Zero Shot Live: Who will survive Agentic AI in India?

Hello, hello, everyone. This is Vidhatri, the producer of ZeroShot. You might have heard me pitch in from the background as Rohin, PGK, and Brady argue away about AI with a whole lot of passion every single episode. But today is not regular programming. This is a special episode of ZeroShot. Last Sunday, the Ken hosted its first ever ZeroShot live event, which was completely sold out by the way. We called it the "Great Unbundling". Some of you were there and I want to personally thank you for showing up. It meant a lot to us here at the Ken. Here's the premise of the event. Agentic AI, AI that doesn't just answer questions, but dust things on your behalf from booking your cap to paying your bills to comparing your loans to placing your food order is going to arrive soon. It's going to arrive hard and it's going to arrive very, very fast. And when it does, it's going to unbundle every layer of value that India's consumer internet companies have spent the last decade building and perfecting. The questions we wanted to answer with these, who survives this AI agentic race and how? We got four guests to help us think to these questions. Murali Krishna and B or Murali spent 27 years scaling consumer tech businesses. He was a president of Xiaomi India and had since had Mintra and eBay. By the way, he has a PhD on how platforms create lock-in. When there's a barrier for customers to switch from your brand to competitors offering. He calls himself a quasi-academic, but he really brought that sharp understanding and contrarianism of someone who spent years studying a subject deeply. Then we had Amod Malviya, who was a CTO flip-cut until 2015. He then co-founded Udan India's largest B2B e-commerce platform. He's now building an AI startup called Pre-Six for manufacturers to go from the drawing board to the production floor. Joining them was Anand Chaube, the co-founder and CEO of Capillary, a customer loyalty platform that powers over 100 loyalty programs across many countries for brands like Indigo, Tata, Domino's, and Marks and Spencer. He thinks about what keeps customers coming back, a question that's going to get even more complicated in an agentic world. And finally, we have Thomas Spen. He's the co-founder of Mahabeli, a Malali restaurant in Delhi and Sairba. When I was in Delhi, I would make sure to go have their onams Sadeya every single year. Thomas is also the joint secretary of the National Restaurant Association of India, where he works on e-commerce policy and the economics of online ordering. He's seen what platforms do to restaurants from the inside and has very, very sharp opinions on the food delivery business. And piecing it all together was Rohan Dharmakumar, the Kent co-founder and CEO and one of our lovely hosts. It was a whole lot of fun being in that room. Hope you have fun hearing it too. [MUSIC PLAYING] OK, let's get into it. Rohan introduced the framework the room would use for the rest of the morning. Systems thinking. Specifically, something called the iceberg model. If there is one skill which is going to be really important in the agentic AI and the AI era, it is the ability to do systems thinking or to view everything as systems. This, typically, one of the things when you look at systems thinking is that what is visible to you is not what the real scope of the problem is. What is really visible to you is just the events that you are seeing. There's a lot of stuff which is hidden much below you, which is why it's called the iceberg mental model. The iceberg model has four layers. At the surface, events, what's happening right now. Below that, patterns. Why does the same thing keep happening? Go a little deeper, and that's where you get structures. What's actually driving those patterns? And at the bottom, the hardest to see are mental models. The beliefs and assumptions that build the whole system in the first place. To look through the iceberg at India's biggest companies, the room used six lenses. And sorry for getting technical, but I'll explain why these lenses matter later. Here's Rohan talking about the first lens. The lens one is what's called the economics of surplus, which is really where is the money? Who's making margins? Who's making commissions? Because if you are either, I mean, one of those companies, you're trying to safeguard those. Very famously, to use one of the most repeated cliches for Jeff Bezos when he started Amazon, which is your margin is my opportunity. This is really that. When you're essentially saying that agents have the ability to cut through layers of middlemen and transact directly, that's what really happening. It's a margin collapse. It's a commission collapse. It's an advertising collapse, et cetera. So then you have to really look at, where does it sit today? Who's making money from advertising? Who's making money from commissions? Who's margins on a product are unsustainable because they have a very large market share, and they're charging a premium for it? This is that becomes a really interesting way to look at the economics of surplus. Now the second lens is all about your attention. Literally the most precious commodity right now. We called it the attention traps and discovery economics. Any app and it's you're looking at it, it's monetizing your attention, either through advertising, or through fees, or through subscriptions, et cetera. It knows what you browse for, it knows what you search for, it knows, and that over time allows it to make money, either by letting companies target you, et cetera, and stuff like that, right? So another way is to look at the existing world lens, say, who knows stuff about me, who knows stuff about my preferences, my purchases, et cetera, and where do those things sit today? And who's actually capitalizing and monetizing on my attention? The third lens is about our usage, our habits. How we fall back on certain apps without even realizing why we do it? Rohan called it friction as a business model. Friction as a business model, right? And I think this point is made like, morally said that we human beings by nature want to reduce friction in our life. So a lot of the most powerful apps and services that we use today, I spoke earlier about phone pay. I continue to use phone pay, even though it's irrelevant, which API I use, because I prefer to not have the friction of trying to remember which app to use. Does it have my last transactions, et cetera? So which are the apps that are currently used-- I mean, in the sense that they have your patronage because you're trying to avoid friction. And therefore, if that goes away, what can you do with it, right? Structural modes in lock-in was one of the other frameworks we used. Now, this answers a question about what makes a business fundamentally unique and hard to switch from? And how do you create that irreplaceability in an agentic world? OK. After Rohin described the lenses, he made an announcement. What we're going to do is that these lenses, we're going to try and apply to some of these companies, right? Now, you might be wondering what Rohin wanted people to pick. Let me tell you, companies. He essentially listed down a set of companies to choose from. The audience had to pick which of these would survive the post-agentic world. We call that the stress test. But the answers were not merely based on wipes. Remember the lenses we were listing down a couple seconds earlier? That's the framework the audience had to use to make their choices. Long loop, but it had a promise. Whichever you guys, you folks, pick. We'll try to go through that list and try to do a live simulation or a live stress test of that company based on those. By the way, the audience wasn't answering in thin air. For the first time, we used an interactive tool called Slido that could be accessed from the KENS app. Asking the room to respond to us in real time. They were answering polls and responding to our questions. And all of that was projected on stage. What that did was give the guests a clear sense of what the audience was thinking. It was interactive. It was innovative. And it was exciting and super fun. [MUSIC PLAYING] Nika, Gro, Lenskart, Kred, Misha, and Rezepay. These were the options. The audience voted. And guess what? 35% of the room picked Kred. Would be fair to assume most of the folks who voted for Kred are Kred users as well. That means one-third of the room raised their hands. The question was whether Kred had actually built something durable for them. Or whether it had built something that looked durable on the surface, right up until agents arrive. Boom, the stress test began, and our guests brought their A game. Mulee first set the scene. In the agentic world-- and I'll apply two lenses on it. Not one, but I'll apply two lenses on it to try and articulate the way I'm thinking about it. about it. Number one is across those four stages of the life cycle right we spoke about aggregation discovery trust fulfillment. Where are you playing primarily? We have beaten all the marketplaces which focus primarily on aggregation and discovery to death and rightfully so because those are for the most part agent replaceable right agents can possibly do it more efficiently than humans can and therefore those marketplaces lose power. That is one part but there is another part to it a business is not just a marketplaces not just aggregation and discovery there are these other assets that the business has some of these are physical Amazon fulfillment infrastructure. Swiggy blankets, dark store infrastructure those are not trivial right so that is one part but more important than that are let us just call them intangibles quite often when we say intangibles we discuss brands but intangibles are far beyond brands of course the strength of my brand is an intangible it does not reflect on the balance sheet it reflects on my market value not on my book. So the brand is an intangible my information know how technical know how etc etc understanding the customer that is that is great intangible my relationships with my partners with my customers with with the entire ecosystem that is another market based asset. I believe companies which are strong on these factors trust and the other intangibles are agent resistant so you have agent replaceable and you have agent resistant and therefore whenever we look at these cases I will apply these to framework are you agent replaceable are you agent resistant. Agent replaceable versus agent resistant catchy words that reveal something far deeper and more interesting so the ability to adapt to tech is of course key but not just that merely argue is that businesses need to become unavoidable let him explain how this applies to cred in his own words. How does a company demand pricing power and this this runs at the core of the economics of any business and and strategies spend like countless hours trying to debate this part and you could debate this you could have a different point of view I think it boils down to three things. Number one you should be unavoidable for a transaction to complete. NVIDIA chips are unavoidable right example and I will take NVIDIA because it is an example where everybody can understand if you want to run whatever hyperscalers and run AI or computer and so forth maybe NVIDIA is largely unavoidable is cred unavoidable perhaps not in the payment space maybe a visa master card in the US largely unavoidable because you have to route to two of them or discover right you largely you have to route to them there is hardly any other option. So are you unavoidable in the workflow if somebody is using you are they switching cost is they switching friction not only am I unavoidable I cannot even switch out if there is a better alternative because so much this person knows and there is that end of me it fundamentally needs to reinvent the way it does business by leveraging it single percent biggest asset which is transaction history of the top 1 percent paying in the consumer economy in this country right I do not believe they are unavoidable there is very little switching friction unless cred coins comfort something and we on a point what they do is getting commodities because paying credit card bills is just that it is just a payment so I would say hey they are in for tough times. Cred is genuinely good at holding your attention the app is designed to keep you in it it has a very premium feel it has got gamified rewards curated offers and the ads do not get me started on the ads they are fantastic. How much does cred manage to monetize your human attention any business that requires you to scroll dwell or watch in order to see ads agents do not scroll they just retrieve which companies have advertising and behavioral profile databases rich data what human preferences very strong right now for credit of course right sponsor result density and discovery monetization right these are I think some of the things around attention traps and discovery economics I feel currently credit is very well placed on this. The room did not push back on this one they were all on board then came friction as a business model and that's when things got a little intense. Friction as a business model in the context of cred can also be interpreted in a slightly different way that the current actors in the in the merchant and the credit card processing space actually do not do a do a great job of engaging with the customer banks or credit card companies. So you know which is what credit you know to whatever degree I understand it because I am not a user I felt below the threshold the they they they they solve that friction it allowed for people to actually be much smarter with their credit card payments. The only real thing is that the information does not belong to credit the information belongs to the user. At this point somewhere in the room a credit user made the case for what the company actually does well it tells you which loans to prioritize it helps you make smarter decisions about your credit that they argued is genuine value that goes beyond just paying bills. That's a brilliant use case right because yeah I've heard this. So that's the kind of use case that the point is that there are two different sites of the information that credit is sort of covering one is the actual scenario of that user second is how the scores typically tend to work and you know what kind of a decision you know might sort of pull your score down or you know etc. The thing is this part of the information is actually not very difficult to find if you go to ask any foundational LLM today they they actually would be able to do a good job of it. If you as a user the information belongs to you can you use a foundation LM to pretty much do this whole setup you know in a seamless fashion yes you can do and that is the that is the sort of issue with the creds model at least to the degree that. Just saying this friction will disappear because this can be done naturally. It's actually easy off loadable to a 20 AI agent. 20 AI agent. There's an opportunity for cred here that they do have an existing trust rate relationship. So to whatever degree before the customers start doing it for themselves you actually make it easier so that the customer does not even have to think about setting up their own agent to do that. This is interesting that you bring up because this that there are two ways to think about this one is the classic trap of incumbents. A large incumbent which is sitting within existing business model you know very it's it's very few times that they think about before someone else disrupts me can I do something to myself. Unem joined into say there are a lot of people and startups would be eyeing for that top 1% of clients which is exactly creds base that meant more competition more stress and the overall struggle to differentiate here's how that conversation played out. And for top 1% of India I think there would be lot more agents that compete for them compete for them because the economics would just work out and there's also good point so this is actually another point that I was thinking because the richest people in India are probably will have the most agents competing for them plus they are also likely to be the earliest users of AI agents because they are more tech savvy as well. The richest people in India are not as tech savvy as you get to be. No I'm I'm I'm talking relatively. That was merely again pushing back on Rohan's statement then Thomas jumped in. Sorry family of law. Yeah they look at wealth management in that perspective I guess yeah agents would be competing because there's a lot more at play right. No but at that level I guess a lot of people prefer a physical agent. Honestly. Now hear me out. This is the main question. Does cred have a structural mode and lock in that will help it survive in the post agentic world. Thomas and a mode has some thoughts. Does cred have modes and lock in history? I guess non portable like you've been paying your what do you do with your history? Why do you want to carry your history? But the that history comes from the like the user has sort of shared that it's not your own mode. The user. You know but what is that history of payment history? Why do you need your history? Nature of no no. I mean my answer clear up it like the fragmentation problem right like if cred has got my my financial history in its full. It's a lot more useful than multiple people having pizza pieces of it. So the lock in over there. No no I get that but where like. like. Why do you need that history? What do you intend to do with it? Why can't you just walk away? - As cred as a consumer. - No, no, no, I'm saying you are the consumer. I'm saying if cred has a lock in on you, right? You can you walk away from cred and write off that history without really losing much? - It's kind of like how we pick phone platforms, right? I mean, you're probably gonna use a Mac for the rest of your life because not just history per se, also it's documentation is certain, we habits. All of that paints a fuller picture. - That's a good example of the difference. - Now, I've told you about the audience, they were really engaged. So at this point, someone in the audience shouted out that cred's lock in could be its rewards. - Some of the airlines have built loyalty programs who are far more valued than even airlines itself. Because on those places, those airline points became a currency and those currencies became portable and became so strong that there was a inherent value in that. You can actually sell that currency and other brands would buy that. From what I understood from cred, there is, there is, there's a very questionable value to those points. (audience laughs) - No, it didn't happen even to work on it, but I think that was the full idea that you can get a deal here, which others will not get on a flip card or they will, that was the intent around it. - Fair, I mean, if one were to really question that and say every credit card company has some kind of a reward program, every airlines company has a reward program. So it would only be a structural lock in if you feel like cred's reward program is so unique and so valuable that walking away from it would mean giving up a significant amount of value, which seems to not. Exclusivity is what you are effectively getting to, right? - Now I want to remind you again about Mulee's PhD about lock ins. So he added on. - Lock in has multiple dimensions. So lock in can be a state of mind, right? Which is behavioral. I'm just used to familiarity, inertia, behavioral, right? I'm used to this way so I don't want to change. That's one lock in. The other lock in is just cognitively where you think about it. It's a state of mind, right? And that's where switching costs comes in. I'm just so used to this. I don't want to try something else. The trouble of moving to something else new is too much I don't want it, right? So this is state of mind. And the third element is more emotional. It's a state of feeling. And what we mean by that is look, I trust this. I'm happy. Why should I move? And when you speak to people about lock in these are the three different factors that come in. So either people are stuck because of inertia. People are stuck because moving out is too tough. Or people are stuck because look, I don't want to trust anything else. I'm happy with this. The test that you need to put up with credit does it pass any of these three tests, right? Is it a matter of habit? Yes. For me for paying credit card, yes, is a matter of habit. Number one. Number two, do I find other ways of paying to difficult, complicated or moving my data to complicated yes, they have my data, they have my past transaction record. For whatever it counts, they know my due dates and so on and so forth. So maybe, hey, that is one reason of switching friction, preventing me from going to someone else. The third thing, do I trust someone else less than that I trust them? Perhaps not. So at least on two of these three counts, in my assessment credit scores yes. At to that extent, yes, they do have some degree of mode. At least, by this framework, you could have an alternate framework, which has a different point of view, and I would respect that. [MUSIC PLAYING] The credit stress test continued on and on. I was in the room with a mic and passing it around, because people had so many opinions. I wish all zero-shot listeners were a part of this conversation. After this, we asked a room full of smart people a question that did not have ready answers. The stress test had done its job. It had given the room a vocabulary, agent replaceable, agent resistance, strong today, week tomorrow. And now, we flip the question. Instead of who's losing, we asked, who's actually built to win? The slide that came up when this conversation was happening was called construction, act three. It listed seven companies. The room hadn't spent much time on. ONDC, Rapido, Sarvam, Geo, Beam, Misho, Delivery. Companies that were either too unglomerious, too infrastructural level, or too early in their journey to have made it into the main act one discussion, except Geo though, which seems to be so many things at once. The answer to who would emerge as a winner and as the dark horse was unanimously Sarvam, with 68% of the audience choosing it. It's PSU hiding in unicorns' clothes. That was Rohan. If you weren't familiar with his voice already, Murli had an issue with Rohan's categorization in the first place. He did not think Sarvam was a dark horse. Models are getting commoditized. And not even there is unease beast. There is 19.9 stroke, 20, while a difference between most models after a point of time. So finding it difficult to grok as to pun intended as to why Sarvam is even a dark horse here. You're saying it's obvious. No, no, no. I'm finding it difficult because it just becomes another infrastructure player. Tomorrow you'll have three other so-called sovereign models. Unless the government is-- No, no, no. But that's the point. That's the point. I'm only-- So, one of the-- Bharatiya and so on and so forth. Unless the government wants to bet on this, right? In which case, this is like an influenced dark horse. But I'm not bet saying, I will make them the champion. Correct. Sorry, continue. I'd much rather say, hey, the dark horse is someone who owns a critical part of the infrastructure, which is not replicable. The counter and slightly nuanced one came in from Amund. It's not about the model he argued. It's about where Sarvam sits and is positioned. The degree to which the LLM has been able to lock itself in the distribution is actually going to have a much bigger play, which is where Sarvam does have an advantage. Let's say across because of the whole sovereignty, I think, which I think is-- That is literally the biggest factor I feel. The sovereignty, I just think, of governments across the world. And especially in India, we'll insist that crucial data stay in India or governments and certain businesses use, lean on them to use our own local models, et cetera. I think that's too big a factor to buy. It is a big factor. And it should be a big factor. I just feel that it would have been great to have more competition. But it is something that is likely going to be in servant favor. At this point, a curve ball was thrown into the discussion. And that was beam. Sarvam was a more visible answer. Beam was not. It just got 13% of the responses. The room was clearly skeptical. But hey, the panelists and Rohan had a contrary in you. Also, tying it back to the government back distribution point that was earlier made about Sarvam. In a post-agentic world, I feel like if you could route your payment through any entity, then beam stands a chance because it's backed by the government. It's got a much good chance. Here's where Amod had a nuanced take on beam. See, beam does not top of mind for people. It doesn't need to be for with agents. Need to be for agents versus whether it can be for agents. It has to expose an agentic interface. UPA SDK to the best of my knowledge, is structured in a way that's actually not very healthy for agents to do because of how it ties into the device. So today. Today. So the only way beam can get any advantage with given the fact that it's low in top of mind score, the only way it can get an advantage in the post-agentic world is to deliberately force itself out of the-- Yes, that's exactly if you remember the early days of UPA. Beam in that sense is-- I mean, I can imagine if the UPA protocol evolves enough to support agentic payments. I can imagine a scenario where beam is actually given access to those features among the early-- Like the early days as compared to the early on. Not possible. Think about it. Like the whole-- if I were at phone pay, five or some year, I would be shouting from the rooftops, how can you say that you're not-- No, no, no, no, no, no. Not only first. The first three, four entities that get access to agentic payments is-- Okay, first three, four entities. Visual constituta, I'm guessing 95% of UPA anyway. Correct. Which will be like basically Google pay, phone pay, and beam pay, and that's actually-- So pretty much, you will see the same distribution as you have with the current UPA distribution. It might get a little balanced out. Yeah, that's my point. That's a little balanced out. Beam is nobody in that. Like, asthma, no, we don't get-- No, I love for beam to succeed. That beam example came out of nowhere, but it sort of made sense when the room reasoned it out. And that was the event for you. Walking through unfamiliarity together. The overwhelming feeling many had before the conversation. was uncertainty. It was messy. There were no clear answers. By the time the conversation was over, there was a clear sense of the storm that's about to come and how we can make sense of it using a first principle's lens even when there's uncertainty involved. That's what the morning was about. A live three hour event where we collectively debated, discussed and dissected how businesses will change in a post-agentic world. This was the Kent's fifth live event. We're doing a lot more of this, more live events where you can experience our journalism and work. If you want to be in the room next time, keep an eye on the Kent's website for what's coming. Thank you so much for joining me today. We'll be back next week with a regular episode of Zero Shot. Hope you enjoyed this. Have a great week ahead. That was Zero Shot, the Kent's weekly podcast on the biggest developments in artificial intelligence. Our hosts and commentators are Praveen Gopal Krishnan, Rohan Darmakumar, and me, Brady Ng. Our sound engineer is Rajiv CN who makes everything sound spectacular. Don't miss our Zero Shot columns which are published every Saturday. We'll be back with more for this podcast next week. [Music]