There is a growing disparity between hype and reality when it comes to agentic commerce. This is not so much a critique of AI agents’ capabilities, or the infrastructure that has rapidly emerged to support them, as it is an indication that this model is still in its early stages.
Many merchants still associate the term “agent” with independent sales organization (ISO) reseller agents, rather than artificial intelligence.
In a recent PaymentsJournal podcast, Hilla Peled, SVP of AI & Data Science at Nuvei, and Don Apgar, Director of Merchant Payments at Javelin Strategy & Research, discussed agentic commerce from the merchant perspective, where the primary concern is how businesses can earn customer trust as this emerging model evolves.
Although questions remain around the model, that does not mean the progression toward agentic commerce is slowing. On the contrary, now is the time for merchants and providers to develop strategies and prepare for widespread AI agent interactions.
Preparing for the Agentic Revolution
Despite the apprehension surrounding agentic commerce, there is little debate that AI now plays a significant role in consumers’ lives. It has become a go-to resource for a wide range of tasks, and commerce is no exception. Shoppers are already using AI to compare prices and discover products, and autonomous agentic personal shoppers represent a natural extension of this trend.
Most merchants are eager to support this shift, but not if it requires compromising funds or expanding PCI scope.
“Merchants aren’t saying, ‘give me a shopping agent,’ but they want to be prepared to the extent that their customers show up with shopping agents,” Apgar said. “Largely, they’re trying to figure out what this means, which standard will prevail, and how they should look at their architecture and their position in agentic commerce.”
Many businesses are closely examining how the technical details will be resolved, including how transactions will settle and which standards will govern interactions.
One of the most important questions is agent ownership, whether an agent is acting on behalf of the consumer, the merchant, or the AI company that developed it. Establishing this responsibility will influence how all parties approach agentic payments.
The ultimate objective is to create trust in the process by ensuring that an agent reliably executes its assigned tasks and delivers an outcome that meets customer expectations. This is no small feat, given the range of fraud and security concerns, including the potential for bad actors to manipulate agents, consumers, or merchants for malicious purposes.
The answers to many of these questions remain unclear, meaning the space will likely experience uncertainty and adjustment as it matures.
“One of the biggest things that we observe is the gap between what customers are doing when adopting the public agents versus what the PSPs and acquirers are willing to take on,” Peled said. “Everyone is now just preparing themselves to the point where agentic commerce will become much wider.”
“We know that 1.5% of purchases in the U.S. have been agentic, which is huge when you think of the current state of agentic commerce,” she said. “At the same time, both customers and merchants and any business across e-commerce today is wondering, ‘What is the next thing they need to do in order to be ready when agentic commerce will explode?’”
Volume Follows Trust
To better understand the current state of agentic commerce, Nuvei conducted a study examining consumer attitudes and behaviors. The study found that only around 1% of respondents wanted fully automated AI purchasing, while 56% said they would never allow an AI platform to spend without their approval, regardless of the amount.
While these findings may appear to challenge the future of agentic commerce, previous technology shifts have demonstrated that consumer preferences can evolve rapidly.
“If you remember in 1995, almost nobody wanted to put a card number online, but we know where we are at today” Peled said. “The demand isn’t for autonomy, it’s for the trust trails that make delegation safe. In payments, volume has always followed trust and the infrastructure decisions are being made now, years ahead of the volume.”
There are clear parallels between the growth of e-commerce three decades ago and the emergence of agentic commerce today.
Along with initial concerns about security and fraud, many believed certain products could never be successfully sold online. For example, some experts argued that online clothing and footwear sales would struggle because consumers preferred trying items on before making a purchase.
Agentic commerce may encounter and overcome similar barriers, but potentially at a much faster pace.
“Javelin research picked out a few categories—including travel, B2B and commodity purchases—that we think will be the first to earn shoppers’ trust using an agent, but it’s not going to take 30 years like it did for e-commerce to evolve,” Apgar said. “The tech is moving so much faster that it will reach a maturity point and a tipping point for agentic commerce much sooner than it took e-commerce.”
A Fundamental Shift in Commerce
The accelerated pace of innovation and adoption has already been demonstrated by generative AI.
“The AI space has been around for 17 years and the pace that AI has evolved in the last 12 to 18 months is just unprecedented,” Peled said. “There was always evolution, but what we see now with the capabilities and with the actual adoption shows that the hype about agentic commerce is not just hype. People are adopting agents because they understand their power and agents are becoming stronger and more capable very quickly.”
This means organizations should prepare for change sooner rather than later, especially businesses serving younger or more tech-savvy customers.
However, merchants should recognize that agentic commerce is not simply a gateway to new markets. Instead, it represents a fundamental shift in how commerce operates.
“The reality is that this is existing purchase volume. AI doesn’t magically give people more money to spend,” Apgar said. “These are existing purchases that are going to go from whatever channel they’re being made in today—whether it’s retail, e-commerce, or mobile—and be converted to agentic.”
“Where is the early impact of that going to be? Which merchants will be impacted the most? Early adopters have the ability to pick up market share from lagging competitors,” he added.
A Trust Problem, not an AI Problem
The potential for early adoption is why many merchants are closely monitoring the evolution of agentic commerce and attempting to identify the point at which it reaches mainstream adoption.
“They understand that at some point, others will eat their lunch if they’re not adopting early,” Peled said.
For merchants determining how to move forward, there are practical steps they can take. The first is to audit existing infrastructure and assess whether it can support orchestration without custody. Next, merchants should build first-party capabilities so they are not forced to reintegrate solutions later.
Retailers should also stay dialed into market developments and regularly reevaluate their roadmaps. Given the pace of change, this process should occur every few months rather than every 12 to 18 months. This will help ensure that merchants’ technology stacks are prepared as agentic commerce becomes more widespread.
To stay ahead of the curve, Nuvei recently completed a live agentic commerce proof of concept. In collaboration with Visa, Arvato Systems, and fashion brand Kings and Priests, the initiative demonstrated an agentic transaction executed through a unified workflow, all within shopper-defined parameters like spending limits and approved categories.
“We launched a live Visa transaction where an AI agent purchased, paid inside the merchant’s own experience and cleared across multiple European issuers,” Peled said. “The hard part was never teaching an agent to buy; that’s what agents know how to do. It was giving issuers and schemes across markets a reason to approve a purchase that no human initiated. That is a trust problem, not an AI problem.”