AI Agents Reshape Payments as Agentic Commerce Moves From Concept to Critical Infrastructure

Agentic Commerce Is Here: Why PSPs Must Act Now to Stay Relevant: By Milko Filipov

Agentic commerce is moving from concept to reality faster than the industry expected. AI agents are increasingly steering how consumers discover products, compare options, build carts, and initiate purchases. While the long-term vision is fully autonomous payments, the near-term reality looks quite different. We are entering a semi-autonomous era: AI agents do the heavy lifting, but the user remains the final decision-maker. This hybrid phase represents a critical window for PSPs. The ecosystem is being reshaped upstream, well before the transaction reaches the checkout page, and payment providers must adapt if they intend to remain part of the journey. Although autonomous commerce is the end state, early implementations from platforms like ChatGPT and Google Gemini still rely on user confirmation before payment is executed. Agents prepare the entire flow, yet the final "buy" action still comes from a human. This subtle difference fundamentally alters how payment intent is created. The transaction originates with the user but is orchestrated by an agent acting on their behalf. This shift requires PSPs to rethink how they interpret intent, how they validate user consent, and how they support purchasing journeys that unfold across multiple contexts before reaching the payment initiation step. The systems that once assumed a human-driven, browser-based journey must now be ready for programmatic, multi-actor journeys led by intelligent agents. A profound structural shift is happening beneath the surface. Large language models are evolving into a new middle layer in the ecommerce and payment ecosystem. Discovery, evaluation, recommendations, cart assembly, and even checkout orchestration are increasingly mediated by LLMs. This is a new type of gatekeeper -- one that sits before the merchant website, before the PSP, and before any traditional conversion funnel. Agents are becoming the entry point, the navigation layer, and the decision engine of commerce. As this layer strengthens, it begins to shape which merchants, PSPs, and payment methods users encounter. The strategic implication is clear: if a PSP is invisible or incompatible with the LLM layer, it risks being bypassed entirely. Meanwhile, merchants continue operating with their existing setups. Some rely on a single PSP; others use orchestration platforms to work with multiple providers. These orchestration layers were built for traditional ecommerce flows -- websites, apps, and direct user interactions. Agent-driven commerce challenges these assumptions. When AI agents initiate and manage the journey, the orchestration rules must expand beyond routing logic. They must understand structured agent metadata, support delegated authorization flows, and accommodate transactions in which the visible customer experience happens outside the merchant's domain. Merchants will expect their PSP -- and their orchestration partners -- to handle this complexity seamlessly. If they cannot, merchants will gravitate toward PSPs that can. If PSPs and orchestration platforms are not prepared for agentic commerce, they face a gradual loss of relevance. Agents will prioritize providers that expose agent-ready APIs, support metadata-rich transactions, and offer verifiable, low-friction consent frameworks. In a world where the agent mediates the purchasing decision, compatibility becomes a determining factor in whether a PSP is even considered. This is an existential shift. The competitive advantage once tied to checkout design or merchant-side integration begins to erode. What matters is whether the PSP can plug natively into LLM ecosystems and enable reliable, contextualized payment flows. New entrants -- AI-native payment providers -- may seize this opportunity if traditional PSPs move too slowly. Despite the risks, this transition opens enormous potential for PSPs willing to modernize. Agentic commerce increases transaction frequency and introduces entirely new categories of purchases. PSPs can become foundational infrastructure for this new paradigm by offering agent identity verification, mandate management, contextual risk scoring, and intelligent routing across payment rails. In this semi-autonomous phase, trust becomes a new currency. PSPs are uniquely positioned to provide the frameworks and controls that ensure agents act on behalf of users securely and transparently. By doing so, they not only retain their relevance but become essential partners for both merchants and AI platforms. To unlock this opportunity, PSPs must rethink their technology and operating models. Agentic commerce requires API-first architectures, event-driven systems, richer metadata capabilities, and fraud engines that analyze agent behavior rather than human-only patterns. Legacy stacks will struggle to keep up. The winners will be PSPs that embrace openness, interoperability, and collaboration with the LLM platforms shaping the future of commerce. The shift toward agentic commerce is happening now, not in a distant future scenario. Even in its early stages, the purchasing journey is being reshaped long before the user confirms a payment. PSPs have a limited but powerful window to adapt -- modernize their infrastructure, integrate with emerging agent protocols, and redefine their role in an ecosystem increasingly orchestrated by intelligent intermediaries. Those who act early will set the standards for semi-autonomous payments and secure a privileged position in the future landscape. Those who wait risk being sidelined as new, AI-native competitors fill the gap.

Agentic Payments: When Money Starts To Think: By Nkahiseng Ralepeli

Agentic payments are moving from slideware to architecture. Programmable money, policy aware wallets, smart contracts as transaction logic, and AI agents that can hold budgets, reason about constraints, and execute on behalf of users are converging into a new payment pattern. The trigger for this piece was a conversation with Karthik, who helps lead Google's web3 work, about where agency really lives, in the card, in the wallet, in the merchant, or in the model that orchestrates them. The intuition is simple, once money behaves like software, and software can plan and act, payments stop being a screen you tap and start becoming an outcome you specify. Tell an agent "keep my cloud bill under $2,000, pay suppliers within 24 hours to secure a 1% discount, roll any excess into T-bills," and let smart contracts enforce terms, while regulated stablecoins and tokenized deposits move value across rails, without asking you to click "confirm" a hundred times. The incentive stack. For consumers, programmable money collapses chores into intents. Subscriptions renew only if usage justifies it, travel bookings hold funds in escrow until check in, refunds arrive instantly when conditions are met, and budgets enforce themselves with real time policy, for example, "groceries yes, takeout twice a week, alcohol not this month." Stablecoins make this portable, a user can carry a dollar, or rand, that clears any day, any hour, while smart contracts gate spend to approved categories. For businesses, agentic payments compress working capital cycles. Invoices become smart objects with terms, early pay discounts, and collateral eligibility expressed in code, while disbursements, rebates, and chargebacks run as workflows with proof trails. Stablecoin settlement shortens time to cash in cross border flows, while tokenized deposits preserve on-us speed and control inside the bank. For platforms and banks, the reward is stickier distribution and lower cost to serve. Service requests mutate into autonomous flows, KYC and sanction checks attach to tokens, reconciliation is event driven, and fraud controls move from after the fact to pre trade simulation, an agent refuses to route value if the risk policy would be violated five steps later. Smart contracts become policy appliances, they do not rely on memory or manual checks, they either pass or they do not. A tangible example, end-to-end. Picture a mid-sized seller on a marketplace. An inventory agent monitors SKUs and commits to restock if price and lead time thresholds are met. A treasury agent holds a spending envelope funded by tokenized deposits and/or a regulated stablecoin balance for cross border settlement. A payments agent scans open invoices, weighs early pay discounts against projected cash needs, checks supplier credentials and trade limits, and selects the cheapest, fastest rail available, RTP in one corridor, stablecoin in another, mobile money where relevant, with atomic delivery versus payment when counterparties support it. A risk agent simulates the path against AML policy and travel rule requirements, verifies counterparty proofs, and writes an on-chain audit artifact before funds move. Settlement executes through a smart contract escrow, releasing funds when the logistics oracle confirms receipt. When stock arrives, a revenue agent allocates receivables to sweep T-bill tokens overnight, and funds a just in time working capital line if inventory is slow to turn. No CSV stitching on Friday, the CFO still sets policy and reviews exceptions, the agents and contracts do the rest. Who is building, and what they seek to capture? Wallets and super apps want to own the intent surface, the place where a user expresses goals and grants authority. Their bet is that policy engines, secure delegation, and a library of atomic payment actions and smart contract templates become the new browser and search bar for money. Card networks and acquirers are integrating programmable settlement, off-card rails like stablecoins and RTP, and agent friendly credentials, so tokenized cards can authorise machine to machine payments with spend limits embedded, while contract based escrow removes disputes at the edge. Banks are equipping accounts with programmable entitlements, tokenised deposits, and guardrails that let agents move value safely inside the perimeter, while providing gateways to regulated stablecoins for ecosystem reach. Custody of contract keys and stablecoin reserves is not a side quest, it is core banking in a programmable world. Payment processors and PSPs are productising orchestration, routing, and compliance as APIs that agents can call deterministically. Cloud and AI providers are building agent frameworks with secure key custody, tool use, policy verification, and human in the loop controls, so enterprises can deploy payments agents without writing a custodian from scratch. Business models that pencil. There is clear revenue in policy and orchestration, enterprises pay for engines that compile business rules into executable flows with guarantees. Transaction economics improve through routing, providers monetise savings against baseline interchange and correspondent costs. Treasury and yield share appear when idle balances sit in tokenized cash instruments under strict liquidity buffers, and when stablecoin reserves earn permissible yield that subsidises fees. Compliance as a service becomes a line item, issuers and banks charge for attestations, travel rule messaging, and per event screening. Developer platforms create marketplaces for certified payment tools, dispute modules, escrow contracts, and invoice objects, with revenue share tied to usage. Premium assurance tiers emerge, verified suppliers only, guaranteed delivery windows enforced by contracts, refundable if policy not met, introducing service level pricing into payments. Design choices that separate durable platforms from headlines. Agency must be explicitly delegated and revoked, with scopes, spend caps, counterparties, and time windows that are machine enforceable, not PDF terms of service. Keys and policies need defense in depth, hardware security for custody, multi-party authorisation for large moves, and policy engines that are auditable and versioned, every change leaves a trail. Chain strategy should be multi rail, not everything runs on a blockchain, but when it does, semantics must be uniform across networks, fees predictable, and failover paths defined, with clear rules for when to use tokenized deposits, when to use stablecoins, and when to fall back to RTP or card. Cash leg certainty is non-negotiable, map use cases to the right instrument, and use smart contracts for escrow and conditional release only where they add correctness, not ideology. Compliance belongs in the flow, allow lists, credential proofs, sanctions screening, and travel rule messages should ride with the payment object, so an agent can prove policy satisfaction before it pays. Contract libraries should be curated and formally tested, upgrades controlled by governance, and emergency pause procedures documented. Human oversight should be designed, not improvised, exception queues, sampling, dual control for policy edits, simulated execution for high-risk flows, and explicit playbooks for reversing or freezing stablecoin transfers on compliant rails. Privacy needs tiered disclosure, counterparties get what they need to complete the trade, regulators and auditors get verifiable views, while public data is minimised through privacy preserving patterns, for example, hash commitments and off chain proofs. Competitive dynamics and strategic positioning. Distribution beats everything. The platform that sits closest to intent, payroll, procurement, accounting, consumer wallets, will set the default agents and contracts use. Banks with trusted brands and enterprise sales can win the policy and custody layer even if a partner owns the UI. Networks with acceptance will remain central, judged by how well they interoperate with non-card rails and contract-based settlement. Closed loops create margin, open loops create scale, the winners will operate a controlled core, policy, custody, identity, contract libraries, with easy off ramps to every rail customers need. Domestic instant payment schemes will handle local volume, cross border and supplier payments will favour stablecoin settlement where it is cheaper and programmable, and tokenized deposits will preserve deposit relationships for enterprise flows. Partnerships, rather than zero sum positioning, will define the fastest movers, banks with PSPs and cloud providers, networks with wallets, merchants with treasurers. Analytical signals to watch? Cycle time from intent to cleared payment should collapse, measured in seconds, with variance tight even at peak times, including cross border stablecoin legs. Cost per payment should fall versus baseline card and correspondent routes, net of compliance and contract overhead, with routing and contract choices explainable ex-post. Exception rates should trend down as policy coverage and contract reuse grow, and when exceptions occur, time to resolution and dispute reversal should be measured in minutes, helped by escrow releases and programmable refunds. Working capital metrics should improve, earlier recognition of receivables, faster settlement, more precise cash buffers, and higher capture of early pay discounts. Fraud and loss rates should decline with pre trade simulation, conditional contracts, and programmable constraints, not increase, and when losses occur, recovery times should be shorter due to freeze and claw back hooks on compliant stablecoins. User satisfaction should reflect less time in portals and more time in outcomes, fewer manual touches per thousand payments, and higher conversion in checkout flows that use instant, policy aware settlement. A measured outlook. The institutions that treat agentic payments as a product capability, not a lab demo, will ship faster. They will anchor custody and policy in bank grade controls, expose clean APIs, integrate RTP, card, tokenized deposits, and stablecoins, and curate contract libraries that encode business rules correctly. For innovators, the question is no longer whether agents will pay, it is who will define the policy engines, the contract standards, the credentials, and the rails they prefer. The firms that decide, govern, and execute with discipline will not only reduce today's frictions, they will open entirely new corridors of access for emerging market users, from instant cross border payroll and supplier finance, to dollar saving with guardrails, to micro insurance that pays in seconds, to credit that is priced on real cash flow, delivered by contracts that keep promises without paperwork.

Autonomous Agents Challenge Retailers to Earn Trust | 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. "Buy in one click" became a shorthand for the ideal shopping experience: instant, invisible, effortless. This required compressing the distance between desire and purchase, removing steps, and, if possible, turning every moment into a shoppable one. But that era is already coming to a close, thanks to the rise of agentic AI. Findings in the November 2025 Payments Orchestration Tracker® Series, a PYMNTS Intelligence collaboration with Spreedly, reveal that AI agents, autonomous systems that shop on consumers' behalf, are poised to take convenience to its logical endpoint. This emerging next wave of digital commerce won't require clicks at all. Consumers will define an outcome, like, "keep me stocked," "find me the best deal under $50," "buy a replacement when this is about to expire," and an agent will simply make it happen. The question isn't whether this capability will exist. It's how retailers will make consumers feel comfortable using it. And that's not an impossible challenge. In fact, the path forward is achievable with deliberate design and clearer than the hype suggests. Consumers rarely shop in straight lines. They respond to moods, moments, small impulses, social cues, and personal preferences that shift from week to week. They browse aspirationally, comparison-shop for reassurance, and splurge unpredictably. They buy some things purely on habit and others only after a deep emotional assessment. That's why a first step can be ensuring any agentic AI systems interpret consumer intent as dynamic and contextual. A system that understands "keep my pantry stocked" should also recognize that brands, budgets, or diet preferences might change; that a sale might justify a bigger order; that a holiday requires more variety; and that certain items still require human approval. After all, consumers don't need to control every step of an automated process to trust it. They just need to understand what is happening and why. The report found that four in five consumers (80%) are more inclined to make purchases when brands provide a personalized experience, making agentic AI systems that are framed as collaborative partners rather than rigid executors potentially more attractive. Read the report: AI's New Age: Building Human Intent and Trust Into Agentic AI Still, one of the most important shifts required to make agentic retail viable is redefining what "control" means for consumers. Control does not require manually approving every action. It requires having the authority to set boundaries, adjust behavior, pause activity, and reverse decisions when needed. In practice, this means giving consumers the ability to shape agent autonomy on their terms. Some shoppers will want agents that suggest but never buy. Others will want agents that complete specific categories of recurring purchases. Some will delegate routine tasks entirely but maintain manual control over discretionary or emotionally driven categories like fashion, gifts, or beauty. The goal is not to automate everything; it is to automate the parts of shopping consumers are most willing to offload, without intruding on the parts they still enjoy. Agentic functionality will succeed when it blends into familiar environments rather than asking shoppers to adopt new behaviors. This continuity helps consumers see agentic systems not as a radical shift, but as an evolution of tools they already trust. Meeting customers where they are reduces friction and accelerates adoption far more effectively than dazzling them with novelty.

AI Agents Turn Issuers Into Real-Time Decision Makers | 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. However, for issuers, retailers and the platforms connecting them, the emergence of AI agents that can complete multistep tasks end-to-end isn't simply another feature cycle. It's the start of a potential structural realignment set to rewrite how shoppers handle everything from finding a product to choosing the best payment method to executing the transaction. "The real transformation happens when payments are built directly into AI-driven workflows," Marqeta Chief Technology and AI Officer Fouzi Husaini told PYMNTS. "We envision agentic commerce becoming just a standard part of how the payment ecosystem works," Husaini said, adding that the issuers who will thrive as agentic AI capabilities mature will not be passive processors but active enablers. This is not a distant-future scenario. Powered by real-time data, flexible payment infrastructure and rapidly maturing autonomous agents, the AI shift in commerce is already reshaping how retailers, issuers and platforms think about their roles in the buying journey. In the next operating system of commerce, the payment isn't the end of the story but the beginning. This represents a shift from user-pulled to system-pushed commerce. As retailers have learned over the past decade, discovery is no longer the problem. Recommendation engines and content-driven shopping behaviors have multiplied opportunities for consumers to encounter new products. What has remained fractured is the leap from intent to purchase. Users still leave apps, break flows or fall out of channels altogether before converting. Each subsequent step serves as an opportunity for friction, abandonment and profit loss. In an AI-native environment, however, the traditional purchase funnel compresses into a single, often invisible action. Once authorized by a consumer, an AI agent handles everything that follows. "When payments are built directly into these AI-driven workflows... that eliminates friction points where consumers traditionally had to leave one experience to complete a transaction," Husaini said, citing Marqeta research that found 29% of surveyed U.S. consumers said they are interested in AI-powered wallets that automatically optimize payment choices based on their spending habits. In the traditional linear flow, the consumer starts at a merchant, checks out, and the issuer processes the transaction. Agentic commerce breaks that sequence. AI agents might initiate a purchase from a banking app, trigger a transaction from a card-linked service, or preemptively select an issuer-optimized payment method. Suddenly, issuers appear not at the end of the transaction, but at the beginning. "As AI agents take on more decision-making, issuers now have an opportunity to start to shift left and be present earlier in the purchase journey rather than at the point of transaction," Husaini said. The shift could have far-reaching implications. For retailers, it means the front door of commerce may no longer be the storefront. For issuers, it opens an opportunity to shape decisions rather than merely process them. For consumers, it introduces a future where their financial tools work autonomously and proactively. As agents take over the decisioning layer, retailers must ensure their products, prices and promotions are machine-readable and accessible. The interface of commerce may shift away from the retailer's own app or website to an AI layer sitting between consumer and brand. Retailers that fail to prepare risk losing visibility and relevance as agents steer consumers elsewhere. "Trust and security become even more critical when we're talking about AI agents executing transactions on behalf of customers," Husaini said, adding that as agents gain autonomy, the systems that secure transactions, from fraud detection to authentication, must evolve as well. The enabling layer underneath the rise of agents may ultimately prove more transformative than the AI bots themselves. Real-time data infrastructure and open, standardized interfaces are becoming crucial. For agents to execute tasks reliably, they need fast, permissioned access to account data, contextual information and transaction capabilities, without brittle or bespoke integrations. "You have to start with things being secure and accessible," Husaini said. That's where protocols like Marqeta's Model Context Protocol (MCP) server come in. Originally developed by Anthropic, MCP is quickly becoming a preferred way for AI systems to access structured contextual data. Marqeta's implementation allows AI agents to directly interact with payment infrastructure, doing tasks like pulling balances, initiating transactions, receiving real-time signals and more, all through standardized APIs. "It creates a direct connection between AI applications and payment infrastructure, dramatically reducing time to market," Husaini said, adding that intelligent credentials, driven by AI and powered by issuers, could increase top-of-wallet use. Husaini also said he foresees "new revenue streams and partnerships" for issuers as they transform into platforms for AI services. Instead of merely providing payment rails, they become orchestrators of intelligent commerce ecosystems, where agents plug in, execute tasks and operate autonomously.

Agentic Commerce Protocol: How OpenAI and Stripe Are Reimagining the Future of Online Transactions: By Milko Filipov

AI-driven shopping is becoming one of the most significant shifts in modern e-commerce. As autonomous agents start handling tasks like comparing products, making purchases, and communicating with online stores, it's becoming clear that today's platforms -- built with human users in mind -- aren't fully equipped for this new reality. To address this gap, OpenAI and Stripe have developed the Agentic Commerce Protocol (ACP), a framework that makes transactions between AI agents, merchants, and customers secure and scalable. In many ways, ACP sets the foundation for a future where autonomous systems can participate in commerce smoothly and safely. Inside the Agentic Commerce Protocol At its core, ACP is a uniform and secure method for handling payments initiated by AI agents. When a user requests a purchase through an AI model such as ChatGPT, the protocol creates a Shared Payment Token (SPT). This token captures the customer's authorization while shielding private financial data -- including bank identifiers and card numbers -- from exposure. Merchants receive the SPT and simply process it through their existing payment setups, meaning ACP integrates smoothly with today's infrastructure without requiring extensive changes. Beyond payments, ACP also includes robust identity verification for AI agents. Only validated, permitted agents can interact with merchant systems, reducing the risk of unauthorized transactions. Every purchase is recorded in an auditable log, giving merchants full insight into agent-driven sales activity. Businesses can also apply governance features such as real-time fraud screening, spending limits, allowed merchant lists, or category restrictions -- tools that are essential as autonomous purchasing scales. And because ACP was designed with concurrency in mind, it supports large numbers of AI agents executing transactions at once, handling everything from multi-item orders to refunds and disputes. Together, these capabilities position ACP as a foundational technology for the next generation of AI-enabled commerce, distinctly more adaptive and transparent than traditional online payment flows. Why OpenAI and Stripe Make a Strong Team The collaboration brings together two leaders in their respective domains. OpenAI contributes the intelligence layer -- large-scale AI systems capable of interpreting consumer intent and coordinating purchases -- while Stripe provides the mature, globally trusted payment infrastructure required to process transactions reliably and at scale. The result is a protocol that is both technologically sophisticated and straightforward for merchants to adopt, accelerating industry-wide readiness for autonomous commerce. Implications for Merchants: Opportunities and Challenges Companies embracing ACP stand to benefit early from a new channel where AI agents influence or complete the purchasing process. Automated transactions may improve conversion rates, speed up decision-making, and increase customer satisfaction. ACP also preserves merchant ownership of the customer relationship -- from order management to customer support -- while enabling AI to drive efficient workflows in the background. However, businesses that postpone adoption risk losing ground to faster-moving competitors who leverage AI-driven interactions to create more seamless customer journeys. In a market where autonomous commerce is becoming integral, falling behind could mean losing both visibility and share. Looking Ahead: The Future of Autonomous Commerce ACP is still in its early stages, and its capabilities are expected to grow. Over time, we can anticipate integration with additional payment providers, expanded partnerships across AI platforms, and more advanced logic for sophisticated purchasing scenarios. As AI technology continues to advance, ACP is set to become a key facilitator of complex, interoperable, and intelligent digital commerce. If you'd like to understand how the ACP fits into the broader agentic commerce landscape, take a look at my earlier article on the topic.