3 minutes
Agentic commerce: building for adoption, not autonomy
Hosted by The Paypers, this webinar reveals how AI agents are transforming commerce and what merchants need to do now to capture the opportunity.
Key points
- Discovery, not checkout, is the live use case. Merchants chasing agentic checkout are solving tomorrow's problem before today's: Product catalogs need to be machine-readable before an agent can recommend them at all.
- Fraud and disputes face a harder problem than payments does. Agent-mediated transactions strip out the behavioral signals fraud tools have relied on for years, and friendly fraud already accounts for roughly 70 percent of dispute volume in traditional commerce.
- Worldpay's guidance to merchants: Assess your customer base and current stack before overhauling anything. Payment infrastructure is rarely the blocker. Knowing who your customers are and how fast they're adopting AI tells you how quickly you actually need to move.
Two weeks before this panel convened, OpenAI quietly shelved Instant Checkout inside ChatGPT and pivoted to an app store model. One product decision captured the state of an entire industry: The infrastructure race is on, but consumer behavior has not caught up. That gap between readiness and adoption framed a webinar hosted by The Paypers, “Agentic Commerce: Building for Adoption, Not Autonomy,” featuring Worldpay, now Global Payments, alongside Commerce (BigCommerce) and Justt.ai.
Moderator Dwayne Gefferie brought together three people building different layers of the agentic stack: Reshmi Suresh, head of agentic commerce at Worldpay, who leads payment infrastructure strategy for agent-driven transactions; Michaela Weber, SVP of product at Commerce, who works on how merchants stay discoverable in an AI-first shopping environment; and Shahar Tal, CTO at Justt.ai, who builds dispute management and fraud prevention for a world where the buyer might not be human.
Merchants are curious, not committed
Suresh opened by drawing a distinction that shaped the rest of the discussion: closed-loop agents, which operate only within a merchant's own ecosystem, versus open-loop agents, which can buy from anywhere. Within open loop, she outlined three levels of maturity: advisory agents that hand off to a human at checkout, agent-driven flows where a human still clicks “buy” and fully autonomous agents that purchase without human involvement. That last category, she said, “is mostly hypothetical” today.
Asked where merchants actually stand, Suresh pushed back gently on the sense of urgency many feel. “I don't think every merchant needs to drop everything and say, 'Oh my God, I have to make agentic happen tomorrow,'” she said. Her advice: Understand your customer base, learn the space, then decide when it's time to act.
"Understand your customer base, learn the space, then decide when it's time to act."
Weber agreed that discoverability, not checkout, deserves merchants' attention right now. She pointed to a taxonomy problem underneath the hype: Each large language model, like each marketplace, wants product data formatted differently, and the models themselves are still restricting how much data they'll ingest. Commerce has one live example worth noting: Lake Champlain Chocolates runs an agentic checkout on Perplexity. But Weber was candid about its scale. “I don't think that's driving material volume overall for our merchants.”
The protocol pileup
Suresh walked through the current landscape of frameworks, from OpenAI's Agentic Commerce Protocol and Google's Universal Commerce Protocol, to AP2's cryptographically signed mandates, to Visa and Mastercard's agent-registration and token systems, to newer entrants like verifiable intent and the Trusted Agent Protocol. She was careful to flag how fast the list keeps growing. “In fact, the last two weeks, we had two more protocols that came into the market,” she said, before adding with a laugh that her own slide had run out of room: “Like, please let's not have any more.”
Asked later whether consolidation is coming, she didn't pretend to know. “I really, really, really hope there is,” she said, pointing to industry bodies like the Agentic AI Foundation as evidence that the will exists, even if the outcome doesn't yet.
Fraud and disputes enter uncharted territory
Tal reframed the panel's most pointed exchange: Agent-mediated purchases look identical to each other in ways human transactions never did, which breaks the pattern-matching that legacy fraud systems depend on. He introduced what Justt.ai calls the agentic dispute rate, treating friction at the dispute stage as a signal for what needs fixing earlier in the transaction flow.
His clearest statement of the stakes: “The liability for disputes really sits with the merchant.” Storefronts built to minimize friction and maximize human conversion are the opposite of what agents need to operate well, and that mismatch, he argued, will only get more expensive without preparation. Roughly 70 percent of dispute volume already comes from friendly fraud, and more than half of merchants still handle disputes manually, a combination Tal called unsustainable at agentic scale.
Suresh added a forward-looking note on where liability might shift. “If the agent makes a mistake in the future, the merchant should not be” left holding the burden of proof, she said, particularly in cases where the merchant has no direct relationship with the end consumer. She sees ongoing industry conversations aimed at rebalancing that equation as new evidence standards, like verifiable intent, mature.
Who owns the customer
Weber addressed a concern she hears constantly from merchants: that agentic commerce becomes a race to the bottom on price and shipping speed once an agent, not a person, makes the comparison. Her counterpoint was direct. “Your website isn't going away,” she said. Consumers still browse, watch product video and read reviews on brand sites, and that behavior gives merchants leverage an agent-only relationship would erase.
What merchants should ask this quarter
Pressed for the single question merchants should bring to their next vendor meeting, each panelist gave a different answer:
- Weber: Ask your commerce platform how it's preparing your catalog to be visible on agentic surfaces.
- Suresh: Ask your payment provider how ready it is for new fields, tokens and fraud signals, and what it's doing to shape the frameworks still being written.
- Tal: Push dispute vendors past roadmap promises and confirm there's real technology behind the claims.
“Payment is not the long pull, it is the discovery aspect. You won't be bought if you don't show up.”
Asked what merchants will wish they'd started twelve months from now, the panel converged on one answer: Get the product catalog agent-readable first. Payment infrastructure can follow. As Suresh put it, “Payment is not the long pull, it is the discovery aspect. You won't be bought if you don't show up.”
The panel closed by pointing attendees to Worldpay's Agentic Commerce Report, which found that while a large share of consumers now use AI to research purchases, only about 6 percent in markets like the US and UK are comfortable letting an agent complete a purchase without them. A second edition of the report is forthcoming in September 2026, and Suresh said she's watching to see how much that trust gap has closed since the last survey.
Related insights
