In the past few months, a race has emerged to define the technical standards for agentic commerce – the systems that will allow AI agents to make purchases and manage transactions on behalf of consumers. When Google launched its Agent Payment Protocol (AP2), quickly followed by Stripe and OpenAI’s Agentic Commerce Protocol (ACP), it signaled that major players were moving fast to set the rules of this new marketplace. To help frame what’s at stake – and ensure consumer trust and transparency are part of that foundation – Consumer Reports and Stanford’s Digital Economy Lab convened a public conversation: The Race to Standardize Agentic Commerce.
This discussion brought together some of the people designing this first generation of agentic commerce protocols. With representatives from Stripe, Visa, and Skyfire, alongside Consumer Reports’ own Ginny Fahs and Stanford’s Dazza Greenwood moderating, participants talked through the technical standards and governance frameworks that will determine how AI systems make purchases on behalf of consumers. What emerged was not a single story, but a revealing set of overlapping concerns, unresolved questions, and points of genuine alignment.
The starting point was straightforward: AI agents are already shifting from merely recommending products to purchasing them. This shift raises immediate questions about how identity, consent, loyalty, and accountability will work when a human isn’t the one clicking the “buy” button. Greenwood framed the conversation around three pillars—authorization, duty of loyalty, and evaluation—and then let the participants show, through their own work, how much of that terrain is still coming into focus.
Defining the Foundations
Stripe’s Product Counsel Kelsey Velemirovich described its partnership with OpenAI on the Agentic Commerce Protocol (ACP) – an open standard that gives agents and merchants a shared language– and its new Shared Payment Token (SPT). Their focus is on preserving the merchant’s control over how their brand and checkout flow appear, while keeping consumers’ payment credentials safely abstracted. Stripe’s perspective is pragmatic: sellers want to remain visible and authoritative even when the consumer interacts through an agent, and agents need a trustworthy way to communicate what the user intended to buy.
Visa’s Jalpesh Chitalia, who leads Growth Products, approached the same space approached the same space from a network-wide vantage point, and discussed how trust is a system-wide challenge. For decades, Visa has governed trust between issuers, acquirers, merchants, and cardholders through a combination of software and rules. Now, Visa is formally adding agents as a new kind of participant in their global network with their new Trusted Agent Protocol (TAP). Their emphasis is on making sure the card used by an agent actually belongs to the person who delegated authority, capturing the user’s mandate in a structured way, and logging the agent’s actions so that fraud can be prevented—and, crucially, so disputes can be resolved by pointing back to what the user explicitly approved. Visa’s answer to the agent question is both procedural as much as technical: “rules will become very important to drive and instill trust in the ecosystem – and to hold parties accountable,” explained Chitalia.
Opening the Door for Agents
Skyfire’s co-founder Criag DeWitt, meanwhile, focused on a simpler but critical barrier: access. Most agents, he noted, are treated by websites as bots or scrapers to be blocked. Skyfire’s Know Your Agent (KYA) protocol helps merchants verify legitimate agent traffic by packaging identity credentials with ordinary web requests. From Skyfire’s view, the immediate problem isn’t how agents pay; it’s how they get in the door without tripping security alarms. “What we see today are agents traversing the web,” said DeWitt. “Merchants need to know who that agent is working on behalf of – or it’s a 403 error. Our goal is to deliver verified information on all parties so sites can make their own trust decisions.” In effect, Skyfire is building the connective tissue that lets agents “walk through the front door” of the web, rather than sneaking in through side channels.
Bringing the Consumer Into the Loop
Against these technically grounded approaches, Consumer Reports’ Director of Product R&D, Ginny Fahs, raised a set of concerns that sit closer to the consumer’s lived experience. “The number one thing that makes an agent work is trust,” she said. “Consumers must trust the agent, and counterparties must trust it too.” Ginny Fahs pointed out that the informational landscape agents rely on is already distorted by advertising, placement bias, and uneven data quality. Studies she cited show that agents can be swayed by where an item happens to appear on a page, or by claims engineered to appeal to large language models. Fahs went on to describe CR’s vision of “conscious commerce” – designing systems that reflect the values consumers bring to real-world purchases. “How do we make sure that the things people care about – ethical labor, sustainability, affordability – are represented in the agent’s choices? Fahs asked. “Those values should be encoded in the market itself.”
This is indeed a critical question we must ask ourselves: how do we know the agent is actually choosing what is best for the consumer, rather than what is most convenient, most lucrative, or most visible to it?
A Field Still Taking Shape
This tension—between securing the payment and understanding the choice—ran quietly through the entire discussion. While identity and consent can be codified, loyalty—the duty of an agent to act in the consumer’s best interest—remains undefined.
Stripe, Visa, and Skyfire are tackling clear and urgent problems: fraud, access, risk, merchant acceptance. But the consumer-level problem of whether an agent is acting for me or acting on me is only beginning to be articulated. Greenwood’s own framing pointed to this gap: identity and consent can be logged, but loyalty is not yet formally defined or measured.
Another revealing tension came up around the future of information access. If agents are the ones “reading” the web, traditional advertising breaks down. Skyfire and Visa both acknowledge that content providers may move toward charging agents small fees for access. That could help sustain high-quality information online, but it also introduces questions about equity, transparency, and the affordability of objective product data—questions that matter deeply to CR’s mission. The webinar did not resolve this, but it surfaced the stakes.
Building the Standards We Deserve
The webinar was not a showcase or a product demo; it was an early forum for dialogue between the people building the pipes of agentic commerce and those safeguarding the public interest. If you missed the webinar or want to rewatch it, you can view it below:
The consensus is clear: the infrastructure of this new economy must be designed with trust, transparency and accountability at its core.
This is only the beginning, and there will be upcoming sessions that explore how to evaluate agent behavior, how to define and measure loyalty, how to handle liability when an AI makes a mistake, and how to keep consumers—not platforms—at the center of the emerging agent economy.
If these questions matter to you – if you want to follow these discussions or contribute to the emerging standards – please reach out at LoyalAgents.org. The rules we set now will shape the future marketplace we all rely on.
