Ep #146: Redefining Web Access for AI Agents ft. Philippe Mizrahi (CEO, Linkup)

Adil Saleh 0:39 Hey, greetings, everybody. This is a deal again, Hyperengage podcast. And we try to hunt people. We try to hunt products that are, you know, just setting apart from this, this web that everybody tries to, you know, ride these days, building ChatGPT, rappers, they call it, they don't call it rappers, by the way, they call it AI first, AI powered platform that's going to change the world the way it approach. And that's why we try to find something unique. And today we have, you know, Linkup with us. I was the chief executive of Linkup. Phil has done pretty good in the past, I would say, two years, and I see that this is a huge problem. That's why we're sitting here. And we've, we've tried to screen, fill and Linkup to, you know, talk to our audience. That how, you know, building API infrastructure for any individual or any businesses in part against any workflow, be it marketing, be it product, be it developers, be it, you know, sales, be it you know, top leadership, you know, everybody needs to have some sort of, like a one click access to an to an AI for any of their problems. So that's why we, we're talking to Linkup today. Thank you very much, Phil, for taking the time. Philippe Mizrahi 1:56 Yeah. Thank you for inviting me. It's a pleasure. Adil Saleh 2:01 Yeah, pleasure meeting you. Likewise. So Phil, now thinking about you three folks, I know that it's a small team, but you know, it's not a small enough problem. I know that you guys have background that is kind of giving you a domain experience on the engineering side of things. So how did this inception of Linkup, that is not too long ago, came together, and how do you guys gel in together? And, you know, actually bought into this vision. Tell us more about about the inception of Linkup? Philippe Mizrahi 2:29 Yes, sure. So, I mean, maybe a bit of like overall context about Linkup. As you said, Everybody's building at the application layer of AI. So everybody's building AI applications. So it's pretty much always the same thing, like you take an LLM, maybe it's the open AI one, maybe it's like the anthropic one, maybe it's the mistral one, because you're French and and then you connect that LLM to some databases, some intel databases. But what people realize very quickly is that if you want to have high quality outputs, if you want your system to work properly, if you want a system to be aware of what's going on in the world and be able to output stuff based on that, you want to connect your LLM in, your AI system to the internet, and to do that, pretty much what we saw people do before Linkup came is that people would build some sort of a janky pipeline where, like, they would take a prompt out of this prompt. They would, you know, like, create maybe 567, different Google API queries, feed that out to Google, or like a SERP API. Then they would take, for each of these queries, the 10 first results, they would go on each of these results, and they would scrape the content of the web pages. Then for like the smart one, or like the technical ones, they would chunk all these different contacts pages. They would break this down, do some notification and then feed that back to the LLM, and so the LLM would be aware of us online, what's on the internet to be able to generate high quality outputs. And back to 2024, last year, when boys, Denny and I started working on Linkup, we're seeing all these people building at the application. Application, they are doing always the same exact thing. And it was always janky, it was always slow, it was always expensive, but they were still doing this, because you have to do that to have a properly functioning AI application. You have to have this application connected to the outside world. And so we're like, wow, like we have to do something better. And we did that in one week. Actually, in one week, we're able to build something. But it was, like, already at par with what people were doing, um and and then, like, quickly we iterated, and we're like, increasing quality, decreasing the latency, decreasing the cost, um. And fast forward to February this year, we're able to launch a version of a system that was ranking number one on the main factuality benchmarks. So this means that Linkup today is the world's best search model for AIS in terms of factualities. So factual, it is pretty much like, are we outputting something that is correct, that is true, that you can trust some ground data, that you can be you can you can use to, like, build your application on top of you can trust to make business decisions, you know, like, to make investment decisions. And so. And so on. And so this is what we built. We started that around a year ago, and we've done that with my two co founders, Boris and Denis. We both, like all of us, are engineers. Boris went through McKinsey, so he's sort of like business guy in the team. And Denis the CTO, he has some interestingly, so he was working at Spotify before, and interestingly, he was the founding engineer of a search startup in the audio space. So he has some experience in search, but in the audio space, which is somewhat different, but some communities as well. And on my end, I have, like, a background in math and AI and I worked as a product manager at Lyft in the US, where I worked on two things, automost vehicles and actually a search engine for installed data. So this is kind of like what brought us to like AI and search. We wanted to build something that I was at the intersection of AI and search, and we wanted to build something that, you know, was sort of like building the one of the core building blocks of the future of the Internet, of the future of the agentic web, as we Call it, because we we feel like essentially, the internet is going to change a lot. It was built for humans. It was, you know, like a web page is something that essentially is for humans, right? And as we have more and more AI agents that are built and that are going to become the one, number one user of internet, this is going to be super important, like, a pivotal moment in the industry, and Linkup and Boris and Danny and I, we wanted to be part of that. We want to be part of, like, building this core infrastructure of this new generation of technology that is going to be, you know, so interesting and so defining in the next, like, 10 years of industry. Adil Saleh 7:42 Okay, so I'm glad that you've, you know, delved in deep with, like, the root of this, this problem, and how humans are going to interact with AI, and how we can make it accessible. Then they try to explore and look for different products, or maybe, if they're a user of existing, let's say, a CRM HubSpot, they are looking for, for any specialization within that, but outside that, it's still a big, big, big problem. So if you, if I look at it for as as a two sides of coin, some from a user perspective, from a customer perspective, like platform or tool build on AI, very much specialized for some of the use cases. Let's say for my use cases, I'm a DTM leader. Let's say I want to, you know, automate a lot of my sales and marketing processes and intent based outbound, if I might use, like, some of the platforms that are pretty much proven, and, you know, a lot of companies are doing that as a product versus if I go ahead and I, you know, I you know, I see I already know my use case. I type in my use cases, I give the some of the contextual information, as you mentioned, like fact based information, to a web platform or search platform that can custom made an API for me that I can integrate for that workflow to be streamlined. So how the outcome has been changing, and how do you see it as a business use case, more of than an than individuals. I know that you guys have raised capital, which is quite congratulations on that, too. Pre seed is pretty healthy. Pre seed, but thinking of some business use cases, some of the, some of the you know, B to B marketing teams or sales teams or product teams leveraging these APIs and getting specialized or maybe tailor made outcomes that are proven and delivering results. I've been some of you. I've been seeing your some of your customers success stories as well. But how do you see it in a longer term view? Philippe Mizrahi 9:39 Yeah, so we're very much like building for the world of agents. We actually sell to agents. Right now, we still have to go through a moment where a human is getting a credit card and paying something in like through stripe or whatever, but in the future, we won't even need that. And we actually did a partnership with a company called Tobit in the US that is pretty much like enabling AI agents to not even require human intervention to use the linked system. And so AI agent in the future, if they feel like they need some access to high quality Internet data, they could go and find us through company, like Linkup or through our own system, and start using us to find some high quality content. And so that's what we're building towards. The truth is like right now, the definition of AI agent is kind of blurry, and nobody knows really what an AI agent is, but what we're seeing is a shift from two. Additional software engineering, where everything is deterministic to fully autonomous agents. And you know, in this line, we're somewhere in between right now where, like, you have some use cases that like pretty deterministic, some use cases that like more, like aI workflows, and we're trending towards AI agents. And the way we think about our API is that it's very much a tool that AI agents can use. They can decide when to use it. They can decide how to use it. We are going to be, you know, like in the toolbox of a lot of AI agents, and in the toolbox can have, like, a lot of different other things, you know, maybe you're going to have access to very special data sets. Maybe you're going to have access to, like, a math API that you're going to be able to, like, compute stuff very efficiently. Maybe you're going to have access to all sorts of APIs and systems, and in the future, all these agents, you know, they will collaborate. And so you'll have agents working with agents to build something together, um, you'll have agents using tools to be more independent and more powerful. And this is going to be like a whole new world, a new economy that is going to be built. And you have like protocols that are going to help build this economy. You have MCP, obviously that is like booming right now. That is this product called, called, it's essentially like an entropy build protocol that is helping agents use tools in a very standardized way. So it's very much it's way easier for agents to have access to different tools and use them. You have Google that has built a to a and a to a is protocol that helps agents collaborate. And it's interesting because, you know, like you can either make agents independently smarter and smarter, which people have been doing so far, but you can also help them work together. And you know, it's very It reminds me of a society like the human society where, like you and I, obviously you can walk to be smarter independently, but you can also decide to collaborate and together, we can do stuff that is super cool without having to, like, take care of everything ourselves. And so, you know, this is very much like this agent economy and this agent ecosystem that is going to, you know, like, form itself with time in the next few years, and that we want to be part of, and and so to your point, about like, specials API, about like all these different services, like, we really think about The future setting, about selling to agents. We want to design something that is like, extremely well designed for agents. And so you have competitors of ours that are like a bit different, a bit more structured. You have companies that do stuff like we do, but really they build that for humans, actually, and we can go back to, like, the competitive landscape and so on. But like, I think it's important to understand in gap, you have to understand that, like we're building for agents. Adil Saleh 13:46 Okay, I'm glad you, you know you already brought to some of the conversation I wanted to have going forward. It says that how you want to have a some sort of marketing position now that you guys are funded, you want fast track, because I know that pace is really important to beat the competition, and if you have the capital, that actually elevates that. So how do you see this product in terms of from the commercial side of things, from the market positioning, and you know, the business end of the economical end of of you three, you know, might be sitting down and thinking down round, because a lot of this is changing and pivoting really, really fast. I know that it's, it's still about, there's one thing that's been common is still about getting, acquiring some new big logos, let's say 1015, big logos. And then that becomes easier to, you know, go top to bottom and and go from mid market, you know, SMBs, and then you you create volume, but at the end of the day, two years, three years down, you still have like, 15 to 20% you know, of your customer base, you know, giving getting you the 90% of your revenue, top end revenue. So thinking about this from the from the commercial side of things, and acquiring these logos and setting the right positioning. I know that the messaging has been pretty unique. You founders are pretty much yelling. So do you see it as a challenge at this moment? Philippe Mizrahi 15:10 So I think in terms of goods market, our strategy is very simple. For the first few months of Linkup, we didn't have a strategy. We're mostly operating like a like a research lab, like trying to get the best product. We didn't care so much about like selling the product, but we were working with people that you know were giving us feedback on how well the product was, was was a behaving and was it helpful? Was it you. Full would people would be willing to pay for it. So that was, like, the first stage of a company. And then as we moved on to the next stage, we're kind of like trying to form an opinion of, like, what is the best way to bring this product to market? And the first customers, the first paid customer that link have had, is someone who found us on Reddit. And I think it was interesting to us, because it's someone who we never talked to, who lived on the other side of the planet, and who very quickly became, like, a, you know, a huge advocate for Linkup, because they love the product. They integrated that in the application and and so that kind of, like, brought along product led growth strategy where, like, what we did was sell to the developer ecosystem. We, pretty much like, made a lot of effort to explain our product to engineers, make it as easy as possible to use. You know, like, have great documentation and build connectors to all the different frameworks that were already in the market. So like a Lang chain connector or lemma index, we are obviously, you know, like build stuff with hugging face with crew AI all the like developer, you know, like orchestration systems, but also like the low code, like no code systems, like make n, a n, stuff like that, to make it so easy to use that people just like, you know, like, learn about it on Reddit, maybe, like, play around with it a bit, and then, like, come on Monday to the company and be like, okay, guys like, you know what? I found something this weekend. This weekend were actually solved the issue in working on for the past four weeks and and that has been going on like pretty well so far. And it's interesting, because if you build a brand in the developer ecosystem, then actually that brand is very strong basis on top of which you can go up market. And so what we're seeing now is like, we have inbound for big corporates from enterprises that want to work with us, because we've built this brand of We are the trusted, very serious, very like, like, stable and and correct a source of information on top of which can build and make decisions that impact your business. And on top of that, it's fast, pretty fast, very high quality, number one on charts, and at a very affordable price, actually. And so if you do that, and do that in a way that like, so easy to use that like any developer can just like, see the value very quickly, then like, even the big corporations will, like, the this will infuse up to the top, and this is where you're seeing right now, like we're seeing, like the bigger companies in the US, in Europe and the rest of the world as well, like to reach out, reaching out to us, and starting to use the products. Adil Saleh 18:46 Interesting. And you're already like, of course, it's initially, it was more product led, and then it was, you know, it's more about, you know, making sure that you do the right marketing. You You know, you stay in the same same social media network, or omnipresent and same network where your customers are sharing time. So you mentioned a lot about about developer teams and all of that. So how do you see other segments? I know you some of the customer stories are for marketing and prospecting as well that I deeply looked at. So I know that it's use case of different adjacent markets, and you have competitions products that are tapping into those as well. But how do you, as a chief executive, how do you see it tapping out for other customer segments? And what is your biggest customer segment at this time? Philippe Mizrahi 19:33 So we, our API, is journalists, we can answer a question about pretty much anything, and we're like, number one on the factuality on journalist benchmarks. So it means like, you want to use, you want to like, a factual answer to anything you can use, Linkup. But where we have a huge spike is on anything related to, like, corporate information, business data, market data, anything in the corporate world is where we're, like, extremely good at and so I see it as, like, we've pretty much, like, trained a McKinsey analyst, and you can ask, like, lifestyle questions to this McKinsey analyst. It's going to great at answering these questions, but where it's going to be especially great is answering questions about your business, about your competitors, about your prospects, about all that stuff. And so obviously this is leading our like strategy, or like marketing strategy, or go to market strategy, etc. Like we see a huge spike in use cases like automated sales, AI go to market qualification of prospects and. Um, you know, like all these different workflows, we also see a huge spike on like business intelligence, analytics, financial intelligence, you know, like we have, like fans making investment decisions based on our data, especially interestingly, like the smaller companies where you have less data available easily than like public companies, for example, but we also have a long tail of different use cases. And, for example, like hiring, it's actually a pretty big one. We have support, we have debt collection, we have a bunch of different, you know, like use cases in the long tail. And you know, like, the fact that we are, like, very strong in one direction, but also, like, pretty, actually strongly and better competition on, like, the rest of the use cases is it enables us to see bumps in use cases. We can see, like, something coming we can see, like, okay, there actually something interesting coming in here. Like, we have abundant in the distribution. We're like, okay, there's maybe something we should invest on. And right now, we've been extremely good at everything corporates, everything business rated and so on. But we can easily, just like, replicate the same playbook in a different vertical. Because, um, what's in the DNA of Linkup is being super factual, like, super trusted, uh, people can count on us to make very serious decisions. But that could be decisions on different topics. Could be, you know, like, right now we're very much corporate, but it could be also like medical decision. It could be a legal decision, etc. Adil Saleh 22:16 Interesting and, of course, facts around like any use case, be it like for for marketing automation, sales automation, prospecting, outbound, anything I mean, you gotta make sure that you have access to the data your platform, your API, that is absolutely factual and it's not opinionated. A lot of these, this wider range AI capabilities and large language models, they need to be prompted very, very contextual, very, very concrete and specific to have these kind of outcomes. So you have that all started sorting pretty much at the back end as an infrastructure, right? Philippe Mizrahi 22:56 Yeah, sorry, I didn’t get the last part of the question. Adil Saleh 22:59 Like you have pretty much done the infrastructure of the back end in the way that it delivers that, that level of, you know, outcomes and efficiency. Because, at the end of the day, it's all about, you know, making it really accurate outcomes, efficient outcomes, be it anywhere like from, you know, developer teams to, you know, you know the sales and marketing and product and you know all sorts of teams. So now thinking about this huge of a problem, like the cost of planning, how you're optimizing the cost on your end for some of your larger customers, I know that it's not it's not easy. So thinking of it as scale, optimizing the cost. How do you see it as a challenge? Philippe Mizrahi 23:42 Yeah, you know, like, this is obviously something that is top of mind for us. We have raised some money, as we said, like, we have raised a pre seed round last, last summer, and so we have money in the bank, but we're not billionaires, and so we have to be very cost efficient. We have to be very smart. And I think this is the case both in our systems. We have to, like, you know, like, think about system design in a very efficient way, and also in the way we operate as a team, right? Because we're pretty lean, we're pretty small team, and we have to be super efficient. We have to, you know, like, decide how we invest our time very carefully. We have to work only on the most extremely urgent problems. And, you know, I have this thing where, like, as we prioritizing projects, we only have so much bandwidth that we can do, like, only the most important projects. And if we have time to do stuff that is not critical, it's almost like, okay, best of that's something that's wrong, right? Because we should not do stuff that is not the most important stuff. And so that's like, how extremely you are in terms of prioritization. And I think this is how we can, like, pretty much beat. We beat, actually, people, like companies, teams that are, like, much bigger than that, than us, and much more funded than us as well. And I think this is, you know, like to what we're discussing earlier, like the only thing we do is an API, like, we're not like publicity, for example, is building the whole suite of products that are like, very well done, very well designed. They have this super nice UI, this super nice app. They have this voice integration. They have all the different things that are like. And I love complexity. Oh, this is a huge inspiration for us. But like, if you think about it this way, like you. Don't have that many more people working on search at perplexity than a bank up, because you have more people at perplexity that are making sure that like, the buttons really work than people working in the search infrastructure. And this is how we were able to like, be super good, right? Like, it's like, because we're super focused. And I think this is how we think about cost, like, we think about prioritization and like, working in the right problems, and only that more than like, Okay, how much is costing, like this table, or like this furniture that I need to buy from my office, you know. Adil Saleh 26:14 Interesting. I was also referring more towards, like, the cost of the API for the llms, and, of course, the calls that you you're going to be processing for, for your customers, especially larger customers for, I was looking at the price. I know that enterprise might price might be different, but if you talk about some of your freemium and you know the plan that is, you know the mid package, how do you optimize cost for like, are you not thinking about going for seat or usage? How do you go about optimizing that as a GTM framework in a longer time? I know it's too early, maybe you guys might be thinking about it, but it may come down on the table, because cost is always increasing when it comes to, you know, using AI and large language models, and they're called. Philippe Mizrahi 27:00 Yeah, yeah, for sure. So I think one thing that makes it even harder right now to predict cost is that, like, costs are moving a ton, especially on the AI side, and so you we've seen, like in the past, like a lot of changes, both in terms of like how intelligence more are, and also like how cheap intelligence becomes. So like a model now is like much more, like, much smarter and more, much more capable than it was a couple years ago. But also like, for the same level of intelligence, it's also much cheaper. And so I think what we need, what we need to do on our side is when we predict costs and we think about operational margin and like, the pricing that we able to like pass through to our customers. We need to think about directionally, what's our hypothesis in terms of how these costs are going to go down, how these capabilities are going to go up. And so we, you know, we need to, like, try to like, continue the line, understand where we think that the industry is going to go, and we need to think about, okay, obviously, right now the highest priority is to bring the top line up, but also do this in a way that gives us, you know, like, healthy and good margin in The long term. And so it's less about optimizing costs for today. It's more optimizing costs for Yes, yes. And doing this in a way, but like, you know, like, we're not wasting our time on stuff that is not going to be relevant tomorrow, right? And so very concretely, it's pretty simple stuff. It's like building system at scale. It's building systems that are not over engineered, having, like, some data layer that is extremely focused on the stuff you need, not building stuff you don't need. And right now, like most of the cost also is, is the team right now. And so you also need to, like, hire the right people and make them work on the right things essentially. Adil Saleh 29:09 Yes, absolutely and being absolutely lean, thinking about, you know, optimizing at scale and at the same time, as you mentioned, making sure that you live in the moment and you work closely with with the customers in hand, and try to do what's high value for you as well as them, in terms of product and development and roadmap, what's what's what makes you excited sitting on the roadmap, on the product side this year, I know that we're just one quarter down, but must be cooking something. Philippe Mizrahi 29:39 To react to what you said. And then I'll share a bit more about like, I think that a lot of people build stuff because that's cool. And obviously it's it's like, what everybody wants to do work on cool stuff. But like, if you want to be lean and cash efficient, you need to work on stuff that is, well, just the stuff that's necessary for your users, and you can cut all the rest. And if you're good at doing that, you'll just work on the most important things, and soon you'll save a ton of money, and that's what you're trying to do as well as we can. To your point on what's cooking. So right now, our product offering is pretty simple. We have a model that is actually two models, a standard one that's the faster option and a deep one that's a more comprehensive option. And so what we're doing now is actually ready to launch a couple more models, one that's currently in private beta, with a select. A few customers that is even deeper model that is going to be extremely good at reasoning and creating lists, because right now, like the world runs on lists, everybody's making list of prospects, of competitors, of everything, and it's not something that is easily done and and so the this model that is going to launch soon is going to be great at doing that. We also going to on the so that's like on very like, high quality, high latency and the spectrum, and all the way on the other side, we're going to launch another model that is extremely high speed, was, like, much faster than anything else we've launched so far, and at a very small cost in quality. We try to, like, keep the quality as high as we can while, like, shipping a model that is extremely fast and can be used in, like, cases where latencies is, like, most one of the most important things. So these are the two main things we we're going to work on and in pilot, too bad. Like, obviously, we, like, spend a lot of energy and a lot of time building the next generation of a model you know, like, this analyst that I told you we trained that is extremely good at, like, looking for the information you need, as well as building this data index that you know, like, is one of the critical sources of information for other queries. And so these are, like the two core pieces of engine that we continuously work on, continuously improve. Index is increasing in size, in creating density and and so. But obviously that takes a lot of energy and time. Adil Saleh 32:20 Yeah, absolutely. You know, it takes a lot of energy and time and and sweat, and I'm glad, I'm glad you guys are three engineers. You know, engineering engineers are always good to have on the team, especially with this evolution in AI and with with all different backgrounds and their thing thinking through things in a different lens. And, you know, who's, who's the leader of the product I mentioned you, you mentioned the CTO. But of course, that is more like customer driven, you know, how do you actually plan the sprints for some of the new features, some of the key decisions you, you're taking on the product? Philippe Mizrahi 32:55 So, you know, like, I think we are very complimentary on the volumes and in the rest of the team as well. Like, everybody's bringing different things to a table, different experience, different skills. But in the way we think about like, planning is very different from like what we do in the big tech companies, you know, like, we don't do yearly planning. We have like, North Star goals. And we think about, you know, like, what are the key projects that we need to put in place to, like, reach out goal. And we have super iterative like, we do this, we have, actually, this monthly period where, like, we plan what we want to achieve in a month. And then we have weekly sprint planning. So, like, very iteration cycle is extremely quick. And when you think about prioritization, it's, you know, like, obviously Boris brings a lot of like, this enterprise, this, like, like business sense, and he's constantly negotiating, like, bigger clients and like, he's negotiating partnerships. So he has a lot of insights to bring to the table when it comes to like, business and product prioritization, I have a pm background so like, I know like, what it means to like, come from the business and and business problem statement, and bring this, formalize this, and bring this into software. And Denis, obviously, he's, like, extremely strong on the technical side. So like, he knows how to build systems at scale and do this in a very efficient way. And so all those of us together, like, we're like, actually working pretty well together to, like, build projects, execute on them, and do this in a way that is quick, that, like, brings us towards our goals. And obviously the rest of the team is, like, super, super, like, involved in that. Like, it's a very small team, you know, like, we're, like, not even 20 people. And so it's very easy to just, like, involve a one. Like, we're very flat. There's no layer of hierarchy, like, there's no management. And so it just means you can always all, like, be around the table and, like, discuss everything and just like, move forward very quickly. Adil Saleh 35:04 Absolutely, because you know bringing the A players first place is super important. And if you if you make the right decisions and bring the right culture from the top line, it follows. The bottom line follows. And you know that impacts you as a culture, and then eventually as a company, and reflects into your product, and then customer service and all of that. I love the way you guys are approaching as pretty lean team, pretty close knit, you know, working together and building something amazing. And, you know, thinking of some product, more than 18. Months having pre seed, a three minute pre product market fit. It's quite a quite amazing job. And I wish you guys good luck, and I keep on looking for for your journey going forward. And I think you're connected on LinkedIn now, Israel have sent you the request. And you know it was pleasure having you on. Philippe Mizrahi 35:56 Yeah. Thank you so much. Thanks for spending the time discussing Linkup with me. I mean, obviously, like, this is a topic that I'm passionate about, and I would spend hours talking to you about the future of internet, about how we think about AI agents and so on. And I mean, honestly, there's so much going on, like, it's one thing that I feel like I'm super lucky to be working on this in this industry, on this type of products. It's not every day that you get to, like, shape the future of internet. And I feel like, humbled and like, super lucky to be, like, actually paid to do that. You know, it's so great. So yeah, but yeah, thanks for inviting me to talk on the podcast and and then stay in touch. Adil Saleh 36:43 Pleasure is entirely mine. Thanks, Phil. Have a good rest of the day. Philippe Mizrahi 36:47 Thank you so much. Take care.