Agentic AI Transforms Commerce as Major Payment Networks Enable Autonomous Transactions

How agentic AI will change commerce as we know it | Fortune

Throughout history, we've witnessed breakthroughs that didn't just improve things -- they've also completely reshaped human experiences. In commerce, similar paradigm shifts have repeatedly altered how people shop and how businesses operate: the first cash register changed bookkeeping; the introduction of the barcode completely changed global logistics; and the arrival of the internet changed our reliance on physical stores. Today, the world of retail and payments stands on the verge of another such monumental moment: agentic commerce. Agentic commerce is a fundamental shift that will reshape how consumers discover, search, and purchase products. In the coming years, shoppers will embrace AI agents that discover products, compare options, negotiate prices, and complete purchases to give them exactly what they want, at prices that work for them. Unlike predictive AI (which helps us forecast), or generative AI (which helps us create), agentic AI takes the crucial step of taking action on a person or company's behalf. It goes beyond traditional e-commerce to a truly assistive experience -- whether on a consumer's own device or through a retailer's site -- that feels like a personal shopper for every consumer. By next year, we anticipate a significant rise in consumers interacting with AI agents to manage their shopping, from initial intent to final receipt. For merchants, the agentic era introduces two new interaction models for connecting with customers. First, merchants can own the consumer experience from product discovery to checkout. This model requires creating a branded, conversational agent that can interact with a shopper's AI agent. For example, a consumer could instruct their agent: "I'm going to the Canadian Rockies in August and am not sure what to wear. Can you recommend a couple of outfits in my style?" The merchant's agent could communicate with the shopper's AI agent to access permissioned shopping data to provide more relevant recommendations based on the consumer's style and budget preferences. The second model involves operating as a fluid ecosystem player, prioritizing the sale no matter where it originates -- from a consumer's agent, another merchant's platform, or from something different altogether, like a social app. Imagine a consumer asks the merchant's agent to purchase a product that is not in its catalog or is currently out of stock. Instead of losing the sale, the agent could interact with other retailers' agents to source the item, complete the transaction, and fulfill the order -- creating a frictionless and endless shopping experience that puts the shopper's needs first. For payments providers, agentic commerce is a fundamental shift in the transaction process. It replaces the standard checkout experience with a direct, automated connection between personal AI agents and merchant AI agents. These smart agents can handle everything from price tracking to fraud detection, making transactions faster and inherently more secure. However, having a human in the loop will be important to ensure that critical, high-value transactions initiated by autonomous agents on behalf of a person are reviewed and verified by that person. To enable these new interaction models, interoperability is crucial, and the industry is championing protocols like the Agent Payments Protocol (AP2), an open, payment-agnostic framework for secure, trusted, and seamless agent-led transactions that already has partners like Mastercard and PayPal on board. With AP2, both consumers and merchants can unlock new ways of interacting with each other. For example, a user could instruct their AI agent to book a round-trip flight and a hotel for a specific weekend with a total budget of $1,500, and then the agent can find a combination that fits and securely book both simultaneously. The shift to agentic commerce is happening now. To take advantage, I believe there are four core areas where executives must focus: There's no doubt the future of commerce will be agentic. The time to build the foundation for this industry shift is now. As AI becomes more sophisticated, commerce will move far beyond transactions to anticipating and managing our needs autonomously. Business leaders who act decisively today won't just keep pace; they will help engineer the future.

Agentic AI: The factory that's scaling finance at Money20/20 USA | bobsguide

The days of "AI is coming" are over. It's here, and it's wearing a name tag. At Money20/20, leaders from TD Bank, Stripe, and Fiserv, alongside NVIDIA, spilled the secrets on how they're going from simple prediction to full-blown, autonomous AI Agents. Here's why your product road map is about to get a massive upgrade. [...] The days of "AI is coming" are over. It's here, and it's wearing a name tag. At Money20/20, leaders from TD Bank, Stripe, and Fiserv, alongside NVIDIA, spilled the secrets on how they're going from simple prediction to full-blown, autonomous AI Agents. Here's why your product road map is about to get a massive upgrade. The financial world has officially reached an inflection point. For a decade, we relied on "predictive AI": crunching tabular data to give us insights. It was smart, but it was essentially a high-powered suggestion box. Now, a new, more thrilling technology has landed: Agentic AI. This isn't just about suggesting an action; it's about the AI taking the action itself, autonomously running core workflows. As Pahal Patangia of NVIDIA put it, this is the rush toward a safer, faster, and smarter commerce ecosystem. The Fraud Fight Gets a Transformer Upgrade The biggest headline for security and risk leaders? Traditional fraud models are officially obsolete. Josh Ackerman of Stripe detailed the new frontier: Foundation Models. Stripe, which processes a cool $1.4 trillion (that's 1.3% of global GDP), built the world's first payments foundation model. Think of it as an omniscient payments brain that sees patterns no human or old algorithm ever could. * The Results Are Wild: In the past year, while industry-wide fraud rates climbed by 15%, Stripe's transformer-based model slashed fraud on its ecosystem by 17%. That's not a tweak; it's a structural defense against the digital bad guys. * The Agentic Commerce Paradox: The new problem? Agentic commerce, where bots (like the ChatGPT instant checkout) make purchases on your behalf. This is a massive leap for convenience, but it creates a trust gap. Stripe is now pioneering ways to actively "underwrite trust" for these agents, ensuring a purchase is made by a good actor, not a malicious bot. TD Bank's Sumee Seetharaman confirmed this trend, revealing they built their own predictive foundation model, PRISM. The payoff: it picked up customer nuances missed by older models, leading to a phenomenal boost in personalization speed and accuracy across the bank. From Wall Street to Main Street: AI for the Little Guy The real economic opportunity, according to Fiserv's Sanjay Saraf, is taking this massive AI power and distributing it to the "edge" specifically, to Small and Medium Businesses (SMBs). "This is the first time I feel that data, data science, and Agent tech stuff can really bring inclusivity," Saraf declared. SMB owners are juggling the front office, the kitchen, and payroll all at once. For them, an AI factory isn't just a cost-saver; it's a life-saver. This democratization happens at the nexus of agentic commerce and embedded finance: Building the AI Factory: It's All About the Plumbing

Agentic Commerce: Excitement, Caution, and Confidence: By Anurag Mohapatra

The next era of digital payments might not begin with instant payments, stablecoins, or a brand-new rail. It might start with something far quieter but equally transformative: software that can make payments on our behalf. As of 2025 we have already seen applications that can search, catlog and suggest items for us to shop. Agentic Commerce, takes this to the next steps where AI can search, decide, and pay within limits we define. And now, two of the world's largest payment networks, Mastercard and Visa, have taken the first step to make it real. Their announcements marked a quiet but historic shift. Card networks are opening their rails to AI-powered agents that can transact securely using the same infrastructure we already trust today. This isn't going to be an alarmist take about the dangers of agentic banking or a glowing endorsement either. As someone who works at the intersection of payments and fraud prevention, I feel both excitement and caution. I believe this next step can make payments more efficient and secure, but it will also test how well we can adapt our fraud defenses to a new kind of participant: the intelligent agent. When I asked an executive from a large American bank about their plans for agent-led transactions, he said it was still too early to discuss specifics. While that is true, anyone following the pace of innovation in AI is aware of how quickly things are changing. Each week brings a new model, a new capability, and a new partnership. Payments are simply the next frontier. So what exactly are Mastercard and Visa promising? In simple terms, they enable trusted software agents to initiate payments on behalf of individuals. Today, if I use a mobile app to order coffee, I am the one approving the transaction. In the future, an AI agent could do this for me. It would know my usual order, check the available balance, look for offers, and complete the purchase without me pressing pay. The key difference is that I would have given it permission or "agency" to act within clear boundaries. To make that possible, Mastercard has introduced the concept of registered agents. Only agents approved and listed in their network registry will be allowed to transact. The Issuing banks will decide whether their customers can enable a particular agent and what controls to apply. Each registered agent can have its own spending budget, transaction limits by merchant type, and time or location restrictions. If anything seems off, the bank can disable that agent immediately without blocking the entire card. Visa's Intelligent Commerce follows a similar model. It focuses on how agents discover, decide, and pay securely using existing network tokens. Both initiatives share the same goal. They do not change the rails of payments. They redefine who is allowed to use them. To understand the potential, it helps to look at what is already happening elsewhere. In China, Alipay and Luckin Coffee have been experimenting with conversational ordering. Customers can chat with a digital assistant that takes their order, confirms the price, and completes payment in one seamless interaction. That is agentic commerce at a single-merchant level. It works beautifully within one ecosystem. Now imagine the same concept scaled across the entire Mastercard or Visa network. Suddenly, we are talking not about millions of payments but billions, spanning every merchant that accepts a card. That is the magnitude of change when networks embrace agentic commerce. As the holiday season approaches, it is easy to picture what this might look like. Imagine building your Christmas shopping list, setting a spending limit, and telling your digital agent to take care of the rest. It scours the internet for the best deals, cross-checks prices, applies loyalty points, and executes purchases within your budget. The shopping still happens on familiar card networks, only now the buyer is a piece of software acting responsibly on your behalf. The reason this matters is scale. Payments only change the world when they scale. Even if only one percent of card transactions become agent-initiated, that represents hundreds of billions of dollars each year. A fraction of adoption will still require banks, merchants, and fraud systems to evolve. This is not a futuristic side project. It is a near-term reality with massive implications for how we define trust and risk. Whenever technology shifts, risk shifts too. The question is not whether agentic commerce will bring new fraud challenges, but how different those challenges will be from the ones we face today. In human-led commerce, the weak link is usually the person. Fraudsters exploit our emotions. They trick us with phishing emails, fake customer support calls, or cleverly timed messages. We rely on one-time passwords, behavioral biometrics, and device checks to confirm that the person transacting is genuine. In agent-led commerce, the vulnerability moves from the person to the software. Instead of persuading a human to click a link, an attacker might try to manipulate an agent. They could inject malicious instructions, steal cryptographic keys, or compromise the agent's runtime environment so that it behaves differently while still appearing legitimate. These are subtle, technical threats rather than emotional ones. Detecting them will require a new kind of visibility. The good news is that agentic commerce is being built with security at its core. Every registered agent will have a verified identity, a cryptographic key that proves it holds permission to act, and a defined scope that limits where and how it can spend. A compromised token is useless without its matching key. A rogue agent that tries to act outside its limits will be declined automatically. If an issuer suspects something unusual, it can revoke that agent's authorization instantly while keeping the rest of the card active. Fraud controls will evolve to focus on consistency rather than intuition. Instead of asking, "Does this person look like themselves?" the question becomes, "Is this agent behaving as it should?" Banks and payment providers will monitor agent IDs, versions, and transaction patterns the same way they monitor device fingerprints today. If an agent that typically executes one small payment per day suddenly initiates dozens from a new network address, that pattern will raise an alert. In other words, we shift from behavioral biometrics for people to behavioral analytics for software. These same principles strengthen transparency. Every transaction initiated by an agent will carry clear data about who started it, from where, under what policy, and within what limits. That level of provenance in knowing who initiated a payment, where it originated, and why is the cornerstone of effective fraud control. Instead of guessing at intent after the fact, risk systems can rely on verifiable data at the moment of authorization. The opportunity for the broader fraud prevention community is enormous. New data points will enter our risk models: agent identity, provider, runtime environment, key fingerprints, and transaction scope. Our challenge will be to interpret them intelligently. Fraud prevention has continually evolved in tandem with technological advancements. We adapted to the rise of e-commerce, to mobile banking, to real-time payments. Agentic commerce is simply the next chapter. Some might still ask whether giving AI agents the power to transact makes payments more dangerous. My view is the opposite. It can make them safer. The traditional internet commerce model relies on stored card numbers, saved passwords, and fragile layers of user verification. Those are the reasons so much of today's fraud exists. Agentic commerce replaces that system with something far stronger. It uses tokenized credentials that can be limited to specific agents, cryptographic proofs that prevent replay attacks, and real-time revocation that stops abuse instantly. It is not foolproof, but it is a significant leap in control and traceability. This moment reminds me of earlier milestones in payments. When magnetic stripes replaced handwritten signatures, people worried that copying cards would be too easy. When chip and PIN arrived, many believed customers would never remember their codes. Even mobile wallets were once dismissed as unnecessary. Each of those transitions felt uncertain at first. And each one, in hindsight, made payments more secure and convenient. Agentic commerce will likely follow the same trajectory. We are at the start of that journey. The rollout will be gradual, and the learning curve will be steep. But I see no reason to fear it. This is an opportunity to reimagine how trust is established in digital payments. By combining cryptography, registration, and thoughtful oversight, we can create a framework where intelligent agents transact responsibly, transparently, and safely. I don't have all the answers about what agentic commerce will look like a year from now, but I'm excited to find out. What matters is that we approach it with curiosity, collaboration, and the confidence that trust can evolve alongside technology. Because for the first time in payments, trust will not just be verified by people. It will be coded, cryptographically bound, and continuously learned. And that is something worth getting excited about.

The Prompt Economy Has Arrived. Now Comes the Hard Part | PYMNTS.com

By completing this form, you agree to receive marketing communications from PYMNTS and to the sharing of your information with our sponsor, if applicable, in accordance with our Privacy Policy and Terms and Conditions. The authors describe agentic AI as a "team of digital colleagues" capable of autonomous reasoning, planning and coordination across departments, freeing organizations from the limits of rules-based automation. Case studies from Hitachi Digital, NTT DATA and Bigblue show how these systems can rewire complex operations by turning fragmented, multi-system environments into seamless, outcome-driven workflows. To do this successfully, organizations must appoint "mission owners" who define and oversee goals, empower AI agents to act toward them and structure work around results rather than departments. The authors outline three imperatives for building agentic AI systems: design around outcomes and clear accountability; unlock data silos through shared business logic rather than perfect data centralization; and develop leaders and governance frameworks capable of supervising autonomous systems responsibly. They stress that agentic AI doesn't demand flawless or unified data. It does demand interoperable systems, clear logic and human oversight. Leaders, they write, must be willing to delegate to digital teammates while maintaining guardrails for transparency, security and escalation. Companies that begin with high-friction workflows, assign ownership and build governance into design will, they conclude, "write the playbook for the operating model of the future." "Agentic AI marks a significant evolution in intelligent automation," the article states. "Guided by human supervision, it becomes an intelligent layer that links scattered processes and cuts through silos into a more seamless flow." In order for agentic commerce to work, it will of course need to be secured. Cloudflare's October 2025 blog post, "Securing Agentic Commerce: Helping AI Agents Transact with Visa and Mastercard," by Rohin Lohe and Will Allen, explains how the next phase of digital payments, commerce executed by AI agents, will require a new trust framework between consumers, merchants and payment networks. To that end, Cloudflare is partnering with Visa and Mastercard to develop infrastructure that authenticates AI agents through cryptographic protocols. Visa's Trusted Agent Protocol and Mastercard's Agent Pay both rely on Web Bot Auth, a Cloudflare proposal that uses public-key cryptography and HTTP message signatures to verify that an agent acting on a consumer's behalf is legitimate, authorized and non-replayable. This system lets merchants instantly recognize registered agents and determine whether they are browsing or making payments, without confusing helpful AI assistants for malicious bots. Cloudflare's approach makes agentic transactions verifiable across the full ecosystem -- developers, merchants, networks and consumers -- without overhauling existing merchant infrastructure. The authentication flow allows Cloudflare to validate agent requests by checking cryptographic keys, timestamps and nonce uniqueness before forwarding a verified transaction. Over the coming months, Cloudflare plans to integrate Visa's and Mastercard's protocols into its Agent SDK, enabling developers to easily manage agent keys, create valid signatures and transact securely using standard APIs. This shared security layer, the company notes, ensures that trusted agents can interact with merchant sites safely while maintaining full compliance with payment networks' trust requirements. As the authors put it, "The era of agentic commerce is coming, and it brings with it significant new challenges for security." Banks are starting to add their own flavor of thought leadership to the agentic debate as well. In a piece titled "Agentic AI: A Strategic Shift in Business Evolution," UBS Chief Technology Officer and Technology Fellow Mitra Heravizadeh positions agentic AI as a foundational inflection point, not a passing trend. For UBS, becoming an "AI-enabled institution" means embedding autonomous intelligence directly into its operational fabric, from decision-making and market prediction to client engagement. Heravizadeh writes that AI agents will function as "intelligent collaborators," executing multi-step tasks once handled by humans and enabling a move from reactive problem-solving to proactive opportunity creation. To achieve this, UBS is re-architecting its digital infrastructure around agility, interoperability and continuous learning, what it calls the core design principles of the agentic enterprise. The post details how UBS is restructuring its operating model to manage and govern AI agents responsibly. IT teams will evolve to oversee autonomous systems through defined permissions, behavioral monitoring and compliance oversight. To scale this transformation, UBS has created an Agentic AI Center of Excellence (CoE) to coordinate innovation and maintain ethical, secure and standardized deployment across engineering, data science and compliance teams. The CoE's mission is to establish a shared language and governance blueprint so AI agents can "communicate, collaborate, and operate together in a compliant, secure, and ethical manner." As Heravizadeh puts it, "Agentic AI is not a future trend -- it's a present reality," one that will shape the next era of financial innovation and institutional intelligence.