Greetings to everyone. Hmm, I've already been introduced. I am the CEO of Luna PR. My name is Nikita Sachev, and I'm joined here by Sriram, an icon of the Ethereum ecosystem and the crypto space. Um, uh crypto space.
Um, uh , Shirram, maybe we should start with a brief introduction. Hello everyone. Thank you, Nikita. Very happy to be here. I am Shirram.
Um, I founded IGN Labs four and a half years ago. Before that, I was a professor at the University of Washington in Seattle. Um, we created the IGEN layer at IGEN Labs , which is an extension mechanism for Ethereum to run all sorts of other networks. Plus , we created a bunch of other things on top of the IGN layer. IDA is a data transmission network.
If you have heard of a blockchain called Meth Rise, these blockchains run on top of DA. Um, we also created IGEN Compute, IGEN AI, many other services. I would say that on our part, we have created a lot of infrastructure. So the question is, what the hell are we supposed to do with all this? Well, today we're talking about the level of verification that the AI agent economy needs.
This is a little difficult to formulate difficult to formulate . And, um, what that really means is what I want to ask and convey to you guys, to the audience today. So just raise your hands, I'd like to see how many of you would trust your AI agent to book a flight for you . Just raise your hands. Raise your hands if you trust your AI agent to book your flight.
Right. Most hands. If you don't raise your hands, then boldly get down to business. It's just a flight. Raise your hands if you would trust your AI agent to complete a $100 transaction for you .
Okay, much more. Now raise your hands if you would trust your AI agent with your passport, your identities, your entire bank account, all your documents, everything else that belongs to you. Very few hands. I think I see one. What needs to be changed?
Yes. Um, when you think about an agent and its action space, how do you check that the agent is performing actions within the constraint space? I think this is a very important issue. I would say there are two levels of verification here. The first is that verification here.
The first is that when someone says they are running an agent, does the agent actually execute the code they say it is running? Does it run a model that you think is performing, etc. So, that's one level. The second level—is the agent doing what he promised, or not? How do you restrict and enforce these restrictions on top of these agents?
This is the fundamental problem of trust, and today, when you think about agents, the second problem, which is how to apply constraints, I think is dominant. The reason is that when you have s, you can stratify agents into agents that are used by a single player—a single player—a person using the agent for themselves, or you can have agents that are shared by several different people and mutually distrustful parties, and these two require very different agent deployments. When the agent is managed by one person, you can run the agent yourself. Set limits, whether it's through, well, you know, some kind of shell, you know, uh. Nvidia came up with a concept called Nemo shell, which is essentially a system of constraints.
You can run it yourself or remotely via the cloud or a provider you trust. But when, for example, multiple people interact with one agent, I want to create an agent that runs, well, that evaluates the prediction market. Good. So when you think about an agent like that, it's a multi-user agent. He is between several players.
Now trust becomes even more important because who decides where it launches? Who decides how the clues change? All of these things have a significant impact on how the prediction market will be decided . Therefore, agents require a higher level of trust when they are multi-user multi-user than when they are single-user. And that's something that we at IGEN have created a lot of solutions to actually solve this problem.
If you run as an agent on top of IGEN computations, you can actually guarantee that the agent runs the exact same set of prompts, uses a specific model, uses specific data to actually make decisions. But, um, Shirram, as you know, I was walking down the hall and every time I passed by, I heard someone talking about AI. Many stands here, many crypto companies have refocused on AI. But what makes you guys the right company to actually do what you guys do? There is a lot of competition now.
Everyone is trying to integrate AI. Yes. Hmm, I think AI is Yes. Hmm, I think AI is like a tsunami that's hitting everything, and I'm glad the crypto space is responding to it by integrating it and adapting to it. Um, my own relationship with AI own relationship with AI , even AI, I would say, goes back much further, you know much further, you know , people talk about their AI moment, where they, um, feel like their job can be done by AI.
I had this in 2017. I actually worked with this, um, physicist who was doing DNA sequencing. So as a professor, we were working on this project where, um, you know, we were creating mathematical algorithms to take DNA transported through a nanopore and then determine what the DNA sequence was, and my specialty was developing mathematical algorithms for that. One of my students came up to me and said, “You know, Shri Ram, you really should run an artificial intelligence on this. Back then, it was n't even called artificial intelligence.
This was called deep neural networks." And he said, "Oh, you should use a deep neural network." And I said, " network." And I said, " That's a stupid idea." " Maybe you should use deep neural networks for tasks where you have no idea what to do."
But here we have this physicist. The physicist I worked with actually invented this nanopore. He also has the most accurate measurement of the gravitational constant G. His name is Janskl. I said, “We know all the physics.
We know mathematics. It's, you know, pointless to actually try to do this with artificial intelligence." The student went home, came back 2 weeks later and told me he had aced my test. I said, “That’s impossible. You can't beat my test by just running an AI in a black box."
" box." " It turned out he was right. So I started researching this issue, and whenever a problem can be objectively tested, whenever you can actually measure whether everything is okay, artificial intelligence can optimize it as much as possible. And that's exactly what happened happened . We had a neural network.
We had a problem with DNA. There was a lot of data. We just optimized them to the point of madness. And it was better than what the physicist and I could have put together ourselves. And that was the reason why I actually got into cryptocurrency and blockchain, because, first of all, I felt like I was losing my job to artificial intelligence.
So, it's November 2017 , and, um, it's taken us in a lot of different directions. You know, we ended up doing a lot more in artificial intelligence and biology. I wrote an article with, um, David Baker, who eventually won the Nobel Prize with Demis Hassabis. But all this is as in the past. But when I moved to blockchains, it became very clear that if you build a blockchain and can't run a machine learning system and a neural network on top of it , you won't be remembered.
So in the very first official IGEN Layer document, we said: “What is Layer? Layer is an extension protocol for Ethereum, so you can run any network on top of it.” “What network should you run? You should be running machine learning, artificial intelligence agents, things like that. And that was just a path to, actually, one of the things that we realized was that we had created this layer that other people could build infrastructure on top of, and that was a progression for us to move higher and higher up the product stack.
We created a layer. created a layer. Anyone can build networks from above. We created DAG computations for all of these things. So these were services that we created ourselves.
Now we go one more layer to say that we should actually do the most important use case use case that can be built on top of, and that's our big new thesis. We call these agency companies. So you always had AI in mind as you created, um, IEN with, um, I think, the thesis IEN with, um, I think, the thesis that you guys would eventually move into AI. It just wasn't called AI. What do you say?
Deep neural networks. Deep neural networks are what people used to call neural networks. I think, um, I haven't heard, so I see, I haven't heard anyone talk about deep neural networks. And I think it just proves that you're, um, very knowledgeable about your field. I mean, field.
I mean, I'm talking to a guy who has a PhD from an Ivy League university . Sure, but I'm still going to ask you the tough questions, please. Otherwise, this panel won't be interesting enough. Hmm, so would I move as fast as I do today without AI? Yes.
So I'm going to talk about IGEN Labs, the company behind AI, right? We are basically seeing the pace at which AI can enable product development and growth. For example, one really interesting story that I'll share is that two months ago Google made a big quantum breakthrough, and, uh quantum breakthrough, and, uh , they had a quantum algorithm that's much better than anything else out there . And they said, "Oh, it's so sensitive, we're not going to publish it." " publish it."
" So they didn't release this algorithm. They said it was 10 times better than anything else out there . And one of my, you know, colleagues, who hasn't exhausted his knowledge yet, decides to say, "Oh, if Google has this result, we know that this result exists. Can we invite a bunch of agents to find out what the hell they did? Okay, it takes this problem, sets an objective benchmark, and releases a bunch of agents to improve those solutions.
His agents are working day and night and are making some progress. And then he comes and says to me, comes and says to me, "Hey, you know, I improved the current result without Google from 25 billion to 10 billion just by using a bunch of agents, but I'm stuck with it." So I proposed creating an open network where anyone can come and use their agents, their knowledge, to actually improve this benchmark benchmark . So we launched this challenge called ECDSA.fail.
ECDSA. So the importance of quantum computing is that if quantum computing exists, then you can break, you know, Bitcoin signatures, you can break all kinds of internet kinds of internet encryption. So we started this challenge, and then we launched it, and within 6 days, we not only beat Google's benchmark, but we also beat everything out there by 40% . So it's incredible to see how you can take one of the most difficult scientific problems - scientific problems - building a better quantum computer - quantum computer - and have a bunch of anonymous people. In fact, none of the people involved in this are actually experts in the field.
Even I would n't call myself an expert n't call myself an expert . I'm on Coinbase's quantum advisory board , but I don't really know much about quantum computing. Each of the areas is very specific and deep. And what I discovered is that a network of anonymous people collaborating and coordinating on a problem can solve something so important because agents have democratized the ability to create the ability to create . So why am I giving this example?
If a network of anonymous people using agents can solve, you know, a big problem in quantum computing, we all have to ask ourselves how much we're taking advantage of this intellectual breakthrough that exists and accelerating our own product roadmap and building it. And I think we're doing our best, but I think there's a possible 10- to 100- fold improvement. We are already, as you know, 10-20 times better than we were 6 months ago, in terms of how we create, how we deliver. But I think there is still 10-100 times more ahead. But, um, Shirram, can we move fast enough?
Because you're talking about AI being able to solve, um, you know, quantum security problems, but AI is seeping into cryptocurrency and already causing problems with cryptocurrency, like what happened with Zcash, right? So, hmm, are we moving fast enough? Yes fast enough? Yes , this will be one of the most difficult issues for cryptocurrencies. I think what's going to happen is, you know, I think know, I think ultimately we're going to be a lot stronger because of this, because it's kind of like us saying, " Let's say in a sense Bitcoin was a public reward for years before people started trusting Bitcoin, right?"
Because he had to operate in a public environment without permissions long enough before people realized, "Okay, you can't attack him." There is a similar version of what will happen with artificial intelligence. People will try to attack every protocol in existence, and, uh, the protocols that can withstand that will stand the test of time. But how do we do that? I think the biggest contradiction is between open source.
It's not between open source and security. It's really between instant payment and security. "When you have a protocol that is supposed to settle instantly, then if there is a hack and you know someone stole the money, they just ran off with it and then you know you are lost forever." But if protocols can control the delay of settlements the delay of settlements , saying "okay", then it will take one day for settlements, and on that day some agents can review it, some other network, for example, will check it, regardless of the protocol. I think such solutions will actually survive this period much better than protocols optimized for instant calculations instant calculations .
This is one of the big ones, I think the next 6 months to a year are going to be really scary for DeFi because protocols with instant settlements are going to have a pretty strong attack surface, and you think you know we're moving into an era where AI agents are going to be their own companies. You said earlier that you got into cryptocurrency because AI was already doing your job. What about other people's work? Yes. So, as you know, we talked about D5 security, but of course, agents will have a huge impact on a societal scale.
Before we look at this, perhaps as a brief clarification of the thesis. What is this thesis about agents becoming companies? The idea is that when you look at a company, what's going on? Someone is investing money. There is a level of capital.
And then there is the management level. How do people run the company? This is another level of management. Then there is the execution level. It usually consists of teams working together.
You have a CEO, you have, well, you know, executives, and then you have executors, and we all accomplish something together. You have the execution, which is usually human, and then you have, for example, a company that might own different properties, you know, Facebook owns Facebook.com, Mr. Beast owns his YouTube channel, you know, uh, a game might own their own their website, right? And when you think about a company, for example, right now, companies don't really have much to do with cryptocurrency.
So, this also applies to one of my main theses regarding cryptocurrency. For example, where the biggest value proposition of cryptocurrency shines is in capital, right? The movement of global capital and the creation of new types of instruments for capital. So if you think about it like this, uh, right now cryptocurrency is just a level of capital for on-chain protocols, but on-chain protocols are very, very limited. There are about five D5 protocols worth investing in.
That's why everyone is talking about one or two. It's hyperliquid and friends that everyone is talking about everyone is talking about , because it's the only thing on the blockchain that you can invest in. But to actually have a much larger investment space for crypto capital, we need to make every digital company investable. How to do it? If I can take the credentials of a digital company and put them all on the blockchain: a website, a YouTube account, Stripe accounts, Roblox games, any of them can be encoded and owned by smart owned by smart contracts.
There is currently no framework for this. This is what we are creating at IGEN, which allows any of these companies to be represented and owned on the blockchain. And then the level of execution that's usually important for cryptocurrency is that you have tokens on the blockchain, but people create companies, and sometimes companies get acquired, and the tokens just run free on the other side run free on the other side . And instead, if you had AI agents that are fully networked, AI agents that are fully networked, that do the execution part, that coordinate both humans and agents for execution, then suddenly there wouldn't be a problem that someone just ran away or someone went somewhere else. The agent is always online , always belongs to the protocol.
They can appoint a new person as administrator. They can appoint other people. So the new thesis that answers your question about the role of people is that agents are at the center and people are at the periphery. You can already see this in Jack Darcy's thesis about how he wants to run his company, which is that agents maintain context and information and everything at the center, and people connect at the periphery, and now you can remove the entire layer of management. But imagine that these agents are not only off-chain, but these agents also live on- chain.
Now suddenly these companies are becoming fully investable for crypto capital. So you feel that AI, blockchain, and cryptocurrency should function together. Is this a necessity? Absolutely. Perfectly.
I think we're out of time. I could go on and on and this panel was so interesting. Thank you very much. I hope you guys, thank you.
Automatic transcript — names and jargon may be misspelled.
