Beyond Bottlenecks: Mathieu Baudet on the Next Era of Blockchain

GM, GM. Welcome to what's going to be another great episode of DevNTell. So if you didn't know, DevNTell is a weekly podcast held right here where we welcome founders, hackers, and anyone in between to come on the show and showcase their product. And today, I'm ecstatic to welcome Linera founder Mathieu Baudet. So Linera, if you didn't know, is the first real-time blockchain infrastructure designed to bring the infinite scalability of Web2 to the cryptographic power of Web3. So if you stick around for today's session, you'll see Mathieu explain the origins of Linera, how it's different, and how it can unlock real-time Web3 apps for the next billion users. Alright, let's get into it.

GM, GM. Welcome to the show, Mathieu. Pleasure to have you on.

Hello, GM Microchains! That's how we say it in Linera.

Gotcha, GM Microchains. Yeah, we're really excited to learn all the great content behind Linera, how it started, why it exists. But before we get into all that great detail, perhaps you'd like to give a brief introduction about yourself if people aren't familiar?

Yeah, for sure. So yeah, on a personal note, I'm French. I grew up in France, studied there, and then I went to California. I had a job at Facebook where I was a software engineer for many years. And then I got into blockchain there. I was one of the first founding engineers of the blockchain project at Facebook. We were like super excited, it was really cool. Did a lot of cool stuff, unfortunately didn't ship for all the reasons that have been explained. But I also managed to do a lot of research there, and Linera is basically a product of a generalization of the research I was doing there with some of my co-authors. And that's the short story of me.

Awesome. And you hold a PhD, correct?

That's right, yeah, in cryptographic protocols. It was kind of the closest thing you could imagine to blockchain back then, but before blockchain existed. But there were only a few protocols back then, it was like TLS and encrypted messages. But yeah, so we were studying a lot of potentially interesting protocols and how to analyze them.

Interesting. And I guess, were you always interested in tech, or was it something you found a love for early in your studies or later in your studies?

Yeah, I mean, I was lucky enough to have a personal computer at home when I was a kid. And you know, honestly, I don't know, that's going to reveal my age maybe a little bit, but it was really at the time where they were just arriving. So I had actually a green screen and then four colors and then sixteen. And so yeah, so I was coding as a kid. I don't know, I just thought it was really cool and I enjoyed reading in the magazines, you know, Scientific American. They used to have a really cool section on computer science, the Game of Life. So I obviously coded that one when I was a kid. And yeah, so I was really nerding out with everything computer-related early on. This being said, professionally, in the French system back then, science was mostly math and physics. So that's what I studied, joined some kind of the French MIT, École Polytechnique, and that's where I started doing computer science, then I did a master's, a PhD. Yeah, so early on I was seeing myself more as a researcher, and then I joined, I was a security expert in the French government briefly, and then joined a startup in Paris, then joined Facebook. And that's where I specialized a little more in software engineering. I wanted to learn how to build like really scalable systems. And you know, to learn that, basically you kind of had to join a big company back then. That's kind of where things were happening. And yeah, a good time there. I worked on machine learning tiers, before machine learning was cool, as you know. These big tech companies running ads, they had very, very large teams doing all sorts of machine learning, not necessarily LLMs, but all sorts of machine learning to run the ads, and they had a lot of tiers. And then also to fight site abuse, like harmful content. So we were running all those tiers. It was pretty interesting.

Amazing. And so with your background in cryptography and then this experience that you've gained at the startup and then at Meta, playing around with ML, curious how you arrived at, basically how your interest in blockchain let's say ignited and came together while you were working on, I think Linera at the time was an incubated project in Meta, correct?

Yeah, so it was actually just, so what happened is when the team Libra inside Facebook was created, the vision was we're going to solve cross-border payments, potentially get billion users by essentially air-dropping the app into Facebook Messenger or Instagram. It was obviously super, super exciting. And so we, with some colleagues, designed a classical blockchain protocol based on the state of the art back then. I was actually doing the initial specifications for the protocol. And so it was a blockchain following a design that you can see in Ethereum and then later Solana, and then obviously Aptos and Sui that came out of it. And with my colleagues, we... blockchain design, they are based on a single chain of blocks and there's always a limit to the TPS. And people like to communicate about it, as a matter of fact. They were like, oh, we reached like 20,000 TPS, now we're doing 100,000 TPS. And so when you have a background coming from Big Tech, so Google engineers, Facebook engineers, they hear that and they're like, what are you talking about? Like you don't build systems with a hardcoded limit in the TPS. That's not how it's done. Basically in those companies, if you do that, your colleagues are going to kind of laugh at you and say, well, good job, but you're going to have to rebuild it again in six months. So there was, back in our minds, we did the job, we did the whole design, but back in our minds, for some of us, there was like, okay, but what happens when we actually get a billion users there? And that's where we started doing research. So George Danezis, the now chief scientist at Sui, who was a professor at London University, he came up with this idea, like look guys, there's this new area of research, people are looking into systems where you don't need a full BFT consensus protocol. You can actually do payments with only certain primitives in the field that is known. And so we looked into that and we built a super scalable payment system that we called FastPay. And it was an academic research. It was really cool, like a very simple protocol. And then it stayed there, basically. So first, turns out we didn't have to address a billion users in Libra because for political reasons the project was basically shut down eventually. And then a payment system obviously is kind of nice, but we didn't have any smart contracts, you know, so it was kind of limited. But for an academic project, it was pretty good. And so yeah, so down the road when I had this opportunity to, so I started to have ideas how to generalize that. And I was like, this is cool, like I could do low-latency DeFi at scale, we could do lots of things that suddenly we don't have a max TPS too. And we can be super low-latency. And yeah, so that's where I created Linera. And then I did some more research and came up with the concept of microchains, which is the core concept of Linera that maybe I can explain a bit more later, but that's the genesis of Linera, yeah.

Amazing, amazing. Yeah, it's very interesting how all that kind of came together and then a lot of that core initial team kind of split off and built these major blockchains like you did yourself. So it's kind of cool, it kind of sort of has the PayPal vibes, you know, like the big brains split off and made all of the big corporations.

No, we had a very strong team. I mean, honestly, I gotta say Facebook did well for the field of crypto by investing in this team. You can see all the projects that went out there that came out of it. L1s obviously, but not only L1s, there's also an L2, a Bitcoin L2, from David Marcus and other projects.

Amazing. Yeah, that's awesome, man. And I guess now arriving at Linera, maybe you can give us a high-level pitch of what Linera is, and then we can get into the details.

Yeah, so essentially the main idea underlying Linera is that we want to be an L1, but we want to bring, you know, we don't want to be an L1 like everybody else. We... there are already a lot of L1s that are specialized, in our view, that are kind of good at DeFi, very specialized, and running everything on a single chain of blocks. And so that has a few consequences, that so you're happy you are a single chain so you have infinite composability, you can do your flash loans, order books, AMMs. And in Linera we're saying, well look, that's not the end of the story. When you're putting all the users on a single chain actually it's going to create some limitations. They are all going after the same resources, they're trying to put their transactions in the blocks. And so if you have... if there's suddenly an airdrop or there's a massive campaign on a particular chain, you might find yourself kind of priced out and not being able to use the system. And for some applications that's just completely fine, you know, but when if you think that crypto is meant to be massively adopted and kind of the next iteration of the internet, you got to compare yourself with what people are used to in Web2. And so in Web2, when I'm using, and obviously that's everything we have, everything on our phone, everything is based on internet services. And so now if I'm... and people are used to things just work. So you are in a traffic jam, you're on Google Maps or Apple Maps, I don't know, every... and there is like hundreds of cars around you also using the same service, but doesn't matter, it just works, and there's no bottleneck or anything, you know, like you can just rely on the systems to work.

And so at Linera we're the starting point is based on, inspired by the research I was mentioning earlier, is to say, well instead of having only one chain of blocks that we optimize to push the limits of TPS with large blocks and parallel execution or you name it, we're going to take a step back and say do we need to have all the users in the same chain? So we certainly want all the users to be in the same ecosystem with unified liquidity and a unified security and unified finality. So we're going to have one set of validators, you know, like it's one system secured by proof of stake. But we're going to structure the block space differently. What we're going to do is we're going to have millions of parallel chains that we call microchains. And so they're going to have like... they're going to be optimized for small blocks instead of large blocks. And we're going to have users potentially having their own chains. We're going to have the opportunity to create tiny chains, microchains for a particular task, such as one game of chess or one micro-auction. And then of course we're going to also have normal app chains, so to run an order book or to run an AMM. So we really generalize, taking the best of both worlds between this academic research I was mentioning and the traditional single chain L1 concept. And now, so when you have a system where you have all those parallel chains and you can spin those new chains super fast on demand, and they have tiny blocks, then suddenly it really opens the door for new kinds of applications. And that's really what we're going to go for in Linera, is really trying to expand the scope of Web3 and provide things, create an infrastructure that really is going to enable the next wave of Web3 applications.

Interesting. Like would you say when you say spinning up parallel chains, is it kind of like a metaphor to auto-scaling but on blockchain?

So, what would be a good metaphor? Okay, so here's one. So in the past, you know, you had supercomputers like mainframes, and everybody was using a terminal to access the... so you had keyboard and screen, and that was a terminal, and the terminal was not computing anything, it was just talking to the mainframe, and all the computations were happening on the mainframe. But then people invented PCs, personal computers, and then suddenly everybody has computation on their own device. And then the devices are in a network, are networked and they can talk to each other and that's kind of the world we live in now. And what we're saying is essentially right now when you put all the users on a single chain you're kind of in the mainframe system, where users they are external to the protocol, they don't have their own block space that they can address anytime they want, they have to talk to an API provider typically, an RPC provider that's going to mediate their relationship with the mainframe, with the blockchain, whatever, and then they also the blockchain has a certain capacity and you may if you are near to it the fees are going to go up. So with Linera we're saying is like we we we're going to give everybody their own personal chain, plus some more like temporary chains, app chains, we're creating a zoo of chains in a way, but the main idea, the starting point is the personal chain. So when I have a personal chain, because it's small, it has only tiny blocks that I create once in a while when I want to do something, we can actually replicate it on my device. And so I'm really part of the protocol, I have blocks and a VM, so I have the ability to execute the blocks on my device, so in the web browser or down the road in the mobile app. And I don't rely on an RPC provider to talk to the Linera system, I actually just interact with the Linera validators directly. I can so I can push blocks if I want to do something, so I create a tiny block in my chain to start a transfer for instance. But I can also receive notifications, so if somebody sends me money or I will my device gonna receive a real-time notification and then pull the data that I might be missing and then talk to the any applications that would be connected to my Linera wallet would receive a notification say hey, you need to refresh your UI because something happened. So the design is we enabling out of the box we're enabling trustless real-time Web3.

So that's the design I'm describing. And yeah, and that was really the main focus for the first chapter of Linera, really looking after real-time applications, real-time use cases, so on-chain games, micro-auctions, managing user intents. So because now I have my own chain so I can also say... I don't know if we can explain more of that, but there's this concept of chain abstraction, so people have been working on providing ways for a user to express a high-level intention to do something. So you say hey I want to buy an NFT on Solana, but you're not on Solana, you... and in fact maybe you don't want to create an account on Solana and buy the tokens and buy the NFT and then move the NFT out. So you what you do is you would like to a system where you can express your intent and say hey, I want to buy an NFT on Solana, make me an offer. And then solvers come in and they say okay, sure yeah we can buy this NFT for you. So in Linera, part of the use case for personal chains is for instance in my personal chain I can advertise an intent or I can create a temporary chain that says hey I'm going to manage the auction system to buy this NFT for this user. And so we do that for one tiny chain in that's going to run isolated from everything else. When you do that, the nice thing, going back to the analogy with the Google Map in the car, when you run a chain in isolation for one particular task, what's nice is then you then you can actually have some notion of a guarantee of service, quality of service, because it's running mostly isolated from everything else. So now down inside Linera, our validators, we can use an architecture that is inspired by Big Tech essentially, where each validator is like a data center that's running that has a pool of workers, a frontend, a pool of workers, a shared database, looking like a small Twitter if you want. And there's going to be, and now we can, because the chains are running independently, we can use the workers to really add capacity anytime. So we can add new workers, each chain is running in one worker. That's how Facebook, all those big tech companies work by the way, like you there's in one data center somewhere there's there's one machine that is managing the queries you're making to your Facebook account. And if a lot of people are using their Facebook account at the same time, Facebook is like I don't care, I'm just adding machines to the system and each machine can handle how many accounts at the same time that are active. And so just by adding more machines I'm supporting more active users. And so Linera validators they work exactly like this, we call them elastic validators. So if suddenly there's like millions microchains that are active, that's okay, we just add the architecture allows adding more workers to handle those microchains. As a result, you if you scale out like well, you can handle the load, you don't have to make people wait or pay more to arbitrage on a scarce resource. Interestingly, you can also downscale, that's another interesting aspect of that is you don't have to pay too many workers if they actually you don't have the like people are sleeping or something like that, so during when it's not a rush hour you can downscale the network. So yeah, so Linera is very inspired definitely by this kind of architecture coming from Big Tech.

And so we were talking about applications, yes. So real-time applications, so micro-auctions is something I'm very interested in notably for user intents. I'm unpacking, sorry, it's a Friday, it's a Friday, I guess I'm allowed to digress, right?

Yeah, go for it, man. Yeah, let's do it.

Okay, okay, okay. So that was a bit of a digression I guess, but yeah, going back to the use cases, so real-time use cases and then recently, what happened is that we also looking into interoperability. And that one is I think it's really interesting because I didn't see it coming basically, it's like it's not something that I initially thought we would be good at. And essentially the reason is everything about a real-time, you had those tiny blocks and were like, oh, this is great we can synchronize them in real-time because they are very cheap, very lightweight, so you see like it's gonna help us be a very low latency system. So that was the initial vision of Linera. But then suddenly for interoperability you realize that because we have structured the chains as this parallel system, in my chain I can create a very fast block, but I can also create a very slow block. So I can create a block that's gonna query the internet and do all sorts of crazy stuff that you would never be allowed to do on an L1 because it's too slow. And in my chain, well, it's gonna take a while, my chain is busy, but everybody else, all the users they they are just fine, they are working in their own chain, they're minding their own business. And so Linera can actually allow users to create very fast but also very slow blocks for and that's super useful for interoperability purposes. Because like I said, slow blocks you can query internet, you can verify ZK proof, so you can do heavy computation, you can download data from a data storage, apply data validation to it. That's that's a partnership we just signed with Walrus, the DA layer, the data availability layer of Sui. So we already like make it possible to download blobs from Walrus, apply validation and then once you have created your slow block where you did all this work and the block gets finalized inside Linera, then you can take the certificate inside Linera like the certificate of finality, and you can show it to the other chains, and so you can prove to the other chains all the computation you did in inside Linera, you know, as telling you something about the block and so you can use Linera essentially as a data validation layer next to existing L1s. And so that's really something we really a kind of use case that we're really interested in, so building hybrid applications, so you have builders they are have their system in an existing L1, maybe Sui, maybe Solana, maybe Ethereum, and then they talk to Linera, we spin a microchain for them, we do all sorts of computation, maybe also real-time games, real-time on-chain games, and then when it's done it goes back to the existing L1.

Fascinating. Yeah, it seems like Linera is quite feature-packed and state-of-the-art. Like, you don't hear that too much from L1s, and you always have to have these L2s or L3s to kind of make up for that lack of scalability that you mentioned at the beginning. And on the topic of builders, how easy is it for say EVM developers to come and build on Linera? Does Linera have its own programming language, or what's the landscape of the tooling on Linera?

Yeah, so my mindset here is we've already innovated so much on the architecture that I want it to be like not innovating on the VM. So we chose Wasm and Rust as a programming language, so Rust is used by other chains like Solana, Polkadot, whatever, and so we so we started with Rust, our infrastructure is also written in Rust so it was a natural choice for our engineers to create the first SDK in Rust. But we are not, you know, we're not opinionated and in fact we're adding support for EVM just now as we speak, I just announced it yesterday because we had our first milestone, we're like, okay, this works like we can do this, we can do this, and it's already on the main branch you can already create tiny tiny demos, we're now going to add all the bells and whistles and and and make it composable with the rest of Linera. So we're going to be a multi-VM system with composability between apps running on different VMs inside each microchain.

Wow, that's amazing. And yeah, I guess fresh off the presses, Linera is now EVM compatible. Awesome.

Look, we started, yeah yeah. So but officially yes yes, we definitely have the roadmap and we're gonna make that available I would say next quarter, next quarter it should be ready.

Amazing. Yeah, looking forward to seeing that. And I guess, we touched on it briefly before starting the show, but we've seen a lot of craze around AI and crypto intersecting. Just curious to hear what your opinions of that intersection is and how you see Linera kind of playing in that landscape.

Yeah, so yeah, we love AI. We love AI. So traditionally I think it's useful to structure the debate like AI plus crypto, so you can have two categories, two ways to see about it to look at it. So one is crypto helps AI and another one is AI helps crypto. So crypto helps AI is actually maybe what investors are the most excited about, I think. Basically the idea is like oh AI is going to revolutionize the world, but it's also dangerous. It's centralized, you could have those large models that are owned by just a few big players. But and here hey, we have crypto, we can create communities, we can coordinate people and the vision would be they they will be able to put their resources in common and train better, better models than than the big players that are centralized. That's one because maybe when you create a community you can people will volunteer in the secure protocol, they will be able to use their private data that would otherwise not be accessible. So there's a very ambitious roadmap where people are thinking that decentralized AI is actually, you know, down the road could be better than centralized AI and by as a result solve the problem, one of the problems of AI. Another thing also is slightly less ambitious but you could also when I query AI inference services like DeepSeek or ChatGPT, they give me answers but I have no idea what they how they got the answers, and for certain use cases that's not gonna work, like maybe a regulators are gonna complain that I'm using like opaque algorithm to impact the life of people or that we don't know privacy like if privacy is respected or something like that. So you could say we're going to look at AI inference services and we're going to add traceability, we're going to add privacy, we're going to add accountability, and turns out that everything we developed in crypto can help with that. So that's also another way crypto could help AI, it's more like crypto technology helps AI. But that's also very interesting and definitely people are are looking into that.

Okay and then and then there's the other side where you're like AI helps crypto and that's how you like run AI inside crypto. So like for instance you you have on-chain AI agents. So you're like yeah crypto is nice but I want to run AI, I want to be able to have those incorporator incorporating intelligent beings on the chain. And yeah, that's also super fascinating. We... and then and then that's one thing and then there's AI used to make it easier to to use crypto. Yeah, that's another another thing where I can talk to my wallet say hey, I want to do this, I want to do this. So yeah, so we looking for the last two, we've been actively working on this. So we have a demo, check out Twitter, we have a demo where we use the we connected an agent, a text, a text LLM, a chatbot to a microchain. So you give a you you tell the chatbot hey, you you know you're on a microchain, this is what you can do with it, and then as a user I say hey I would like to do this or what can I do and then you get you get some answers hey I can transfer funds, actually the video is kind of fun because you say hey I'd like to transfer fund to Bob, and the chat is like okay but I need the key of Bob, man, what's the key of what's his address? And so that was kind of and then another in the video also my colleague is like oh I'd like to transfer to myself, and the chatbot is like well why why do you want to do that? That's that's a bit silly, why do you want to do that? So they chatbots are fun. Yeah, so we have and so the here the good thing about Linera is we worked really hard with this this new architecture that I was describing where as a Linera client I have a replica of my own microchain, and I'm that makes the Linera client act as a local as a trusted data provider, RPC provider for the Web UI or the AI agents. So long story short, so as from the point of view of an AI agent, I'm you're talking to Linera, you're talking to the Linera client, it looks like a just a normal Web2 service that happens to be running locally so you can trust it. And that's very convenient. So we can abstract away a lot of the details of blockchain, and basically we're already halfway, I mean like just even before connecting the chatbot, we've already abstracted away a lot of details in the initial architecture of Linera and that makes the integration with AI agents way easier. So we were pretty happy actually to put together this demo in a couple of days using that and it's and it's secure because you you talk to your wallet. So the the agent is talking to your wallet, so you can trust the answers. It's really neat.

Another thing we did, so going back to the other kind of applications running AI inside the transactions, we we also have a really good this is also a very good application of Linera because the slow what I what I when I when I was mentioning the slow blocks. So inside a Linera transaction, in my chain I can I can I can query AI inference services. So Linera is not going to have GPUs like on and running machine learning models inside our validators because that's a different area, different expertise and just different project entirely. But we can definitely query tiers of AI services from a Linera transaction. And that's something that's very easy to do for us and pretty hard to do in normal blockchains. A normal blockchain will not let you query external services from a transaction for a variety of reasons, one of them is it's not deterministic. Another is another reason is it's slow, it would ruin the block time of the other blockchains. The and the determinism is something that for us, well obviously it's so determinism is so in our case because we execute the blocks right away, if there is determinism we'll see it right away and the block is is invalid. In other other blockchains if they let you put a transaction inside a block that is non-deterministic, then down the road when they are executing the block, the transaction is already final, it's already included, and that ruins the design. So non-determinism is actually very much harder to deal with in other blockchains.

Yeah, I agree and in even in AI in itself that's like a whole it's a whole company basically, right? But the long story short, that's another example where integrating AI services we find out was actually pretty easy for us. Didn't we didn't really thought about it initially but as soon as we realized oh we can actually query other services from Linera transactions, so oh wow we can do AI. So we have a partner that's providing all the AI inference that we just announced and it just works. Amazing. That's cool, get to kind of chain different transactions together with blockchain, it's pretty neat. Unfortunately we're pretty much at time. Last question for you for the folks that you've inspired today to start building with Linera, what would you recommend the best way for them to to get started?

Yes, so we we have a great community. So you can find us on Discord, reach out and and get involved. We have weekly hours that are also really fun where everybody comes in and we have our moderators and head of community Danny that are like putting a show putting up a show, inviting interesting guests every every week. So that's how you can find us and how you can get involved.

Amazing. Yeah, and with that, just want to thank you so much, Mathieu, and everybody else who's watching the show today. It was a very interesting and fascinating talk. I know I'm going to be deep diving a little bit on Linera myself after all the great content you gave us today. And for folks who have been watching us live, I've dropped the link in the chat where you can claim your attendance badge for watching us live through BitBadges. So collectibles or anything, definitely check that out. But yeah, just bringing it back, thank you so much, Mathieu, I really really appreciate your time coming on the show today.

Yeah, for sure, that was a pleasure. Thanks for all the great questions, it was great to be here.

Yeah, my pleasure. And yeah, and wish you one of wish everybody a very happy Friday, happy weekend, and we'll catch you back here for another great episode of DevNTell next week. Right on. Take care, guys. Bye-bye.