Supercharging AI Possibilities: The Next Frontier in Crypto | Sreeram Kannan - EigenLayer

[Music] awesome and we are back everybody uh this next speaker uh is somebody that many of you guys probably know because he's been building one of the largest protocols in all of crypto the last few years uh the first time I actually met this gentleman uh was a few years ago back before Ian lay had started and I didn't understand what it was and so I ignored it and I ignored him in my telegram DMS and that was definitely a mistake and so I recommend you guys learn from my mistakes you guys listen up to this next top because next up we've got shiram talking about supercharging AI possibilities the next Frontier in crypto thank you everybody hey hey hey hi everybody super excited to be here I'm Shri Ram founder of Ian lir founder and CEO at Igan Labs which is the company building the Igan L protocol today I'm going to be talking about what is the role of crypto X aai is this really a thing is this a scam what is going on what is the fundamental Theses where are we today where we can go on what role ethereum and Aon can play play in this okay so first I want to open with is this really a thing is this crypto AI really a thing is there a thesis for why we should be taking this very seriously on the one side we have crypto we're all building here has an apps problem you know we say where are the users where are the apps what is the utility of this whole thing is it just meme coins financial nalism and speculation is there something more crypto has an apps problem AI on the other hand people building very interesting new intelligences has a trust problem how do you trust these things how do you trust what this AI agent says how do you trust this whole Paradigm so I I think there is a match made in heaven between crypto and AI crypto can bring trust AI can bring intelligence and applications one way to think about it is Agents of the new apps if agents of the new apps you know bringing agents and crypto together can solve each of those problems another way of thinking about it is a very technological lens think about what is AI general intelligence on the other hand what is crypto crypto is Unstoppable computers if you can take take Unstoppable computers and mix and match them with artificial intelligence artificial general intelligence you will get Unstoppable artificial intelligence which is a crazy interesting new novel combination so I think we should take this seriously is Unstoppable intelligence going to work in human interest how do we make sure that these systems are useful and productive the other aspect of it is Agi and blockchains blockchains are verifiable computers can we build verifiable intelligences trustworthy verifiable intelligences okay that's the high level Vision I think there is something here both at the technical technological level as well as at a product level right crypto market cap is $2 trillion AI market cap is $10 trillion when these meet you can actually get real ignition but where are we today why do we see so much skepticism among the crypto people of the agent thesis because was so first let's take this example what do we see today when we see like onchain agents showing up like what are these onchain agents basically an AI agent hooked up with a crypto wallet you you give an ethereum wallet to like an on to an agent and say oh this is the onchain agent but really it's not that different from connecting an AI agent with a web to hook like a stripe API okay is this what we're here to build a centralized agent running interacting with ethereum is that the interesting thing I don't think so this is neither Unstoppable you can shut down this server and then the agent dies not verifiable you don't know what this dang thing is going to do in fact it has a private key if you try to give money to the agent to trade or something you can lose your money this is custodial technology we're not here to build that definitely not here in ethereum okay so what can we do is this even theoretically possible so the first thing to observe is smart contracts are unstoppable and verifiable this is really ethereum's contribution to humanity how do you build Unstoppable verifiable code take a code write it throw it on top of ethereum runs on its own runs as predicted that's why Ian lir you know the host was mentioning today that he met me before the first time I came to eat Denver 3 years back I didn't even know you need to have a ticket to come into the main event I was standing outside like you know tracking down people with you know looking at their t-shirts and seeing you know which company they are from and seeing if they will listen to me and today we here one of the main sponsors and one of the biggest protocols on ethereum this is possible because ethereum's Unstoppable verifiable and we can just build a protocol throw it on top of ethereum borrow trust and scale it to tens of billions of dollars it's not possible without that can we actually accelerate that by saying smart contracts actually become smart they don't have much intelligence you can put code inside it but it's not that you can't put expressive complex code in it but theoretically you can put a little bit of AI inference code in a smart contract and theoretically it's possible to build Unstoppable verifiable AI another way of thinking about it smart contracts are the minimum viable agents right smart contracts they just run on their own and they do stuff and so like can we actually make them really intelligent they're just not practical today because you can't write you know AI inside this but here is our realization and why we are super excited at igen layer Igan lay makes autonomous verifiable AI practical not just possible but practical I'm talking about 2025 you can have Unstoppable verifiable AI agents okay how do we build this so the first thing about I layer is what do you build on top of IG everybody knows IG lay may be staking reaking something like that but reaking is just the starting point of IG layer the core idea of IG layer is to be able to build the verifiable cloud you know you should be able to deploy applications on top of this cloud and be able to use Services we call these Services autonomous verifiable Services avss they are the ones that absorb the trust from the eat staking and other staking that happens on aen layer and get these Services which are complex infrastructure services that can serve any application so why do we call them a avss autonomous because they run on their own you can actually continue running this service you don't need to trust anybody because anybody can put up more e and continue running these Services they are verifiable because if the stakers go Rogue and try to attack the protocol they're going to lose their deposit so that makes them verifiable okay so now this that's the Paradigm the Paradigm is anybody can come and run this service run it in a way that's Unstoppable autonomous and verifiable now you can start offloading intelligence you know you can think of like decomposing so you look at it here there is a smart contract of the AI agent sitting inside an ethereum smart contract and this smart contract unlike you know agents today which have a private key and they can rug you if you lose your private key instead these smart contracts of course don't have any private Keys what you do is you say like hey here is a smart contract of an AI agent the smart contract takes actions if they are certified by this AVS the smart contract takes actions if they're certified by the Zs and you move the inference offchain but it is still verifiable onchain this is a massive transformation now we can think of this factorization you can have an onchain contract on ethereum which just does the basic bookkeeping maintain assets and so on offchain you run this ABS runs all the time autonomously and verifiably okay what can we do no you can just run inference as an AVS verifiability because of the slash ability byon layer and data availability when an agent runs it produces a lot of data you need to have this data be public you need to have this data be public for anybody to verify that the agent did what it said it did so this is called Data availability in ethereum and I da is the most scalable data availability layer 50 times what ethereum's blob store can do today 10x any other data availability layers you may have heard about in the market okay so igen da is a hyperscale data availability layer what we did we didn't invent this technology for scaling da we took donk Rod's Don Shing which is on the ethereum road map just built it out of protocol you know we discuss a lot here you know in E Denver the main topic that I see is how much can we accelerate ethereum is ethereum going too slow no you can build any service imaginable as an AVS today and accelerate ethereum this is my call to all of you go build new interesting things as avss including a that's what this talks about there are other ways to do verifiability you know using you you know everybody re-executes everything in fact you see like people say oh Solana is very scalable we don't need this ethereum modular road map all this stuff is it really can you run a hundreds of thousands of AI agents on salana you cannot you cannot you need hyperscaling we need hyperscaling you cannot achieve hyperscaling when you have redundant execution every node does the same thing in salana in ethereum today but in the roller per what we are planning for is to not have every no and everything already using I and I layer you can actually achieve it okay when you think about these AI agents an agent is much more than just AI inference what is an AI agent an AI agent is something that can actually take actions not just be intelligent but be able to take complex actions when you think of these AI agents when you want to build a verifiable AI agent you need to verify not just the model but every dependency that it has if you don't verify these dependencies you really don't get a verifiable agent this this we know right for example we say a is a verifiable smart contract but if it's dependency an oracle is not verifiable then of course the smart contract is not verifiable because you can get rugged by the worst of the things that you trust so if you want to verify any agent you have to verify the orchestration the model the tools this is the agent runtime picture from anthropic and what we've done is think about how to actually make it verifiable okay now that you have this verifiability what can we do with it you can actually mix and match components to build really really really powerful agents I said really three times I mean not just for effect but I really mean it it so what do I mean by really really really powerful agents and they're verifiable they have a brain okay what what is the brain of an agent you know we think and like come up with Concepts it's about inference you're running predicting what needs to be done it's about training you need to train update your model it's about collecting data see it says verifiable in each one of these why because if an agent trains and updates itself and it's not verifiably updating itself you can get rugged so you want to make sure every single step here is verifiable verifiable inference verifiable training verifiable data verifiable benchmarks you want to know whether grock 3 is better than chat GPT how do you test this you want to test an agent like an operator you know from open a is it better than some other thing how do you know these benchmarks is llama 4 better than llama 3.2 how do you test these things verifiable benchmarks GPU markets this is the brain but brain is not enough you need sensing you need the ability to understand what's going on in the world you need a mechanism to understand what's going on on Twitter what's going on inside your Uber app what's going on in the real world at in here at e Denver you need real world oracles you need application oracles you need internet oracles all of these every single one of these boxes already being built as avss autonomous verifiable services on agen we have 100 plus plus avss being built in 50 plus categories each of them aore new infrastructure service so what we saw when we enter into the space was anybody who's inventing new infrastructure had to build their own layer one so you know sometimes we joke here in ethereum about allt ones and like and so on but there is a necessity when you want to invent something new you have no place to express it you have to find a way to express it and that's what the whole alt L1 where but what we've done in igen layer is let innovators invent something which is as fundamental as a new layer one but don't have to reinvent the wheel on all the other things just build the new thing that you're building is it inference that you're building is it training that you're building is it a GPU Market that you're building is it an internet Oracle you can get you can build all of these things on a common security substrate that's what igl does what you can do is mix and match all of them on igl now this agent which has has the ability to perceive and has the brain also has the ability to satisfy intents normally when you think of intents you say like oh maybe humans have intents and AI will satisfy it but it could also be the other way around Ai and agents have intents and we satisfy it so it can put out bounties that we go and satisfy it can have privacy layers trusted execution envirment fully home Maric encryption multiparty computation it can take take actions triggered on events co-processors data storage the scope of what verifiable agents is is enormous the scope of what verifiable agents is is enormous you can just mix and match these services to build verifiable agents you also can do ID like imagine if an agent is a person it needs to verify the counterparty that it's dealing with you need proof of human proof of device proof of agent agent reputation each one of these new and interesting layers can be built on I layer this is one of the reasons we think of Ian layer as the verifiable commitment Cloud what is the verifiable commitment Cloud you know many of you like I said maybe heard of I lay as like staking right stakers and operators come together and now you can build these avss on top and these avss serve applications there's also in the middle we call Primitives you know you have commitments verifiable Services verifiable apps and these apps just come in and mix and match these avss and like I said agent is just the new app so if agent is the new app agents can come and compose these services to build verifiable agents we also have the core Primitives like I pointed out when you want to run an agent you want to publish ta publish data to a common verifiable da that's what Igan da is we also have other very powerful features like ion forking which I'm not talking about today but these perimeters together give the expressivity for services to build new and interesting things like these AI agents um okay so if we look at this picture so we have many categories of abss that agents can compose to build new and interesting things what are being built today so you know this is a category map of avss autonomous verifiable services that can be built on top of aen layer they're built on top of aen layer and some of them are being built you see names here some of them are not yet being built so it's a call to action all of you brilliant developers here you know people joke that oh ethereum is infra land without these powerful infrastructure you cannot build you cannot express the level of agents that we need to build Unstoppable verifiable agents so several Cate here many many interesting players this is not Theory land many of these avas is already being built and deployed on aan L today you know here's a a very small sampling of what these services are layer lens building AI benchmarks I can Benchmark agents open Ledger building data markets you can buy data you can sell your data to these agents which can then pay you for your contribution verifiable models being built by neural fabric even triggered actions sentient you know a huge project we're collaborating with them on building an AI justice system you know when you think of smart contracts you think of what is expressable in simple code but when you want to build complex contracts you need adjudication for it witness real world location I can actually attest you could there's an app by witness called Infinity watch you can go in and prove that you're in this location to a DEC centralized group of 6,000 nodes running around the world running today on agir silence you can actually take your secret key and give it to a network which gives you guarantees of privacy opacity when a tool call is being made you are calling door Dash you're calling open AI how do you verify these API calls you can do it today on opacity so the category of what is being built is mindboggling and very few people understand where this is going it's not just that these Unstoppable wable agents are interesting on their own accord they also accelerate the foundations of crypto how do you think about intents how do you think about smart contracts how do you think about governance how do you think about decentralized signs in the presence of AI it just transforms the entire landscape I want to end this by saying what we're building at Igan layer is the verifiable commitment Cloud a cloud where you can make commitments that are verifiable and enforceable by code and so the project is much much much broader than reaking and I think it can inject a new Lee of energy into the ethereum uh ecosystem today with that i' would like to thank you and if you want to find more information please you know take a picture of these QR codes and I'll leave you with this question with fellow inventors here what are you going to inv event what are you going to build on aan L today thank you so much [Applause]