AI agents are rewriting e-commerce: Latin America moves faster while Europe confronts the rules of autonomous shopping
Latin America enters the transition with mature instant-payment and tokenisation infrastructure, while Europe faces a different challenge: reconciling autonomous transactions with a sophisticated regulatory framework designed for human-initiated payments.
For three decades, e-commerce has been designed around one fundamental assumption: a human being is sitting in front of a screen. The consumer searches. The consumer compares. The consumer adds an item to a cart. The consumer enters payment details. And ultimately, the consumer presses the button that completes the transaction.
Artificial intelligence is beginning to break that model.
The next transformation of online commerce is not simply about using AI to recommend products. It is about delegating parts — and potentially almost all — of the purchasing process to software agents capable of understanding an objective, finding alternatives, applying spending and delivery constraints and initiating a payment.
For Harald Valdivieso, CEO of WAIA, this transition is already moving beyond theory. And the contrast between Europe and Latin America provides a particularly revealing picture of what comes next.
On 11 March 2026, Banco do Brasil and Visa completed Brazil’s first transaction initiated by an AI agent using Visa Intelligent Commerce. The transaction took place in a controlled production environment, with Visa providing authentication, tokenisation and security controls. A day later, Santander and Visa announced controlled agentic-commerce transactions across Argentina, Brazil, Chile, Mexico and Uruguay.
Those experiments illustrate a broader shift: the interface of e-commerce is beginning to separate from the infrastructure that actually executes commerce.
“The question companies should be asking is not which protocol they need to join first,” Valdivieso told EUBizNews. “The important question is whether their architecture is ready for a world in which the customer interacting with their systems may be an agent rather than a person.”
From e-commerce to agentic commerce
Traditional online retail revolves around websites and applications built for human eyes and hands. Agentic commerce changes that relationship.
A consumer could eventually instruct an AI assistant to find a product meeting specific requirements, compare suppliers, verify delivery dates and complete the purchase if predefined conditions are satisfied.
That means the storefront itself may become less important.
For retailers, the decisive issue increasingly becomes whether machines can understand their catalogue, inventory, prices, delivery conditions and business rules without having to “read” a conventional website like a person.
The transformation is therefore deeper than adding an AI chatbot to an online store. It requires commerce infrastructure to become structured, machine-readable and executable.
Shopify is already moving in this direction. Its Catalog API structures product information so AI agents can query it, while the Universal Commerce Protocol, co-developed with Google, provides a common framework for agents and merchants to interact across the purchasing journey.
For Valdivieso, this is one of the most important changes companies need to understand. The website is no longer necessarily the front door to digital commerce. The catalogue itself is becoming an API for machines.
Three layers are emerging behind AI shopping
Valdivieso argues that the emerging agentic-commerce ecosystem is easier to understand when divided into separate layers.
The first is product discovery and catalogue infrastructure: allowing an AI system to understand what a company sells, at what price and with what availability.
The second is commerce and checkout orchestration: the infrastructure through which an agent creates carts, communicates with the merchant and turns a purchase intention into an order.
The third is payment authorisation and trust: proving that the consumer actually authorised the AI agent to spend money within specific boundaries.
These layers can overlap technologically, but they solve different problems.
And some of the world’s largest technology and payments companies are attempting to establish standards around them.
Google and Shopify introduced the open-source Universal Commerce Protocol (UCP) in January 2026. Google expanded its strategy at I/O in May with Universal Cart, designed to operate across merchants and services including Search, Gemini, YouTube and Gmail. Google says its Shopping Graph contains more than 60 billion product listings.
Meanwhile, OpenAI and Stripe developed the Agentic Commerce Protocol (ACP), launched as an open standard in September 2025 to connect AI agents and merchants through programmable purchasing flows.
The emerging architecture is therefore unlikely to be controlled by a single interface.
That matters because companies preparing for agentic commerce should avoid assuming that today’s most visible AI shopping application will necessarily become tomorrow’s dominant commercial gateway.
Google is building the trust layer
One of the most significant pieces of the emerging infrastructure is Google’s Agent Payments Protocol, or AP2. Its purpose is not simply to help an AI find something to buy. It addresses a more fundamental question:
How can a merchant, bank or payment processor prove that a human genuinely authorised an AI agent to make a particular transaction?
AP2 uses tamper-resistant digital mandates to create a verifiable connection between the consumer, merchant and payment processor.
Those mandates can define what the agent has permission to purchase and under which conditions.
In April 2026, Google transferred AP2 to the FIDO Alliance, the industry organisation associated with authentication standards such as passkeys. AP2 version 0.2 also introduced support for “Human Not Present” transactions based on previously authorised instructions. Google additionally developed Verifiable Intent with Mastercard to create a tamper-proof record of user-authorised agent actions.
For Valdivieso, the move reveals something important about the strategic battle surrounding agentic commerce. Google does not necessarily need to own every shopping interface if its standards become part of the trust infrastructure underneath them.
The prize may ultimately be less about controlling the digital storefront and more about establishing the rails that other companies rely on to prove identity, consent and authority.
The checkout is only the visible part
The first wave of agentic-commerce enthusiasm focused heavily on making purchases directly inside conversational interfaces.
OpenAI and Stripe launched Instant Checkout in ChatGPT in September 2025, initially supporting purchases from US Etsy sellers and announcing plans to extend participation to Shopify merchants. The underlying ACP standard was designed to allow merchants to remain merchant of record and continue handling payments, fulfilment, returns and customer relationships through their existing systems.
But Valdivieso argues that focusing exclusively on the checkout button misses the hardest part of commerce.
A successful transaction still requires accurate stock information, taxes, fraud prevention, fulfilment, refunds, customer service and reconciliation. AI does not make those problems disappear.
“Commerce has a lot of boring operational complexity,” he said. “That complexity is exactly what determines whether a demo can become a production system.” That distinction has important consequences for retailers.
Companies should not interpret agentic commerce primarily as a new sales channel requiring another checkout integration. They need to treat it as an architectural transformation across their entire commerce stack.
Europe’s problem: autonomous technology meets human-era regulation
Europe enters this transition from an unusual position. The European Union has built one of the world’s most comprehensive digital and payments regulatory frameworks.
That provides substantial consumer protections but also creates difficult questions when the entity executing a transaction is software acting under delegated authority.
Existing European payment rules were designed around human users authorising transactions. Agentic commerce complicates that assumption.
If a consumer tells an AI agent to purchase a flight when the fare falls below a particular price, for example, the consumer may not be present when the transaction actually takes place.
Who authorised the final payment?
How should that authorisation be recorded?
What happens if the agent interprets the instructions incorrectly?
Who bears responsibility when delegated spending limits are incorrectly applied?
And where exactly does the AI agent sit within the chain of regulated payment participants?
These are no longer theoretical questions.
The regulatory clock moves more slowly than the technology
The speed mismatch has become increasingly visible to financial regulators.
Speaking at the European Central Bank Forum on Central Banking in Sintra on 30 June, Bank of England Deputy Governor Sarah Breeden warned that AI is reshaping finance at considerable speed and argued that financial authorities must adapt as autonomous systems become more capable.
The United Kingdom has already moved to address the issue more explicitly.
On 14 July 2026, HM Treasury opened a consultation on modernising payment-services regulation. Among the innovations explicitly identified by the government are agentic payments, alongside tokenised payments and Open Banking.
For European companies, the implication is significant.
Technology capable of executing increasingly autonomous commerce is arriving before legislation has fully established how responsibility, consent and liability should work in those transactions.
According to Valdivieso, businesses should not interpret that gap as permission to postpone the problem.
They should do the opposite.
Consent could become the defining issue
In traditional e-commerce, consumer intent is comparatively easy to understand. Someone selects a product and confirms a payment. With autonomous agents, consent can become conditional and temporal.
A user might say:
“Buy this product if it falls below €500 before Friday.”
The final purchase could take place two days later, while the consumer is asleep.
For the merchant and payment provider, the crucial evidence is therefore no longer simply that somebody clicked a button.
They need a reliable record showing what the consumer authorised, how much the agent was allowed to spend, which conditions applied and whether the final transaction complied with those instructions.
This is why Valdivieso believes the mandate concept being developed around agentic payments deserves attention even from companies that are not implementing AP2 itself.
The underlying principle — cryptographically verifiable, auditable consent — could become fundamental to trustworthy autonomous commerce.
Latin America starts from a different position
If Europe’s advantage is its sophisticated regulatory architecture, Latin America enters the transition with another asset: payment infrastructure that has already undergone a profound digital transformation.
The region has spent years accelerating instant payments, digital wallets and tokenisation.
That creates fertile ground for agentic transactions.
Banco do Brasil and Visa’s March transaction showed that AI-initiated commerce could operate through existing authentication and tokenisation infrastructure in a real production environment.
Santander and Visa expanded the picture by testing controlled agentic-commerce transactions across five Latin American markets.
Mastercard is pursuing a parallel strategy.
In December 2025, the company announced that Mastercard Agent Pay would expand across Latin America and the Caribbean during 2026. Its model uses Agentic Tokens, permissions and spending limits defined by consumers, with verification and traceability incorporated into the payment process.
One statistic illustrates the region’s preparedness particularly clearly: Mastercard says nearly 100% of its issuers in Latin America are already enabled for its tokenisation technology.
Latin America can move quickly — but speed creates another risk
The region’s advantage does not mean its regulatory questions are solved.
Rather, Europe and Latin America face almost inverse challenges.
Europe has highly developed regulation but must adapt rules built around human actions to increasingly autonomous software.
Latin America has payment infrastructure capable of supporting rapid experimentation, while specific rules governing AI agents remain comparatively undeveloped.
That creates an opportunity for faster adoption — and simultaneously makes security architecture more important.
For Valdivieso, companies cannot wait for regulators to solve fraud, delegation and accountability.
Controls must be embedded in systems from the beginning.
That means spending limits, transaction-level traceability, clear agent permissions and human intervention for decisions with greater financial or operational consequences.
Europe and Latin America are heading toward the same destination from opposite directions
The comparison between the two regions reveals one of the most interesting characteristics of the emerging AI economy.
Europe possesses sophisticated institutional and regulatory safeguards but faces the challenge of adapting them to machines capable of acting independently.
Latin America has built highly dynamic digital-payment ecosystems that can accelerate implementation but needs to ensure governance catches up with technical capability.
Neither model has solved the entire problem.
And both regions can learn from the other.
Europe’s emphasis on consumer rights, authentication and accountability could help establish global standards for trusted agentic commerce.
Latin America, meanwhile, demonstrates how modern payment rails and widespread tokenisation can reduce friction when new forms of digital commerce arrive.
“The infrastructure and the regulation are moving at different speeds in both regions,” Valdivieso said. “The difference is where the bottleneck sits.”
What retailers should be building now
For WAIA, companies should avoid betting their entire e-commerce strategy on whichever AI shopping interface appears dominant today.
Instead, Valdivieso identifies a more durable set of priorities:
Machine-readable catalogues: product information, pricing, attributes and availability need to be structured so agents can interpret them reliably.
Real-time inventory: an AI agent cannot execute commerce safely if the stock information it receives is outdated.
Auditable consent: businesses need to know what the customer delegated to an agent and under which limits.
Traceability: every relevant decision and transaction should leave a verifiable record.
Spending and risk controls: autonomous purchasing requires explicit boundaries.
Human escalation: higher-impact decisions should have mechanisms allowing people to intervene.
Protocol independence: commerce architecture should be capable of interacting with different agents and standards rather than becoming dependent on one platform.
The goal is not to predict whether Google, OpenAI, Stripe, Visa, Mastercard, Shopify or another company will establish the dominant standard. It is to build an e-commerce operation capable of working with whichever ecosystems ultimately prevail.
The competitive advantage will be invisible
The arrival of AI agents could ultimately change what companies mean when they talk about optimising an online store.
For the last two decades, optimisation largely meant improving the human experience: faster pages, better search, fewer checkout steps, clearer product photographs and higher conversion rates.
Those priorities will not disappear.
But companies may increasingly need to optimise for a second customer: the machine acting on behalf of the human customer.
That machine does not care about a beautiful homepage.
It cares whether prices are accurate, inventory is current, product attributes are structured, delivery promises are reliable and transaction rules can be executed programmatically.
This could shift competitive advantage away from the visible storefront and toward the underlying architecture.
The next e-commerce revolution is not another website
The most consequential change in agentic commerce may therefore be almost invisible to consumers.
Google, Shopify, OpenAI, Stripe, Visa, Mastercard and other players are building different pieces of infrastructure through which AI systems can discover products, communicate with merchants, establish consumer authority and execute payments.
Some standards will grow. Others may disappear. New ones will emerge.
But the direction of travel is increasingly clear.
E-commerce is moving from a world where software helps humans shop toward one where humans can delegate commerce to software.
Europe and Latin America are entering that transition with very different strengths and weaknesses.
Europe must ensure its extensive consumer and payments protections can function when the buyer is represented by an autonomous agent. Latin America must transform its head start in digital payments into a trustworthy ecosystem without allowing regulation and risk controls to fall too far behind.
For businesses operating across both markets, Valdivieso’s conclusion is less about choosing the technology winner than preparing the company itself.
The businesses best positioned for the next era of e-commerce may not be those with the most impressive AI shopping demo. They will be those whose catalogues, inventories, payments, consent systems and business rules were already designed well enough for machines to understand and execute them.
