Daybreak - If Razorpay is right about AI, you may never open a payment app again

Episode Date: April 9, 2026

At a fintech conference in February, Razorpay showed a demo. A user ordered food on Zomato by voice and paid — without opening a checkout page or a UPI app. No friction and no redirects. Just a job done end-from-end.The same week, OpenAI quietly rolled back its own in-chat shopping agent.Razorpay is calling this the biggest disruption to payments since UPI. But agentic commerce raises questions that a demo can't answer — around trust, fraud, consent, and who's liable when an AI spends your money.Is India ready for that? Is anyone?Hosts Snigdha Sharma and Rachel Varghese, speak to The Ken reporter Mutasim Khan. Tune in.Buy your tickets for the Zero Shot event here.Daybreak is produced from the newsroom of The Ken, India’s first subscriber-only business news platform. Subscribe for more exclusive, deeply-reported, and analytical business stories.

Hi, Rachel. Hi, Sninktha. So, I have a quick question for you. When was the last time that you actually enjoyed shopping online? Like, you know, finding what you want, comparing prices, especially the whole payment process, you know, waiting for the OTP, putting your UPI pin. Honestly, that's probably the least fun thing about shopping. Right?

Now, imagine if I told you, you just have to say what you want out loud and it's done. There's no app, there's no checkout. out. There's no pin. That sounds like an ad. Are you trying to sell me something on daybreak? No. Actually, this is something that happened on a stage here in Bangalore just a few weeks ago. And as expected, everybody in the room lost their minds. And I think by now you know what I'm talking about. It is called agentic commerce. And it is basically when an AI just handles the entire shopping journey for you. Yeah. And it sounds incredible. But only until you

start pulling at the threads. Exactly. Because as it turns out, there are a lot of threads to pull here. For example, is the agent actually working for you or is it working for the platform that you're shopping on? Those are two very different things and that changes everything, right? And also, what happens to all the data it's collecting about you every time it shops on your behalf? Who has access to that and who actually owns it? And what if it buys the wrong thing, it could be the wrong size, the wrong date, maybe the wrong merchant, who's going to be responsible? We don't have an answer on that, by the way.

Yep. And that is what puzzles me because all of this almost sci-fi level of advancement in tech that has happened in such a short span of time, which means also that so many things can go wrong, right? And yet, we barely have a clue about who is accountable for what. Right. And meanwhile, Razor Pay, which is one of the country's biggest payment gateways, is betting its $700 million IPO on this.

And this is also happening when a giant like Open AI just killed the exact same idea. So as it turns out, two of our colleagues here at the Ken, Wally Vikram and Mutasim Khan spent weeks inside this very story, and Mutasim is here with us in the studio to talk to us about it. Welcome to Daybreak, a business podcast. from the Ken. I'm Snigda and I'm Rachel and every day of the week we bring you one business story that is worth understanding and worth your time. Today is Friday the 10th of April. Okay. Thanks so much, Matasim, for coming to daybreak again. Thank you. Always good to be here. So Matasim, your story started with this very interesting demo, right, that happened at the FTX Razorpe event where somebody just said,

Zomato order Chai and Samosa from Chai Point right and there was no pin required nothing was required from the user and the order just went through for someone who's listening to this whole thing for the first time what just happened like it sounds like a party trick

it does it does and that is how the room received it also you had a bunch of like applausees and ooze and haze and the room was amused generally so what actually happened is well, so right now in status quo when you want to order from Zamato, you usually have a whole exploration process where you type in the search bar, you select out of whatever available options you have. And then there is a whole payment process where you are redirected to a

UPI app and then you enter your pin, etc. They basically collapsed all that into just one single process with a chatbot where you are the chatbot what you want. The chatbot sort of selects the best according to its own selection logic and then it asks you to confirm once and you do that and that's it. It takes care of the payment on the

back end and then your only interaction with the app then is through engagement with this chat board, that's it. How does this happen? Like is what I want to understand. Like have you given permission in advance? Like how

is that payment going through Okay, so there are two separate processes here, right? One is the payment process and the other is the search and whatever selection process. For the payment process specifically, which is where I think Razor Pay comes in, plays a more central position. There are basically two ways in which this can happen. So one is this feature by UPI, which is called UPI Reserve Pay. Basically, what it allows you to do it is you can authorize a certain vendor. by vendor, I mean a Zomato or a Swiggy.

You can authorize it for a certain set amount. So, for example, $5,000. I can through my UPI app, authorize Swiggy or Zomato to debit $5,000 from my account automatically without me having to route back to the UPI app. So that is sort of the architecture that they are using at the back end to make these payments. Right. And so the context is basically this is what they call an agentic state. stack in payments, which is basically that they are deploying an AI agent who does all of

the other things for you. And your role is only limited to like telling it on a very bare minimum basis of what it is that you're looking for. And then it does everything else. So it's an agent that is doing all the clicking and all the searching and everything at the back end. So you bring up this very interesting like a comparison, right, with what Razor pay is doing and what is going on in the world outside of India when it comes to.

AI agents, right? And you talk about how open AI cancelled its feature that it was, that it had with Walmart and Target and all. And the, you know, the surrounding story for that is that they're getting ready for their IPO, and they're ditching all of these like side quest things that they've had to get like IPO ready and focus more on their enterprise like products. So I was wondering why Razor Pay is also getting ready for an IPO. So why is, you know, Razor Pay embracing agentic payments, agentic AI in this particular time when one of the biggest AI leaders

in the world right now is not doing it at the moment. Okay. I think that is a very important question and sort of sits at the center of the story that we did also. But to begin with the context is that open AI is a very different kind of company than Razor Pay. What Open AI had done is that they had made it possible to order stuff directly through their chatbot.

So they did an integration with Walmart and Shopify and all these marketplaces. And then you didn't have to open the Shopify or Walmart app. You could just ask the chatbot. The chatbot would relate to an agent who would then do everything else for you. A bunch of things happened there. One, not all marketplaces were sort of ready to let external chatbots enter into their apps or marketplaces. And this is where something called an MCP comes in, which is a model context protocol. So if your agent wants to interact with sort of another ecosystem,

they do it via this MCP. So the MCP mediates in some sense what the agent does with that ecosystem. So a couple of things happened. One, not all marketplaces were giving full autonomy to these agents, simply because if you have all of these agents coming from external third party apps, it sort of takes away traffic from your own website. And that means a lot of the ad. avenue that comes in or a lot of the sponsored listings.

I don't think these marketplaces have figured out how to place them in this new agentic future. To what happened with Open AI specifically is even on the supply side, a lot of merchants did not sign up with this. So the number was something like, I think just a dozen merchants out of like thousands of merchants actually signed up to this thing, which means that, you know, you couldn't do much. You mean, you did not have a lot of options to order from the. chatbot itself.

Now, why that happened? I am also not sure. Maybe, you know, the merchants did not think it was worthwhile or they were not comfortable with the technology, whatever. But coming back to why Razor Pay is doing it, I think a couple of reasons. One is this whole race of trying to position yourself as AI first. And even in the way that Razor Pay positioned it, they did not talk about it as a feature or a new product.

They talked about it as a sort of a re-imagination of what Razor pay is and the payment. government's ecosystem. I think it is partly that they know that the world is anyway going to move towards this, will move in this direction finally. And so they're just like jumping on the bandwagon, but also that they're preparing for an IPO. And it's probably that positioning yourself as an AI first company probably fetches you a better multiple in the way people are looking at AI right now.

Got it. Right. We'll get to Razor Pay again and how the role of Razor Pay will. also changed. But before that, you know, Matasim, you mentioned Ali Pay in China, right? And how they managed to scale agentic payments

to, like at some point in February, in one week, you said there were 120 million transactions, right? So it looks like Dave managed to pull it off. How did that happen? I think so, even in my conversation with people

when I was doing the story, the China context is very, very different because Ali Baba is a whole monopoly, right? It's a lot of different services just bundled under this one ecosystem. So probably this external internal friction there is not as much, right? You are in the Alibaba ecosystem and your sort of your agent is engaging within that whole ecosystem. Right.

So probably it worked out better there. But even other than that, even in China, for example, in the same week that we did this story. A lot of these Chinese companies had banned OpenClaw. So OpenClaw is another one of these agentic technologies that you could use. For example, it latches onto your OS, so you could use it in your phone, right? And there was a couple, there is something called Dubao. I'm not sure I get the name right, but this is another China local agent, which also a lot of these platforms like AliPa blocked. So sort of what I saw was this same story where marketplaces want to have agents, but they want these agents constrained within their platforms.

Right? So, I mean, they want agents, but without the full agency is what I said, because people still haven't figured out what it is going to do with their business models. Because like I said, right, your engagement with a lot of these intermediary app completely changes. For example, now in this system, you don't need to open Google Pay or phone pay every time you're making a payment. You don't need to engage with the exploration page of an Amazon or a resume.

matter. So, this is sort of broadly what I've been seeing is that for these big marketplaces, they are under threat of a lot of their user engagement changing, like radically. So then there is this thing where they would rather have agents inside their app with the idea is that they will reduce friction, like you don't have to do much, rather than having agents from outside. Because then, you know, that agent, external agent has a much larger set of vendors that they can compare with. Right? So now if I want a chai and I'm using a third party agent,

we'll compare across Zomato and Swigi and maybe in the future across the Kirana store, whatever, the chai store nearby might have its own merchant agent. So then your traffic is sort of getting diluted, right? And companies are still trying to figure out how that plays out in their business model. So that was broadly what I saw as happening in the larger context. Just quickly before Rachel gets to her question about going back to Razor Pay, I want you to help us understand how the role of Razor Pay will change. And also these UPI apps, what about GPA, what about phone pay, what's going to happen to them if they're not even in the picture anymore?

Like, how is it going to work? Okay. So for Razor Pay, actually, there are a couple of things. One is that this, what we saw is a consumer-facing agent, right? Which you and I will use in interacting with these apps. But Razor Pay's agentic stack is much wider than that. For example, a lot of their agentic use cases are also based on the vendor side of things. So, for example, for a Swiggy or Zomato, for handling things at the back end.

For example, you have a lot of these failed payments, right? And then you have to raise these like resolution queries. Or you need, you know, more like convenient way of handling data. Or for example, your call center. So there was a very interesting, again, a very interesting demonstration in the same. event where if you were to go to, for example, Zomato and you were to add something to your cart and then you, for whatever reason, did not want to order and you close, the agent will attempt to call you and then will tell you, oh, you had this on your cart, what is the problem, why you're not

ordering it? And so it can help in those card convergence. So their agent... For the customer, that's annoying. Yet another phone call from another... In fact, in the event, it was even more dystopian in some way because the agent was actively negotiating with the person. They offered a $500 discount also. I mean, I think that was purely for the demonstration purposes.

But yeah, so the point being that their agentic stack is much broader than just what happens in a consumer-facing thing. They're also like doing this for companies who are on the marketplace side of things and not on the consumer side of things.

Got it. So along with how, you know, things are going to change for our UPI apps, you point out that this could actually be an opportunity for banks, right? Because now that UPI apps, like they're kind of moving downstream in the whole interaction, this could be like because banks haven't really been able to compete with these payment apps, this could be an opportunity for them.

So can you explain why that is? Okay. So first a little bit of context on where UPI apps sort of sit in this whole equation is that earlier, I mean, right now, you had to open the UPI app every time you want to make a payment. But presumably that's not going to be the. case in an agentic future because at the very best you just have to authorize it once for whatever vendors and then you don't need to open them anymore again so the user interaction with these apps one becomes a lot lesser and also changes like fundamentally in some ways but for the banks

now the idea was and we spoke to a senior executive at a at a local Indian bank the idea was that what differentiated UPI apps from bank apps was a superior user interface experience. But now that that user interface is not that prominent in the process of things, then they felt like, you know, then you could have a lot more competition from these bank side apps also. Because if all UPI is doing is just like helping you do some kind of authorization or handling things on the back end without there being much user interface. presence, then the idea is that maybe these bank apps can sort of enter that space and that

market can become much more fragmented. Right. Also, I think at this point, it's very important to draw this distinction between the two different types of agents, right, when it comes to agentic commerce. One, like you said, is the one that shops for a platform, which is like in-platform agent, right like the Zomato example that you gave and the other one is the neutral agent which will work specifically just for you

which will be looking across platforms can you tell us the difference between these two sure so fundamentally the difference is that for an independent agent the logic is the logic that they use for searching or giving you options whatever is not tethered to the app at all in the sense that if I have an external agent, they don't really care about what benefits a Zomato or a Swiggy.

So their loyalty would be to you. Their loyalty ideally should be to the user and what the user wants. But once you have an agent that is like tethered to the app's own catalog, then what ends up happening is that the app can code in or the business can code in its own requirements. And then those will also play a role. For example, like very, very simple example would be. sponsored listings, right? How they could work in this new thing is that

sponsored listings are embedded into the logic of what, how an agent searches when it is searching on, on Zomato, right? So, you might ask it to search for X, Y, Z parameters, but in its calculation, it will add another, like another parameter which has to do with the app's business logic. The idea is that that is not the case with the independent agent. So in that sense, the argument was,

an independent agent if it is allowed to work that way. Because again, these independent agents again have to go through that marketplace's MCP to even make that search. Right. So the idea was that an independent agent from a consumer standpoint is much more objective in their search. But marketplaces don't seem, it seems that they do not want independent agents yet because of what it might end up doing to their business models and how customers are

interacting with that. Just a quick follow up to this, for a consumer, what would the difference mean in terms of the kind of data and permission you would have to give to a neutral agent, an independent agent who's working specifically for you versus an in-platform agent? I mean, that I think, is still a very, like a convoluted question to answer. So we had some conversations with people who are working in the regulation space and they seem to agree that no one has really thought about this because it is a very sort of complex problem to think about even in terms of the DPDP, the Data Protection Act, how that is going to feature into this. What kind of information is the agent allowed to retain and remember how much of that information is available to companies that are running these agents, etc. don't have a very clear answer on that. Also, I don't think I'm like very well educated on that subject. So, yeah, not very sure.

Sure. I think in general, I mean, I don't think there is a lot of information and clarity on this. True, true, true. And that is what, I mean, that is what even came up in conversations is that the technologies is just moving way too rapidly. And the regulators are not, have not caught up yet. or broadly are even not trying to catch up. So, for example, one of the conversations I had, they were,

they were sort of trying to say that, you know, the regulators such as the RBI or even NPCI, whatever jurisdiction is, it has, has to catch up now because technology is going to scale, adoption is going to scale. And if regulation is not there, then you're looking at a lot of, like, a lot of problems. For example, with fraud. So even right now, when you have,

a human in the loop in all of these processes, right? With the UPI app, with everything, search and confirmation, the fraud rates are still very, very high, especially considering that India is like relatively new to this kind of like tech infrastructure in their daily purchases, etc. So, if regulation doesn't catch up and you don't even understand what is happening with these things, then it's going to be a problem. For example, if, so this is one example, if I want to book a ticket to a Taylor Swift concert. I tell my agent, okay, 12 p.m. the tickets go live. Please book a ticket for me.

It has to be a front row ticket and whatever price bracket. And the agent says, okay, and ends up booking a ticket which is not front row. Right. Whose fault is that? Is it me? Is it the agent? In that case, who is accountable? Is it the bank that?

Who is the customer going to relay this complaint to? So all of that, none of that is really clear right now in the current regulation. landscape. So you also talk about, so this we're talking about like a future, right, like with so many people are using agents and it's almost become like second nature at this point. But you also mentioned that there is a cognitive cost to use an agent, right, on the consumer side. Like what does it take for consumers to actually start using agents? So can you explain, you know, what that means here? Sure. So there is a trade-off the way we see it. One is, sure, the agent is reducing a lot of friction in the sense that you have to do much.

much less work. But that only works when you want to order something like a commodity product. Like it works for a chai and samosa because how different will a chai and samosa be from like sold at different places? But when it comes to more like more differentiated purchases, there are two things. One is do you trust that the agent has made the correct choice or is showing you the best choice? Like if I want to book a ticket, I want to book an indigo ticket. and I tell it that, you know, book the cheapest ticket,

it gives me one option or two options. Do I trust that it actually booked the cheapest ticket? Because even a 500, 600 delta for me is a huge delta, right? So that is one. The thing with trust, whether I trust the agent's output. The second also is that I have a long-formed habit. Like, it's a decade-old habit the way I'm interacting with these platforms, the way I'm used to searching, the way I'm used to deciding.

Then there is also a cost to switching from this habit to another habit. habit. So the trade-off is whether the sort of the value delta that you get on choosing an agent, whether it is able to justify your trust as well as the inertia with your habits and whether that happens is to be seen. And we will see those in adoption numbers over some time. Right. Matasim, there's one more thing about, you know, these permissions, right, that you as a user need to give when you're using a genetic. commerce, basically the agent is acting on your behalf and there are dozens of permissions that you're giving for future transactions, right?

Which are, I don't know if they're individually reviewed. You call this continuous consent in your story, right? McKinsey is calling it third actor problem basically when a non-human entity is making transactions that existing law is never designed for. What does that mean for someone who just wanted to buy groceries, you know, from a platform? Yeah, so the problem of continuous consent basically is that in the current term, the human is in the loop at all points in time. You are authorizing and permitting everything that the app is doing.

But in an agentic world, the consent stops after a point. And then all the subsequent actions have been taken by an autonomous agent, which you had an intent but did not have consent. You did not have a chance to even give your consent. So I think for a normal consumer, like we said before, it all comes back to what happens if something goes wrong. You know, and the idea also is that UPI, for example, NPCR UPA have spent so long building trust in their payment infrastructure. Now, let's imagine a world where this, you know, agent ecosystem scales and you have frauds or you have some kind of hiccups. That really risks eroding trust in payment infrastructure itself or erode.

troding trusks in how these agents work. So for a normal person, I think that is what it comes down to is if their trade-off is justified. And if, you know, if they have bad experiences, especially right now in this phase where everybody is just getting introduced to this kind of thing, that can have a sort of long-term impact on how people are even looking at these payments, especially for a large part of India, like I said, who are still getting calibrated, still getting used to these new technologies and new ways. of purchasing and ordering, etc. So to go back to, you know, all these marketplaces that aren't allowing agents right now, right?

Amazon stopped. I think perplexity is comet. And, you know, I think the founder also called it out to say that it's because Amazon can't bombard customers with like continuous ads. So, I mean, we've been talking about this a lot also because of the zero short events. that's coming up, right?

Like what the agentic future looks like. And, you know, agents are not going to browse. They're not going to have like impulse buys. They're not going to fall for like dark patterns or whatever. So what does this mean? I mean, this is a future that is going to happen. So what does this mean for these marketplaces? How would they have to redesign or recalibrate to make sure that they are agent-friendly as well

and not just human-friendly? Yeah, I think that is at this point an entirely open question. I would really like to do a more thorough investigation to see how these marketplaces are actually thinking about what they will do. And that's the content of the zero short event, obviously. But for right now, I think what these marketplaces are doing is installing these agents within their apps. Right. Probably, I mean, you're catching up to the larger, whatever is happening in the larger ecosystem. And also maybe it's a pilot.

It's a sort of pilot to see because this is not a replacement, right? This is an additional feature that is going to be on your Swiggy and Zomato app. And I think these next few months or let's say the next couple of years are going to sort of give these marketplaces more idea about user behavior and how their engagement with the app is going to change with these agents and purchasing with these agents. So that is still, I think, probably a very, very open question. but like I said, a part of it would be to figure out how do you incorporate your business logic into the agent's choices, right? And that's a whole different problem for them to figure out with ads. I'm actually not very sure how you have ads because obviously Chad GPT talked about having ads on their own platform. So what happens with that and whether now you know you have new UI where you still are going to be having ads popping up here and there.

But the idea is that things are going to change. And so these marketplaces will have to sit down and sort of brainstorm on what happens, what the implications can be and how they have to move forward with this. That is for sure. Also, out of curiosity, I was wondering, like, we are building this, like, bifurcation between agents over here. Like, we have these native ones that are native to certain apps and websites. And we also have these so-called neutral ones.

but considering like something like an open AI are also being are also trying to put ads on their platform doesn't it also challenge the whole neutral fallacy a little bit if open AI is being paid by certain companies to put like advertise their products but possibly an agent would choose something else entirely because it's a cheaper better option so you know that that seems like a definitely but the thing is when we are talking about an independent agent it can exist also independent of an open AI or a cloud or an anthropic. I mean, I can locally create an agent on my own laptop with one of these open source libraries or something.

So the idea is just to sort of these are two extremes of a spectrum and it helps with sort of like a thought experiment. I mean, obviously the reality is probably going to be somewhere in the middle. But I think that bifurcation is useful just to understand the trade-offs on the spectrum and which side this might go to. But yes, it's just so many things happening and we don't know how this is going to turn out. Okay.

We're kind of running out of time, but I have two questions. One is a slightly philosophical question, Matasam, because since you worked on this, right? And there was somebody who was a former payments executive who told you, you know, there have been countless experiences in my life where I'm in a store wanting to buy something,

but the paradox of choice. made me drop out. And this is where agentic commerce comes in, right? But also, on the other hand, being human means having a choice, right? What are your thoughts on this

as someone who deep-dived into this subject? I think it again comes back to what kind of... Sorry, I'm making this very mundane, not staying in the same philosophical paradigm, but it comes down to what kind of choice it is that you're making.

if I want to order a chai samosa or if I want to order a pair of socks or just something like a very, very kind of mundane kind of thing, where differentiation doesn't matter even more. Sure, I think the paradox of choice is just adding more cognitive load. If I want a samosa and I have like 100 different options, you know, that is a lot of cognitive processing I have to do. But when it comes to things that, you know, are not this kind of chai samosa type of products,

then I think the paradox of choice is actually the privilege of choice, where you know you actually are able to choose. And I think with an agent, so when I am, for example, browsing through something, it is not like I am using like very, very set logical, definable parameters, right?

There's a lot of like intuitive things, you know, something I might like based on whatever experience, something I might not like. So that is going to change. So in a sense, okay, I mean, you might not have as much choice,

but then you also don't have sort of maybe as much freedom in choosing what you want. Again, it's a trade-off between effort and probably just like the freedom to, you know, in a free-flowing manner, choose what you want to choose. Yeah, that is. Yeah, like, I think as of now with the way things are in terms of regulations, right, and how they're not being able to catch up what you said about privilege of choice can soon be taken away, right?

Just last question, right, Matasim, you know, McKinsey said in a report that three to five trillion dollars globally by 2030 will be spent on agentic commerce, right? And raise a pay, as you said, is betting its IP on this. If you had to name one thing that is most likely going to be the cause that will derail agentic commerce in India, what would it be? I mean, I think at least from whatever investigation we did for this story, it is going to be this aspect of trust. Because if you have experiences with these agents where you are losing money, especially in an economy that is still very price sensitive, I think there's just going to be this looming bad experience with an agent, which is why regulations need to calculate. So from a purely adoption standpoint, that is probably going to be probably going to be the thing that can really take it off its rails. But I think from more like from more business standpoint, which is from the standpoint of these vendors who are using this agentic stack for a lot more, like automating their daily business operations and customer relations, etc. I don't think there is much that can derail it.

I mean, everybody is already moving towards it and I think people are just going to go ahead with it and find out what happens later. All right. Thank you so much, Matasim. That was very insightful. And we're looking forward to your follow-up story on this. Hi there. Before you leave, I have a quick message for you that is in line with what we discussed in today's episode, actually. You see, AI agents do not browse. They don't scroll. And they don't succumb to dark patterns. They compare every option, find every hidden fee, and root every transaction to the cheapest, fastest and most rational choice, all of this in milliseconds. So if your business model depends on human friction, on human attention or inertia, the next 24 months will be the most consequential of your career.

Zero short that Rachel mentioned earlier in our podcast is the Ken's first live event dedicated to imagining these futures, rigorously, provocatively and without the PR gloss. You will be interacting with founders. and operators who are already thinking about this. And you will influence the agenda and the direction of this conversation. And you will walk out, upgrade it. It is happening this Sunday on 12th of April at Bangalore International Centre at 10 a.m. This one is worth waking up early for.

Ticket link is in the show notes of this episode.