Artificial intelligence (AI) assistants can already recommend products, compare prices and automate routine purchasing tasks. The next step is agentic commerce, where AI systems do not just suggest purchases but initiate them on behalf of users. This shift introduces new requirements for payments infrastructure. When autonomous systems begin initiating transactions, the card networks, issuing platforms and authorization controls that power those payments must evolve to support machine-initiated commerce safely and intelligently.
Digital Commerce Is Becoming AI-Assisted
AI is rapidly transforming digital commerce from assistive, user-directed experiences into increasingly autonomous purchasing flows.
Consumers are beginning to trust AI not merely to guide purchases but also to complete them.
With agentic commerce, consumers can instruct autonomous assistants to identify products, evaluate options and complete transactions across multiple platforms on their behalf. For example, a consumer could ask an assistant to buy a pair of jeans under $50 from selected brands, and the assistant could complete that purchase.
Many shoppers are already interested in these types of transactions. PYMNTS Intelligence research finds that 48% of consumers are at least somewhat interested in AI agents doing their grocery shopping or planning their meals for them. The same share would let an autonomous assistant manage their subscriptions, and 44% are somewhat interested in using the tech for buying gifts.
Some industry players see agentic commerce becoming mainstream technology this year.
Momentum is building quickly. Mastercard, for instance, is advancing its Agent Pay capabilities using Google’s Universal Commerce Protocol (UCP). Visa, meanwhile, has partnered with major AI companies to enable agentic credit card payments. During the 2025 holiday season, the payments giant announced that its partnerships had already led to the completion of hundreds of agentic transactions. Based on this momentum, Visa predicts the technology will enter the mainstream by this year’s holiday season, with millions of consumers using autonomous shopping agents.
In agentic commerce, the payments layer is central to ensuring secure, authorized transactions.
As AI assistants evolve from advisers into purchasing agents, the payments layer becomes the point at which intent is translated into action. Without embedded payment capabilities, agents remain limited to making recommendations rather than completing transactions, underscoring the central role of payments in enabling agentic commerce.
As a result, the demands placed on payments infrastructure are expanding rapidly. Systems must accommodate higher transaction velocity, more complex orchestration across providers and new forms of interaction that extend beyond traditional checkout flows, reflecting a shift toward more dynamic, agent-driven transaction models. Moreover, they must do so while maintaining secure, trusted execution in environments where transactions are initiated by software rather than users.
These requirements are exposing the limits of infrastructure designed for simpler, linear transactions. As agentic adoption accelerates, such limitations will increasingly set the stage for breakdowns in flexibility, control and security.
Legacy Payment Infrastructure Breaks in Agentic Commerce
Systems built for human-initiated transactions fail to keep pace with automated, cross-platform activity, limiting control and increasing exposure to fraud and compliance risk.
Legacy payment systems lack the flexibility for agentic commerce.
450%
Increase in dark web community posts mentioning “AI agent” over a six-month period compared with the prior six months
Traditional payment infrastructure was designed for transactions initiated directly by human users, not autonomous systems operating across platforms and at speed. As agentic commerce emerges, this creates a structural mismatch between how transactions are generated and how they are processed.
Many legacy systems rely on monolithic architectures and rigid implementation cycles that cannot adapt quickly to new transaction patterns or evolving use cases. While modern networks are already enabling AI agents to initiate and complete transactions in real time, legacy platforms often require months or years to deploy changes, limiting their ability to support dynamic, machine-initiated commerce.
This lack of flexibility becomes more pronounced as agent-driven systems generate high volumes of parallel, cross-platform requests rather than sequential, user-driven interactions. Infrastructure not designed for these patterns can introduce friction through rate limits, inconsistent responses or authentication failures, preventing transactions from completing reliably.
As a result, legacy systems struggle to evaluate transactions in real time, enforce granular controls or adapt to continuously changing behaviors—capabilities that are essential for enabling secure, seamless agentic payments.
Legacy systems pose an urgent security threat.
The shift to agentic commerce also introduces new security challenges that legacy systems are not equipped to address. Traditional fraud detection models were built around human behavior, relying on historical patterns and manual review processes that cannot keep pace with machine-speed activity.
In an AI-driven environment, adversaries can launch highly targeted, adaptive attacks at scale, using synthetic content and automated workflows that evade conventional detection methods. Visa reports a 450% increase in dark web posts referencing “AI agent” over a six-month period, underscoring how quickly fraud tactics are evolving alongside these technologies.
Agentic commerce also introduces new identity challenges. Payment systems must verify not only users but also the agents acting on their behalf, requiring new approaches such as delegated authorization, programmable spending controls and “know your agent” frameworks. Without these safeguards, agent-initiated transactions can be difficult to distinguish from malicious automation, increasing the risk of fraud, compliance failures and operational disruption.
Intelligent Spend Platforms Enable Agentic Card Payments
Intelligent card platforms are designed to enable flexible, high-volume machine transactions while strengthening fraud prevention and maintaining reliability in evolving commerce ecosystems.
Intelligent card platforms address the limits of legacy infrastructure.
Paymentology research indicates that next-generation card infrastructure is well-suited to agentic payments. By giving financial institutions (FIs) greater control over how transactions are authorized and managed, these platforms enable issuers to define spending through programmable limits on transaction types, merchants and behavior. They also support real-time decisioning, allowing each transaction to be evaluated instantly against defined controls and parameters. Tokenization strengthens security by protecting sensitive payment credentials while enabling agents to transact without exposing card details.
46%
of card issuers say they selected their main issuer processor because it offered tokenization.
This combination of control and security is a key differentiator for card providers. PYMNTS Intelligence research finds that nearly 46% of issuers selected their main issuer processor for its tokenization capabilities, while roughly 35% cited spend controls.
Cloud-based, API-enabled infrastructure supports scalable, interoperable agentic payments.
Application programming interfaces (APIs) allow modern card platforms to connect seamlessly with digital ecosystems, enabling interoperable, agent-initiated transactions across platforms. Cloud-based infrastructure underpins these capabilities, providing the scalability needed to handle high volumes of automated, machine-initiated transactions without sacrificing performance or reliability. Such environments also support continuous iteration, performance monitoring and multi-agent coordination across disparate networks.
These infrastructure capabilities are already being put into practice through major industry partnerships. For instance, Fiserv recently announced plans to incorporate Mastercard’s Agent Pay Acceptance Framework into its platform to authenticate and settle machine-initiated transactions. PayPal and Microsoft have partnered to integrate agentic commerce capabilities into Copilot Checkout, reflecting a broader shift toward API-enabled, cloud-based payment ecosystems supporting real-time transactions at scale.
Together, these systems enable high-throughput, agent-initiated payments without compromising reliability or performance.
Agentic payments are redefining how authorization and transaction context are managed.
As agentic payments move from concept to real-world deployment, they are revealing limitations in traditional authorization models. These models assume the individual initiating a transaction is also the decision-maker and liable party. When software acts on a user’s behalf, that alignment becomes less direct, requiring systems to interpret not just credentials but also transaction context.
Existing payment messages often lack visibility into how a transaction was initiated, what constraints were applied or how decisions were made. Without this context, issuers may face challenges in evaluating risk, approving transactions and resolving disputes. As a result, modern platforms are evolving to provide richer transaction data and more flexible control layers, enabling issuers to better understand, govern and manage agent-initiated activity without introducing unnecessary friction.
Future-Proof Payments: Build the Infrastructure for Autonomous Transactions
Agentic commerce is arriving faster than most payment systems are ready for. As machines begin to initiate and manage transactions, FIs and payments providers face a clear choice: They can adapt their card programs now or risk becoming a point of friction in an increasingly automated ecosystem. Preparing for this shift requires more than incremental updates. It calls for rethinking how control, security and decisioning work when the user is software acting on the consumer’s behalf.
PYMNTS Intelligence offers the following actionable roadmap for FIs and payments providers preparing for agentic commerce:
- Modernize authorization. Move beyond static rules to dynamic, programmable controls that reflect user intent, context and agent behavior.
- Strengthen identity and trust layers. Ensure every transaction, human or machine-initiated, can be tied back to a verified user with clear, auditable consent.
- Invest in real-time decision-making. Enable systems that can evaluate and approve transactions instantly, even as volumes and complexity increase.
- Tokenize. Protect sensitive credentials while allowing agents to transact securely across platforms and environments.
- Build for interoperability. Use APIs and flexible architecture to connect seamlessly with merchants, networks and emerging AI ecosystems.
- Scale with cloud infrastructure. Support high-frequency, machine-driven transactions without sacrificing speed or reliability.
The institutions that act now will help shape how agentic commerce works. The rest risk being left behind as intelligent systems redefine how money moves.
