A digital assistant can already do a lot of the work that once belonged to people.
It can search for products, compare suppliers, book travel and draft invoices in seconds. But the moment that assistant is allowed to complete a purchase, something fundamental changes.
That step from suggesting to spending is where agentic commerce begins, and where the familiar rules of digital checkout start to bend.
In traditional eCommerce, trust is often established at a single point in time. A shopper logs in, authenticates and checks out. Agentic commerce breaks that sequence apart. An agent can search, select and initiate payment over time, sometimes with minimal human input, sometimes with oversight layered in. The transaction becomes a process rather than a moment.
“In an agentic world, there has to be some form of continuous validation of agent behavior,” said Rubail Birwadker, Visa’s senior vice president and global head of growth, in a conversation with PYMNTS CEO Karen Webster. “Human in the loop is not going anywhere away anytime soon.”
Advertisement: Scroll to Continue
For Visa, that reality reframes the problem. The question is not whether agents will participate in commerce, but how to make their participation feel as natural and as trusted as today’s digital payments. The answer lies in simplifying the path from intent to transaction for merchants and building a trust layer that can move as fast as agents do.
Making Checkout Disappear Without Losing Control
Visa’s ambition for agentic commerce is not to introduce a new kind of checkout, but to make checkout recede into the background. Birwadker framed agent-driven buying as an extension of behaviors consumers already trust.
“It connects really well to checkout,” he said. Over time, payments could become “the next level of one-click checkout, perhaps even without a button.”
The interface may change. It could be voice-based or conversational. But the outcome should feel familiar. Fewer steps. Fewer interruptions. Fewer moments where intent fails to become a transaction.
That experience matters just as much to merchants as it does to consumers. Sellers are already seeing a shift in how demand arrives. Merchants are encountering human shoppers and agents side by side, often with more information in hand and greater leverage on price, Birwadker said. That dynamic can reshape traffic patterns and put pressure on brand positioning.
What hasn’t changed is the merchant’s need for simplicity. Supporting multiple agents, payment methods and technical standards introduces friction behind the scenes, even as the front end is meant to feel effortless.
Birwadker called integration “a huge nightmare” for merchants trying to keep pace.
The trust layer enabling agentic commerce must be easy to adopt and easy to scale, with “a low-code, no-code way for merchants to do this consistently, globally, across all agents,” he said.
Knowing When an Agent Is Acting With Permission
As commerce becomes more agent-driven, identity becomes critical. Merchants need to know who or what they are transacting with, and whether that entity has the authority to act.
Visa’s response to that challenge is know your agent, or KYA. The concept centers on verifying an agent’s cryptographic identity and establishing what Birwadker described as a “root of trust.” That root links an authenticated consumer or business to the agent and to the credential the agent uses, including the boundaries of its authority.
Visa has launched a Trusted Agent Protocol that allows a verified agent to pass a signature to the merchant, Birwadker said. That signature can carry parameters such as spend limits, merchant categories, timing and behavior patterns.
If that connection can be built “in a standardized fashion that works consistently across everyone,” he said, KYA becomes practical at scale rather than another bespoke integration merchants must manage.
The goal is not to slow transactions down. It is to give merchants confidence that an agent arriving at checkout is legitimate and properly authorized, without introducing new friction for the customer.
Designing Trust for a New Attack Surface
Agents also change the security equation. In an agentic environment, Birwadker said, “you have a new entity sitting in the middle between a consumer and a merchant.”
That intermediary role introduces new risks. Agent impersonation. Unauthorized permissions. Synthetic merchants. Data tampering. High-velocity automated fraud. Each represents a variation of familiar threats, amplified by automation and scale.
Visa is addressing those risks by incorporating agent-specific signals and behavioral patterns into real-time risk evaluation. Transactions are assessed continuously, rather than relying on a single authentication checkpoint.
The intent is that consumers should not experience added complexity simply because an agent is involved.
“If we do our job right,” there should be “no real discernible difference” between traditional eCommerce and agent-driven commerce, Birwadker said.
Why Standards Matter as Agents Multiply
Even with strong identity and risk controls, agentic commerce cannot scale if every platform defines trust differently. Birwadker said he expects fragmentation early on but sees standardization as inevitable.
Merchants operate globally. They need a consistent way to recognize a trusted agent regardless of which model or platform represents the buyer. Visa’s position is that core trust standards should be established through industry bodies, so they work the same way everywhere.
That does not mean bespoke integrations disappear.
“Individual protocols will exist just like custom pipes exist today,” Birwadker said.
However, those connections should be complementary, not isolating, so they do not fracture the ecosystem into incompatible systems.
Tokens as the Control Layer for Agentic Commerce
Tokenization sits at the center of Visa’s strategy for making agentic commerce manageable and secure. Visa already operates tens of billions of tokens, originally created to reduce the exposure of sensitive data across the payments ecosystem.
“Purely for cards, we really do believe” in tokens as a foundation of trust, Birwadker said.
Over time, tokens have evolved beyond security alone. They are now configurable and capable of representing authenticated identity and carrying permissions. In an agentic context, that flexibility allows a token to be bound to an agent and surrounded with specific payment and spending controls.
Equally important is what happens when something goes wrong. Instant revocation and lifecycle management are important because they allow permissions to be withdrawn quickly without exposing underlying personal data, Birwadker said.
Building the Path Forward
Agentic commerce is still in its early stages, and integration remains a gating factor. Visa is working with companies building agent experiences to define what end-to-end flows should look like, with a focus on embedding authenticated card network credential tokens directly into agent protocols.
Fragmentation is likely to persist in the near term, but rising transaction volumes and real-world edge cases will drive convergence, Birwadker said.
Trust in this model is earned incrementally, he said. Verified agent identity. Explicit and enforceable permissions. Credentials that can be controlled and revoked.
“It all comes down to trust,” Birwadker said, adding that the aim is that the protections people already associate with a Visa credential apply whether a human or an agent is doing the buying. It’s the only sure way that agentic commerce can scale without adding complexity for merchants or the ecosystem that supports them.
For all PYMNTS AI and digital transformation coverage, subscribe to the daily AI and Digital Transformation Newsletters.
