Payments firms build AI commerce rails, but ecom platforms stay guarded

Payments companies are building rails for autonomous AI-led shopping, but ecommerce platforms remain wary of losing control over discovery, recommendations and margins

Digital payments companies in India who are privately building rails for artificial intelligence (AI) agents to autonomously shop and transact are growing uneasy at the lack of urgency among digital commerce platforms to open up their flows.

 

Fully autonomous commerce, from discovery to checkout, is at least two to three years away, according to senior payments executives — who caution that if digital commerce platforms stay closed to AI agents, the whole model may struggle to take off.

 

“Across events and conferences, it is the payments companies who are talking about agentic commerce and not the commerce players. We don’t hear marketplaces say their inventories are agentic first,” a senior executive at a digital payments company quipped.

 

The reluctance to go truly agent-first is not accidental. AI agents are disintermediating by design — they bypass the recommendation engines and discovery algorithms that platforms have spent years and fortunes building, and which are at the core of their business model.

 

This is one of the problems that India's payments executives are brainstorming for: How to convince platforms to allow agentic commerce.

 

“Agents will bring a disintermediation of commerce platforms. That’s why large platforms may not necessarily like it. Why would any user go to a commerce platform if the agent is going to get a good deal, discover items, and relay feedback according to my choice and taste,” a senior fintech industry executive said, adding that most of the platforms’ effort towards agentic use cases is largely concentrated among chatbots alone.

 

Payments companies executives spoke on the condition of anonymity as they were privately in discussions with commerce companies to understand bottlenecks that the industry was facing.

 

AI agents for users could originate from merchant apps, consumer-facing payment apps, AI assistants such as ChatGPT or Claude, or dedicated agentic platforms such as Hermes and OpenClaw.

 

In theory, these agents hand power back to the user, and they do it by compressing the discovery and checkout timelines.

 

A shopper hunting for the right pair of shoes can spend hours weighing designs, sizes, brands, prices and discounts — and that is on just one platform. India alone has a dozen digital marketplaces and despite that a user might not get the best offer.

 

An agent, handed the same request, could do all of it in minutes, and — if properly authorised — complete the purchase within pre-set limits.

 

The savings is not just in time. Agents also soften the impulse buying that human shoppers are prone to. Where a person gets steered towards a queue of 'recommended items' they may not really need, a focused agent simply doesn't, and every purchase it rejects is a sale the platform loses.

 

An agent could also be empowered to monitor prices of products and services continuously, and initiate purchase at a best possible price, again squeezing out margins for commerce platforms.

 

At the core of it, the frustration among fintech platforms is that while they are preparing agentic rails, they are not seeing volumes just yet.

 

More economic value, they argue, will come from enabling agentic use cases end-to-end, while payment becomes a settlement function alone.

 

“A lot of the economic value is moving up the value chain even before payments. Eventually what is payment becoming? Payments is just a settlement function. All have to move up the value chain,” a third person from the fintech industry said.

 

To be fair, commerce platforms are also contending with real security risks: agents can be manipulated, and their vulnerabilities probed and exploited. Opening their systems to AI agents is, on safety grounds alone, a genuine gamble.

 

Last year, e-commerce giant Amazon sued AI platform Perplexity demanding the latter stop allowing its AI browser agent, Comet, to make purchases online for users, according to a Bloomberg report.

 

The e-commerce giant accused Perplexity of committing computer fraud by failing to disclose when Comet is shopping on a real person’s behalf, in violation of Amazon’s terms of service, the report had stated.

 

That said, while agentic commerce grabs headlines, not many users have been able to utilise them to their fullest extent yet. Executives believe the future of scaling use case would involve improving the overall user experience.

 

“Companies are investing in chatbots today, but in what direction? It looks like after the chatbot, the rest of the checkout experience remains the same. Eventually, they are operating on their own catalog and not necessarily expanding the search which makes it more manual work,” a fourth person, who is in talks with e-commerce platforms in India to enable agentic flows, said.