The next transformation in banking will not be a simpler app or a friendlier chatbot. It will be the moment when artificial intelligence stops answering questions and starts executing financial decisions.
This new stage is known as agentic finance: systems capable of interpreting intent, comparing options, using tools, making decisions within defined limits, and completing transactions. It is no longer about asking an AI how much we spent this month, but about asking it to pay a bill when the timing is right, find the most affordable loan, renegotiate debt, or handle a supplier payment without going through five different systems.
The difference is enormous. In one case, AI advises. In the other, it operates.
The shift has already begun. A global Cambridge survey on AI in financial services shows that 52% of institutions in the sector are already piloting or deploying agentic AI, while 23% are in more advanced stages of scaling or transformation. In other words, more than half of the global financial system is already experimenting with technologies that do not just predict, but also coordinate, decide, and act.
In payments, the impact could be even greater. According to McKinsey, the global payments industry processes approximately 3.6 trillion transactions and generates around $2.5 trillion in revenue. If AI begins intervening in even a fraction of that flow, we are no longer talking about an interface improvement, but rather a new decision layer on one of the economy’s most sensitive infrastructures.
The central challenge will be trust. For an agent to move money, being intelligent is not enough: it must be authorized, auditable, and constrained. Verifiable identities, spending limits, tiered permissions, decision traceability, manipulation detection, and clear rules defining when an agent can act independently and when it must stop will all be required.
That is why it is no coincidence that Visa announced a collaboration with OpenAI in June 2026 to enable secure payments within agentic commerce experiences, nor that Mastercard is developing standards that allow agents to buy, pay, and operate without undermining trust in the system. The question is no longer whether AI-powered payments will exist, but who will control the credentials, permissions, and accountability when those payments take place.
For Latin America, this conversation is especially relevant. The region combines high levels of digital adoption, fragmented financial systems, persistent informality, and a significant need for financial inclusion. In this context, AI capable of organizing personal finances, comparing products, automating payments, or explaining complex terms could expand access and reduce friction for millions of individuals and small and medium-sized businesses.
But it can also amplify risks. Poor financial recommendations do not affect everyone equally. An authorization error can be critical for a small business operating with limited cash flow. An opaque credit decision can reinforce existing exclusions. And a poorly protected agent can become a new gateway for fraud, phishing, or manipulation.
The risk is already visible. The FBI reported that cybercrime losses reported in the United States exceeded $20 billion in 2025, with investment fraud among the leading categories. At the same time, deepfakes, synthetic voices, and hyper-personalized scams are making it increasingly difficult to distinguish between legitimate instructions and fabricated manipulation.
For this reason, agentic finance should not be viewed as total automation, but as responsible delegation. The goal is not to surrender control, but to redesign how it is shared.
The bank of the future must prepare for a world in which many operations are no longer initiated by humans clicking a button, but by agents acting under human instructions. This will require more robust APIs, stronger data governance, auditable decision engines, granular permissions, real-time fraud prevention, and interoperability.
Customers will not ask for an omnichannel experience. They will ask for problems to be solved. And whoever captures that intent first will not always be a bank: it could be an AI assistant, a digital wallet, a commerce platform, enterprise software, or an agent embedded in virtually any environment.
The opportunity for financial institutions will be to become the trust network on which these agents can operate. Because in finance, autonomy without control is not innovation—it is risk.
The wallet of the future may not have a will of its own. But it will have delegated agency. And when artificial intelligence begins moving money, the question will no longer be what a machine can answer. It will be what it can do on our behalf.
