One of the biggest differences between being rich and poor is having people to do things for you. When a person moves from being poor to the middle class, they might get a cleaner or take their clothes to a dry cleaner; when they become rich, they might get a driver, a private chef – an entire entourage devoted to making their lives easier.
There’s growing excitement about the rise of AI-powered agents that act on behalf of consumers – not just during product discovery, but right through to purchase. Search-like queries on ChatGPT might only represent 1% of those on Google, but 1% of the colossal global search market is a huge number of searches, with potentially millions in revenue.
Originally conceived as digital concierges to simplify search, these agents are now making actual purchases, and they’re doing it without ever handing control back to the shopper.
Large payments players are laying the groundwork for AI-based commerce. Visa recently launched a Digital Credential Innovation Hub to explore new identity models for agent-based transactions, while Stripe confirmed it is developing secure transaction capabilities for AI agents. And just yesterday, Google revealed plans for an AI agentic checkout for shopping — a move that confirms this shift is no longer speculative, but imminent.
But beyond the optimistic headlines lies a more complicated picture. What happens when an AI agent makes a purchase for you? And more importantly, what can go wrong?
Managing Director, PSE Consulting.
What’s Really Going On Behind the Scenes (Usually)
Let’s be clear: there’s no single “standard” for how agentic shopping works – the process is still evolving, and different platforms take different approaches. That said, here’s a common flow we’ve observed in early implementations.
When a consumer uses an AI agent to shop, the process is superficially simple, but technically intricate. First, the user saves their payment card details — including full PAN, CVV, expiry, billing, and delivery addresses — with their chosen AI platform.
Shoppers are unlikely to be buying very inexpensive items like a pizza or very expensive items like a new car. They probably won’t be using it for goods with a heavy visual emphasis, where part of the enjoyment is browsing until something hits you – clothing being the best example. They are likely, at first, to use agentic AI to help them decide between relatively expensive products that are difficult for non-experts to understand: let’s use a good pair of Bluetooth headphones as an example.
The agent, which could be powered by ChatGPT, Google, TikTok Shop, or Amazon’s AI initiatives, uses natural language to respond to a shopper’s request. Just like a shop clerk, it will ask questions to refine results: how much do you want to spend? Do you want over-ear or in-ear headphones? Are there any features like noise cancelling or waterproofing that you need? It can then refine results and present purchase options.
Once the shopper decides a payment process begins that will be mostly invisible to the shopper:
- The shopper stays in the AI interface and never visits the merchant’s site.
- A “Buy” command within the agent UI triggers the agent to autofill the checkout form on the merchant’s site.
- The merchant receives the full card details as if a human shopper were typing them.
- The agent submits the order, and confirmation is sent via both the agent and the merchant.
Critically, the merchant is unlikely to know that they’re dealing with an agent rather than a human. This introduces risks, because if anything goes wrong — an incorrect item, a delivery mix-up, or pricing error – the shopper must resolve it directly with the merchant, even though they never interacted with the merchant’s website themselves.
In other words: don’t talk to me – talk to my agent.
Known Risks (So Far)
There are several emerging pitfalls already evident:
Security vulnerabilities: In January 2025, Chinese AI platform DeepSeek was hacked, exposing users’ stored credentials. The centralization of payment data in AI agents makes them lucrative targets.
Susceptibility to scams: Fraudsters may design sites specifically to trick agents into completing fake checkouts.
Ambiguity in liability: If an agent misplaces an order or inputs incorrect details, it’s unclear whether the AI provider or the consumer bears responsibility.
Poor compatibility:
- It may not support alternative payment types like PayPal, digital wallets or bank transfers (which account for roughly 45% of eCommerce volume in the EU).
- It can’t easily handle additional checkout steps (e.g. seat selection, delivery slots).
- Struggles with card declines, especially in international transactions, where decline rates can be anywhere from 5 to 30%.
In markets like the EU or Japan, legal requirements around Strong Customer Authentication (SCA) mean that consumers must approve each card transaction, making AI-led flows problematic or non-compliant.
The Bigger Picture: Are We Witnessing a Commerce Revolution?
Beyond the immediate risks and logistics, the rise of agents raises fundamental questions about the structure of digital commerce.
Will this model gain traction with consumers? It could fizzle like voice commerce and Amazon’s Dash buttons, which failed to take off due to trust and usability issues. Or it could explode, much like the rise of marketplaces or in-app mobile buying. The answer depends on how much value consumers place on convenience, and how well AI agents can overcome trust and control issues.
If AI agents become the preferred interface for eCommerce, the web as we know it may fragment. Why visit a merchant site at all, when your agent can do the work? This shift could drive the development of Model Context Protocols (MCPs) – AI-optimized data layers that replace websites altogether. Some merchants may respond by blocking known agent IPs or designing checkout flows that frustrate automated systems to force direct interaction. Industries like marketing would fundamentally change as it becomes more important to engage with AI agents than human beings.
Meanwhile, platforms like ChatGPT will need to find ways to monetize their newfound influence. That might mean charging merchants referral fees, sparking the emergence of a new SEO-for-AI ecosystem. But such monetization introduces new questions about trust: if your agent is taking commission from merchants, how unbiased are its recommendations?
As we stand on the edge of this transformation, one thing is certain: the infrastructure of digital payments is being rewritten. The question now is whether consumers – and merchants – are ready to follow their agents into this new era.
And I, for one, will be watching closely – either directly, or via my agent.
LINK!
This article was produced as part of TechRadarPro’s Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro
