Microsoft and Mastercard are laying the foundations for a new stage of electronic commerce in Latin America. Now, AI agents not only recommend products, but execute payments under tokenization schemes and spending rules previously authorized by the user.
This is a market opportunity that, according to estimates by McKinsey and BCG, could reach US$5 trillion globally, transforming the regional digital economy.
During the webinar “The Agentic Commerce Revolution: How to Take Advantage of the $5 Trillion Opportunity in Latin America” organized by Finnosummit, executives from Microsoft, Mastercard and the fintech Skyfire detailed how the so-called agentic commerce is evolving from the experimental phase towards operating models based on tokenized payments, robust authentication and programmable budgets.
Agentic Commerce
Table of Contents
Hao Wang, Principal of Payment Product Management at Microsoft AI, explained that the company recently launched solutions such as Brand Agent and Copilot Checkout, integrated into its copilot ecosystem. Through Brand Agent, the user no longer “searches”, but rather “converses”. While with Copilot Checkout, the user no longer “pays” (manually), but rather “authorizes” the intent. Its objective is for the agent to function as a digital “concierge.”
From his perspective, “experience is no longer based on search.” In other words, instead of performing manual searches, the user can delegate a permanent intent, such as keeping a certain product available at home. The agent then monitors inventory, recognizes preferred brands and executes the purchase when appropriate.
However, Wang stressed that Microsoft focuses on the discovery and intent layer. “At Microsoft we can help the user discover and understand their intention (…) but to facilitate the transaction we need Mastercard, which is the one who helps settle the operation and complete the payment so that the purchase really happens,” he explained. That is, the execution of the payment and its settlement fall on the financial infrastructure.
Tokenized payments as a structural requirement
From Mastercard’s perspective, agent-driven e-commerce requires clear technical conditions to operate safely.
Guida Sousa, SVP Digital Payments for LAC at Mastercard, pointed out that the current model is not one of total autonomy, but rather supervised delegation. The user retains control, especially when registering the agent and defining payment permissions.
In this environment, tokenization becomes a central element. Transactions executed by agents should travel as tokens, not as exposed sensitive data, in order to reduce the risk of fraud and improve authorization by issuers.
In addition, authentication – including biometrics and passkeys – is part of the architecture necessary for the ecosystem to trust transactions initiated by software. The logic, according to managers, is that less friction facilitates adoption, but without tokenization and strong authentication, trust does not scale.
Programmable budget
As Microsoft redefines the experience and Mastercard secures the payment infrastructure, Skyfire introduces programmable budgeting as the third component of this alliance.
Randall Davies, Head of Business at Skyfire, explained that autonomous agents require a different model to the traditional KYC (Know Your Customer). «KYC exists to answer three questions: who is transacting, who it represents and whether it is authorized,” he said.
In an autonomous trading environment, the actor is a software agent that can operate continuously. Therefore, the “Know Your Agent” (KYA) concept not only verifies identity, but also reach, according to Davies.
“KYA focuses on verifying the agent’s platform (…) what permissions they have, what their spending scope is and whether their behavior can be audited and revoked when necessary,” he explained. Furthermore, toHe added that, in practice, this means that the agent can operate with a wallet with predefined rules.
You can authorize a maximum weekly amount, limit it to certain businesses or establish a specific validity period. If you go outside of those parameters, the authorization may be revoked before the payment clears.
Latin America: infrastructure and opportunity
Deployment in Latin America faces structural challenges. Davies noted that this is because the payments infrastructure is fragmented and there are more than 20 different regulatory frameworks, which complicates the scalability of the model.
However, the region has high penetration of digital wallets and real-time payment systems, which can facilitate adoption if tokenization and interoperability advance on the acquiring side.
In the medium term, executives agreed that agents will change the way users discover and choose products in e-commerce. However, the financial part—the settlement and clearing of payments—will continue to operate on the current infrastructure, now strengthened with tokenization, authentication, and programmable rules.
In this context, they also agreed that the advancement of e-commerce with AI agents does not depend only on the model recommending better, but on combining intelligence with tokenized payments and spending limits defined from the beginning.
It should be noted that, although Microsoft tools are already operational globally, their impact in Latin America depends on the adoption of local businesses. The real push comes from allowing AI to not only ‘suggest’ purchases, but to execute them from start to finish safely, thanks to Mastercard and Skyfire putting the financial locks (tokens and spending limits) so that the user can delegate their card without fear.
The key for experts is not to automate without control, but to allow decisions to be delegated while maintaining clear rules.
The main image was created by Mobile Time with AI.
date: 2026-02-14 00:03:00
Keep reading