In the third episode of Conexión Fintech, we were joined by:
Hao Wang, Head of Product Management, Microsoft
Randall Davies, Head of Business, Skyfire
Guida Sousa, SVP Digital Payments LAC, Mastercard
Our CEO, Andrés Fontao, hosted this episode of the series.
The conversation explored one of the most significant structural shifts facing the financial ecosystem today: the transition toward Agentic Commerce.
More than an improvement in user experience, Agentic Commerce introduces a new operational layer—autonomous agents capable of executing financial decisions under programmable delegation.
For Latin America, this represents a strategic opportunity. But it also presents deep challenges around infrastructure, governance, and trust.
The Three Models of Agentic Commerce
There are three primary architectures:
Agent-to-Site
This is the model closest to today’s commerce. An agent navigates a merchant’s website the way a human would—searching for products, selecting items, adding them to the cart, and completing the purchase. It represents intelligent automation built on existing infrastructure, as it does not fundamentally transform the underlying payments system.
Intermediated Agent-to-Site
In this model, a trusted intermediary is introduced. For example, Microsoft Copilot can understand user intent and discover products on their behalf. It can curate inventory and facilitate checkout within the same platform.
This architecture adds an institutional layer of trust. From Microsoft’s perspective, it holds the strongest near-term potential in Latin America. Why? Because scaling in the region depends on two structural factors:
Trust
Infrastructure
In markets where digital fraud remains a significant concern, consumers and merchants are not yet ready to delegate payments directly to autonomous agents without institutional backing.
Agent-to-Agent
This is the most advanced and forward-looking model. A consumer’s AI assistant negotiates directly with the merchant’s agent by exchanging structured data, validating inventory, and executing transactions.
It is an elegant and efficient model, but it requires maturity in digital identity, delegated authorization, and protocol standardization.
As Hao Wang explains:
“AI will automatically track inventory and place the order on your behalf. More importantly, AI could even find a coupon for you if milk happens to be on sale that day. The discovery phase will change; however, settlement, backend processes, and financial rules—banks, credit card settlement—will remain. The process will simply become faster and more convenient for the user. In the future, as a customer, you may have multiple agents: one for shopping, another for travel, and another to help you book a restaurant reservation.”
— Hao Wang, Head of Product Management at Microsoft
From User Experience to Structural Infrastructure
Agentic Commerce does not eliminate traditional financial infrastructure.
Authorization.
Clearing.
Settlement.
Risk management.
All remain essential.
Innovation occurs at the discovery and decision layer. Responsibility remains anchored in financial infrastructure. Microsoft enables discovery and intent. Mastercard enables secure settlement through tokenization and advanced authentication. Skyfire introduces verifiable identity for agents.
The system evolves. It does not disappear.
Infrastructure, Delegation, and the Emergence of “Know Your Agent” (KYA)
Agentic Commerce elevates traditional financial infrastructure rather than replacing it. As the Fintech sector in the region now captures 46% of total technology venture capital investment, innovation is shifting toward the discovery and decision layer.
The fundamental shift is programmable delegation. To secure this process, Randall Davies introduces the concept of Know Your Agent (KYA). In his words, real utility will arrive when agents “not only recommend, but can act, transact, and optimize autonomously on behalf of consumers or businesses.”
Tokenization and Authentication as Anchors of Trust
Guida Sousa emphasized that for this model to function in our region, security must be structural.
“All transactions must be tokenized, as this significantly increases security and reduces fraud,” Sousa noted. She also highlighted the asymmetry in ecosystem readiness: “In Latin America, the vast majority of issuers are ready… the gap lies more on the acceptance and acquiring side.”
Programmable Wallets and the Evolution of Risk
Agents operate through constrained wallets that include:
Tokenized funding sources
Predefined authorization rules
Detailed audit trails
Immediate suspension capability
Issuers will retain their authorization engines, but they will need to evolve to distinguish between:
Direct human behavior
Authorized machine-to-machine activity
This introduces a new level of sophistication in fraud analytics.
Protocols and Interoperability
We are currently in an early stage of development, where multiple emerging protocols coexist. This is far from a winner-takes-all market. The strategic position is clear: the future will be built on openness, interoperability, and a layered ecosystem architecture.
While standardization will emerge organically over time, the immediate priority is to enable continuous innovation without slowing technological adoption.
Regulatory Implications
Institutions face new requirements:
Transparency in automated decision-making
Oversight of AI models
Explainability
Agent risk management
Although fully autonomous regulatory agents are not yet standard in the region, assisted compliance automation presents a clear opportunity.
Adoption will be gradual and will include human oversight.
The Preparation Roadmap for Latin America
Building “agent-ready” infrastructure requires:
API modernization
Broad tokenization penetration
Advanced authentication (biometrics, passkeys, FIDO)
KYA frameworks
Adaptation of risk models
Governance embedded by design
A 3–5 year consolidation window aligns with historical adoption cycles in payments.
Conclusion
Agentic Commerce does not replace financial infrastructure. It elevates it.
As highlighted by the experts in this episode of Conexión Fintech, trust, identity, and security become even more critical in an environment where AI agents execute autonomous decisions.
In Latin America, the intermediated model—supported by strong infrastructure, verifiable identity, and robust tokenization—will likely serve as the entry point for mass adoption. But the pace will depend on how prepared the ecosystem truly is.
Listen to the Full Episode
To dive deeper into the conversation with Microsoft, Mastercard, and Skyfire:
🎙️ Listen to the podcast episode