Today's episode of Zeroot is actually very special. I know you've been saying that every week, but this week is actually very special. Brady, do you want to tell us why it's a very special episode?
Well, one aspect is that it's very, very forward-looking. We're really looking at the future. So, what are we looking at?
Okay, the other thing is where I'm recording from. I'm actually recording from not from the Ken office. I'm recording from another office in Kormagla and it's the office of Razer Pay. I'm going to begin by telling you why I'm here. For almost a decade, the most important object in Indian commerce was the checkout page. Then UPI came along and made the act of payment feel so simple. But you still had to open the app, search, scroll, filter, add to cart. Every step is where merchants lose the sale. Razer Pay's bet is that this entire choreography disappears. A customer wakes up and says 'I want to make halva tonight.' Then someone figures out the ingredients, bills the cart, checks the pantry, and makes the purchase—all inside the conversation. The company processing $180 billion a year, powering Zomato, Swiggy, Cred, calls this agentic commerce, and its CEO thinks it's India's next leap after UPI. So today's episode is not about the little things; we're assuming this future is real and asking what it looks like. How does it work when a machine picks the restaurant, when commerce moves to a chat window? Who wins and who changes? For this conversation, we have Harshil Mathur, co-founder and CEO of Razer Pay. Welcome, Harsh.
Thank you, Pravin. Thanks, Brady. Nice to talk to you guys.
I just want to tell our listeners that Razer Pay did not set any conditions on this conversation. They didn't tell us what to talk about or what not to talk about. We're just here to talk about agentic commerce. The only thing we're not going to discuss is Razer Pay's business and future projections because they've just filed a DRHP and are in a silent period. Despite that, we're going to talk about agentic commerce a lot. Thank you, Harshil.
Yeah, great to talk to you guys. I'm very excited about agentic commerce in general, so always happy to talk about it.
Okay, let's begin. I'll start with a broad question. You once said that chat is the starting point and AI should finish the job—say it and it gets done. Can you walk us through what this looks like 2-3 years from now? From start to finish, how do you imagine it happening?
Today, when you want to do something new, your gateway to the internet starts with search. But increasingly, that gateway is becoming LLMs or chats. ChatGPT is the second largest user base in India. I personally start with ChatGPT for questions like how to make halwa. The difference is that an agent is a smart gateway—it understands your intent, personalizes responses, and can go much deeper. It can do the job for you as long as you give it the right tools. A couple of years later, I could open a chat window and say 'I want to make halwa today.' It knows my past habits, suggests adjustments for my diet, then creates a cart on Swiggy with my preferred brands, and asks if it can pay. The key is that it's extensible—you can do this in Hindi, on voice, or via a third-party platform. Two years out, you might not even open anything; you'll just speak to a smart device. That's the level of extensibility that makes it beautiful.
The way I see it, agentic commerce is already unfolding in places like China with Alipay, but there's a habit problem. People try it when there are giveaways, but the habit doesn't stick. How do you think about the journey to get to the seamless experience you described?
That's a great point. Experiments in the US haven't taken off, and China hasn't gone that far either. The reason I believe it will take off in India is that we don't have entrenched habits. India has 450 million UPI users but less than 50 million regular e-commerce users. That's less than 10% penetration. 70% of Razer Pay's GMV comes from just 15 million users. So there's a massive untapped market. Indians shop conversationally and are voice-first and multilingual—that's how we buy offline. E-commerce hasn't replicated that. Agentic commerce solves that by allowing you to chat and ask questions. There are no entrenched habits to fight; the natural Indian way of shopping is finally possible digitally. That's why I see a bigger opportunity for India than anywhere else.
Interesting. I wrote about this long ago—I called them 'India's California users,' a tiny set of people driving most purchases. Let me also pull back and talk about Razer Pay. It powers 55-60% of India's online payment gateway market, with FI25 revenue of about 3,783 crores. Now let's get into the milestones you've built over the last 6-8 months. In October 2025, you piloted with NPCI and OpenAI to order groceries from Big Basket via chat. In Feb 2026, you powered agent payments on Claude with Zomato, Swiggy, Zepto using UPI Reserve Pay. You also built an MCP server to generate live UPI links mid-call. Then Agent Studio, built on Anthropic's Claude, helps merchants with abandoned cart conversion, dispute response, subscription recovery. You also launched natural language reconciliation where merchants can upload a bank statement screenshot and get instant reconciliation. And you're running in-app agentic commerce pilots with Zomato and Swiggy. So you have multiple surfaces: chats, merchant tools, in-app, voice. What are these building blocks for? How have you built all this?
We believe that two years out, most day-to-day operations will be powered by agents. As an infrastructure company, we're building the layers to support that world from a commerce perspective. One side is connecting to agents like ChatGPT and Claude so they can initiate commerce on behalf of users. The other side is that people still come to apps like Swiggy, but instead of searching, they can talk to an agent inside the app. We're building both sides because we're not betting on a single end state. The Reserve Pay protocol allows authorization once and then agents can buy on your behalf without you having to open your phone every time. On the merchant side, we provide tools via Agent Studio so businesses can build agents to manage reconciliation, refunds, and recovery without deploying more people. Those are the two broad sides.
It seems like you're hedging on all sides—supporting purchases via external chat agents, internal app agents, and merchant-side tools. Is that deliberate?
Yes, nobody knows exactly how this will play out. We want to bet on the future being driven by agents, but it's early to call the shape. As an infrastructure provider, we built all payment surfaces—web, mobile, links—and different segments adopted different ones. Similarly, in the agentic world, different flows will suit different needs. For deep travel planning, a broad chat interface might be better; for specific purchases like monthly rations, an app might be better.
So the analogy is a nervous system—you're building the underlying system regardless of the body's shape. Is that fair?
Neither am I a biology student, but that's a fairly good analogy.
Okay. Thinking about small businesses, restaurants, mom-and-pop stores—they risk losing their customer relationships with agentic commerce. It seems like agentic commerce favors existing platforms and could wipe out small businesses, like Walmart did.
