Why DeFi agents beat humans and curators

Transcript

All right, waiting is over. Um So, pleasure to be here. My name is Gautier. I'm the founder at ZaiFi. And uh this is my sixth edition at ECC. And today we're going to talk about why DeFi agents can beat humans and curators. So, a bit about us and me. I've been building products since 2021 around account abstraction in DeFi. And I'm going to have ask you guys a question today is who is using DeFi uh on daily basis? All right.

For those who didn't raise their hands, well, you guys are very lucky because uh honestly, DeFi UX management and security still sucks today. Uh I still hate, you know, interacting with different protocols, uh moving my funds between chains. And agents and AI uh is solving this. It's abstracting all the complexity and making it accessible for users. But let's just see what is really an agent if you really think about it. It's something that sense its environments, something that thinks. The thinking part could be LLM-based or just purely deterministic. And also it's something that acts based on these things.

So, the agent will serve a huge market uh in the next years. There's going to be $1 trillion of stablecoins on chain. And the agentic-driven economy is going to go between 3 to 5 trillion dollars of total volume by 2030. Uh and out of all this uh stablecoins and and volume, there's going to be only 80% of uh these stablecoins generating yield. Now, let's look at how, you know, uh AI has evolved. Eventually, we've seen the GPT-3 models for researchers, uh technical people. Then we've seen Claude, uh ChatGPT consumer assistance. And over the last uh couple months, we've seen uh you know, personal agents like Open

Claw. Um and there's been a huge surge around this. And it's going to be the same for DeFi. Everybody's going to have their personal DeFi agents to do certain things on chain. And even agents are going to act autonomously and do things as well. Now, if you look at the uh total TVL around on-chain capital allocation, which is mostly handled by uh curators and humans, it's around 20 billion right now that are allocating funds into DeFi. Out of those 20 billion, only 0.5% is actually handled uh in uh around DeFi agents, right? Which is really small.

Um it went to 250 million, then went down to around 60, 70 million. So, the industry is still very nascent, but um there's a lot a lot to to to go to with the the total 20 billion that's out there. So, now let's look at the agentic space as a whole uh on chain. So, we've seen different types of of agents like trading agents such as Banker who, you know, some of you are probably already know. It's been really interesting. It was a an agent's wallet connected to an LLM uh through Telegram. So, you could just talk to it, say, "Hey, I want to I want to do like DCA on on on on ETH. I want to have a limit order.

I want to launch a token." So, this worked really well because they were able to to surf on the on the wave of Open Claw and people looking to create projects uh in in the crypto space. So, people were launching tokens through Twitter um through through their agentic wallets. And recently, we've seen Coinbase launch their agentic wallets, OKX, uh MoonPay. So, we're really starting to see uh things rise up and having different types of agents. So, there's trading agents, there's betting agents. I think you all know also all that's around sentiments, XBT, uh DeFi Lama as well.

And then this is the category which has gained the most uh attention and traction. Uh it's the yield agents. So, this is where we fall under. Um so, it's basically agents that look at different lending markets and try to find to find the best rates for users to generate yield on. It's also agents that monitor uh different risk and different markets to make sure that users always get uh the best risk-adjusted yield across DeFi. So, um now I'm going to introduce you ZaiFi and how um we envision this. Um So, it's it's pretty simple. Uh it's it you can see it like um like an SMA, right? Uh so, a separately managed account uh in

in traditional finance. But uh managed managed by an agent. So, what I mean by that is that you just come with your wallets. Uh we create a smart account for you. So, it's basically like your your own smart wallets. And then you sign session keys. So, the session keys is basically you giving policies to a specific agent to do certain things. So, you're basically giving constraint um aspects of contracts they can interact with. So, for example, if the agent is compromised, he can't just send funds to another wallet or do any type of weird uh transaction. So, once that is done, user just deposits USDC. And in here, this is where the the the ZaiFi agent comes into play. So, it looks at the size of your wallets. It looks at the

different opportunities. Uh and then it allocates your USDC uh into the best yielding opportunity. Um now, if you have like a bigger amount of funds, it can look at all these opportunities and split your funds into different vaults to basically increase uh your APY and also de-risk your position. So, it's also something that monitors the market 24/7, monitors the risk. And um we've built um so, we built a consumer app where, you know, um any money anybody can come, create their wallets deposits.

Uh we've also built an SDK so for neo banks, uh centralized exchanges to provide uh yield agents to their users. But we've also been really focused on providing uh a yield agent for also other other agents, right? And this is going to be a huge use case in in the following months because there's going to be agents on chain that's going to have a treasury and that's going to want to generate yield on that treasury. So, I would like to talk about how like we also enable risk monitoring. So, we've seen the recent hack with Resolve.

And this is where actually gets uh you know, we're not we're not expecting to see that the product would make sense on this specific vertical because as soon as our agents uh detected a depeg, uh it would just look at the collaterals in each underlying vault. And if USR was in those vaults, the agent would directly rebalance the user's funds into another position. Um so, that was really nice because it was really showcasing how like an agent can also uh save users' funds. Um we've seen we had the same event with Stream Finance where we saw, you know, that there's liquidity uh coming out of the uh opportunity and then agents were rebalancing the users' funds.

So, so far uh we've processed over $2 billion of funds um over four chains um and over 13,000 wallets. So, these wallets are continuously rebalancing users' funds across chain and across protocol and and paying the gas for it. Um and so, we don't eventually like list all all the pools that's out there. This is where the it's a So, I would say this is the this is the human part where like we add and we do do due diligence on each underlying opportunity that we add uh we add for the agents to interact with.

So, how do we uh remain efficiency, right? So, we we can rebalance wallets up to 24 times per day. Um we don't just jump into higher APY. We look at IRMs. We also um have MEV protection. We simulate transaction just to make sure that if your funds go to an an opportunity, it doesn't change the APY. Uh and then we split the funds also to to to make sure that you're you're always getting the best yield. And something we we really put a a high focus on is adding and removing opportunities. So, um basically, this is the how how it works. So, you have the the UEA, the session key. Then the ZaiFi deterministic agent prepares the

transaction. Um and then we have uh this this target registry, which is basically something we can uh we can change from a multi-sig. So, we can change the the the target registry contracts that the agent can interact uh the the target that the contracts that agent can interact with. So, basically, this uh enables a flow where user just signs once the session key. And we have the right to add and take off opportunities if the user decides to. So, this enables to this enables us to always remain efficient, to always make sure that the user gets uh the best risk-adjusted uh opportunity across the market at all times.

So, this is a a user who deposited 10K. Uh here you can see all the different uh opportunities that that the agent can interact with. Um and this user, for example, was rebalanced 64 times over the last 3 months uh generating over 7% APY. Um so, yeah, all this is is all all the transaction you see, the gas is abstracted. User doesn't have to interact with uh each underlying opportunity. This is all done autonomously uh for him. So, in other words, you know, ZaiFi is a sort of like invisible force which just, you know, does everything for you.

Um while you also keep some sort of control and some customization of what the agent can do. So, it auto rebalances, auto compounds, claim rewards. And it's your wallets. You're not like pulled into a vault with like thousands of other people. And you're just like giving away your money uh giving control over your money to some to some curators and you can personalize everything you want directly as well. So, yeah, we were in an offsite and and this is a funny story, but we see Zifi agents as gnomes. So, gnomes were like mystical creatures that used to handle wealth in the Swiss mountains and in an invisible way.

And yeah, I like I like to compare this to to to to the gnomes and we've done this now for for smart wallets, but we're also providing the same technology to to vaults. So, basically what we're doing is that we're wrapping the smart wallets that we created that users create into a vault. So, that um in other words, you can deploy a vault in a couple seconds and then handle it directly through our agents. So, it's really abstracting all the complexity of creating a vault, handling it, handling the risk and and all the different aspects of of you know, integrating protocols into a

vault. Making sure that you know, there's there's less there's always monitoring the risk. So, we're basically abstracting all the that complexity into basically a vault that that that's powered by our agents. So, this is where it gets interesting. The NAV is updated immediately. Deposits and withdrawal are automatic. And all the strategies and all the the the protocols are directly integrated. So, you don't have to do all these things. So, basically now anyone can create a vault in a couple seconds, handle it through a smart account and have all the automation and risk monitoring that most curators do out there.

So, yeah, I can go quickly over this, but the way it works is very simple. User just deposits into the vault. Then once the defense are deposited, they go into the smart wallets and then we have an oracle who always monitors what's happening on the smart wallets, updates the NAV to the smart smart wallets that updates to the to the vault. So, as soon as a user wants to withdraw, indexer just listens to the to the different vault events and then just withdraw the funds from the smart wallets to the vault. So, yeah, this is quite nice because a lot of partners wanted to to work with us and to enable basically our our agent infrastructure to work directly from a

vault. So, this opens up more possibilities for partners, also for curators and also a simple way for anyone to just create a vault and we could see see that also just from from a simple tweets. So, this is it, folks. There was a last slide thing saying thank you for listening and yeah, it was a pleasure. >> [applause] >> Any any questions? Yeah. What are the things that can go wrong and sort of how do you sort of mitigate that risk?

Well, I mean the things that that's can go wrong is is um but we've seen with resolve resolve hack and and and stream finance, you know, agents can you know, we have 13,000 wallets. So, sometimes they cannot act as fast. So, now we're just improving the efficiency of this and making sure that it's it's going to act even faster in the future. That's one aspect. Um but besides that, like the way we build it is is to make sure that you know, like even if the the agent is compromised, it can't like just you know, send funds to another wallet.

The only funds where it it can it can send to is your original wallet or the different destination contracts that's in the that's in your registry, basically. So, yeah, this we we try to think a lot about different edge cases, but for now that's that's only things we see. Maybe there's some other things that's uh you know, we didn't see, but that we believe that right now this is like something which is really secure and efficient. Are Steak House, Gauntlet and every other curator is going to create their own agents in the next 36 months? That's a good question.

Um they could they could do so, but I mean there's a long road way to go before doing so. So, yeah, we'll see. We'll see maybe. Yeah. Um so, this might be a personal hot take. But in terms of the lending market optimization, do you see that as mostly rule-based or is it cuz it's kind of predictable or in which circumstances do you see maybe like a LLM can be doing better?

Others? Yeah. This is a great question. So, in the early days we actually were using LLMs to to basically do everything, but then we realized it was not you know, was hallucinating. It was it would be too risky. So, we need to have something which is always working you know, 100% of the time. So, I mean it when you're doing like lending markets and just you know, rebalancing users' funds like USDC between different markets like you know, the liquidity is there. You can't really like front run because every wallet has like his his own like customization, his own way of doing things. So, if you stay mostly like doing money markets and lending, then

there's no way really like to front run the thing itself. If that answers your question. All right. Well, thank you for listening. And have a nice day.