Proxy commerce remains a work in progress, its promise of automated purchasing at odds with the practical limits of infrastructure, trade control, and overall model reliability.
In a recent chat interview on Monday, Artificial intelligence solves Founder and CEO Dan Wagner told PYMNTS CEO Karen Webster that the destination is clear even if the industry is still assembling the pieces. A goal is a system in which consumers express their intentions once and agents handle discovery, evaluation, and implementation without repeated input.
He also stressed that merchants resist open access for a reason. “You don’t know my products,” he said, describing the retailers’ point of view. “You’re not separating me from my clients.”
At the same time, he acknowledged that consumers will not give up on AI interfaces. “The consumer doesn’t want to deal,” Wagner said. “The consumer wants to go to ChatGPT and ask the question.”
This tension defines the present moment. Retailers must engage in agent-driven merchandising while maintaining control over customer relationships and product representation.
Advertisement: Scroll to continue
Wagner identified two paths toward ambitious results. One involves direct participation between consumers and commercial agents. The other most important path is agent-to-agent interaction, where AI systems transact on behalf of users and retailers simultaneously.
“We will replicate a richer experience, through generative AI,” Wagner noted, describing the shift away from catalog browsing toward guided conversational discovery.
Toaster problem
As for the current headwinds, Webster described her own attempt to test the system through them Buy a toaster Through the artificial intelligence interface. The experiment revealed the gap between discovery and implementation.
“The toaster failed on two things,” Webster said. “First, the brand I wanted to purchase was not returned, and second, I was unable to purchase it.”
She added that current systems act as a smart search function and not as a real trading engine. Wagner agrees that the industry is no closer to solving these shortcomings, pointing to structural flaws in how the models work.
At the heart of the issue is accuracy. “There is a fundamental disconnect between the hallucinations that these models inherently drive and retailers’ anxiety about opening the doors to their warehouses,” Wagner told Webster.
This disconnect creates operational risks. Retailers don’t want agents misrepresenting products or making orders that lead to returns and customer dissatisfaction. “The best person to sell sneakers to a consumer is the brand owner,” Wagner said. “But not ChatGPT, which is specialized.”
From research to agents, the long arc of change
Wagner framed the present moment as the final step in a decades-long evolution. He began his career in the early 1980s, digitizing newspapers to make them searchable at a time when most people still relied on physical archives.
“When I started, people didn’t have computers,” he said. “The IBM PC came out.”
From there, commerce moved from physical stores to online search and then to e-commerce platforms. Generative AI represents a further transformation, changing not only where transactions take place, but also how people interact with digital systems.
“What you’re actually doing is talking to a probabilistic mathematical algorithm that guesses the next word,” Wagner said. “But for all intents and purposes, it’s a conversation.”
This conversational interface becomes the entry point for commerce, displacing search bars and category navigation.
Agent-to-agent commerce and the scale problem
The effects are most evident in interactions between one agent and another. Wagner described a scenario in which a single query triggers hundreds of simultaneous orders across retailers.
“It will interrogate 500 stores every time I order something,” he explained. “Instead of it being two or three, there will be 500 (traders) in real time.”
We would love to be yours Favorite source of news.
Please add us to your favorite sources list so our news, statements and interviews appear in your feed. Thanks!
This level of activity imposes new demands on infrastructure. “Every e-commerce platform on the planet needs to scale to manage this load,” Wagner said, adding that current systems are not designed for such scale. Webster pointed out the structural uncertainty this creates. “We don’t really know the answer at this point,” she said, referring to whether proxy trading will function as a marketplace, a channel, or something entirely different.
Gatekeepers, not gatekeepers
Wagner’s response to these challenges revolves around controlling traders. He said retailers should act as gatekeepers, managing how agents access and display their products.
Against the backdrop of this need, and with a nod to his own company, “we built an agent infrastructure for the dealer to talk to agents to control the narrative,” he said.
Wagner was clear about the need for new infrastructure. He said: “The new method requires a new system,” considering that traditional e-commerce interfaces are approaching obsolescence.
He extended this argument to payments. “They are sitting on old rails,” he said of current systems, noting that they were not designed for the scale and flexibility required in an agent-driven environment.
Rezolve introduced its own payment framework, Rezolve Pay, as part of a broader effort to bridge the gap between discovery and transaction. The company is also investing in blockchain-based infrastructure to support large-scale, real-time interactions.
Growth and positioning
The company’s recent performance reflects the growing demand for this approach. Solution reported Strong revenue growth and expanding enterprise adoption, with a growing contracted revenue base and deployments across verticals. Wagner said the momentum stems from retailers seeking a structured way to engage with proxy systems.
“I’ve never seen anything like this before,” he said, referring to the pace of adoption, as the company has more than 950 enterprise customers in sectors such as retail and hospitality.
What comes next?
The unification remains unresolved. The industry will need common frameworks to control how agents interact with merchants, manage identity and complete transactions, Wagner said. He also stressed that Rezolve’s model is software-as-a-service, building on previous shifts to cloud-based commerce platforms. He said the difference lies in size.
“Agents don’t go one at a time,” Wagner said. “They go everywhere instantly.”
This fact will determine which players will get the power of attraction. Retailers who can accurately represent their products in agent-driven environments are likely to gain visibility, while those who cannot risk being marginalized.
Wagner left little doubt about the direction of travel and its effects on trade. “What’s happening now is on a whole new scale of extraordinary possibilities,” he told Webster.
For all of our PYMNTS AI coverage, subscribe to our daily newsletter Amnesty International newsletter.
