Agents as Companies: The EigenCloud Thesis | Sreeram Kannan & Gajesh Naik

Transcript

Agents become the new companies. That's our thesis. Because when you have an agent and property rights, suddenly they become ownable, they're new kinds of digital assets that are highly valuable. Hey everyone, quick disclaimer before we get into today's episode. Nothing said on Bell Curve is a recommendation to buy or sell securities or tokens. This podcast is for informational purposes only and the views expressed by anyone on the show are solely our opinions, not financial advice. Our guests and I may hold positions in the companies, funds, or projects discussed.

All right, everyone, welcome back to another episode of Bell Curve. This is going to be a very fun episode. Uh today I'm joined by Sriram and Gajesh of EigenLayer or Eigen Cloud, rather. Guys, welcome. Pleasure to be here, Mike. Like I've been here a few times, but you know, really excited to share what we've been up to and what we're doing next. Thanks for having us, Mike. Me, too. I'm pumped. Uh yeah, Sriram, you're a veteran of Bell Curve, potentially one of the the all-time >> [laughter] >> guests, both in terms of quality and and number of appearance. And the last time we had you on here, we did a whole season on restaking. Uh that's what we were mostly talking about. Uh we're going to be talking about something completely different uh today.

EigenLayer has a super exciting new focus around the agentic economy, but maybe before we get there, let's rewind the clock. Talk to us about restaking, what ended up happening. Uh yeah, talk to us about all that. Amazing. Yeah, so the um Actually, like even maybe like I'll go one one step before. One of the reasons I got into crypto is, you know, before before this I was working on genomics and using mathematical algorithms to do like genomic stuff. And um I know I'd do mathematical statistical algorithms to decode like how a DNA should be sequenced, all these kinds of things. And there was this episode where one of my students came to me and said

yeah, this is 2017. Uh and he said like, "Hey Sriram, like can I just use AI to solve this problem?" So this is, you know, at that time it wasn't really called AI, it was called deep neural networks. And then I'm like "Okay, that doesn't make any sense. This is like physics and math and like why you're trying to just use black box algorithms?" He said, "No, no, I'm very excited." He just went home. Two weeks later he comes back to me with, you know, his box and he's just like trained his neural network to like do this DNA sequencing thing and it beats my algorithm, which I had been designing for 1 and 1/2 years till then. Including with a physicist who actually invented that nanopore.

I'm like, I couldn't believe it. I'm like, "Okay, this is crazy. This can't be true." And, you know, people talk today about whether AI, AGI, they're going to lose their jobs and all that. I felt like, you know, back in 2017 uh this box, which I had massively underestimated, kind of like comes back to me with what I had been thinking pretty deeply for many years. It just beats me. And then I'm like "Okay, so stop and take notice. Like the world is shifting. You need to understand that."

So that was my AGI moment, right? Like, you know, all of us go through this moment where like, "Oh, I'm very proud that I can do this this thing." I don't know if you'd seen that Dave White from Paradigm had written a very this is maybe like 3 months back on like, you know, knowing to do math and doing this stuff and like, you know AI does it better than him and then he's like "Oh shit." You know, it's like a little bit of a dying is what he >> [laughter] >> And all of us are going through that right now, but I felt like that that is when it happened for me. It's 2017 November.

And part of the reason I got into crypto is because blockchains and cryptography are the only things that are stable in a world where AI can do anything. It can do my work, your work, but it can't break cryptography. It can't break a blockchain. So this was kind of like in the kind of ancient origin story of like my shift into crypto, into blockchains. And we started EigenLayer with the goal that hey, look at a blockchain, I can't build whatever I want inside it, right? Like for example, I can't build a machine learning algorithm inside the blockchain. Like that's not a thing that you can do.

Because it's just meant for financial transactions. You have these highly constrained programming environments, highly constrained scale. So you can't do what you want. But I want to do all kinds of fun stuff, just throw machine learning on top, just run like a game on top, like all these things, and we couldn't do it. And so our solution to this problem, this was the problem statement that oh, blockchains are interesting because they're robust against an AI, AGI era because they can form the coordination substrate for us, which, you know, nobody can break. And but blockchains are not programmable expressive enough. So what can we do about it? And on my like took a wind a winding path, which ended up in we are

building EigenLayer. And you can see even in the name, like EigenLayer is your own layer. You should be able to build whatever you want. It's not a restaking layer, it's not a staking layer, it's not universal staking, none of that, right? It's called EigenLayer. You should be able to build what you want. In fact, like I was just showing somebody my like the first white paper, which was never published on EigenLayer, you'll see like, "Oh, how can you run machine learning inside a blockchain? How can you run authentication inside a blockchain? How can you put credentials inside a blockchain?" These are the kinds of questions that we asked. We are like, "Oh, maybe if there's a standard layer which lets anybody supply trust and shared security, anybody can build their layer on top." That was our goal.

Turns out as we were trying to build it, like our the the input that we needed was staking, right? So we said, "Okay, staking secures everything. Can you just re like use the Ethereum stake to run all kinds of other networks?" And so that was the whole idea of like EigenLayer connecting stakers to these AVSs, these services. And the first service we built was on top of this was EigenDA, which was the data availability service. So this is kind of like the origin story. And we did not I did not expect or think that the staking was the most important thing. The staking was an input, the AVSs are the output, the demand side. Like I think last time, you know, a year and a half or whenever we

did the restaking episode we talked a lot about like the demand side. Like who are these people? Who are why are they building AVSs? And like the mental model was always like, "Oh, blockchains are going to be big because they can coordinate." And what are they coordinating? They're they're letting more people like What is What is going to be built as AVSs is, you know, more general data, more general compute, more general machine learning, more general authentication, all these things, right? So that's that was our thesis. And as we kind of like built the product, one thing became very clear is this the market structure that we had constructed for EigenLayer was restaking staking on the one side and then like AVSs on the other side.

And AVSs were themselves like infrastructure protocols, right? This is like bridges, oracles, AI like networks, all these things, right? So they're infrastructure. And you know the market size in crypto for like all the development activity that's happening is not big enough to accommodate a massive infrastructure segment. This is just like the the kind of like the phase and evolution of crypto and where we are in that that journey. I I think it's too early to have hyper modularization, which is what really like EigenLayer was trying to do. It's to say like, "Oh, every infrastructure service can be like a SaaS and you can

specialize and just kill that one service." It's too early in the kind of evolution of crypto to actually go hyper modular, like hundreds or thousands of these modular services. The the market size is not there, the revenue is not there, the apps are not There are five apps in crypto, right? So with five apps, you should not be having hundreds of infrastructure services. So we were building for a wildly optimistic world. That world hasn't yet manifested. We're here because we believe that we can still manifest it in the future, but it's not here today. So this was kind of like the learning from EigenLayer. The learning from EigenDA was we built this data availability layer, we had a hyperscale. In even in

EigenLayer before I move on, like in restaking, you know, there were other people who were trying to enter and like, you know, with pretty significant like, you know, backing and so on. And I think one thing we did well is to kind of like be the very clear market leader of the category. Like we created a category, we are the leader, we won the category but it's not clear the category has short-term value. Like so that's the summary of what happened to EigenLayer and AVSs. If we had started our like bet from oh, we are a restaking protocol, we could have very quickly pivoted into like, "Oh, we have all this stake, why not run some vault protocol on top of this stake? Why not do like other financial services and like things which other you know,

uh players in the market are potentially doing is like saying, "Okay, this is just a vault protocol. You can do like all kinds of other like financial services on top." Because we started from EigenLayer means like you are actually like we want to solve the developer problem, we want to solve the problem of expressivity, programmability where we can build machine learning, data, all these things on blockchains. We didn't go that way. We took the other way, which is, "Okay, if the market's not yet ready to like accommodate hundreds of infrastructure services, can we just build those services ourselves?"

And so EigenDA was the already the existing archetype and then we doubled down and said, let's build Eigen compute, let's build Eigen AI, off-chain compute, off-chain AI. We looked at the cloud and if you look at the cloud, what is the most important thing that's happening? Data, compute, AI. Once right, so there's Amazon AWS that has S3, which is the storage service. There's a compute service called EC2. And then there is an inference service called bedrock. That's what everybody uses. It's S3, EC2 bedrock. So we just built those natively on Eigen layer. And so that's Eigen DA, Eigen compute, Eigen AI. So that's what we did.

We said, okay, let's go build that. And that was the kind of thesis of Eigen cloud, build these services. And these services um enable um like um these services uh Eigen DA, Eigen compute, Eigen AI actually um uh enable an application builder, not an infrastructure builder. Directly come on our person build whatever they want. And so we back massively reduced the barrier for somebody to do that. You can just write a program like you would write to deploy to a cloud.

Write it in and you package it into a docker container and send it to Eigen compute. It just run it off-chain. But runs inside a trusted execution environment, back up all the data to Eigen DA. So that even if that node you run it on a single node but back it up on a DA so that if that node goes away right, somebody else can pick up that archival data from Eigen DA continue running the compute. So you get the core properties of blockchains, which is the service continues running, not dependent on a single point of failure while actually getting the efficiencies of a single central node running it.

So that's the kind of like core architecture of Eigen cloud. Okay, so just to re-sum uh you know, the re-staking So basically Eigen layer started it to solve um some of the constraints of building things in blockchain environments. Um maybe the re-staking uh the re-staking side of things was a little bit far in the future. That like whatever it is the demand didn't necessarily materialize. You built your own first-party applications uh such as Eigen DA. And the way that you view Eigen cloud is kind of an extension at least of the original mission, right?

And maybe drawing some inspiration from the uh you know, the cloud companies today in terms of the focus on data, compute, AI. And that's kind of the stack that you're building for Web3. That's correct. Exactly. So and and so we call this cloud scale crypto trust, right? So we want to get to like cloud scale. So this is why we are kind of militant about scaling the data throughput where we don't want to like run it at like a thousand X cost overhead to like what you could do on the cloud. You should do it at pretty similar basis. But you should still get the the blockchain's properties of unstoppability verifiability censorship resistance, all these things that we care about.

And so how do we manifest that? What are the right architectures to do it? It's a little bit like the roll-up road map in some sense, right? So in a roll-up you're decom- decompose into compute and and data and like data is stored on a decentralized archive but a roll-up sequencer a single node is running it. But you know, if that node that node can't lie because it's whatever verifiable. But any other node can always come and pick it up and continue on. So that's the whole roll-up thesis. We just take that and like expand it to like far beyond the reach of current roll-ups, which is only to do financial transactions but actually you can do arbitrary compute.

So uh a couple more questions on uh this side of things before we get to the um the new Eigen cloud and AI sort of architecture and vision. And you know, does the demand for DA? Uh it sounds like DA is still a core part of this new uh this new vision like I'm even just looking at Celestia as well like you obviously like their kind of bread and butter was DA and you know, was you know from the retrospective was Do you think that it was early for DA? Do you think that ultimately that's never going to be a super large market or do you think that this new direction is going to necessitate you know, a ton of very high throughput DA?

>> Right. Right. Great great question. I think the the So one of the problems with the I was talking about hyper-modularization before, right? One of the problems with the way we modularized things like even just DA and execution it was we [snorts] massively scaled the data throughput. Hoping that other roll-ups will scale the compute throughput. And so like now it's up to like you know, Arbitrum or Optimism or somebody else to do that as opposed to us doing it. And that was kind of like the benefit of the architecture as you just get specialization and different people can do different things.

But what happened is the main use case of crypto remains like just you know, financial transactions whether it's stable coins, trading, like these these use cases. They don't really need like you know, uh gigabytes per second throughput. Like it's just not you know, even if you some We did a calculation when we launched our 1 gigabytes per second offering on Eigen DA. I think we can fit like 100 or 1,000 visas or something. Like it's pretty insane. And so which basically means like okay, you're not going to get >> [laughter] >> Right. You're not going to get limited yeah on that throughput side.

>> You're not limited by this thing. And so like what we found is we were building for a world that is not yet manifested again. And I think that's the same for Celestia and us. And like inside the DA game again we did really well. We have the highest total value secured. We have the highest throughput. Like we have all kinds of metrics if you just take that as a category and see like how we did at Eigen. I think we did very well inside that thing. But it's just now that the first instance of a mega ETH is actually able to like use this throughput. We were just looking at Rise on mainnet today.

>> Right. Right. >> Rise is running like right now at 200 megagaps per second. And they use Eigen DA. But it's just starting to happen. When there's one or two like that's not a market, that's just like you know, proof of concept. Like we're still in the proof of concept era of like high throughputs needed on the DA side. So that's the that's that's our read of what's happening. The way we solve it is Eigen cloud like if I'm running more complex AI agents, other kinds of applications on top of Eigen cloud and they all need data throughput. We just use it ourselves, right? So we use it in our own agent stack. We use it in our uh like we have to go one level up to figure out oh, what are these things

which are not financial transactions per se. What other things need verifiability, unstoppability that actually can benefit from something like Eigen DA. And so that's that's what we'll talk about next. But this is this journey has been like pretty illuminating. We just keep marching up from we started from all the way underground like oh, you need stake. Now you need AVSs. Now you need Eigen DA, Eigen compute, Eigen AI. And then we go one level up. So it's a kind of crazy crazy journey that we've been in trying to kind of figure out the product market fit for and just even manifesting my own hope and belief for what crypto can be which is much bigger than financial

applications, which is what we started with. Which is what we're marching towards, which is not yet here. So that's that's the short summary of our march. So you know, it's very interesting cuz we're have we're talking at this moment where most people have stopped believing that crypto is useful for anything outside of financial applications. Uh I know that's Actually Chris Dixon wrote something which is along the lines of what he's always said, you know, in like his read-write-own thesis that this is a new you know, computing environment where you can do a whole bunch of stuff. And I thought it was really I I came out with this a couple months ago and he got aggressively downvoted, you know, like how how dare you suggest that this is only good for financial like that's why we lost all this money. Yada yada yada.

Um and you know, it's interesting when you kind of look at the potential opportunity set for Eigen. Um one direction and I heard you talk about this before for um but I wouldn't have didn't I don't know if you said you didn't have the option to do this or I don't know if you considered like Eigen has a huge balance sheet. Um and one way that you could go maybe is like leveraging the balance sheet of Eigen. Um so and then the other way and these aren't mutually exclusive, right? But you could also say like hey, we want to double down on the original mission. We just got the layer wrong. And maybe you're doing both but I'm curious how you internally thought about those two different opportunity sets.

>> Yeah. Um like maybe I don't know, there's so much in DeFi, right? Like providing a liquidity sleeve for RWAs that want to you know, get in and out and some stuff like that. So how did you evaluate the two different Yeah, that's a great great question. And I think this was something we thought about quite a bit before narrowing down to where we are going right now. Um the thing I found is the the most important thing happening like you said, Mike, is that institutional finance is meeting crypto. It's happening. It's the moment's right now. If you're in that game like it's it's it's interesting. It's exciting.

Lots of things are happening. It's people payments, stable coins, real world assets. You know, forex moving on-chain. All these kinds of things are happening. The thing we felt was in in that world building technologies of trust and verifiability is that the rate limiting factor for BlackRock issuing new RWAs on the blockchain? Nah, not really, right? So we would have to basically say oh, like among all the things that we build, which is an enormous number of things the only thing that's really valuable is this re-staking vault that we have.

And then go like focus on that. So it's not my DNA. It's not our thing. We're not Wall Street people. We're not financial people. It almost happened incidentally. The whole re-staking thing was an input to create this output. In fact, like the people who staked with us were all betting that AVSs are going to explode. That crypto has an infra problem. Eigen's a way we're going to solve these infra problem that's going to create all these AVSs. They're going to have tokens and whatever and we're going to get get exposure to this. This was kind of like the the thing people bet on us for.

And that world didn't quite manifest itself, right? So, it's we were too early. Infra middleware like has very little value in all of crypto right now. And it's because there's very few apps. And when I go back to like the institutional finance intersection crypto, like it's interesting and it's definitely happening. But we are not like, you know, are we going to win that space? We'll be a player in that space and it's not very exciting for us to do that. So, what So, the question became like you know, you ask this question like, oh, there was this thesis that crypto's only good for finance. Like where you know, where are we with that

thesis? I can maybe like go into that a little bit. But if do you have any comment on on my previous claims or statements? No, no, no, I uh I agree. I I I mean, maybe it's my DNA as like a founder, but I think of companies in terms of DNA, right? I mean, just very logically, you start with a mission, you hire a bunch of people that help you align that mission and yeah, sometimes I feel like people just look at a company like, well, you could do this. And it's like, okay, but I hired a bunch of people to do this, right? With this very specific skill set, like you built something a position tool. Also, people tend to have beliefs, right? They're like, I'm excited about this original mission and vision and um yeah, I think people sometimes externally look at companies,

um especially maybe sometimes more the investor perspective is you'll find success via It's almost like they look at it like a puzzle piece and if I just get the strategy right, whereas yeah, I tend to look at it much more from the human first, motivating people and, you know, organizing and motivating I think is like the the key anyway. But that's Absolutely. I completely like I can't agree more and the like definitely at Eigen like one of the things we have been successful at is building a very mission-driven org, right? Like where you know, and you go and ask people, they'll say like, oh, what what do you believe and they'll say like, coordination, open innovation, like things like that. Those are the kind of first principles that we get people to like come and show up for.

And suddenly say, oh, we're just doing a finance just like, you know, a new world protocol, like it's not going to fly. Firstly, I'm not excited and then like nobody else can be excited after that. So, basically that's absolutely right. So, there's a founder, team product founder, team, market fit rather than just like the tactical opportunity. We will explore the tactical opportunities like, you know, with partners and other people can or you know, hey, you know, create vaults on Eigen or whatever, right? But is it going to be the main focus for us at Eigen? The answer is no, right? Like that's just not our DNA. So, maybe um and maybe Gajesh, this is where I could tag you in here. Um all right. So, we compared Eigen Cloud in the stack to Amazon right here in terms of data,

compute, AI. Let's just start with a really basic question of why would a developer choose to build on Eigen as opposed to leveraging the solution that exists in the market? Yeah, sure thing. So, currently going um if you want to have a secure environment with like all um like all the verification layers which we currently provide, we we allow you to verify if the code is running correctly, is the app running on like does the app have its unique uh private key which only that code can access. All those components is something you can get on Eigen Cloud in like 5 minutes. It takes 5 minutes to like go and pay for it and go and deploy your first app on Eigen Cloud. And uh yeah, we we we're just simplified entire

stack by giving all these verification tools like off the shelf. So, one use case which I think will shine a light on what Gajesh just said is imagine you want to run an agent. Okay? And I think this is also like the agent was one of our like shelling point examples that we built the entire stack towards. Because what does an AI agent need? He's data, he needs compute, he needs infra. So, like that's kind of like the grounding principle for like why we built these three blocks. And if you if you want to build an agent, you have like the following options, right? So, I can just deploy my agent on my own laptop. Like the open claw movement, a lot of people like just have MacBooks, Mac minis, just take their

claw and throw it there and just runs and like you can have fun with it, right? Like that's option one. It's a personal agent that you want and you know, it runs locally. The second option for a personal agent is, oh, I don't have enough memory in my computer. I don't want to keep it running on my laptop. I just borrow a cloud instance, you know, Amazon or Google or whatever and then throw it on top of it and then it runs or Digital Ocean. It's cheap. Go deploy it there and it'll run on your behalf. You own the root keys of this thing and it's fine. So, that's how we believe personal agents will run. So, deploying like a personal agent, single player agent onto Eigen Cloud is a counter pattern. You don't do that. It there's

no utility for you, so you should not do that. So, this the points that Gajesh was mentioning just trust, verifiability, correctness, unstoppability, like the core value proposition of blockchains don't shine through. Like you want to control it, you deploy it, you own the root key. You trust Amazon, nobody's worried about Amazon's rugging them. Like, you know, they're not in that cypherpunk universe. So, people are really happy. They just deploy and then that's for personal agents, that's the correct pattern. But when you go beyond personal agents, when an agent is not acting on behalf of just Mike or Gajesh or myself, then you get into the question, who runs this agent? Where does it running? Even

if you say, oh, it's running on Amazon, who controls the root key of this developer instance? Somebody has the root key. They can change everything about this thing. So, just Amazon's not providing trust to run an agent in a trustless verifiable way. No, they're just saying like there's some developer who has a root key. Like that's the person who has the control of this. So, we'll run into a problem. Okay? Like imagine Gaj and I got into a bet and we're saying um you know, the AI agent is going to adjudicate who won the debate, right? Like, you know, uh then and you know, they that's a very sensitive thing. Do I can I change the prompt? Can I change the model? Can I run it many times? What else can I do to

like ask that agent to give me a favorable result? There's a lot of degrees of freedom to actually like manipulate it to like get what I want. And so, Gaj will be like, oh, [ __ ] um you're not running this [ __ ] Like I would I'll be like, oh, you should not run this because you're the counterparty to me. So, when you have multiplayer agents, when you have multiple players, there's no single party that actually owns the agent, then you get into this thing of that you run it. Who How do you trust it? How do you make sure it's verifiable? How do you make sure it's unstoppable? It just keeps running, you know, without us needing to worry about it. And that's where our Eigen Cloud comes in. So, in the use case of a single user doing a personal agent, no

utility. When you have multiple players, potentially untrusting parties with each other, that's when all of blockchains, right? Like honestly, that's when it becomes interesting, right? When you when you have a single trusted party, you don't need these blockchains. You just need them when you have mutually distrusting parties. So, that's the use case. An agent. Anything in between multiple players. I can give some examples and use cases, but that's really the kind of use case where it's interesting. You know what also is I was going to ask you here is um I feel like there is a interesting dichotomy here in between the what like the words are important. And verification, I think a lot of people in crypto now when they hear verification, they're like, ah, do we really need But

you know where that's a really sexy word actually is in like MasterCard and Visa. They really like that word because the uh what the challenge and the pain point is is like it's kind of like equivalent to dispute resolution in a sense for them. It's like, hey, uh if an agent did this, how can I prove like and it messed up, where can I like where's the versioning? Can I go back in time and understand where uh someone made a mistake so I can assign a sort of guilty party. And I think um you know, when I think of the architecture that you're building here, I mean, I have a bunch of questions on how it's different like the commerce and financial rails differ on on crypto rails versus what exists, but am I roughly thinking about that in in the right way? Like this That's absolutely

correct. So, trust and verifiability matter when the stakes are high, right? And the consequence of a wrong computation is lost money. I think that's the place where it's the most clear and I think that's why MasterCard, Visa, these kind of examples come to mind. You know, part of our own evolution, I communicated like what happened between EigenLayer, EigenDA and this cloud thing. And I think one of the things we've done in the last like 6 months is think a lot about what are those use cases? What is the one like canonical category of use cases where trust and verifiability matter. Because Mike, like you said, many people hear trust verifiability and they're glaze over. They're like, okay, this is some cypherpunk people talking about I don't trust anybody for anything, right? But

you definitely see that in in money, like when you move money, like this is why Bitcoin became a thing, right? Like I don't need to trust anybody. Like I own my money. I don't need to trust a banker. I don't need to trust anybody else with it. That was very important. And in finance, which is the next evolution the way we think about it, it's obviously important where like, you know, I'm I'm I'm I'm doing some kind of settlement, escrow, like I'm settling a prediction market, these kinds of things, you need verifiability. But part of what we've been thinking a lot about is where is where is the verifiability unlocking brand new markets that have never existed before or couldn't exist before because trust and verifiability are missing. And so I'll I'll get into that.

That's really like our core fulcrum. And so I I view this like our own journey is kind of like like I said, from the bare bones infra, okay, you need stake, you need AVSs, you need like I can do I can do I can do I can do we did that and we're still like not yet at like, you know, success is because we need to go one level up to see like that is this trust and verifiability most material. And the thing I used to hear from everybody is actually nobody cares about trust. Okay. And then I would ask them oh, like will you take your retirement account, your 401k and put it in this like uh you know, anchor protocol?

They're like, are you idiot? Are you an idiot? Like and so what I hear from that is when money is at stake people are of course sensitive to trust and verifiability. In fact, trust is the permissioning mechanism. Right? You say like, oh, I trust BlackRock, I trust Fidelity, I trust like, you know, whatever these five things. Among them, I'm not going to differentiate anything because they are trusted. And among the trusted things, I don't care about like who's doing it, I'm just going to use them. But trust is a permissioning mechanism. But if you're not trusting, you're not even in the game, right? So you're just out of the game.

And what are the things that are out of game that we can bring to the center of the game? And like what is this new direction that actually puts a spotlight on this hard problem that I think we've solved, which is how to endow trust and verifiability to arbitrary compute data on AI. And why is that material? So that's like maybe the next shift for us. And also like so trust facilitates commerce. The The other thing too that's really important that I don't think is well understood either sometimes on the payment systems like why does Visa command the margin that it does? It's not like the technology requires that.

It's um you know, it's like chargebacks and things like that and dispute resolution and who get So having a very detailed record is it's it's necessary to um facilitate uh to facilitate commerce and people taking action on chain, but it's like the what I just like the weirdly what really people want is they want a record of who messed up if someone messes up so that there could be liability towards that party. All right, like that actually is kind of the foundation of trust. It's like everyone knows who's acting Accountability. Yeah, accountability accountability is the exact right way to describe it. Um Okay, so uh Gajesh, I'm going to pick on you again here. Um

help me understand the sequencing of this market. Uh actually actually before we get there, uh talk to me like explain like I'm five. Uh I kind of understand this idea of like everyone's very hyped up on this idea of uh agents using stable coins. I understand at a high level it's like the API system feels very clunky and you know, they're not going to be able to you know, yeah, get through a bank's security process, but like help me understand the different challenges of um you know, commerce and agents leveraging crypto rails versus traditional rails. Yeah, I think uh the stable coin stuff like is definitely like really helpful for you to like make payments and stuff, but you can't do you you can't do more than that. Like it's like if a agent is given a credit card and a stable coin, I think the the only differentiating

factor is that the APIs are better for stable coins, while credit card they have to go on a browser and then go ahead and like input it. So it's just it's just API level thing and I don't think uh sorry, I think that like Stripe and all these companies will build like a agent level API for you to just like put your credit card and you I bought it. >> already I I we can see for example, Ramp is saying, oh, you can create a sub credit card for an agent with its own spending limits, authorization, whatever, right? Like and Stripe is doing a lot a lot of work in agentic payments. So we think that you know, when people think about crypto and AI, we they think like, oh, there's going to be an agent, agent's going to use money and like money's going to be

on the blockchain. And that's the main thing that crypto and AI kind of like intersected. Like that is the the common understanding. I'll say like even deep insiders, right? Like this is what they're talking about. Oh, crypto like AI is you going to use blockchain money. But I find that mostly it's incremental benefit. It's not a 10x, it's not a 100x. I can use a Stripe API like I just said. The Stripe API uses a bank account. They're all digital already. So I can just give them that. It's just this authorization and containerization, record keeping was not yet there and like all these guys are doing more, Visa's doing more, Stripe's doing more, Ramp is doing more. So you're going to get that.

Because you know, agentic commerce is going to be big, so it's not they're not going to just like sit around. Of course, the global interchange fee and like the whole mechanisms better with crypto. So like maybe they use crypto in the backend, but it's just become abstracted like, you know, glorified uh complex backend databases for settling and reconciliation and and like things like that. Uh so it is happening, it is valuable, but is it like a big space for like some new trillion dollar asset class or anything like that? We don't believe that. We think it's interesting, it's good, it's a use case for crypto. We're going to it's going to send more transactions our way in crypto. Like so it's cool.

Um the thing that we believe that agents don't have Actually there's a there is a pod the pod that I actually like to quote between Microsoft CEO Satya and uh Varkey. Right? And Varkey is asking Satya like, hey, you know, what do you think about this AI? Is this a new species? Like what is this thing? And then he says he doesn't question that AI is going to get enough capability. This is already 6 months back. And he's not going to get enough capability to be like super intelligent or whatever. Satya says, our legal infrastructure doesn't allow an agent to own property to create a company to take liabilities.

Right? Like these things are unique to humans. An agent can't do it. They can't own property. They can't issue equity. They can't take liabilities. They can't do all these. Without doing these things agents are only delegates of human agents. So humans are the principal, they're the agent, they just go and do that. And that's the if you think about what a blockchain is a blockchain allows a program to own properties. Right? Like the EigenLayer program which is some code, right? The EigenLayer program owns some 10 billion dollar of property, right? Like sitting on the blockchain, owns these ETH and other assets.

And people put it in because, you know, this program operates in some deterministic known way. So that's that's what's happening is blockchains are the solution to the problem Satya is raising. Which is how are agents going to own property create companies take liabilities, do all these things that are simply impossible in the current system. Because there's no framework for it. Like what does it mean that an agent owns like a website, right? Like what does it even mean? Actually like we did this thought experiment, you know, um maybe, you know, Gaj, you can say a little bit about like how

agents can own property. Yeah, sure thing. Yes, so I think we So we've built a couple of agents now and Saura is one of the agents we recently released, which is a completely sovereign agent which owns its own own Twitter account and then makes posts from it. And the >> There was a meta like in crypto like nearly 2 years, maybe 18 months back when, oh, you know, AI is tweeting and so on. What's the difference between that and Yeah, yes, it's actually uh pretty simple. So we we basically just take the same code. It's it's almost very identical code. The the main thing we get you got from EigenLayer is the way to attach the identity of the agent I mean the the wallet which is the

identity of the agent to the code it is running. So only the only the code which is running can access the the private keys and the private keys and now have now access to all these accounts. Like you own like like a Twitter account? >> Yeah, like a Twitter account, Substack account. And that's how we we ended up building AgentKit for anyone to go and launch their own agents and like let it own a Substack account. Like anyone can go and type out ideas. So so so this is an agent that can own things. Yes. And so this mapping uh Mike, I found like is when if you just for example go and ask an AI guy, okay, I want your agent to own a Twitter account. How would you do it, right?

And then you think about it and you're like, oh, I'm going to give my open claw, right? Like, you know, whatever agent that I'm running, I'm going to give it [clears throat] my Twitter credentials, right? That's what they would say. And you could say like, oh, that's a perfectly reasonable thing and many people do it and you know, it's a pretty common thing to do. But we believe that's not ownership. The reason it's not ownership is the person it's that person's credentials. It's the person's laptop. They've given it access. They can revoke access anytime. So that agent doesn't own the credentials. The person owns the credentials.

So yes, it's great like you can have a delegate for you. Like I said, for personal agents, you don't need all our fancy frameworks, right? So, you just need you just give your Open Claw access to your Twitter account, and that's plenty. But, when you think of an agent actually having ownership rights, when you think of how human ownership rights work, let's say Mike owns a house, right? So, registered in some New York county, you go and ask like, "Hey, you know, Mike owns this house." Then, you say, "Oh, they first identify you." They say like, "Oh, this is Mike." And here in this in the registry, it says only he has this house.

And so, he can only transact it, and he can only like, you know, sell it and control it. Like that's the the property rights layer that exists there. To actually give an agent like a property right is actually crazy. Like if you think about it, how do you give an agent only this agent can access this thing? Why? I mean, the agent's running somewhere, so it must be able to do whatever. No, no, no, no, no, it's not possible. Only the agent can access this thing. Really? Like Like think of a smart contract. Like that's what a smart contract does. It's just a piece of program that controls that money and like whatever credentials and assets that can live inside that smart contract.

And but smart contracts are not powerful enough to put an AI agent into it. Like I can't just take an AI agent. Like it's runs You need GPUs, you need like the languages don't work, none of that. And you can't slam it into like a blockchain. What we do is by running everything on Eigen Cloud, it's like a smart contract. We call it off-chain smart contracts. Right? So, Eigen So, you throw a piece of code on Eigen Cloud. It's like a smart contract just running off-chain. But it have the properties of the smart contract. It's unstoppable.

>> [snorts] >> It's verifiable. Just that code runs, no other code can run, and no node can shut it down. Like that's the output properties that we can we can guarantee. And then suddenly, the agent owns the property. Okay? And the agents own the account. This has not been possible. This has not been possible anywhere. It's a crazy new thing. And blockchains enable this. So, that's our main insight. And help me understand Okay, so I understand the idea that like to go back to the Twitter account example, you know, if you give your, you know, Open Claw bot you know, access to your Twitter account, that's not them fully owning it. Why is the idea so exciting for Eigen Cloud that these agents can

own it? What's the financial opportunity here? Yeah, that's a great, great, great question. Like when you think about um Once an agent can own interesting properties, whether it's a website, whether it's a Twitter account, whether it's an Instagram, the agent becomes an economically interesting actor. Right? And then you may want to own shares in that agent. Because it's not a agent coin, which has nothing to do with anything.

It's an agent that actually owns properties. It owns a website, it controls a Twitter account, it has its own business model. Now an agent is an exciting economic actor. It's an independent autonomous economic actor that in its own right becomes interesting. And once an agent owns stuff, the agent becomes ownable. Right? Then you want to own the agent that controls interesting properties. Just like you want to own Shell because Shell owns, you know, uh oil fields or whatever equipment are needed to process oil or the distribution.

Right? It's the same reason that this agent now owns very interesting things. Then, agents become like companies. Once an agent owns stuff, I know, if you go back 200 years and said, uh oh my god, we're going to be just companies are going to own more assets than most countries. You'd think I'm a lunatic. Right? Like I say like, "What What do you even mean?" Like And but that's where we are. Like hundreds of trillions of dollars are controlled by companies, not nation-states, right? [snorts] And So, in the history of humanity, that's a very crazy thing because it's a new structure.

It's all accumulating capital because it's a very efficient structure. Company can have long-term goals. It's durable beyond somebody's lifetime. A company can like, you know, um maintain property rights, appoint new people, and then keep moving. Right? So, it's a very powerful entity. And agents become the new companies. That's our thesis. Because once they have an agent and property rights, suddenly they become ownable. They're new kinds of digital assets that are highly valuable. Um I so very infrequently get to insert this random historical fact, but since we're talking about this, do you know the original concept of limited liability

in corporations back in Roman times? If I quiz you like, "What's the No, no, I don't know the Roman times. I only know the like the Dutch East India Company in Amsterdam. So, this was way This is far predated that. There was no concept of limited liability. They did the most companies actually were uh the way they farmed out taxes actually, they basically you had to there was like another province or something like that. Some rich individual up front paid the government, and then they went and went to the province, got to collect the amount of taxes, and the difference they could keep. Uh but the way that they structured these organizations uh because a lot of the people that uh took part in them were like senators and things like that, and they didn't want to be personally liable, they would actually put a slave in charge as the

nominal owner of the company. And because they're not uh they're legally property, the you were ultimately protected liability from the ultimate beneficial anyway. Isn't that like a super Yeah, that's so funny. You know, it's not Yeah. Uh-huh. No, I just had an interesting thought. It's crazy, right? Marc Andreessen uh on the a16z podcast two days back, he was saying that the last human job is to be a scapegoat. I thought so I thought you were going to say a sales guy. I I personally >> [laughter] >> think this is that sales guys are the last ones to ever get uh discriminated by No, BUT IT'S EXACTLY A Roman pattern, which is like the last human job is to be a scapegoat. Apparently he's got the first one as well. Yeah. So, they're going to appoint me to be their

front guy. So, like So, so Sherrod, help me understand too. So, okay, I very much understand like we have First So, first of all, there's going to be way more intelligence in the world than there's human intelligence. It would make sense that you need alternative structures to, you know, organize and support that well. Just in the interim, what is the difference? What actually the the lead into the the comment is like, if you have an AI agent, for all intents and purposes, they are labor for you, right? So, what is the difference from a economic value creation perspective of let's say I create a bot that owns a bunch of Instagram and Twitter accounts and creates a bunch of great content, and you sell advertisements against it. That First like, what is the economic difference of

an agent uh like fully owning those properties versus me controlling the agent owning those properties? Yeah, great. So, if So, maybe like Alex explain the stack that is needed for the agentic companies a little bit, and then like throw light on the difference between like a personal agent, right? Or even like you have a personal LLC, right? For which you have an agent that you employ to do your work, which I think is a big pattern we're seeing is the rise of the solopreneurs, right?

Single persons. They start an LLC. And like, "Oh, yeah." They even recently there's uh uh what is this um GLP company that made 400 million? I forgot them. Oh, is it the one which is worth a billion? Yeah, yeah. No, oh, it's uh it's a medical company. Yeah, it's a medical company selling GLP drugs. So, that's an example of a single LLC that they have, and they basically use agents to like scale and uh spam and whatever they do. Uh But the the the thing is So, so before we draw the comparison, maybe like going over the stack a little bit is helpful.

Because when you think of what a company is made of, you have the capital layer. Some people own shares. It could be a single person. It could be multiple people. Then you have the governance layer, right? The governance layer is, "Okay, how do these different people that own the thing and other people have economic and other governance rights into the underlying structure?" And then you have the execution layer. Execution layer is who actually takes this governance inputs and then actually does the work, right? Like go and plow the fields or do whatever work is needed to actually move the company forward. And then finally, you have a property rights layer, which says like the company owns properties, right? Whether it's a website, whether it's facebook.com, like, you know, Mr. Beast owns his YouTube channel, whatever, right? So,

you have these property rights that are owned by this company. So, company is this capital, governance, execution, and um and uh property rights. Like that's the four like layers that exist in a company. And the amazing thing is crypto has already made the capital and governance layer into software, right? Capital is just like coins moving around, tokens and shares and other things moving around. It's already just pure software, and crypto has already played a massive role in like transforming it.

Governance, like of course we have We We know you can encode like DAOs and other things. You can encode like complex governance rules, way more than, you know, regular company can. So, you have the governance layer. Governance layer can also be in software. The thing that has been missing in crypto is the strong binding on these two layers. For execution, when you had a token, you always depended on a team. Right? And what we saw like at many projects is the team got acquired, the team quit.

Teams like burnt out. Like whatever, right? So you have lots of problems that are quite good, right? They're there, but they're not there. Like you have all these problems. So the team couldn't be bound to the token in any kind of rigid manner. It's a weak binding. It's kind of there. Like it's reputation. Oh, you know, you moved away after 8 years and you feel like you robbed me. Like all these things because the contract is unclear. What is the relationship? There's no legal contract. There's only social and expectation. So like a token binding the execution layer, the team has been missing. And I I would argue that this is the number one problem for why crypto tokens have sucked.

Is the binding between the tokens and the uh team has been weak. The execution layer has been weak. Uh finally, the property rights. For a crypto project, you can only really have property rights for on-chain cash flows. And that basically like narrows you down to L1s and DeFi tokens. That's all that like has on-chain cash flows. And so what that does is restrict the category of ownable token objects in crypto. Like yeah, you can own L1s, you can own DeFi protocols. And there's like 10 of them. That's it.

So that's uh that is where we are in crypto today. And but the superpower that crypto created is made capital into software, made governance into software. What we do is by using AI agents, we make execution into software. Like you're not betting on, you know, Mike or Gudge to do anything. You're betting that this agent is going to run the company. Okay. By converting execution into software, you you make that layer into like undruggable, whatever. Like it's just inherent. It's programmed into the company.

And the ownership layer, you also program into software. And this is a the the the example to think of is Molt Book, right? Like Molt Book was this AI service which became very big. There was a coin. What was the coin? Dollar Molt? Yeah, Molt. I think it was launched on Base. Yes, that's right. This is the This is somebody launched it on Base. And like, you know, that the guy came and claimed his own like royalties from this coin. And um And then Molt got acquired by uh Meta. I think pretty favorable deal, right? Like eight figures, whatever, right? And what happened to the token?

It died. It had a pump. And then like, you know, but because the token has nothing to do with anything. But imagine the token owned the Molt Book website. Imagine the token owned the GitHub a GitHub repo where the Molt Book code is written. Imagine the token owned the IP to the word Molt Book or whatever, right? So if the token can actually own like some of the most important assets that belong to this, then suddenly for Zuck to acquire Molt Book, he needs to talk to the token holders because it's that's where the thing is. You can take the person away because but you've already assumed there's no person, just the agent that is actually running it.

So for the very first time, I think for the very first time, you have capital, governance, execution, ownership, everything is just software. This has never ex- I mean, other than in Bitcoin, it's never really existed before. Always had some other like oh, like the EF is going to push the whatever like the protocol forward, right? And we, you know, we have our own interactions We have it different ways. But that's like you always have some external actors or like you know, Eigen Labs has to push like Eigen token forward, right?

Like it's even top projects like Union Ave have like unclear relationships between the token and the team. So this is kind of like a core problem in crypto. What we do is by slamming the agent in the center and giving the property rights directly into the to the smart contract, it removes this problem. And then create for the very first time a software stack which is like we make the entire company into just pure software. It's a new kind of a firm. Whether we call it a company or entity or whatever other name that we can call it, right? Or a network, right?

But but that's what it is. It's just end-to-end just software. And when it's end-to-end just software, you can invest in it because you know end-to-end it's trustworthy, verifiable. So the trust and verifiability that we're attacking is actually the IPO market. Think of a company. Like if it's Mike's company, Mike's LLC, can he attract global capital? Should he be able to? Because you can change the agent anytime. You can change your accounts anytime. You can sell your account anytime. Unless we have all those like lawyers breathing down your neck and Deloitte auditing you every quarter, like we're not going to trust you.

I You know what also Sorry, go go. I had a thought. >> stack basically you just like have everything is programmatic, trustworthy, verifiable. So you can IPO like when you're $10,000. So there's not even a concept of a team or a founder. In fact, this is this is the craziest thing that like I think we are saying. We're not saying like a token launchpad or something. Oh, like it's, you know, founders are coming and launching coins. Like no, that's not what's happening. It's It's just an agent with its own properties building its own life, right?

Like and it's ownable because it has these in, you know, a private data and state. It has like, you know, con- owns certain properties building games, whatever other things on top of it. So that's the vision. And why it's different from a single person LLC is you don't have access to global capital because you have human slow legal forms of trust that take many years to IPO, that takes like millions of dollars a year of cost. What that does is if you're not a $500 million company, it's just not even under like, you know, consideration.

Right? And what happened in crypto with the ICO meta where people like, "Oh yeah, you know, I'm going to crowdfund on the internet." But without that trust and verifiability layer in the middle, they all just unravel, right? And we saw it in front of our eyes whether the Believe launchpad or whatever other experiments people did or the Molt Book which we just talked about. It just unravels because you don't have any strong rigid structure of ownership. And this is where I think you guys are doing a lot of work on creating token transparency, all these things, right? Like like other other things. This is a technical programmatic solution to a similar problem.

Yeah, I I mean, it's I spent a lot of time thinking about this at the moment. It would be a lot easier if you didn't have people in the loop. Uh so I envy some of your solution. Um guys, this has been a ton of fun. I'm sure this is going to spark a lot of questions. And hopefully, um you know, I can it remains I think at the forefront of kind of like most interesting, most futuristic, and you know, very dreaming positivity-oriented projects in crypto. So thanks for taking the time to to chat. >> Oh, absolutely. It'd be It'd be great if we can do like a uh 5 minutes where you can deep dive a little bit into like how Sovra or maybe Sovra [clears throat] would be a good example. Perfect. Yeah, that'd be good. Sounds good. Yeah. Okay, so this is Sovra. This is the first

sovereign agent which was built on Eigen Cloud and I think first ever. It has its own Twitter account. First sovereign agent by any chance? >> [laughter] >> Yeah, we define a sovereign agent as an agent that can own properties. And this is the first one that Gudge built. Mhm. Yes, so this has right now it has made over like 700 uh posts around like Can you click on one thing like >> Yeah, so this is uh There are like a bunch of different things. I can actually open the Twitter account where you can actually see why it's making those tweets. So Twitter recently published a release where we can't uh go and code it directly, but now there's like something is going and code tweeting and generating it. Yeah, it's doing the same thing for some of the

tweets. And it has been going on for last So it It picks up what is the most recent recent interesting things. And just like a New Yorker cartoonist might and then like draw a cartoon around it. Is that right? That's awesome. That's right. So like when you see things like this, Mike, like the thing that you wonder about is is this really an agent or is this Gudge masquerading as one? Right? Like I That's the That's the right question to like ask and think about. Um Yeah.

>> [snorts] >> Yeah. And >> So you So you could you see like what was going into the thought process of each of these images basically? That is right. So there's like this entire monologue which is which is just posted of like what I was thinking about. So right now it's thinking about like the board founder raises like 100 crores for for AI startup, things like that. So uh it can do that. There's also the agent also has its own earnings. So it does these auctions basically where you can pay it to go and create a cartoon for you. >> It's like product placements? >> Yes. Okay. It's exactly like product placements, but now anyone can get that cartoon here. And with this and through donations, it is able to get like more than $2,800.

So the the thing that, you know, even if you see the internal monologue, Gudge, like I'm pretty suspicious. Maybe you prompted this these dialogues, right? When when you see AI XBT or something, like I'm always wondering which TG group is it in and like what is it picking up from? Yeah. How do you like show Mike here that this is, you know, actually doing it on its own? >> That's a great question. So like right now I would just copy the tweet which I want to verify and paste it here. And here you go, we have a we have a signature from the agent we generated here and if you don't trust this website, you can go on >> So wait a minute, what is it showing?

It's showing that this content not censorship or classification whatever it's showing that that content is signed by this agent's cryptographic address. >> Exactly. Now my question to you would be like how the hell do I know this this is this agent's cryptographic address or you're controlling the script? Because you know to an outsider like you know Mike can have his open claw have a cryptographic address and say Yeah, so that's where that's where EigenCloud comes in where we have an official way of for you to go >> Is it now screenshotting the EigenCloud site? I think so, yes.

Mike, can you see that? Yeah. I can see. Awesome. Amazing. So you can go here, paste the message and the signature and you can verify oh it was generated from one of the addresses One of the addresses, but how do I know this address has anything to do with what? Yes, that's so basically EigenCloud gave you that address from from the mnemonic based on the code. Ah, so like this this is the EigenCloud dashboard which what does it show? Can you walk through So yeah, it shows that your code currently is verified so you can go to the GitHub and see like hey is it running? Which code Which code it's running. So you can actually see that inside EigenCompute device have a

container the specific GitHub this commit this Docker image is what is running. Exactly. So if I want I can just open that code up and then see like what is it saying? But this website also gives you more information, right? Which is inside that code is created this EVM address. And so that address not with anywhere else, it's like an address that is locked inside this container. Yep. And so you know that whenever you see the signature it's this coming from this agent. Exactly. But how do I know this agent doesn't take instructions from a secret master that or somebody else? Exactly. So that's where the comes in

maybe you're not actually like you wouldn't read the codes, that's where we provide an AI trust report which gives you an exact summary of what admin privileges we have like I can like right now you can see that we don't have anything except like gas sponsorships and like if there are any unnecessary bids things like that. This So this is basically an AI audit >> Yeah. of the container that is running. >> Exactly. >> So you can see what backdoors exist and what might be the problems and what So for the very first time you can check that an agent did something. Mike, this is so interesting that like it's such a simple claim oh my my agent did it is so easy for me to verify myself.

But to show somebody else that oh this agent did this thing is so hard. And so that's the problem that we solve. We solve the problem of agent identity, agent actions, agent credentials and by solving all those problems we make the agent be able to own things and to be then become investable itself. So that's that's what's going on. It's a very complex stack under the hood to do what we would think of as a very simple thing outside.

Guys, it's super cool. It's also a very difficult problem right to prove that you know not only that the code is running in the way that it's supposed to be running but the ownership is a super interesting new layer. So very cool stuff and where where can folks go to find out more about this or to Yeah, they can go to eigencloud.xyz and launch their own agent right now. So so there is a website where you can go launch a new Substack kind of agent yourself. It'll have its own Substack account and it'll just like keep writing articles on its own all these things. All right, well we'll link it in the show notes and folks if you want to create your own autonomous agent give it a give it a click. So then over there we need more New Yorker cartoons so guys appreciate you. We'll have to do

an update in in a couple months. Amazing. Thanks so much Mike, really appreciate it. Cheers guys.