ARTICLE: The silent battle for "ownership" of the customer in agency commerce

This is the last article in a series of three that I'm publishing here on NeoFeed about agentic trading . The first described how an agentic transaction happens end-to-end, with the engineering of the two tokens. The second dealt with Know Your Agent , or KYA, an emerging category that defines how an agent is identified and verified to operate on the network.

This concludes the series by addressing the remaining, and perhaps most important, dimension for those who need to make strategic decisions in the next two years: who captures value in the new model.

The ongoing dispute between the layers of the ecosystem is real, it's silent, and it will define much of what agency trade will be like in Brazil in five years. It's silent because each layer presents its initiative as a neutral technical contribution, when in fact it is shaping a strategic position. I will try to dismantle it.

When reading material about agentic commerce, the focus is usually on the protocols: Visa 's Trusted Agent Protocol , Mastercard 's Verifiable Intent , OpenAI 's Agentic Commerce Protocol , Google 's Universal Commerce Protocol , Agent Pay, and Verifiable Credentials.

Technical debates are intense, and there are legitimate reasons to disagree about architectures, especially regarding security. But the dispute that will determine who wins the game isn't on that technical level. It's about who controls the relationship with the client.

Four layers are under tension. The first is that of the AI platforms: ChatGPT , Gemini , Claude , Microsoft Copilot , and Perplexity . This is where the customer actually speaks with the agent. The second is that of the card networks, which operate the payment rail and are building the networks of trust where agents are recognized.

The third group consists of issuers, banks, and fintechs that deliver cards and hold mandates. And the fourth group consists of merchants, who actually sell the product and whose relationship with the customer is being reconfigured.

Each of these layers has its own strategic assets and ambitions. These ambitions are partially incompatible. Hence the dispute.

Those who are closer to the customer concentrate more value.

The central question that each layer asks internally is simple: If the customer delegates a purchase decision to an agent, which layer of the ecosystem has the most influence over that decision?

The response tends to be one where the customer actually speaks with the agent.

Therefore, AI platforms are in a structurally privileged position. When someone says "ChatGPT, buy the wine that the restaurant sommelier recommended for tomorrow," the interface from which the order originates has a disproportionate influence over which wine is chosen, in which store, with which card, and under what payment terms. It's not absolute power, because the agent operates within a mandate defined by the client. But it is substantial influence.

"The dispute, in practical terms, is about who gets closest to the customer without being disintermediated by the other layers."

There are important counterweights, however. The card network controls who operates within its network. Without registration, without a verifiable mandate, the transaction does not take place. This gives veto power and the ability to define rules to Visa, Mastercard, and Elo .

The issuer controls the credit and the mandate, and can decide to block agents who do not meet the bank's criteria. The merchant controls the catalog and delivery. Without a machine-readable catalog, the agent has nothing to buy; without delivery, the purchase is not completed.

In practical terms, the battle is over who gets closest to the customer without being disintermediated by the other layers. And it's a battle where the initial position favors AI platforms, but where the other layers have assets to resist.

The risk for the retailer is becoming a commodity.

The layer that has the most to lose in the short term is that of the retailer. For a simple reason. In traditional e-commerce, the retailer controls the storefront. They decide what appears, with what prominence, in what order, and with what promotion. SEO exists so that the customer can find the retailer, but the final ranking is influenced by dozens of commercial factors under some control of the retailer.

In agent-based commerce, this changes. The agent doesn't see visual layout. It doesn't respond to banners, videos, or sophisticated layouts. It compares structured data.

Adyen and Checkout.com have been advocating for the concept of a merchant-owned mandate for several months. Essentially, their position is that the mandate should reside within the merchant's domain, not the card network's or issuer's. Their argument is that if a third party defines what the agent can buy, that third party has more power over the customer than the merchant themselves.

CMSPI, a payment analytics firm, articulated the point clearly in a recent diagnosis: if a third party defines what the agent can do, that third party owns the client, and the merchant becomes a commodity.

"The agent doesn't see visual layout. It doesn't respond to banners, videos, or sophisticated layouts. It compares structured data."

The migration from SEO to AEO — Agent, Answer, or Agentic Engine Optimization , depending on who's calling it — is the technical symptom of this risk. Retailers who relied on Google for discovery now depend on agents. This means structuring a machine-readable catalog with exhaustive product data.

And it also means competing not for human clicks, but for the algorithmic criteria of the agent. Who ranks first, who is considered more reliable, who offers the smallest margin of error, who has the best delivery reputation. These are different criteria from those that have dominated e-commerce until now.

For large retailers, the adjustment is difficult but feasible. They can invest in agency infrastructure, partnerships with AI platforms, and differentiation through a strong brand that survives algorithmic mediation.

For small retailers, the risk is greater. Marketplace platforms that offer agency readiness as a service can capture an even larger share of the retail market, precisely because they offer small retailers the infrastructure they cannot build on their own.

The accrediting agency has its own dilemma.

I mentioned in the first article that the payment processor hasn't left the game. It maintains its role in processing, managing chargebacks , and maintaining relationships with the merchant. But its strategic position is more fragile than it seems.

Edgar Dunn, in a diagnosis published in March, made the point directly: payment processors that remain stagnant will lose merchants to payment processors that position themselves as agent-native .

The reason is that the traditional functions of the acquiring bank are being redistributed. Capture remains the same, but fraud has shifted to the card network, which validates the mandate, and to the issuer, which sees the context of the authorization. Discovery has migrated to the agents. What remains for the acquiring bank is increasingly commoditized technical infrastructure.

"What remains for the accrediting body is increasingly commoditized technical infrastructure."

There are solutions, and they are strategic. One is to deepen the capabilities of agent readiness with retailers: helping to structure catalogs, integrate with open protocols, and position themselves well in agent search engines.

Another option is to position itself as an aggregator of payment methods, including Pix , Pix Automático , and eventually stablecoins . In other words, to stop being just a card issuer and become an A2A issuer.

The third is to build an intelligence layer on top of the traffic it processes, identifying behavioral patterns of agents who buy from merchants in the portfolio, and offer this as an analytical service.

Whoever makes any of these adaptations in the next two years will come out strong. Those who stand still will become a technical commodity.

Here, one observation deserves special mention. The international PSPs already operating in Brazil or entering the market—Stripe, Adyen, and more recently, Checkout.com, which obtained a license from the Central Bank and is starting operations in the country—are positioning themselves as intermediaries between the agent world and the traditional world, and this position provides a concrete competitive advantage over local payment processors.

They have a more agile technical culture, a shorter adoption cycle for new protocols, and historically enter new layers with a speed that local acquirers take longer to match. If the agentic design depends heavily on this technical translation in the next two or three years, these players could capture a larger share of relevance in Brazil while Cielo , Rede, Stone , and Getnet are still calibrating their response.

It's not an inevitable scenario, but it's a real competitive driver that local acquirers need to treat as a strategic priority, not as a routine technical adaptation.

The agent chooses the best card for each purchase.

There is a specific effect of agentic commerce on issuers that deserves attention and illustrates a broader trend. In traditional e-commerce, the customer has a preferred card, chosen out of inertia, habit, brand, or points program, the economics of which they have never actually followed.

This default card accounts for a large portion of the customer's purchases and sustains predictable savings for the issuer: interchange revenue, revolving credit interest when the customer is late on payments, annual fees renewed without question, and cross-selling of other products.

In the agent system, the default disappears. The agent, configured to optimize, chooses the card that offers the best benefit for that specific transaction with each purchase. It could be one card for airline tickets, another for groceries, another for fuel, and another for international purchases.

The inertia that sustained the default becomes a variable to be eliminated. It's worth reflecting on how many financial decisions today follow a similar pattern—subscriptions that renew without comparison, money sitting idle in accounts without yielding returns, insurance policies that are extended without competing quotes. In each case, there is inertia generating revenue for someone. The agent, by optimizing, redistributes this revenue.

But there's a deeper point, and that's where the real competitive challenge for issuers lies. If the agent chooses the best card, the definition of what is best then depends exclusively on what they can read. And the agent doesn't see metallic design, doesn't respond to the perceived status of the brand, isn't swayed by VIP airport lounges. They see structured data. Points per dollar, cashback percentage, exchange rate, interest-free installment period for each retail category, effective acceptance at the merchant, quality of the rewards program.

"The choice between cards no longer happens when opening the physical wallet, but instead happens inside the agent's booth, milliseconds before each purchase."

The attributes that issuers compete for today in advertising and marketing are largely invisible to the agent. A card with an opaque points program, whose rules are hidden on a screen inaccessible via API, loses out to a card with a worse program, but one that is displayed in a machine-readable format.

A credit card that offers three interest-free months for a specific category, but whose terms only appear on the bill, does not compete against a card with inferior terms, but which are documented in a structured way.

The practical consequence for issuers is that they will need to do, for the sake of card benefits, what retailers are needing to do for catalogs: make them machine-readable. Points programs with clear rules and an open API. Available credit limits that can be checked in real time.

Special installment payment conditions documented in a structured format, by MCC. Predictable approval history, without unexplained refusals that harm the card's ranking by the agent. Those who do this first get on the shortlist of cards preferred by agents; those who don't, lose market share, even without losing customers.

The choice between cards no longer happens when opening a physical wallet, but instead happens inside the ATM, milliseconds before each purchase. It's a silent shift, but with significant consequences for the credit card economy.

There is also a dimension that is silently shifting. Chargebacks , which traditionally deal with disputes over who made a purchase, are now addressing a different question: was the agent operating within their authorized mandate?

If so, the responsibility lies with the merchant or the issuer, depending on the type of problem. If not, a new attribution issue arises, involving the agent's operating AI platform. This is a new class of dispute, which the card network protocols are still developing, and which will appear in significant volume as soon as the model scales.

The migration to A2A opens the game again.

There's a longer-term scenario that's often underestimated in current analyses. The card networks are building their agent-based commerce on top of the card payment system, where they have a dominant position. That makes sense. But the customer isn't loyal to the card payment system itself. They're loyal to the agent executing the purchase. If the agent, based on criteria like best price, best conditions, best experience, chooses to pay via Pix instead of card, it happens automatically, without the customer needing to think about it.

In Brazil, this scenario is closer than in other markets. The automatic Pix system is already operational as a programmable mandate for recurring payments. Payment Transaction Initiators (ITPs) authorized by the Central Bank , which have been operatingin Open Finance since 2021, have the consent, identity, and initiation infrastructure ready—on a rail that is already natively A2A. Stablecoins are emerging as an alternative rail, especially for cross-border transactions. Pay-by-bank is gaining ground in the rest of the world with regulations like PSD3 in Europe.

Recently, an ITP authorized by the Central Bank announced an MCP server running directly on the Pix rail. MCP, short for Model Context Protocol , is an open standard created by Anthropic in 2024 to connect AI agents to external services, and was adopted across the industry starting in 2025.

"The Automatic Pix system is already operational as a programmable mandate, and extending the logic to incorporate a more granular scope is a technically viable path within the existing architecture."

Making Pix accessible via MCP means, in practical terms, that any agent can initiate a Pix transaction on behalf of the client, within a previously authorized mandate. The architecture is shorter than that of the card, since Pix is natively A2A, bank to bank.

This movement is a record of what is already in operation, not a projection of what may come to be. Part of the agentic A2A design has already moved beyond the conceptual stage: in small volume, but with architecture functioning from end to end.

When volume begins to migrate from cards to A2A, the competition between the layers fundamentally changes. Card networks lose some of the structural control they have today. Agents gain decision-making power over payment methods. Acquirers that position themselves as A2A aggregators capture a significant share.

Merchants who accept all payment methods have an advantage. And issuers who treat Pix Automático as a strategic payment method, and not as a competitor to cards, have time to adapt.

In other markets, this scenario is a five- to ten-year timeframe. In Brazil, there are reasons to believe it could be faster. The Automatic Pix system is already operational as a programmable mandate, and extending the logic to incorporate a more granular scope—categories, brands, transaction limits—is a technically viable path within the existing architecture.

Considering the Central Bank of Brazil's history of iterative innovation and the Brazilian pace of adoption of new payment methods, it would not be surprising to see "Pix with an agency mandate" enter the regulatory agenda in the next two or three years. Companies' strategies today need to consider this possibility. Those betting everything on being dominant in the card payment system may be exposed if that system changes.

Brazil needs to decide before the game is over.

The series concludes here, with some observations on what is specifically at stake in Brazil.

I have the impression that the Brazilian payments ecosystem is in a more favorable position than it recognizes to influence the design of agency commerce. It has relevant structural assets: a consolidated Pix system, advanced Open Finance, robust digital identity providers, agile regulation, and a historical willingness to take a stand before the market consolidates standards.

It has institutional capacity through the Central Bank of Brazil (Bacen), Pagos, Abecs, Febraban , and ABFintechs. It has a large enough market to dictate terms. And it has ITPs in production, with orchestration infrastructure—agent, identity, mandate, bank, transaction—already implemented and running.

The window of influence doesn't stay open indefinitely. International standards are being defined now, in 2026, with major participation from American brands and big tech companies, and organizations like NIST and CSA. If Brazil wants to influence, it's now, not in three years.

"As with any structural transition, the final question is not technical. It is political, in the broadest sense of the term: what kind of agentic trade do we want to have in Brazil?"

Some decisions seem more urgent than others. Local authorities need to define their role in relation to global protocols, not just passively adopt them. Brazilian digital identity providers need to decide whether to enter KYA — Know Your Agent — in a coordinated or fragmented way.

Retailers need to build a collective stance, perhaps through associations, on the merchant-owned mandate as a defensible principle. And ITPs need to decide whether they operate merely as technical connectors or build a layer of intelligence on top of the traffic they orchestrate.

As with any structural transition, the final question is not technical. It is political, in the broadest sense of the term: what kind of agentic commerce do we want to have in Brazil? One in which a few global platforms capture value and local actors become supporting players? One in which foreign brands determine identity rules for the entire world? One in which the Brazilian retailer maintains a strategic position and the Brazilian customer continues to have real choices?

The answers to these questions are not written. They are being written now, and Brazil can and should be part of that writing.

I conclude the series here, thanking the readers who followed the three texts. The questions I raised don't have definitive answers—the topic is too new for that—but I hope I've helped to organize the picture, illuminate the pieces, and suggest directions worth paying attention to. The conversation continues in the coming months, and I will continue to comment here on what is relevant.

* Edson Santos is a payment methods specialist with over 25 years of experience. He is a partner at Colink Business Consulting, and a strategic advisor to companies in the financial and technology sectors. He is the author of "From Barter to Financial Inclusion" and co-author of "Payments 4.0 — The forces that are transforming the Brazilian market".