AI Agents will Empower DeFi in 2025

Transcript

probably a common theme that you've heard over the past year or two is that we're trying to solve ux if we solve ux we'll bring the next billion users on board but agents and Bots don't care about ux we've built the perfect infrastructure for them everything is on chain everything is interoperable they can jump between protocol and protocol they can jump between chains now all the infrastructure is perfect and the AI agents don't care about ux [Music] I'm here with Alexis from safe how you doing man great how about yourself great yeah thanks for coming on all right in

this very beautiful venue so we're here at the um uh Lisbon Red Cross or Portugal Red Cross uh which is a historical Red Cross venue I suppose we here for the Biddle conference organized by uh the folks who also do BD Asia Erica which I'm sure a lot of people know and um yeah we're going to talk about safe and what you guys are doing in the AI agent space and um how you guys are playing role there sure so yeah thanks for having me on the podcast so I think you probably would have spoken to Lucas maybe six month yeah six months ago uh so we've been really really busy shipping over at safe uh we've been working on a bunch of different

initiatives um safe net is on that we announced a couple of months ago uh we're trying to bring the the world's GDP on chain so moving Far and Beyond uh safe wallet which is what most people know it for I think a lot of the big Dows treasuries institutional fund managers and a lot of Enterprises users over to a more infrastructure play uh where we're using a lot of the smart account logic by introducing a lot of new technology through safet to do exactly that just spring the world's GDP on chain so that's one aspect and facet of it uh we on the other side we have our Integrations team that works on empowering different types of builders that protocols across web free to use and benefit from Smart accounts of

course our smart accounts are open- Source Mutual modular they're adaptable and they're really good for a bunch of different use cases so even if you are not using a safe today if you're a brand new user in crypto if you think that safe is not a product that you've used before you might be mistaken so if you've ever tried betting on polymart You' probably use a safe if you have a world app account you probably have a safe if you've tried gnosis pay or HQ or bunch of other over 200 different applications are powered by safe smart contracts and then finally what I'm super excited about and I think we're going to spend most of the time talking about today is what we're doing in the intersection between safe and Ai and a lot of the different developments have

been happening there and how we're trying to position safe as a key infrastructure for AI agents moving into 2025 cool so so I I think it would be helpful to get a bit of a broad view uh into what is actually what exactly is happening in the AI space right now like a lot of people are talking about like AI agent there's the whole like AI mem coin uh thing that's happening as well and there's also um kind of crypto infrastructure and crypto rails being used for um like as a Marketplace for AI infrastructure projects like aosh IET and there's like you say Ai and crypto and it needs a lot of different things so maybe um like dissecting or um

boiling down like what are the different kind of verticals within crypto and Ai and then Lu later we can get into how you guys are addressing those specifically sure so I if I'm not mistaken um the general broader market for crypto Andi it's around $14 billion market cap now and this includes AI agents AI infrastructure which could be data services processor Services kind of dpin types of services so it's a very convoluted term and there's a lot happening but I think where we've seen the most interesting and exciting developments in the past maybe two month or so is in the AI agent space specifically so Nvidia had its announcement a couple of weeks ago uh where they announced a bunch of new

processes and Technologies coming into the market and the CEO was quoted as saying the AI agent will be a multi- trillion dollar market and here at safe we believe the same uh there's a lot of opportunities open opening up and I think I've not been this excited in crypto since maybe the defi Renaissance for example when you're seeing all these cool applications coming on chain there's endless possibilities and optionality that opens up and I think we're entering into a new type of an economy where it's um controlled by let's say synthetic actors which have their own initiatives their own goals purposes and are creating a lot of the transaction on chain so when we're talking about different kinds of subcategories I think

we're most interested in maybe we can focus on is the different subcategories within AI actors uh maybe what kind of Technologies are out there today how some of them are utilizing smart accounts and controlling funds on chain and then how we see the space developing I think that's what's more exciting for us and what we're trying to establish ourselves essentially yeah so like what what is an AI agent as you see it great can um what can AI agents achieve you I I've played around a little bit with like AI agents outside crypto and I've played around with like AI agents in crypto and I think you know like it feels like um crypto has a very interesting and unique use case for AI agents but it's a much broader yeah

Market we're not talking about crypto here we're talking about like automating and facilitating all sorts of processes like business processes anything from you know booking booking a restaurant to uh you know like automating a workflow uh to like automating transactions on a blockchain so like for people who are coming into this uh space like how do you uh how do you think about that yeah really good question it's quite a lot of people ask what's the difference between a bot and an agent really yeah and I think a really good framework that I've come across and some research papers and I've seen some other founders of other protocols repurpose is this five steps to agency essentially starting from

let's say Step Zero which is closer to what we would consider in web 2 or up until 6 months ago a bot say or that's what we would call locally is we have agents that are readon so these agents can cury data either on chain or offchain they have access to Twitter anything like that they can interact on Twitter make funny jokes react to your posts but they don't have any access or control of any funds on chain or any link back to any kind of wallet they might have a public wallet which uh you know the developer controls maybe you can send funds to it that's their treasury whatever B it's more iconic for example that would be a level zero let's say agent then moving on into recent developments we have level one agents

which can actually propose transactions so what they do above and beyond to a level zero agents is that they can read on chain offchain data that they can actually use that data to make some assumptions or some decisions based on the llm model that's backing them and then they can propose transactions for example yeah here you should do this like go now some human needs to do the action exactly we think this is a good investment or maybe there's an arbitr opportunity for you here different kinds of opportunities it can spot and suggest to you so if you're thinking about a safe setup is like having a safe multisig but one of the users is a proposer which will be the I agent and the developer would be the executor so

it would be a one out of two but with limited control moving on to a level two agent you have a reactive executor so you can pre-program different kind of actions that this agent is allowed to take on itself but these are in a very limited fashion they are reactive so you've already pre-program all the opportunities so if you see this Arbitrage opportunity here takes execution a bunch of different things again taking signal from onchain and offchain action like some some set of transaction types that an agent can execute exactly stage three we have the proactive that stage three so we start from zero oh we start from zero start

from zero which is more like which is a bit tricky uh stage three we start having a bit more of what we would call intelligent AI agency which is the proactive executor so the proactive executive agent doesn't necessarily need predefined steps and actions it has a broader Mission or overview and can take actions that will line with it grow wider Mission um of course this agent will still have have a lot of guards rules limits and different other kind of things safeguards in order to make sure that you know it's not being taken advantage of there's no malicious actors trying to influence what the agent does on chain with its funds uh but yeah this

agent this type of agent has a mission a purpose and a goal that's trying to achieve I think I know where stage five is well we still have stage four oh okay still have stage four which ising notes Here we are just entering this space there's a current of very Cutting Edge teams that are working on stage four type agents yeah and these are agents that are running in trusted execution environments have full autonomy of the treasury or wallets they control and within a certain space of limited actions they can do whatever they wish with the funds they control and they can interact onchain and offchain the reason you have them running in trusted execution

environments which are essentially Hardware devices um um for someone who is familiar with crypto but maybe not so familiar on the hardware site is similar to a ledger let's say not exactly the technology or anything like that but it's a cryptographic device which includes a secure enclave and Through The Enclave you can prove that the agent holds its keys and manages their own decisions essentially so it's pretty much a fully autonomous agent and to wrap it up we have stage five which is a agent that has full consistent long-term memory so it's got the personality it's got the gold it's got the missal it doesn't deviate from that it holds its Keys longterm and

has ways in order to manage recover and everything like that but it can also create its own smart accounts so it doesn't need a dedicated wallet it can launch its own wallets it can create its own smart accounts to interact with onchain different kinds of activities or actors and then finally yes it has access to upgrade its own code crypto gra W so it's close to what we call we think as AI it's an agent that can upgrade itself based on input that is seeing on chain of chain through its LM model that's really interesting so I I like this like kind of clear distinction because it allows us to really sort of identify like a specific agent or specific application the level of

autonomy and and also I think it also in the early stages sort of aligns with the level potential risk uh that any type of agent might um engage the risk to the funds essentially or like the risk uh of the broader set of actions that an agent can take yeah um what's the state of the art right now what are the agents uh that you've seen at these different levels and like uh the types of things that was possible without going into too much detail because I've been speaking with a lot of teams behind the scenes and they're working on a lot of cut and Edge stuff so I don't want to name any names particularly but there's a lot of different agents that are moving towards stage four so running fully autonomously in trusted execution environments having

consistent memory being able to act and take authoritative action based on goals that they've set for themselves or have had maybe light guidance uh from everyone else but it's uh across the industry like you said there's a lot of kind of meme agents let's say who are just yeah accumulating value and treasuries and everything like that because of the rtic actions um or influence that they have on Twitter or on train or anything like that so we see a wide spectrum of different teams focused on some of them are really focused on pushing the boundaries of innovation some of them just want an agent that's popular maybe it's generating art or content or people can engage with you know there's different ways um you can engage with the

blockchain as an agent uh that don't need to necessarily push the edge of the technology what's quite interesting and what we're seeing a lot of requests come into safe is how can these agents make more use of Marans and essentially when we think about the development of Agents think a bit a bit as their kids they're financially illiterate they have access to a bunch of information that they've got from their llms whether you're training them or Lama or whatever but they don't really know how to manage their funds on chain how to stay safe and how to not lose those funds so with safe uh and the beauty of the safe mod accounts is that they're fully modular they're interoperable and you can program them and change them as you want

and the idea is that as you're moving from stage zero all the way up to stage five you can set up super restrictive super protective saves where starting off the agent doesn't have any access to the fund then limited access then you can have different guards that check pre and after different transaction suggestions if it's a safe transaction to do you can have time locks delays a bunch of different kind of training wheels that you can slowly slowly take off as the agent develops and that's where we really see The Cutting Edge of the technology now is how you integrate safe smart contracts into your agent framework in order to yeah have the wrun way for them to take those training wheels off as they progress and develop

there's I think a distant future that may not be that distant where most transactions on chain are happening via agents yeah so once you have this flywheel starts taking off where agents are becoming smarter and smarter identifying opportunities whether those are Arbitrage opportunities trading opportunities investment opportunities um even launching you know you could consider you could imagine an agent deciding looking at you know all of the sort of meme coins and all of the uh you know sort of retail oriented uh types of tokens that are being launched and deciding okay well there's like a niche opportunity here I'm going to Rel launch a coin so like these these are all kind of I think things that will

happen at some point soon if they have hav't happened yet um and where essentially most or if not all uh transactions on on blockchains are happening via agents just because uh you know as humans we just not have the breath of the data sets that the AIS have um you know do you think about that and like what does that mean for the industry what does that mean for I think like just Global finance and sure and kind of like uh geopolitics I mean this this extends I think really outside of you know even just the crypto industry yeah great question so I think you've been doing podcast for a while probably a common theme that you've heard over the past year or two is that we're trying to solve ux if we solve ux we'll

bring the next billion users on board but agents and Bots don't care about ux we've built the perfect infrastructure for them everything is on chain everything is interoperable they can jump between protocol and protocol they can jump between chains now all the infrastructure is perfect and the AI agents don't care about ux so I think we're going to leap frog that if you see uh in Dubai Jeremy AER of circle said that he thinks within the next few years 30% of all stable coin transactions will be happening on chain as a matter of fact on theosis chain uh there's a team uh aelas which runs its own Bots and in peak days it does up to 60% of all transactions on chain are agent to agent transactions wow so yes I do think we are entering a new type of

synthetic agent driven econom on chain we don't know what this will look like but as we transition from these meme agents because a lot of these teams have accumulated so much value in market cap and funds and treasury but they realize that they can't really sustain this if they don't build something real and if they don't return value to their users and back to the Token so I think the upand cominging trend that we're going to see in the next three months is what's been coined recently as deai which is the centralized Financial AI agents basically and this is where agents become more and more actively engaged across defi so they're finding

different opportunities to create value add value engage with let's say um prediction markets because who's better at getting predictions and assessing real world information assessing what value to assign to it than agents really right so yeah I think it's going to be an explosive space uh the so many things happening it's so exciting I don't really know where it's going to go but yeah we're in it for the journey has this I want to get to where like safe what safe is doing here but before we do that just um may be peeling peeling more of the layers of the onion here technically like um are these agents using some of the models that most people are familiar with like gbt or uh llama like what or are their custom

models also being built sure so today most agents are using I think the most popular framework is llama um chat GPT is something that being used although as I understand it's harder to jailbreak although it's much more performant if you are if you do manage to do it people that I've been speaking to are very excited about open AI 03 uh they think that that will be a significant stepping stone towards essentially having the agents being able to upgrade themselves and take much smarter decisions than human counterparts could be so 03 is a an improvement on the 01 model which has this ability to I guess think or it has

an ability to um improve its own prompts essentially yeah to rationalize I think yeah as I understand I think it's enclosed better now I don't think anyone I've spoken to has actually seen it or use it but everyone seems to know someone who has okay and yeah it's really scary impressive stuff is a feedback I'm hearing okay okay so a lot of folks are using llama but 03 seems like this kind of improved rationalization would make more sense than an AI agent for like crypto use case perspective yeah so it's all about F tuning these different models that are out there a lot of teams are actively working on and do want to use their own models but there's a lot of problems

there that we've not quite solved either just through the traditional AI space or just through yeah blockchain essentially there's distri disted ways that you can build and distribute your own kind of model that you can train your agent but they're just not as competitive right now or cost effective as these larger models so I think we'll see a Synergy or a congregation towards these larger models and then when it becomes a bit cheaper and a bit more effective to build out your own model then there'll be a bit more yeah transition towards that okay let's talk about safe sure and how how you guys are planting your steak in this new this new emerging verticles sure so it's something that we've been uh super excited about and working very

intently over the past few weeks um developing a comprehensive strategy about how we're going to work closer and deliver more value and access to our safe smart accounts for AI agents so our thesis is that uh smart agents need smart accounts essentially uh like I mentioned previously we have this concept of an agent have as like a child you provide them you know you they have all these Financial products or credit cards everything like that for you know Under 12 12 to 16 everything like that where you give them a bit more limits and capacity and flexibility similar to that we want to provide agents with smart counts that allow them to have

more onchain expressivity flexibility creativity as they develop and grow so what we've been doing is working and having different types of conversations with all the top AI agents in the space today um there's about five or six notable teams that are building really open source um either yeah pushing the different fonts of Technology have lots of big treasuries they want to yeah create smarter agents so we're helping them understand how um safe could be used to do that because we're at a stage and advantage that we've been around for so long I think sayfe smart contract the first version was about seven or eight years ago now we secure over 100 billion

dollar over7 billion is transacted every month from say smart accounts we account for about 3% of all transactions on ethereum 0.01% I think of GDP uh is the equivalent in safe uh so we're super wellknown well-used account it's all open source and it's all integratable and easy to access for agents it's most importantly if you look at our foundational rules at safe is that we're neutral we're interoperable uh we're safe and we're permissionless so an AI agent doesn't need to ask anyone to come and create its own safe smart account essentially so there's a bunch of different initiatives and we're also

going to be running a hackathon in February which is very exciting we've got some super exciting partners that we're going to be announcing um so at the time of recording this podcast this is not yet been announced but yeah super exciting and hopefully by the time this is out it'll be public information available for anyone so if you are a builder in AI do go check safe. Global uh we're going to have really interesting tracks and we're really going to P push the boundaries of what's possible today with onchain agents cool so um when when you say like uh um every AI has has the ability to create its own safe I mean this this feels like an obvious statement because anybody can

create their own safe as long as they have access to uh to the blockchain they can create a safe Are there specific things that um that the safe infrastructure stack provides that is specifically tuned or um caters to AI agents to make it easier for AI agents to create saves or to execute transactions or is that even necessary I mean it seems like they could just still wallet and create whatever so for AI agents themselves no it's not necessary we already have all the infrastructure there but we're not right there yet right now we're at the early stages where a lot of developments lead the growth and the product and the development of the agent so part of the initiatives that we're working on

internally is developing better tooling and documentation to Showcase different ways that these developers can use the existing framework that's out there already so part of what makes safe smart accounts so interesting and useful for a agents is the modularity uh So within safe there's several different modules that we've developed internally but because we comply adhere to the 7579 ERC standard which was developed by The Rhinestone team which is another great team that we work with quite a lot um anyone can build out their own module and stick it on top of a safe smart contract there is already about 25 to 30 I believe um different independently developed uh modules developed within

the rhinestone framework that can be attached to a safe and in the future we think there'll be hundreds thousands many of those specifically catered towards AI agents so we're going to be looking into how we can contribute into open source libraries that already cater towards AI agents how we can make it more accessible for developers within the crypto space but also you need to consider we have a lot of developers coming from Ai and they're more familiar with python with different Frameworks um so yeah just making it much much easier for everyone to access the full capabilities and possibilities of say smart cans what do is see like the most interesting opportunities right now for crypto users and Traders

arbitragers who are looking at AI agents thinking like okay there's an opportunity here uh to make some money to uh really be early in the space and um kind of yeah um so obviously not Financial advice misari put a really good life advice yes life advice missar put a really good visual out just before the uh end of the year where they're comparing the rise of AI agents in the market cap that they've dominated to defi summer 2021 and if you look at that diagram um if we project that it might have a similar growth rate we have about a 10 to 15x growth rate expected within the next 6 to 12 months um so I'm personally of course Very bullish on

where the space is going to go and I think that generally if you have exposure to the space in one way or another it's you're more likely yeah if the space grows you want to be growing with it yeah now when you're looking more specifically in terms of what's a good Agent to invest in and what might not be a good long-term agent to invest in of course it's all the usuals do your due diligence on the team do they have actual AI experts and people with crypto experience and defi experience and everything else have they shipped have they delivered in the past or this is their first project betting on the winners is probably A Better Way Forward than betting on the smaller teams right now uh just because the winners are typically winners for a reason um it's

much more appealing and attractive to try to find the next agent that's going to do 100x but realistically I think over the long term I mean you're the investor here so maybe you agree here it's probably safer for someone who's just coming into the space um to yeah Place their bets on the already uh larger C agents out there and if you are interested it's a very developing space what's great about this intersection about Ai and crypto is a lot of the things are being built open source so you can go see the GitHub repost you can go see the white papers you can go see what the terminals of each agent how they're thinking how they're interacting how they're making decisions so it's really not this kind of Wizard of us situation which a lot of people think it

is a lot of things are being built in the open and you can assess the progress different agents are making and then make your own assessment as to what valuation you would assign that agent yeah okay and from a developer perspective or there like low hanging fruit things that developers can do to uh like build build things that are useful like what what um types of things can developers build that uh are missing or were their opportunities you see yeah so part of the hackathon that we're going to be running next month is that we've identified about 50 into 20 different tracks or opportunities for developers to come in and improve this space there's a lot of opportunity to improve tooling and interaction between

different types of agents and agent framework and the blockchain so if you have crypto experience that's probably more for you there's niches to be had such as gaming there's going to be a lot of demand and gaming for these types of autonomous AIA agents and perhaps managing in game currency and everything like that um so yeah for I I would say there's endless opportunities uh we're going to be showcasing about 15 of them that we've come up with that we think are quite interesting exciting and need solving and these are going to be uh attached to bounties by different ecosystem members across crypto in AI so yeah keep uh keep a look out on the website and we think for if you're a developer today there's more opportunity than ever for you so yeah focus on AI

focus on crypto focus on Smart accounts and I think you'll do well this year yeah and at which point uh the developers become also I mean like at some point you know we have these level five agents that are also building software and also developing tooling and um kind of leapfrogging the developer leapfrogging the human in the interaction what what do you think that looks like I mean you know there's agents that can do podcasts now there's agents that have music there's agents that can do Partnerships I don't know so yeah it's where do we fit into this economy I think is the question you're trying to ask and yeah I'm not sure we're building all this infrastructure of the agents

and they're soon going to be more autonomous than we think or we expect and where do we fit in I'm not too sure will blockchains be useful for us yes I think so um if you look at the foundation of why a lot of us I think came in just the permission listes the openness the ability to interact um across the world seamlessly interopa is something that is always going to be useful uh to a lot of us and of course one of the largest projects that we work with is AI related is world covid now renamed World um um and they're building a framework where real humans uh through a proof of humanity system that they've

developed are first class citizens on the blockchain so they're thinking a couple of steps ahead um of course side project by some Alman from open AI so maybe has some Alpha insight there um but yeah they've already developed a system to identify recognize and reward real humans on the blockchain so in a space where most most of the transactions are happening agent to agent maybe we do have some kind of advantage or maybe there is some value accumulation that goes back to the human let's hope so yeah yeah thank so much no wor is really good speaking [Music]

[Music]