MCP is the new front end for the agent internet

Ladies and gentlemen, welcome back to another episode of Beyond the Bubble podcast. Today we have with us a very very special guest, Pierre Louise Theron, uh who is the founder and I believe CEO of Alpek. He is a repeat founder who's already built and sold a company to Fortune 500 companies. He studied engineering at EOL Central Paris where he co-ounded Streamroot in 2013 with two co-founders. Streamroot invented peer-to-peer video delivery tech that sat on top of CDN's. That's a mouthful, but we'll go deeper into that in just a moment. Uh he uh the app let viewers share content with each other instead of all hitting the same server. During the 2018 FIFA World Cup, their tech kept French streaming platforms running when millions tuned it to watch. Again, this is very very useful considering there's a World Cup going on and um you know, it's expected that over 5 billion people will be tuning in at some point during this World Cup. Uh they went through the Techstars Boston platform in 2014, raised $6 million, were acquired by Century Link uh which is now Lumen Technologies. In 2019, Pier Louie stayed at Lumen as VP of product running their global CDN for 3 years in July 2025. uh he reunited with his stream route co-founder Nicolay Rodunov to start Alpic uh which really is spearheading uh the MCP system or the platform that uh you know that has seemingly emerged after the AI age these days it's all about MCP every app every um you know AI tool that I'm hearing about they're talking about MCP over the last couple of weeks uh I got my hands dirty in building uh sort of AI agents and being building AI first apps and I've now suddenly uh realized the power of this technology. Uh Pia, first of all, thank you so much for taking the time out and joining us on the show. >> Thank you very much for having me. Looking forward to it. >> Uh just as a quick starter, I want to get a sense of what Alpic does. Um what are you guys up to? How do you see the the current AI space evolving? Yeah, I mean RPI is is basically we're finding RPI because of the vision that the internet is going to change dramatically and that's where the Google search bar has been the entry point of the internet for the past 20 25 years now. uh but now the entry point is changing is becoming u AI chatbots, AI agents and the way we are consuming the internet is going to be less and less through websites, static website and mobile application and more and more through these AI agents and we believe that all the companies in the world whether they sell products, services or just uh present presenting their product online will need to talk to to AI in a AI native way and what they have today is not enough. Websites, APIs have not been made for uh this type of uh communication. So this is why we built ALPIC to uh power that revolution uh and create uh tooling and infrastructure for people that will need to build this new interface and we'll need to deploy them, secure them, monitor them, scale them and we want to be the home for that. >> That makes a lot of sense. Um I want to sort of double down on that and and get a non-technical understanding of that. Right. So right now my understanding is if you consider let's say the average chatbot I'm not even going deeper into agents uh independent agents but I'm saying let's say chat GPT has probably agents behind it that go out and search for different data and content. When I query chat GPT today normally my understanding is it does some sort of a search it goes and reads certain content from different websites and then it comes back and then sort of tells me about it. Right. So you mentioned that there aren't there isn't enough infrastructure that enables for a lot of these agents to come in and uh consume their content. How is that different? Because charge GPT already does that again on a on a very basic level. >> Charge GPT is already doing some amazing things. But perhaps let's double click on how it is actually working. uh the to to make it simple the way this model is going to give you an answer is based on like first the pre-training it's been like a lot of discussion explanation about pre-training is basically feeding them with as much data as you can the whole internet basically and they're going to be smart about it so they're going to be using that to answer you the second step is everything which they've done around like post training reinforcements reinformance learning which is basically taking some specific data sets that's going to make the model smarter about one specific topic or a few specific topics. So that's kind of everything that they have in their brain. Now because they are not trained every day, they also need to have access to the internet to what's what's going on, what's up there, what's the news, what's the latest product, what's the latest pricing and everything. And the way they're doing it today is basically mostly through web search and it's basically Google. So what these guys have been doing is that they've been indexing the web. So they spent the past couple of years building web crawers that are crawling the internet um and indexing as much as many pages as they can. So when the model needs to have access to the internet, it basically does a web query web an index query to that index to find the right pages. So imagine the model pinging a gigantic databases and gets 10 brewings like Google give you when you're doing a search and the model is going to go through these 10 or 20 budings read them all super super fast and aggregate that info and give it back to you. Uh so that's basically the way it is working today which is again amazing. Think about where we were three three years ago. Uh, but it's pretty dumb as well. It's like reading something trying to make sense out of it. And also web pages are not at all made for uh for models. They they're not made for agents. They've been made for humans. So it's like >> UI, ads, pictures, buttons, everything that we've been building to make the best human interface possible for the past 20 years. but it hasn't been at all built for this uh for these agents. Um so when you're looking at like for example let's say a very concrete example I'm looking to book a flight which is like the most typical use case everybody's been using in the agentic world for the about three years. I'm need to go to New York next week. Please find me the best options. you know, we're spending so much time on that that we we hope soon it will be completely solved by an agent. Um, today the way a model will be do will do it will just again like I said a few websites and try to get some links and giving you links in the conversation. So then you go to the website. I believe that a more AI native world is emerging where first everything will happen within the chat interface or within the agent interface whether it's a chat or not will see and that the redirection to the links of the website are just something of the past and that the model could potentially directly be connected to the backend system of booking.com to the backend system of expedia to the back end system of Delta Alliance and find the best option for me and then offer me a few options and I will book directly within the agent and it will be just done. Obviously, well, we'll see how the payment stuff is is is done and so on. there's there's a bunch of problem to be solved still but the experience I think will be more AI native than it is today by just like hey I found a few links now you can go to these links and try and try the book uh right now so this is exactly why I believe AI native protocol like MCP can really change the internet because now we have an AI native protocol that can connect backend system directly with L&M It's AI native. They speak the same language and they can go way further in the agentic revolution uh than just sending a few links so you can visit websites. >> That makes sense. Uh for MCPS, I mean that's it's it's model context protocol. So my understanding is it's designed for the models to contextualize what any particular um software or tool or or or app is essentially doing. I'm curious though because technically for the lay man the API was supposed to do that right so you had the front end which was the UIUX focused very human uh uh uh way of interactions and then you had the APIs the software layer which could essentially do all of the same things um but you did not need like beautiful visuals for it was all sort of you could code it essentially >> now with with agents coming in and and really championing coding in the start with cloud code and now every other agent doing that. Why do we need MCP when APIs already exist? It's it's a really good question. Uh and uh is actually I think when MCP came out it was it was originally at the beginning not very well understood and kind of people were using it for everything and anything. There were like people saying like well now API is dead I do everything MCP like it's it's over. Then there's been a phase where like MCP was dead because it was so bad for so many reasons. uh and now we kind of going through a little bit of maturation where we start to understand okay what is it for when is it actually useful when it is not useful so I try to to make it clear u APIs have been designed mostly to present an exhaustive view of what the back end can do. you present a list of capabilities that is pretty extensive um that's like a catalog of all the things you can do. It's like a big menu, right? >> And then based on that big backend catalog, you build something that is predictable with UI that's going to call one or a bunch of APIs at the same time to go to the next step of the process. Um, the thing with MCP is that we call it MCP servers and there's the MCP clients, but at our peak, we really see MCP as a new front end. It's a front end for agents. >> Okay? >> The way you're designing an MCP server is to be the the interface that you're going to expose to the agent, but the agent really behave in a way like a human. Well, if you give him a massive API with a massive documentation, every time is coming, he will need to discover and understand how all these APIs are working, which will be a big mess. Even though they are really smart at coding with MCP, you're designing them with an intent in mind. You're not designing them to be exhaustive. you for example take going back to the flight search use case that I took earlier when you're designing an MCP server you're designing an a a tool which going to be flight search that might be coding six APIs in the background to gather all the right information and send back to the user pictures uh price uh flights airport so any information that might be coming from like a pretty hairy and complicated API stack on the background and the model it would be way more efficient to do it that way than having the model just crawling all the API existing and trying to figure out and make sense out of it every time is coming back to it. Mhm. >> Um, so there's that's very important point. There's the there's a really nice discoverability aspect as well with MCP where the model can discover MCP uh and and learn directly when they use it. But there's also some bad things about MCP. That's the reason some people called it uh called it dead a few months ago. um one of them being the management of the context window because with MCP you tend to send a lot of context back to the model and if you start connecting a zillion MCP servers it's going your model is going to be completely lost. So that's one of the big problem but that but that's an engineering problem right you can it's like when you're going on your browser you're not connected to all the websites in the world right you you you're going through some search uh iteration until you actually connect to one website so this is something that's uh getting uh slowly but surely solved and actually the the fact that models are really good at coding like you said earlier uh there's something that's been implemented by Antropic and Cloudfare which is called code mode which is a way of actually using code to call the MCP instead of using the entirety of the of the context and and the uh uh and the protocol communication. So there's a lot of innovation still. It's definitely not perfect. uh there's this I think there was a lot of um iteration needed to be where we are right now and there's going to be a lot more coming but what's amazing is the community that's been created around it that are trying to push push it forward improve it uh from all the big companies in the world uh and also smaller companies like us that are contributing quite a bit >> no that makes sense uh I want to go deeper into the the technical side of things. You mentioned the server and the the client side and how do you sort of how does that sort of end up happening? But um just as a last question on on really the way that you imagine the future to look like and and the way that I sort of talk about that is there's this new sort of very exciting agent that's come out. It's called OI. it uh it sort of automates your family's um you know very complicated systems. So for example, I uh I have a 5-year-old son. I have a 1-year-old daughter. We're a family of four. And increasingly, particularly for middle-aged families, young and middle-aged families, your family life looks like a nightmare. There are football matches. There are tons of utility bills that you have to pay. There are tons of these uh classes that kids have to go to. You have a social calendar. Um, you know, the wife has the calendar, you have a calendar, kids have their own schedules. So, what this agent does is it manages everything for you. It's your um your fifth member if you have a family of four and it keeps a track of everything. It's very exciting the way that they sort of created it. The launch video was interesting as well. Um, and this was a very sort of direct to consumer, very sort of daily life use case, right? It wasn't because a lot of these use cases that were coming in originally were very B2B, very commercial, very enterprise. >> Yeah, of course. >> And what I the moment it came out, my thought was how does the agent have access to a bunch of these tools and services? Because it's basically saying that is going to manage your entire life. When you're actually beginning to manage families, the horizon is really wide, right? Because you're connected to the kids school as well. You're connected to a billion apps. They're taking drum lessons online. They're connected to that as well. so on and so forth. I'm curious and and and the first thing that I that I imagined was if this is what's happening where I'm just I have this personal assistant in my life now. Um and maybe I have a bunch of personal assistants. I have a work personal assistant and a family personal assistant and a friend's best personal assistant. Then maybe that's the person that I'm talking to. I'm no longer going to Google figuring a bunch of these things out. I've essentially automated that process and I'm just now directly everybody is that executive billionaire, you know, back in the day because that executive billionaire wasn't sitting down and googling. They had an assistant. They'd be like, you know, listen, figure out a good restaurant and figure out a time for that and based on what you know about me, make sure you don't screw it up, right? So, so now you have these assistants doing it. Um and and and the whole user experience of going to Google, figuring something out, read a billion blogs, then actually uh finding out what you have to do, coordinating with that particular organization or whatever, and then finally managing it. I think that's the future that I'm beginning to see now where the entire interaction of how we how we sort of interact with technology is going to change maybe 10 years from now. A Would you agree there? Um because that's a foundational shift. That basically means a lot of the top companies like Google and Meta and so on and so forth, the way that their consumers work, the advertising revenue, so on and so forth, a lot of that will fundamentally change. That's trillions of dollars of value evolving into something entirely new. And secondly, how does then MCP as a technology come in play as this experience is sort of being built? >> First of all, I need to check out your early assistant because I also have two kids, one five, one one years old. So, uh my life is getting wild right now. So, check it out. And I actually did some open close stuff for myself uh which is which is kind of fun to uh to to set up. Um listen I believe website are dead. I think the whole uh headless movements that started recently around the SAS world uh it's like headless are SAS are going headless. So it's you don't go to the dashboard and the web and the web dashboard anymore. I think this is just getting started. I think it started with SAS application dashboarding which are awful and a pain for all of us to learn and use to the entire internet where basically my two kids again I don't think they will ever go to a website like when they're grown up. I think they just ask whatever agent they're using. Uh I still I am unclear whether it's going to be one super agent or like a thousand agents. Um it's probably going to be a mix. My vision is that we always end up in a couple of platforms. So back in the back in the 80s and the 90s we ended up in Windows and Apple. Then we ended up in like Google and then we ended up in Chrome and perhaps Firefox and IE. and then we ended up in Android and uh uh and iOS. So this is why I tend to believe that we're going to end up in a couple of platforms as well. Uh and that's the the whole platforming aspect will probably would be something like Taj and CL and perhaps one or two others depending on the regions of the world. Um, so that would go more into like a one agents that rule them all. And this world where you have kind of your super crazy smart assistant that can then be connected to the entire your entire life for the entire system and do stuff for you. I I I believe it 100%. Now the timing I'm not sure. we start to see it emerging which is super exciting but I still think we need to fix and figure out a bunch of stuff. So now going back to how does it work because yeah the vision I think we can call all agree but how it actually happened technically behind the scene >> um I think we're going to have a mix for a very long time of what's available right now which is websites so broading agent web search all of that will be part of the mix because well your kids college or kids high school might not have an MCP anytime soon. So the model will will need to figure out a way to have access to this information and at the same time we'll see emerging AI native interfaces that are to be are going to be directly plugged in a more like streamlined way and efficient way less less token intensive as well. Um, and these two are going to be living together for a little bit of time until kind of an a agent first internet or agent first interface are fully taking over. Um, and that's that's that's really exciting because that's exactly why we we're building this company. Um, and uh there's still like so many problem to solve. So yeah, I mean I I this is exactly the way I see it and I think it's going to be so exciting and I can't wait to see my kids making fun of me when I'll tell them you know back in the days I was typing on Google through 10 website and then I was doing another spreadsheet when I putting all the links so I can share it with my friend and they can give me feedback and then we were actually booking the RBNB like that like they were like oh my god why like it that sounds like the middle ages. >> No, that makes sense. Um, in terms of the tech the the the technical aspects you mentioned, you know, MCP server and then in clients and and so I'm curious about if I correlate this let's say with the early stages of the internet when for the vast majority of the consumer WordPress was really what championed their entrance onto the onto the digital layer or the internet layer. And again for WordPress itself, you had to a host it somewhere. Um, and then obviously you had to ensure that it was properly exposed and people could could discover it. We're going to talk about discovery discovery in just a little bit, but I just want to get a sense of how does MCP work? Like if I have an app or I have a website or I have a small business, what do I like what do I do? How do do I do I uh build an app on top of my website and I make sure that there is some sort of whenever someone comes in there is a way in for the agent to do its agentric things uh as an MCP server. Um what how does that sort of work? give me a technical sense of and and by technical I mean um what all infrastructure is required to make it work for the average user for the small and medium enterprise for the non-technical pizza shop that just wants to be available on the internet to to agents. Yeah, I mean it's it's a really good questions like okay the everything that you just said sounds cool now how do I do it and that's usually the discussion I have with my customers like okay how do we make it happen now um so first going back to the client server thing MCP is is is is an protocol built on top of HTTP so at the end of the day you have kind of the same logic where you have a client to server communication that goes over HTTP um so think clouds. I think a good comparison is like your Chrome browser is now TGPT and your web server is now an MCP server. It's kind of a to make it like kind of a simple. Um the the MCP server itself, it's basically uh transformation layer between your back end and the model. So again, I said that earlier, you need to design it with an intent. You need to design it with like what exactly you want to put it in there. Uh but you also might need some off the shelf things like you said like WordPress, you just like have like a few blocks at the beginning and you were making your website overnight, which was absolutely amazing. Um so with the with with that in mind um the the things you need to you need to do it really depends on who you are. So first going from like the most technical person there's like open source framework out there including the one that we built at Alpig called Skybridge which is an open source framework to build this MCP application. uh and you can use your coding agents and use we have a skill uh and you can ask him like well you know what here's my API key for my back end I want to build an MCP app that does XYZ just do it for me and then you can iterate on that you can play with it make it prett changing the flow kind of uh as you were bu as you would build a front end or or a mobile app. Uh once that done, you need a place to host it because again it's a web server. So uh you can use one of the your favorite tool that you're already using to deploy your application or you can use some uh new technology like ours which is basically a cloud uh platform where you can deploy and host uh your uh MCP uh servers and your MCP application. And we're providing a whole bunch of tooling that are going to be very very key for this future development like a playground analytics understanding the user intent helping you being discoverable which we'll talk about discoverable later. So that's that's the what we're building our RP the open source framework so you can build your own custom MCP and the hosting uh platform. But if you're small and medium enterprises, there's people that built on top of this framework as well some really cool kind of offtheshelf uh tooling that you can just use and say, you know what, here's my API key. Just build me like a offtheshelf super simple carousel like map map like type of tooling. Uh and you can get going with with this type of of tool. uh and the team at nud seal n the seal have been doing like an amazing job there there's other players as well out there and uh and it's super exciting because that's that bring that technicality and because technicality is going down so much is absolutely incredible it's like you can >> an average person can build something very complex very easily uh and I think it's so amazing that it brings back the power to the people with the ideas uh instead of only giving it to the people that have the technicality and the knowhow. >> Makes sense. Um let's talk about discoverability now. And the reason why I think it's extremely important is because I feel like the current mainstream um or or or rather mainstreaming of MCP as a technology >> Yeah. >> is stuck between a chicken and an egg. And it's it's for a vast majority of people the only incentive they have for enabling MCP onto their websites onto their businesses I would say is is discoverability. The reason why Google worked the only exclusive use case that early stage internet had for a vast majority of businesses and people was the fact that that is that became the initial touch point of the world. And so you could Google pizza and it would tell you, you know, what are the closest pizza shops, what are people using? And so the aspect of discoverability was what led to, you know, millions and hundreds of millions of of of businesses and websites being created. The way the reason why I say it's a chicken and egg right now is because barring the whole technical AI first sort of community, the vast majority of people do not really care about what's going on. And in many ways, a lot of these AI tools and technologies end up being a novelty. It's like a nice to have because just because you can do it and for technical people it's like bragging rights. Oh, you know, I have a MCP ready system built that an agent at some point can use it. But >> how is the agent even finding it? Do I have to go to my customers and say stop using Google now start using this particular technology? That's complicated and and and costly for me to change my users behavior. I'm a pizza shop. I don't want to change that technological experiences. Right. So tell me a bit about how the discoverability in this new world um will potentially look like um and and do you expect because right now even with Noodle Seed for example you know they're building these sort of apps on on chat GPT and Gemini and so on it's still by and large this idea that you can have your app but the only way people are going going to be using it is because is is you're going to go to your pre-existing distribution and you're going to say okay now use our app directly unless there is a system where when I talk to an agent uh let's say when I talk to chat JPD it immediately prioritizes it looks out for a bunch of different uh uh MCP servers it takes data from there and in an open world environment and therefore now I have ranking to do where I say Okay, my MCP should come in before their MCP, you know, so on and so forth. So, tell me a bit about discoverability. Tell me a bit about uh competition within the space in terms of multiple people having their MCPs. And then lastly, tie it up with how do you expect that critical mass to come in? Because I feel like the moment discoverability is unlocked, this industry is going to be on fire uh just because of the if nothing else just FOMO uh that would be there. Yes. >> Well, I think we are right now at the equivalent of the internet pre Google. Meaning that before Google, the only way to find a website was to have billboards in the street saying go to ww something.com. >> And that was that was the only way. It's like people were making ads in papers. for uh internet businesses and you were making hides in the streets because that was the distribution. Those are how you finding people and then Google arrived and this whole thing completely exploded. I mean Google and Yahoo and a few others uh created that search and it completely blown blown blew out. So I think this is where we at right now. It's like exactly what you said. Why should I even care building an MCP server if no one even knows that it exists? And we're still very much in that chicken and egg situation. I would say except for B2B businesses because for B2B businesses, SAS, HR system, CRM, ERP, whatever you name it, all the SAS in the world, this has already started because their users are spending most of their time in cloud and GPT every day or cloud code or codeex or whatever they're using and they want to use their tools through agents. So it's not only cool to have an MCP server. I think it's a question of being alive a year or two from now uh if you don't do it and it can be MCP CLI whatever you name it APIs but no one is going to use your dashboard in a few months from now. So this is super encouraging because that kind of show us what's already happening super super fast in the B2B world which where we have a ton of customers already as well and that the big wave of B2C hasn't really took off right now and the reason is exactly what you said is that the platforms where most of the users are meaning cloud Gemini JPT haven't unlocked except for cloud we talk later. Dynamic discovery. Here we call it dynamic discovery. Say, "Hey, I'm looking for pizza. Here are like three ways of buying pizza. The only one who actually now unlocking that a little bit is entropic with cloud. In cloud, I actually made the test just this afternoon." So, it's funny. I was doing a demo with a customer. It's like, "Hey, I'm looking for a pizza. Deliver it to my office." And it literally gave me three connector options say, "Hey, you can use Uber Eats, you can use Global, which is a European competitor, and you can use another one." And what's amazing is that he gave me that as a first answer without doing any web search. He didn't go to the internet. He didn't try to find things on the website. He just his first answer is like, "Hey, I just looked into my connector registry. Here are three options to do it." And if I say no, then it's going to go to the internet and try to find options. But if I say yes, boom, it's going to pop up an MCP connection and just give me options to deliver that pizza to my house. And which is interesting because we see that one of the leading model provider is prioritizing AI answer AI native answer over web search and that's the first time we see that in the history of MCP. It's like first I'm looking to what kind of connection I have to I can go and try to find things on the internet. Um it's super exciting. We actually have one or four first customer called kiwi.com. uh they are like uh they were the first online travel agency uh to be available in cloud and to be honest when we talked to our customers say well why don't we put it on code but that was eight months ago and nobody was using code but for coding at the time so it was just a bet and since the dynamic discovery uh has been activated so the behavior I just described before that traffic has been literally exploded. Meaning that people are discovering it a lot. They're connecting it and they're using it at a scale we could not even imagine just three months ago. So I think this is a good good indication of the wave that's coming. I think it's a the discoverability is a it's a hard problem to solve. Uh I don't think I think it's it's harder than Google. I mean again all comparison part the Google algorithm is was just absolutely bananas 20 20 years ago but the user experience is harder >> like the the search thing we we had I mean it was like links and and the list. Now, how do you present that to people that are used to have a chat answer and now you can have like potentially connectors and you can have potentially ads because obviously if you have discoverability you might have ad in the mix. How do you make sense of all of that in a chat interface? I think the UI UX is very very complicated. Uh, and that's not the only one, but one of the reasons I think some of the chatbots are afraid of it. Um, and they are going slower than than we could have anticipated. For example, Chad GPT came out with their app store like eight months ago and they still do very little uh dynamic discovery. So, I mean, I think it's a very very incredible topic. How do you make it happen? What's the UI? Is the UI going to be generated on the fly? Is the UI going to be controlled by the brand? Can you mix a bunch of connectors to have just one UI? Uh how does the brand appear if you start mixing brands in an answer? Uh how do you mix chat and widgets because the UI won't disappear completely? There's so many I mean questions that are still up in the air which are really wide open for like startups and innovators to to take over. >> No, that makes sense. And I think while we we we we do say that you know there's going to be a one and one a couple of major players who going to win eventually like sort of Google but I think more more than those couple of winners I think there whoever wins is going to enable an ecosystem that is going to enable millions of new use cases new opportunities new problems to solve. And so, you know, people are so stuck on one winner takes all mentality. They're failing to realize that the the the winner is really just building the foundational layer. What comes on top and not and later is where the real value is. But, um, I'm going to bring this through a wrap-up as well, but I'm I'm just sort of curious. You mentioned dynamic discovery in terms of you know if I'd ask a certain question it is going to go into its existing database of existing MCP servers and it's going to recommend something and based on that I can sort of click on it and then I can start a new user experience where now I'm within the app and I'm sort of using it as a as a widget within the sort of chat interface. Now that principally works with Chad GBT was the same thing. You know they came out with this with their app store and the first 100 apps were easy. You know it was Spotify, it was Uber, it was this or that. Um but then what happens when there are a million apps in the same category? >> Yes. >> How do you rank them um in terms of what appears first? The the advertising play is the simplest I think right now where just pay us money and we'll put you uh we'll make you appear. If people are interested, they'll they'll go for it. It's it's the more organic discovery which is the more complicated part where again I think pizza shop is still easy. It's just taking your location and then potentially just looking at some reviews and saying, "Okay, some there's a middle ground. Let's rank it accordingly and and you can optimize it for every person." But for a vast majority of other use cases that are not geography dependent, how do you sort of ensure that the two three options because Google could give you 10 options and then you could say next next and you could actually go to 100 options in this user experience when you only have the opportunity to rank like two or three maximum options. >> Yeah. >> Is there is is is the has the ranking been solved or at least an early signs of how ranking might go? um in that discovery phase once it's no longer just the brand name businesses when when for a use case you have to completely come across some some business that you have no idea exists and then you have to start interacting with it. >> Yeah, I mean I I I I fully agree. There's like 8 million apps in the iOS app store and there's like a zillion websites. So the problem is coming. There's now about 1,500 uh application in JPT. Uh there were like only I think 100 in February. So it's it's it's growing really fast and the problem is kind of almost already there. I would say the the discoverability piece would probably the early signs of that we've been looking deeply at everything that entropic and GPT has been implementing. Uh there's a couple of early signs. So for example, by scrap scraping the uh entropic API, we've seen a few APIs appearing a few weeks ago. uh one is actually rank. So what a rank is well they've been ranking all the application one way or another. We don't know exactly how but there's there's one API that return to the model rank. Um another one is trending. There's like a feedback coming back with trending which is what we've been exploring is the m over the past seven days how many hits how many new connections you had from users. Um, another one that we've seen emerging in JPT is that you have this u little little sums up sums down that is appearing just nearby the widget which can be also a way of proactively proposing applications because they are loved or less loved by others. So that's kind of early signs of what they're trying what they iterating which is very basic right if you compare that with the Google algorithm that's that's very very basic but this is exactly this is where we at um then you can imagine I think what's interesting is that these models know us even more than Google which is crazy right because Google used to say like Google and Facebook they know everything about me. Uh, it's crazy. But now these guys, they know really everything about us. We're we're we're talking about our personal life. We're talking about our friends. We're talking about work. We're talking about the new products we want to buy. We're talking about our kids going to school and stuff. They know literally everything. So, I think they have a gold mine of using in a smart way. Obviously, because if they do it without any privacy, it's going to get wild. But if they do it in a smart way, they can use some of that data to really go and pick the right product and service uh for you. That being said, I think there's going to be uh it's going to be harder to be on the top because I don't think the model will bother going through 25 or 30 or showing showing you 25 or 30 options. I think like you said it might be showing three five but I don't think it's going to be like 15 on the page and then like a 100 page that nobody never really go to. Um so I think there's a prime to being there first because you're going to learn more and if you're good you're going to pick up. So I think that's why there's some for more as well from the industry because there I I definitely think there will be a prime to be there first. Um and I think there will be a prime to really understanding deeply this technology and iterating on it. For example, one one thing that uses as well uh entropic and JPT is the the the description of the tool. So being super crisp on the way you describing uh your uh your MCP server might be a new SEO type of uh thing. Um, so yeah, again I think the discovery bot is is is just getting started. We're just scrapping scratching the surface. Uh, and and the people that crack it early when it becomes big are going to win a lot. >> That makes sense. One last question in in terms of this. Look, a lot of the way that we imagine these sort of MCPS particularly with a UI widget of sorts um is is through a lot of these major chat bots. So chat GBT claude um Gemini we're saying you are already interacting with them and as you interact with them you're going to discover let's say an app and then you're going to interact with that app and you need a widget for that. The way that that interaction actually is evolving is for specific use cases as I mentioned earlier. For example, OI exists for managing your family affairs. There are legal agents. There are, you know, a bunch of different specific use case agents that you're interacting everywhere. You're interacting with them on messages. You're interacting with them on WhatsApp. You're interacting with them wherever you can essentially, you know, send or receive a message. And that sort of again convulates the entire problem because now the interaction layer is no longer just text and uh and an interface around let's say chat GPT right where widgets are allowed some of these agents you're actually interacting with through voice maybe you're calling them maybe you're just saying okay you know um so so it's very difficult to create a unified user experience where you can throw out widgets and this and that and in many ways you you're still with widgets you're still saying there's at the end of the day a human at the end of the loop right so the human is going in and and and they're making that decision. You're giving it options, but the human will be the one who has the agency to decide what to do. Increasingly, we're now going towards a point where the human is just talking to the to that agent and the agent is interacting with the MCP for for all intents and purposes. Do you think we're going to begin to see third-party discovery apps in the first place um that will be agnostic to these platforms and can actually feed into um any agent or agentic experience to begin with? Again, what I mean by that is yes, a chat GBT might hit 800 million um users, but right now what's happening with a lot of companies is they're saying, "Okay, we'll deploy your apps on Claude and Gemini and Open AAI." Sure. But then what about Ollie? Does Ollie go to Claude and uses the API and the tokens to then find you? And then how does that technology sort of work? Because I and as a layman, I understand the chat interface. How does the API technology come into play with MCPs as well? Right. So, so how does that sort of is does that make sense to you what I just said? >> It makes sense. I I think that's that's one of the beauty of having a protocol is that hopefully you don't have to build something new for every new interface out there. Uh and it's like the again going back to the early days of the internet. Uh back in the days, your websites aren't really rendering the same way in Firefox and Google Chrome and Internet Explorer. and you have to you had to tweak a little bit your website to to to have the same the same user experience. So hopefully people converge around uh uh uh one or two protocols and there's going to be protocol wars and and then one's going to leave and another one's going to die. Um, regarding the user experience, it's kind of funny because I think I think WhatsApp has probably the highest potential to win there because uh the whole world I mean a big part of the western world is on WhatsApp. Um, and your friends are WhatsApp and more and more your work on WhatsApp. I'm now closing deals on WhatsApp which actually has any sense. uh and um so I think they have they are a huge opportunity to also be able to be a place where this agentic experience happen. >> Funnily enough, we we made a demo a week ago to actually render an MCP app in an iframe inside a WhatsApp chat. So it's a bit clunky and hacky, but we made it work. Um so yeah I I think the at the end of the day there's two things that's going to be important. First is to have an AI native interface >> one or several depending on where the industry is going. to so you don't want to you want to be sure that everyone can find you whether it's like holy clubot uh WhatsApp agent uh chp cloud uh two investing in being discor and making sure that whether you want the chatp ecosystem the credit ecosystem or like an open world ecosystem where any cl agent can potentially find you you find a way to be there and to be the one that these guys are finding and not your competitors. Um, and I think any and again this is nothing new when you're building a mobile app is the same, right? >> It's exactly the same. No one's going to find it you if you don't have like a good discovery process, a good app and you have like something that is compatible with the iOS uh ecosystem. So I think it's just an extension. This is much more uncertain and it's going way faster which is one of the reason our customers are are getting like overwhelmed uh sometimes because they say but what should I do like this is changing every other months there's all this surface should I build for my website should I build for chpt should I build for WhatsApp or should I build for Slack like everyone is we're still figuring out again my conviction is that we probably end up in a couple of platforms That's that's that's more like a it's not really a bet for the companies. It's more like a gut feeling because at the end of the day, if it goes in another direction, we'll we'll just adapt as a startup. >> But my gut feeling is that people usually are lazy and they just end up in two or three places where they'll feel comfortable. Uh and I think this is going to end up that way. Future might be different. We'll see. >> Makes a lot of sense. P Thank you so much for taking the time out. This is super super insightful. I look forward to seeing uh how you sort of build the rails for these inter agentic interactions. Thank you so much for putting the time out. >> Thank you. It was amazing. Great discussion and thanks for having me. >> And for all of you guys, let me know in the comment section below. Have you started building uh your MCP servers and your agent interfacing layers would love to know more about uh that experience. But nonetheless, it was say zedi. You're watching Beyond the Bubble podcast by Noodle Seed Studios. Thank you so much for watching and I'll see you in the next