Why 2025 is The Year Of The Agents | Illia Polosukhin - NEAR Protocol

okay everybody agents is too good to move on from so we're going to can keep it rolling um this next segment is called why 2025 is the year of Agents I don't think our speaker needs any introduction he's considered one of the founding fathers of modern AI um Co yep let's give him some claps already let's do it make him feel comfortable co-founder of NE protocol and n i Ilia puken thanks okay just you know few technical difficulties as usual but thank you for every everyone for being here not at some happy hour already um so let's talk about why 2025 is year of Agents if you haven't you know already heard 50 AI speeches this week so for those who are not familiar with me uh I before diving into blockchain I was AI researcher I worked at Google research across question answering machine learning and natural language understanding and I was on a team that developed the original Transformer architecture that's now used across effectively every machine learning you're using now we started as AI startup uh we were teaching machines to code and uh we were actually coordinating a lot of people to really uh collect training data and reward them for the for their work and we were missing some way of coordinating them some way of paying them globally and so that's kind of how we got to blockchain we haven't found anything that would satisfy our needs and end up building our own so that's how near protocol was born now for past over a year we've been working uh a lot more on AI side and and bringing this two together now I think it's important to understand that AI is a final platform shift what you know started as just you know fun tool you can use with uh natural language to me always seemed like the way we're going to interact with Computing is going to completely change if computers are able to code if computers are able to write programs themselves they're able to effectively execute actions and make decisions there you don't actually need you know layers and layers of software you don't even need layers and layers of companies and decision makers who would delegate variety of decisions and we see this not even kind of in business we see this in science as well right more and more uh actual research is done with AI tooling and AI researchers are getting rewarded um with Nobel prizes and so the paradigm shift that's happening is from this aggregator dominated internet where we effectively interfacing through Google through Amazon through Instagram through uh door Dash that are really monetizing our attention right and trying to capture that and trying to maximize this and then they routing us to different Silo apps where we need to effectively give more information and then they trying to monetize on it we're already transitioning starting to to this way of Agents interacting with websites right we have operator we have few other things where they starting to interact uh on your behalf uh they starting to act they're starting to make transactions but if you look after that you can imagine an internet where there's no websites there's no apps it's really just uh agents talking to agents and achieving the outcome you're desired Without Really um having anything in the middle right actually making decision executing on it so you know being chill guy just using AI instead of apps so that's really where we're going now the challenge uh so before the challenge just to give you a visual uh you can imagine your assistant on the background just running constantly making decisions and executing them directly right booking flights getting con metal concerts uh as as soon as they open up as well as rebalancing your uh stable coin pools between different uh borrowing protocols based on the you know what's the best trade so this is coming this is not that far away now the challenge is everybody right now uses one of the few API providers and this API providers have effectively ultimate control on what how you can use them and what you can do so this close Source companies are effectively can turn off access to anything you're using right now they can turn off the whole country and if we assume the kind of the whole internet will run on this new way where AIS are interacting they can effectively just you know turn off multiple digits of percentage of GDP uh from access to to the AI technology now cool thing is we have open source AI maybe not AV system but uh we have been you know people been open to share their models uh with the ecosystem now the challenge is it's highly unsustainable right uh companies that are actually doing this are effectively burning so much cash to train this models and do research that they're not able to make money back when they open sources right because nobody is paying them back for for the work they've done so we need a different model and so we're calling this model user owned AI right this is where instead of AI optimizing for some profit of some company or being kind of you know freely shared with everyone and hoping uh there's some economic model later we have an AI that optimize for well-being of individual users and their economic success and there's incentive alignment and structure that really Powers this system where everybody the developers the researchers users the services can actually get rewarded from doing that so that's what we call in user on AI now let's think through like what agents that uh we need to have uh must kind of what what they should provide right so it's important that as we're sharing more and more of our information and data that this agents are confidential right right now this agents the assistants will not be able to help you if they don't have all of the context of what you're doing access to email calendar medical records banking crypto but you won't be I mean especially in this crowd we will not be sharing this information with you know centralized companies you also want them to be verifiable you don't want something that uh can effectively run arbitrary model can change underneath effectively can censor or modify the outcome you want to be monetizable so you want people who building them be able to make money you want them to discover each other right hey how do I know how to do this well I don't need to have all the tools implemented I don't need to have all the apis I can just reach out to the agent that specializes in that and so we need a standard protocol how they can interact pay each other exchange data and information obviously we want them to transa on crypto but you know world still runs on Fiat so we need to support both and we need a way to access Global liquidity even Beyond web 3 traditional Finance you know uh invoice financing Etc so we really want this agents to be on your side but then plugged in into the whole ecosystem of agents and uh services and goods now how does apply to blockchain well one of the thing is is that AI is effectively the final state of chain abstraction right if you have been you know uh kind of last year this Denver we talked a lot about chain obstruction kind of how that is the next wave uh that you know we saw play out last year but now we already seeing a way to interface with kind of variety of tools in crypto just directly from natural language and kind of Rich interactions you can generate uh with AI so that's really what's coming next uh a simplified interface now on the other side we introduced this New Concept of how you can build for web 3 right if you have a verifiable confidential agent you can now build a lot more that you could before with just smart contracts right you can build something that can read data from Twitter run AI model uh you know maintain a local database that completely confidential and then still push all of this execution back to a Smart contract and transact across different chains right so check out the shade agents which is kind of this concept of really programmability that you can bring to your applications uh with ease now now just as a high level you can think of like AI serving almost as a front end to all of the rich interactions we're creating and chain obstructions in the middle right it's an the interface that really plugs in all bits and pieces with intents change signatures Etc and you have a scalable blockchains on the back end really facilitating the value movement and uh asset tracking like actually being a ledger so that's kind of the inter the stack that is uh created so that's a lot of talking let's actually look at some example right so this is an example where uh you have a specialized agent that knows how to uh that knows how to shop for audio equipment so this is a very uh kind of you know specialized agent maybe has context maybe has additional knowledge now you as a user will not be using individual agents you should be interacting with your assistant who will be routing you depending on the request you're doing to a specialized agent under the hood right so for example you want to shop for a uh headphones and your AG your assistant first you know will try to answer the question and provide you some context so you actually know where you're going now when you give it like hey actually uh I want for example cheaper headphones the next thing it does it actually goes and reaches out and finds what are the different specialized agents who can do do this request and by talking to that agent directly it actually gets a number of uh options which then user can choose from so this is a rich interface that gets propagated back to the user user chooses now you know the the sound Sage agent requests to fill in the address where to deliver and through that uh kind of you know communication that happening on the background you don't really see the actual interactions between your assistant and sound Sage you have this process of effectively purchase executed and so now it requested the payment right so again a very structured interaction uh for requesting payment so you can pay as your crypto wallet and obviously through the kind of intent liquidity you can pay with any asset uh from any chain uh for for this uh for this purchase so kind of this is the idea you have effectively UI assistant that's an an interface to the rest of the internet now this is you know interesting uh but what's more important is how this go going forward and specifically how this is coming to the users and so uh we have sweat wallet one of our partners who already integrated this into their own application uh kind of and they're going to be rolling this out uh to the test users where you're able to initiate their uh assistant agent called Mia who knows both everything about you about the context on your phone as well as about sweatcoin and can go and navigate to other agents in the ecosystem right so you have now ai system that is Rich and contextual that can take actions both within your wallet and even within your phone to really interact kind of richly with the with uh the rest of the internet so the idea is with sweat wallet you're already getting you know millions of users who going to be starting to use your agents so as you build more agents right they're getting more and more usage from this this assistant becomes more and more useful so you have a flywheel where more agents mean more use cases more use cases means more users are spending more time with this assistant the assistant becoming smarter as well so this is kind of the idea of this of the system that we're building now how big is this transformation right this is just like a conservative assessment of this agent Network becoming a real thing where you're effectively removing all of the digital ads all of the aggregators all the payment processing fees because we're switching to the uh crypto rails and you have your AI assistant effectively as interface that navigates across internet you obviously can do all the financial services as well so the rich defi ecosystem that's been built in web 3 right fully chain abstracted really allowing user to be kind of on one side in control on the other hand have an AI assistant really helping them and running in a background to execute this action now this is not you know a single company a single uh even ecosystem effort and reality is we need a whole uh group of people coming together and really to build this out and so I'm excited today to announce a open Agent Alliance uh we really want to bring together all of the different part Partners to build together towards this Vision where AI assistants and agents are working together in the opening ecosystem really building this user own AI now this is not just a bunch of logos everybody's building different component of the ecosystem everything from Hardware from inference uh we have you know obviously agent Frameworks that are enabling this we have folks like f who building trusted execution environments to ensure confidentiality we have sweat and B and doubleganger who offering their assistant application to uh really bring users into this ecosystem so um let's have folks on on stage and uh let's introduce this [Applause] Alliance hello hello all right we have this uh amazing crew coming in and uh yeah so I'm really excited I mean we're here to you know introduce this idea and really excited to have all of you as well building and contributing as [Applause] well all right thank you very much guys it's all about photos man what what it's all about marketing at this point you build a big Vision you bring marketing and then you know obviously we all Builders here so we actually build real technology but now now it's all about marketing so I'm really excited this is my talk and uh I hope everyone can try this out and start uh contributing as well as I'm sure all of you are Builders and if you're not we actually have a no code agent Builder platform as well which you can use to uh to build your own agents thank you [Music] [Applause]