Imagine asking an artificial intelligence (AI) assistant what looks an otherwise simple enough question, “what to order for dinner”, and seamlessly proceed to the “order is on the way” stage, within the same conversation. That’s the world of agentic AI payments that Razorpay, India’s largest payment aggregator, envisions. The fintech, at the India AI Impact Summit 2026, has announced agentic payments on Anthropic’s Claude. It is, as Rahul Kothari, who is chief operating officer at Razorpay tells HT, “commerce being re-architected”.
This is the coming together of Razorpay’s Agentic Payments feature, India’s UPI infrastructure led by the National Payments Corporation of India (NPCI), and Claude’s conversational intelligence, in a medley hoping to turn everyday AI conversations into real, completed purchases. Alongside, Razorpay is also partnering with global agentic software creation platform Replit to help AI developers integrate monetisation options in apps they build.
The Anthropic AI-led payments chapter comes a few months after Razorpay rolled out agentic payments on OpenAI’s ChatGPT, as well as connectors for Google Gemini. Kothari believes that in the new age of AI commerce, trust design is going to matter more. “We are moving towards a world where checkouts will become invisible, and therefore trust on payments will be important,” he says.
Crucial to agentic payments aspirations is UPI Reserve Pay, a feature that allows users to reserve a set amount of money from their RuPay credit card, a bank account on UPI or a pre-sanctioned credit line, and make payments from that pool. Razorpay and Anthropic’s agentic commerce chapter presently has Zomato, Swiggy and Zepto as partner platforms.
“UPI really is the secret weapon for agentic commerce too. It lets you block funds for an agent without having to share payment credentials. No other real-time payments or card system does this natively,” says Kothari. Razorpay is dominant in India’s online payment gateway space, with over 55% share.
The AI chatbot, when asked more generic questions such as “order dinner for two under ₹800 from my favourite restaurant”, will be able to search options on Zomato and Swiggy, and complete payment post a user’s confirmation. Kothari insists the core thought behind integrating agentic payments to work with popular quick commerce and food delivery apps, is to redefine how intent becomes an economic action, in an envelope of familiarity.
Regulatory conversation is inevitable
The current payments systems are designed on one foundational assumption, that a human decided to make this payment. The liability frameworks, dispute mechanisms and fraud prevention all are based on this assumption. Agentic payments break this, entirely.
Kothari points out that in previous months, their learnings about agentic payments trends points to a heartening trend — users don’t really fear making AI payments. But he draws a line in the sand. “They do fear losing control, and the challenge for us is to solve real time tracking, instant revocation and user-defined spend limits,” he says.
When an AI books a flight for you and the airline charges twice, who is liable — you, the AI platform, the payment processor, or the airline? When there’s no checkout page, consent is legally undefined. When an AI misinterprets a user’s instructions and buys the wrong thing, is that a fraud dispute, a product dispute, or something the existing rule books simply don’t answer, for now?
The question regulators will have to define with an answer at some stage is, when AI makes a transaction error, who is liable?
“India is best positioned to lead here because the Reserve Bank of India and NPCI have always been innovation friendly. There must be guardrails. UPI was a regulatory innovation, and the same method is applied in the AI commerce as well,” Kothari says, before pointing out the complexity that “AI will keep evolving and you can’t really write rules for every AI agent”.
“Regulators in India will take the lead in how governance is supposed to happen,” he hopes, suggesting mechanisms such as user based consent for payments above a certain value per transaction, ability to revoke a permission to pay, and liability change must be step based.
India’s digital payments trajectory doesn’t need much in terms of an illustration for the uninitiated. NPCI official data indicates UPI transactions clocked ₹28.33 lakh crore in value and 21.7 billion in volume in January — highest monthly statistics for UPI. Alongside, India’s credit card spends at ₹2.12 lakh crore for 552 million transactions, just shy of the ₹2.17 lakh crore record set in September.
For agentic payments to succeed, in terms of adoption, Kothari believes the payment infrastructure must be platform agnostic. “The AI layer, whether it is ChatGPT or Gemini or Claude, is just the interface. The payment system has to work the same way regardless of which AI is on top,” he says. The term he uses is “non-negotiable”.
Razorpay has maintained significant momentum in previous months, including receiving the Offline Payment Aggregator (PA-P) licence from RBI, for collecting in-store payments, last month.
