, early use cases tend to land in two pots. First are goods that are easily price comparable like a Nike t-shirt, where the consumer has no retail loyalty and is just looking for the lowest price this is very close to the simple search processes currently offered by Google. Second, but more interesting, are the “considered but confusing” category: higher-value products where specs matter more than aesthetics, and where the average shopper lacks deep expertise. Think of purchases that are technical, not obvious – like Bluetooth headphones, routers, or cameras – products where specs matter more than style, and expert advice makes a difference.

In those moments, the agent steps in like a savvy sales associate. Powered by platforms like ChatGPT, Google, TikTok Shop, or Amazon, it parses natural language requests and refines the search through follow-up questions. What’s your budget? Do you want over-ear or in-ear? Is noise cancellation important? Waterproofing? Based on the answers, the agent narrows the options and recommends products. These recommendations are based significantly on the consumer trusting the agent to provide the best advice, which presents challenges for advertising-led models which have historically reduced consumer trust.

Once the shopper gives the green light, the checkout process unfolds – largely out of view. The buyer remains within the AI interface, never visiting the merchant’s website. Behind the scenes:

• A simple “Buy” command within the agent’s UI triggers the agent to autofill the checkout form directly on the retailer’s site
• The merchant receives the card details as if a human had typed them
• The agent submits the order, and confirmation flows back through both the agent and the retailer

In most cases, the merchant doesn’t even know an AI was involved. This layer of abstraction introduces real friction – but only after the purchase. If something goes wrong, whether it’s the wrong item, a pricing discrepancy, or a delivery issue, the shopper is left managing the fall-out.

Invisible interactions, real-world risks

Several risks are already coming into focus:

• Risk of security breaches: in January 2025, Chinese AI platform DeepSeek suffered a hack that exposed users’ stored credentials. Concentrating payment data in AI agents creates lucrative targets for attackers
• Vulnerability to fraud: fraudsters are designing websites specifically to trick agents into completing fake or fraudulent purchases
• Liability ambiguity: when an AI agent messes up an order or enters wrong details, it’s unclear whether the consumer or the AI provider is responsible
• Integration challenges: many AI agents don’t support popular payment methods like PayPal, digital wallets, or bank transfers – which make up roughly 45% of EU eCommerce transactions according to research by PXP. Complex checkout steps such as seat selection or delivery scheduling often break agent workflows. Also, according to Pagos research, handling card declines is problematic – especially in cross-border transactions where failure rates can be as high as 30%
• Identification: in regulated markets like the EU and Japan, Strong Customer Authentication (SCA) rules require consumer approval for each card transaction, creating a compliance barrier for fully AI-driven checkouts

Are we on the verge of a commerce revolution?

Beyond these immediate technical and legal hurdles, AI agents raise deeper questions about the future of digital commerce.

Will consumers embrace this shift? It could stall like voice commerce or Amazon’s Dash buttons, both of which faltered due to trust and usability concerns. Or it could ignite explosive growth similar to the rise of marketplaces and mobile in-app buying. The outcome hinges on how much consumers value convenience and how effectively AI agents address trust and control issues.

If AI agents become the dominant way people shop online, the web itself may fracture. Why bother visiting a merchant’s site when your agent can handle everything? This could accelerate the creation of Model Context Protocols (MCPs) – AI-optimised data layers that could replace traditional websites altogether. Some merchants might push back by blocking agent IPs or complicating checkout flows to force direct customer interaction. Entire industries, including marketing, would need to rethink how they engage, prioritising AI agents over humans. Advertising-led models such as Google’s may also need to fundamentally change.

As this transformation unfolds, one fact is clear: the digital payments landscape is being rewritten. The real question is whether consumers and merchants are ready to follow their AI agents into this new era. And me? I’ll be watching closely – whether through my own experience or, perhaps, through my agents.

For more startup news, check out the other articles on the website, and subscribe to the magazine for free. Listen to The Cereal Entrepreneur podcast for more interviews with entrepreneurs and big-hitters in the startup ecosystem.