Why banks and credit unions should start planning now for AI-initiated commerce, programmable money, and a new fraud and control framework

Agentic payments are emerging as part of a broader shift from AI that informs decisions to AI that executes them. The interface becomes conversational, the checkout becomes programmable, and the payment becomes one step in a machine executed workflow. This has implications for the role of financial institutions in payment credentials, merchant acceptance, fraud models, liability, and how value moves across card, account, and increasingly tokenized money rails.

The institutions that move first will have a better chance of becoming the permissioning and trust layer for consumer-controlled agents rather than being relegated to invisible funding accounts in someone else’s ecosystem.

Agentic payments could change the definition of checkout

A customer no longer is required to complete a purchase journey directly. Instead, the customer expresses intent, sets conditions, and delegates the work to an AI system. That system can research, compare, negotiate, decide, and execute within defined limits.

Traditionally, a customer searched for an item, visited a merchant site, reviewed details, entered credentials, and authorized payment. In the agentic model, the customer might simply say, “Find the cheapest refundable flight, book it, and choose an aisle seat,” or “Renew this membership if the annual price is below a defined threshold.” The agent evaluates options, checks policy constraints, and completes the transaction.

The broader point is that agentic commerce is starting to look less like a product feature and more like a potential structural market change. It is moving from the lab into planning discussions across payments, digital banking, ecommerce, and treasury.

Why now?

Several conditions are converging at once.

The underlying AI systems have become more capable of planning and executing multi-step workflows. What was recently a purely prompt-based experience is evolving into systems that can reason across tasks, access tools, operate under policy, and autonomously act.

Market demand is clearly building. Consumers are already using generative AI for product research, price comparison, and purchase assistance. The report The State of Fintech 2026 from Fintech Brainfood, reinforces that same market trajectory, highlighting that AI-assisted purchasing affected 1 in 6 Black Friday purchases and that AI-driven clicks are generating materially better conversion than traditional search. That combination, rising usage plus stronger commerce outcomes, is the kind of pattern that tends to attract infrastructure investment quickly.  Additionally, McKinsey & Company projects agentic commerce will hit $3-5T globally by 2030. $1T in the U.S. alone.

Why this matters to banks and credit unions

If merchants begin optimizing for agent readable offers, structured checkout data, and machine consumable policies, then discovery and conversion may increasingly happen outside the financial institution’s existing digital channels. If networks and platforms become the default credential, rules, and authorization layer for agents, then the institution risks becoming little more than a silent funding source.

The strategic opportunity for banks and credit unions, however, is substantial. They already manage identity, risk, transaction controls, and regulate money movement. In an agentic model, those capabilities become even more valuable. The institution that can define what an agent is allowed to do, how far it can go, when human intervention is required, and how auditability is preserved, which may become more important than the institution that simply holds the balance.

Stablecoins are a natural fit for agentic payments

Stablecoins are not the entire answer to agentic payments, but they are unusually well aligned with many of the emerging use cases.

Agentic payments work best when money can be embedded into software logic. Stablecoin wallets can enforce spending policies, interact with smart contracts, support escrow, release funds on completion, and generate a machine verifiable payment record inside the same workflow that triggered the transaction. That is much closer to how autonomous software wants to operate than traditional payment flows built around human present checkout steps.

Stablecoins offer lower cost and faster settlement. That matters in cross border, API native, and high frequency use cases where traditional economics can be unattractive. Recent market estimates suggest stablecoin payment activity more than doubled in 2025, reaching approximately $390 billion to $400 billion, suggesting commercial use is real and growing.

Visa has reported growing stablecoin settlement activity, and providers such as Coinbase are explicitly positioning stablecoin wallets as native spending tools for AI agents. More broadly, programmable payments are becoming more relevant as AI systems move closer to direct action.

That does not mean cards disappear. It means the future is more likely to be card credentials for some contexts and wallet-based stablecoin credentials for others. In other words, the strategic issue for financial institutions is not cards versus stablecoins, it is which credential belongs in which delegated payment context.

Fraud models and merchant controls will have to change

This may be the most underappreciated implication for the banking industry.

For years, fraud teams and merchant websites were trained to treat bot like activity as inherently suspicious. That made sense in a world where non-human traffic often meant screen scraping, credential stuffing, account takeover, or abuse. But in an agentic commerce environment, some machine-initiated behavior will be entirely legitimate.

A valid customer agent may compare dozens of offers, revisit sites repeatedly, parse product and policy data, trigger checkout flows programmatically and behave in ways that look remarkably similar to the automated patterns merchants historically tried to block. Legacy controls such as CAPTCHA, anti-scraping defenses, bot mitigation rules, and session anomaly triggers were not designed for this world. As a result, those controls may begin excluding or degrading legitimate agent purchase activity.

That has two major consequences.

First, fraud models will need to distinguish between malicious automation and authorized automation. The old logic of human equals good and bot equals bad will fail. Institutions will need richer trust signals around agent identity, permission scope, transaction intent, wallet or credential provenance, runtime integrity, and human override rights.

Second, merchant acceptance models will need to evolve. Merchants that continue treating all autonomous behavior as suspect may end up blocking real purchase intent. In effect, some of the control mechanisms built to protect ecommerce could become obstacles to the next legitimate commerce channel. That has implications for false positives, abandonment, liability allocation, dispute handling, and merchant conversion economics.

This is one reason the network and protocol players are emphasizing trust frameworks and standardization rather than just payment rails.

What banks and credit unions should do now

The window for positioning is open now, but it will not stay open indefinitely.

• Develop an explicit agentic payments point of view. This should cover retail, small business, treasury, operations, fraud, and digital channel strategy. Treat agentic payments as a strategic topic for 2026 through 2028 rather than as an innovation side project.
• Invest in the permissioning layer. Financial institutions should be thinking about transaction caps, merchant category restrictions, step up authentication, time-bound authority, revocation rights, audit trails, and receipt level explainability for consumer- controlled agents.
• Prepare for a dual credential world. Card credentials will remain important, but stablecoin and other digital asset wallets will likely become increasingly relevant in programmable, low cost, cross border, and machine to machine contexts. Evaluate both.
• Modernize the way you think about fraud. Agentic traffic will break legacy assumptions. Fraud teams should start planning for trusted agent identification, behavioral scoring for autonomous transactions, runtime and wallet risk, and revised dispute models.
• Rethink merchant and digital experience assumptions. Structured product data, agent readable policies, API ready checkout, and machine consumable terms will increasingly matter. In the same way that mobile friendliness once became table stakes, agent readiness may become the next baseline for digital commerce.

The Bottom Line

The payments industry has always evolved when a new interface changed how purchase intent was expressed. Cards changed the point of sale. Ecommerce changed checkout. Mobile changed authentication and credential storage. Agentic payments will change initiation itself.

Software will increasingly move from assisting financial decisions to executing them. When that happens, the payment becomes a programmable capability inside a delegated workflow. Stablecoins fit naturally into many of those workflows because they are programmable, efficient, and viable for very small value transactions.

The protocols now emerging are early signs of how this market may organize itself. And the institutions best positioned to win are the ones that become the trusted control layer between customer intent and autonomous execution.

Reach out to SRM and our AI practice lead and author, Larry Pruss, to have a conversation about this emerging and impactful payment technology.

Published by Larry Pruss on April 15, 2026.