Visa's new research with PYMNTS frames tokenization not as a security feature but as the structural layer that makes checkout invisible in agentic commerce. The credential is the commerce layer now.

When an AI agent buys something on your behalf, there is no checkout page. No card number field. No CVV prompt. No billing address form. The agent authenticates, the token resolves, and the transaction settles. The consumer sees a confirmation. The checkout, as a discrete step in the purchase journey, simply does not exist.

That is not a future state. It is the architecture Visa is building right now, and its latest research with PYMNTS Intelligence makes the case explicit. The Prompt Economy Tracker: Tokens, Trust and Transactions, published in collaboration with Visa, frames network tokenization as the structural layer that makes agent-driven commerce possible. Not the security layer. Not the fraud prevention layer. The commerce layer.

Tokenization replaced sensitive card numbers with secure, network-verified substitutes. That was its original purpose, and it worked. Tokenized eCommerce transactions see 6 percent higher approval rates and 30 percent lower fraud compared to non-tokenized transactions. The business case for security alone was clear.

But the numbers tell a different story about where tokenization is heading. Visa now has over 17.5 billion tokens globally, more than triple its physical card count. Token growth hit 44 percent year over year in Visa's fiscal Q1 2026. The company provisioned roughly one billion tokens in its first five years. It added 14 billion in the next five.

That trajectory is not explained by security alone. Tokens are proliferating because they do something card numbers cannot: they travel. A token can be bound to a specific device, a specific merchant, a specific wallet, or, crucially, a specific AI agent. It can be refreshed, revoked, and scoped to a narrow set of permissions without ever exposing the underlying account. Merchants receive valid payment confirmation without viewing card data.

Guest checkout, the friction-heavy fallback for consumers who have not stored credentials, is vanishing. According to Visa's Q1 2026 earnings data, guest checkout has dropped from 44 percent of transactions in 2019 to 16 percent today. Among the top 25 sellers, it is under 4 percent. Some 96 percent of transactions with those merchants require only a simple click or biometric confirmation.

The Prompt Economy Tracker identifies four use cases where tokenization enables invisible commerce: chat-based shopping, usage-based refills, subscription updates, and linked purchases across multiple merchants. In each case, the pattern is the same. The consumer's intent is captured through a prompt or a preference. The AI agent executes the transaction. The token handles authentication and settlement without the consumer touching a payment form.

PYMNTS Intelligence describes this as consumers becoming "their own personalisation engines," where prompts replace clicks and decisions unfold in real time. The token is what makes that possible at the infrastructure level. Dynamic tokens, credential-on-file frameworks, and agent-native tools enable scalable, interoperable AI payments across merchants, apps, and devices.

All four use cases converge on three elements: identity, payment, and personalisation, connected through network tokenization. This is not a coincidence. It is the same convergence we mapped in our analysis of the identity-payment merger reshaping fintech. When identity and payment collapse into a single credential layer, the checkout step becomes redundant. The token carries both who you are and how you pay.

Rob Cameron, global head of Visa Acceptance Solutions, told PYMNTS that tokens now carry privacy-preserving signals derived from shopping history and preferences. Consumers can choose how much data to share: remain as guests with minimal exposure, or allow purchase history to flow through so the agent makes more informed decisions. "One of the key tenets of Visa and our tokenisation," Cameron said, "is for the consumer to have the control on what they share."

When Walmart revealed that conversion rates inside ChatGPT's Instant Checkout were three times lower than on its own website, the lesson was not that AI commerce failed. It was that checkout is the point where loyalty programmes, bundle logic, fraud detection, tax compliance, and customer identity converge. Abstracting it away broke the experience.

Tokenization solves a different problem. It does not abstract checkout away from the merchant. It removes checkout from the consumer's view while keeping the merchant's infrastructure intact. The token resolves through the merchant's existing payment rails. The processor verifies it. The network settles it. The consumer never sees a form, but the merchant retains full control of the transaction.

This is why Fiserv's decision to adopt both Visa's Trusted Agent Protocol and Mastercard's Agent Pay Acceptance Framework matters so directly. As we covered in our analysis of Fiserv's dual-network integration, the processor layer is where protocols become production infrastructure. When an AI agent initiates a payment, the token resolves through the processor, authentication is verified, fraud checks execute, and the payment settles. Fiserv supporting both networks means the infrastructure handles routing based on the credential presented, without human intervention.

Visa's Trusted Agent Protocol binds tokens to specific agents. A payment can only be initiated by the right agent, for the right purpose, at the right moment. This is scoping, not just security. It turns a token from a static credential substitute into a dynamic permission layer that governs what an agent can and cannot do with a consumer's money.

The infrastructure gap in agentic commerce is real. As we documented in our analysis of the dispute crisis facing agent-initiated transactions, the systems for handling what goes wrong after an AI agent makes a purchase barely exist. Chargebacks, refunds, liability allocation: the plumbing is not ready.

Tokenization does not fix all of that. But it addresses the foundational layer: how the agent transacts in the first place. Without tokens, an AI agent would need raw card numbers, exposing credentials across every merchant interaction. With tokens, the credential is scoped, revocable, and traceable. If something goes wrong, the token can be invalidated without cancelling the underlying card. The blast radius of a compromised agent is contained.

Visa now processes 300 billion transactions annually for risk assessment across 14,500 financial institutions and 175 million merchant locations. Its transaction count has tripled over the past decade. Over 100 partners are engaged with the Visa Intelligent Commerce platform, with more than 30 actively building in the sandbox and over 20 agents integrating directly.

Visa CEO Ryan McInerney framed it clearly during the Q1 2026 earnings call: "The core of our consumer payments business is the Visa credential. It is the connection point to the Visa network." When the credential is a token bound to an AI agent, the Visa network becomes the rails on which agentic commerce runs.

The broader market confirms the trajectory. Global contactless and invisible payment volumes are expected to exceed $9 trillion by 2026, driven primarily by mobile wallets, in-app transactions, and embedded checkout. North America alone processed over 95 billion tokenized payment transactions in 2025. The infrastructure is ahead of the demand, which is exactly where Visa wants to be.

Visa is betting that the credential layer is the durable competitive position in AI commerce. Not the chatbot. Not the discovery layer. Not the agent framework. The token.

The logic is straightforward. AI agents will proliferate across platforms. Discovery will fragment across chatbots, voice assistants, and autonomous software. But every transaction, regardless of which agent initiates it or which platform hosts it, must resolve through a payment credential. If that credential is a Visa token, Visa sits at the centre of every agent-initiated purchase.

Visa Flex Credentials, which allow a single payment identity to shift dynamically between debit and credit based on context, add another dimension. Roughly 20 million Flex Credentials are active, with 20 additional issuers expected in 2026. An AI agent choosing between debit and credit for a given purchase, optimising for the consumer's cash flow or rewards, is a natural extension.

The trust gap in agentic commerce remains real. Consumers are willing to let AI agents research and recommend. They are slower to trust agents with their money. Tokenization narrows that gap by giving consumers granular control: which agent, which merchant, what spending limits, what data exposure. The token is the consent mechanism.

Visa is not alone in this positioning. Mastercard's Agent Pay and Verifiable Intent frameworks pursue the same structural role from the issuer side. The card networks are converging on the same conclusion: whoever controls the credential layer controls agentic commerce.

The question is whether the rest of the industry, processors, issuers, merchants, builds fast enough to make invisible checkout the default before consumer trust catches up. The infrastructure is ready. The tokens are issued. The protocols are live. What is missing is volume, and volume requires the kind of trust that only works when the consumer never has to think about how the payment happened.

If the token is the trust layer, and the trust layer is what makes checkout disappear, who captures more value in agentic commerce: the networks that issue the credentials, or the platforms that host the agents?