AI-led shopping is here, but wait for fully automated purchases

UPI payments still require customer authentication, and any new framework needs the central bank’s nod
Shopping cartons labelled with faces that look like robots
Getty Images/Unsplash
Published on: 

Fintech companies are building infrastructure for agent-led commerce, as artificial intelligence (AI) tools increasingly help consumers discover products and make purchases. However, delays in regulatory approvals and mandatory customer authentication have kept fully automated purchases out of reach, especially on the Unified Payments Interface (UPI).

Agentic AI refers to artificial intelligence systems that can make decisions, plan, and execute actions to accomplish a task on its own, rather than just responding to user queries.

Razorpay, Cashfree, Pine Labs, and PayU have each launched agent-compatible checkouts, protocols or merchant tools, betting that AI assistants such as ChatGPT and Claude will become a major route through which consumers discover and buy products. 

Actual agent-initiated purchases, however, remain limited, and the National Payments Corporation of India’s (NPCI) framework for such transactions through UPI has been put on hold, impacting their wider adoption.

Agent-led traffic currently accounts for around 1% of a merchant’s total traffic but could rise to 10% over the next 12 months, and to 50% in the next two years, Reeju Datta, cofounder of Cashfree, told The Financial Standard.

"Today, most AI agents work through browsers. They navigate websites like human users, read and interpret information, and take actions such as clicking buttons or filling in forms. This approach has helped make AI agents more capable and useful, leading to wider adoption," Datta said. But websites and checkout pages remain designed largely for people, which means merchants and payment firms will need to adapt.

According to Datta, traffic from AI agents on the checkout pages tracked by Cashfree had increased 16-fold since the beginning of this financial year. Final payment approval, it said, stays with the user.

“Last year, there were hardly any agentic AI tools that could help customers purchase items and make payments. That is no longer the case,” said Viren Inaniyan, cofounder of Tru Commerce, a Bengaluru-based agentic commerce infrastructure firm.

India’s agentic AI market is projected to reach $1.73 bn by 2030, growing at a 54% compound annual rate, according to a PwC India report released last year.

A flowchart showing the agentic checkout process

Globally, card networks moved first. Visa and Mastercard began building agent-specific payment systems in 2025 around tokenised card credentials, which allow an agent to buy on a customer’s behalf without ever seeing the actual card number. Some US banks already enable such transactions.

India has not reached that stage. UPI payments still require customer authentication, and any new framework needs the central bank’s nod. At Global Fintech Fest in Mumbai in September, Visa demonstrated agentic card payments with 10 Indian partners, including Razorpay, Cashfree, Pine Labs, and PayU, built on existing tokenisation rather than a new rail.

On UPI, the groundwork has come in stages. In October 2025, NPCI, Razorpay, and OpenAI piloted agentic payments on ChatGPT, with BigBasket and Vi among the first merchants. In February, Razorpay and NPCI extended the model to Anthropic’s Claude, with Zomato, Swiggy, and Zepto at the India AI Impact Summit. 

Both rely on NPCI’s UPI Reserve Pay, under which a customer approves and blocks a spending limit for a merchant upfront, so that purchases within it do not need a PIN every time.

In June, Pine Labs launched P3P, which it claims to be the country’s first agentic payment protocol on UPI. It builds on pre-approved mandates and verifies agents before they transact. In September, PayU launched Agent HQ, a platform that lets small merchants become discoverable and purchasable through ChatGPT, Claude, and WhatsApp without writing code.

However, the more ambitious step stalled when NPCI, which was supposed to unveil the Unified Agentic Protocol (UAP) in September, put its rollout on hold to focus on concerns around user protection and security.  

The UAP would let AI agents be registered, verified, and authorised to transact on UPI without changing the underlying rails, with low-value, frequent purchases such as groceries expected to be automated first.

Queries sent to NPCI and the RBI did not receive any response at the time of publishing this article.

The stakes are large. UPI processed 25.47 billion transactions worth ₹31.53 lakh crore in October, according to NPCI data, up 23% in volume from a year earlier.

Still, Inaniyan argued that payment companies are ahead of demand. “The real bottleneck is the demand coming from AI platforms such as OpenAI, Claude, and other AI assistants,” he said.

As agents become the interface between consumers and online retailers, payment companies will need to ensure transactions can be initiated and completed securely without compromising on existing authentication requirements.

This could also reshape how shoppers find products and how sellers compete. 

Mohit Bedi, former senior executive at Axis Bank and cofounder of fintech startup Kiwi, said merchants stand to lose the most. “The issue will be that the merchants will suffer, because the customer will not come to the website. Merchants generally try to upsell or cross-sell.”

For payment companies, Bedi sees an extension of the business. “Ultimately they have to evolve. They realised online was coming, so they started doing payment gateways instead of POS machines. This is an evolution of payments.”