For more than two decades, eCommerce innovation has been shaped by a familiar gravitational pull: customers go out into the digital world, navigating search bars, filters and marketplaces to find what they want.
Payments, fraud, loyalty, routing and every subsequent layer of the stack has evolved around that outward-bound customer journey.
But Adam Hiatt, vice president of fraud strategy at Spreedly, told PYMNTS that gravity is about to flip. And it’s all thanks to the rise of agentic artificial intelligence (AI), systems capable not only of reasoning but of autonomously taking actions such as shopping, booking, comparing and purchasing.
Agentic AI “inverts the responsibility of the purchasing experience from one where you have a customer going out to the merchant to saying, ‘Hey, you come to me with what the proposal is,’” Hiatt said.
In this model, intent becomes an actionable asset. The agent’s understanding of the consumer across areas such as airline loyalty, preferred brands, environmental constraints, dietary needs and more all becomes part of the shopping architecture.
“We’re obviously only in the very first inning of the AI agentic driven personalized shopping,” Hiatt said. Yet even in this early stage, he sees “this idea that the personal intent behind shopping can be put into these personal agents” as the core transformation that will remake the infrastructure of commerce.
“That personalized history of preference and need will be implicitly reflected in what those systems are able to do,” Hiatt said.
Intent Becomes a Transactional Asset
The shift Hiatt described is deceptively simple: rather than customers visiting merchants, the consumer’s historical preferences, loyalty relationships, constraints and behaviors will be encoded into a persistent intent layer hosted within agentic AI systems like those emerging from technology giants.
Imagine, he suggested, a user asking a conversational agent to plan a hiking trip to Nepal. “Based on your history of where you’ve been, what you’ve done in those environments,” the agent could construct a multi-merchant basket spanning flights, gear, insurance and accommodations, and present it for approval.
The user’s main role, Hiatt said, is “as simple as saying, yes, I want that.”
This is not simply convenience, but a profound restructuring of who initiates the commercial relationship.
The personalization dividend is obvious, and the result is that consumers using agentic AI commerce solutions would no longer navigate a sea of irrelevant options. Instead, their history and loyalty would travel with them.
But deeper personalization also increases exposure. The same intimate knowledge that enables agentic AI to streamline decision-making also expands the attack surface for fraud and privacy abuse.
Trust, Adoption and the Long Arc of Consumer Behavior
One of the more serious concerns surrounding agentic AI is the potential for entirely new classes of exploitation.
Hiatt cited “agent phishing” as a stark example of this evolving threat landscape, saying it is an umbrella term for adversaries attempting to manipulate or extract data from the agents themselves. And as more payment methods become integrated into these systems, the consequences of compromise grow exponentially.
“Bad actors will have more opportunities to essentially phish the agents,” he said. “They’re not phishing you the human, but they’re going to say, what can we draw out of these agents that have been given permission to either do things on a consumer’s behalf, or at very least represent their personal information.”
Along with fraud fears, consumer trust in AI-driven shopping remains shallow. Research from Spreedly and PYMNTS Intelligence shows that more than half of consumers are hesitant to use AI-powered shopping assistants, with 41% reporting outright distrust and only 4% expressing complete trust.
“These technologies are nascent,” Hiatt said. “Behavior is currently very opaque, and how do you know what’s actually going on?”
But history suggests that resistance fades when value becomes undeniable. Hiatt pointed to early online dating as an analogue.
“Twenty years ago, it was perceived as odd and kind of bizarre,” he said. Yet today it is standard, even dominant. The same pattern holds for eCommerce itself: “Go back 25 years and you’ll see that there is reticence to put a credit card into a website. But now … we vault them, we trust that those … will be protected.”
Behavioral adoption, ultimately, is typically less a matter of demonstrated trustworthiness. As infrastructure hardens and high-quality merchants deliver reliable experiences, there will eventually be a shift in consumer perception and behavior.
How Merchants Should Navigate the First-Mover Dilemma
If agentic AI becomes the primary interface for purchasing, it is likely to rearrange the competitive chessboard of commerce. LLM-powered platforms — OpenAI, Google’s Gemini and successors — may become the dominant discovery layer, analogous to Google Search in the early web.
“It puts a new set of players into that buying, navigating, intermediating experience,” Hiatt said, noting that discovery is no longer about information; it’s about decision-ready action.
For merchants, this creates new opportunities. Without rigid funnels and one-size-fits-all checkout flows, the interaction between agent and merchant can become more conversational and dynamic because agents enable deeper, more contextual exchange than traditional web or app interfaces.
Still, for agentic commerce to scale, Hiatt stressed that three pillars must be solved: protocols, identity and transparency.
Perhaps the biggest unsolved problem is identity. If an agent can autonomously purchase, negotiate and access sensitive data, the ecosystem needs verifiable credentials, permission boundaries and hardened authentication flows.
“How is an agent known to be a legitimate agent?” Hiatt asked.
This challenge intersects with the rising risk of agentic phishing and LLM jailbreaking, which could expose sensitive personal and financial data. “It’s a big deal and it’s not a solved problem,” Hiatt warned. Without robust identity frameworks, consumer adoption will stall.
Agentic AI is still young, but its trajectory is unmistakable. Decision-making will shift from human-driven exploration to agent-mediated proposals. And the competitive field of payments and commerce will tilt toward players that can interpret, protect and fulfill intent at scale.
“That inversion of the flow of information,” Hiatt said, is going to change everything.
