Sergej Kunz (1inch) - Rethinking AMMs: Why Today’s Liquidity Design Won't Work  at Scale

[music]

So, hi everyone. My name is Sergey Kunz. I'm co-founder of 1inch. And I have a question to you. Who runs own AI bot on own Mac mini or Mac studio at home or in your your own office?

Put your hands up. 1 2 3 Three people. Okay. I will start with something different and come to this topic a little bit later. I'm going to speak today about the broken AMM architecture what is used for DeFi.

We should rethink AMMs. And I will explain why today liquidity design won't work at scale. I have 100 billion question. DeFi holds a huge volume of capital. But how much of it is actually working?

I ask in the previous presentation, this guy who presented the dashboard with all the trades in the Uniswap V3 pools. We have seen the movement of the price and the liquidity. But the question how much is really utilized? It's very important and we got the answer at dune.com/1inch/idle.

You'll find a dashboard what explains how much is really used in the liquidity pools. If you look in the in all liquidity pools, we have a s- spread between 84 and 98 percentage of liquidity what sits idle. It's a waste of liquidity. And it's because it's idle by design. The liquidity forms the price.

So, if you look into Uniswap V3 with 1.4 billion US dollars of TVL, 87 percentage of liquidity sits idle, doesn't work, doesn't live 90% of the time. To the best day usage, we get 13 percentages of of usage of the liquidity. In the Uniswap V3, and this is the biggest one right now. Curve, even Curve with stable token liquidity and this nice platform from Michael sits idle with 91 percentages.

It's broken. No one speaks about that. I I'm going to change it. So, the LPs also losing a lot of money by facing impermanent loss. To fight impermanent loss, you need to generate enough fees.

So, you need very fancy strategy or you need huge amount of volume non-toxic volume what's coming to your liquidity pool, so you can earn on that. Most of the liquidity providers are in loss. And this is not what will scale and adopt the traditional finance to use our test. So, if you have 300,000 US dollars as LP, you would like to provide it into liquidity pools for the most volume pairs. You need to decide on what protocol I provide it.

What gets more volume? Like maybe Uniswap because Uniswap would like to route their own volume to own protocol to not benefit all the other protocols like Balancer, Curve, and others. So, you need to spread your liquidity, you need to fragment it. So, depends on your pairs you select, so different pairs multiplied with multiple venues, you have dozens of pools. You fragment your 300k into 15k pieces.

And even some of this liquidity is sits also idle because this this pool is not big enough to get good amount of volume. The industry favorite metric is wrong. We should switch from TVL to TV TVU. The TVL shows how much the team the project was capable to attract capital what is locked in smart contract custody. Doesn't mean that it generates fees.

It's just there. And based on our previous slides, we see that it's just sitting there and it's idle, doesn't work. What about TVU? Total value unlocked. So, this should represent how much capital is actively usable when trades happen.

So, measures the productivity of liquidity. Capital that earns, not that capital what sits idle in the pool. The fixes architecture act manner, not incremental, means we don't need better pools. We don't need new protocols, new Curve V3 protocol. We don't need new Balancer V4 or Uniswap V5 or a lot of hooks in Uniswap V4.

We need to rethink how capital connects to strategies. So, we should rethink in front in terms of um We need to move from pools to registries. Remember this 300,000 dollars split of liquidity across 20 pools. What if we could back all six pairs in all venues at once? And I'm not speaking about quantum computer or any algorithms what will run on quantum computer.

I'm speaking about a new architecture what introduce a registry, not pools, a program what is strategy, what can empower all liquidity positions in same time. It's non-custody. You can support multiple strategies. And the very important thing is your assets stays in your wallet. You don't have to trust the next smart contract pool what can be hacked and drained by hacker.

You don't have to give your assets away to a custody smart contract. So, the architecture contains um accounting of virtual balances. And it allows to avoid any token deposits, withdrawals from any smart contracts. So, you approve once your assets on your wallet, your ERC-20 tokens. You deploy strategy what defines what price range, what fee you charge.

You can even have very complex strategies. You can have also very simple strategies. For example, flash loan. You can say it works like a like intent. I'm okay that my USDT can be taken for a flash loan and I want to be paid with five bips.

Who says that 30 bips pool on a pair of Ethereum USDT is the right amount of fees this pool this this pool should charge? Who says? Maybe it's 25 bips, maybe 21 bips. You should be flexible in this and your decision should be made by proofs and data and prediction based on the trend and the current market situation. So, strategies should be as programs, not contracts.

You can compute off-chain a set of intents. I want to sell my Ethereum at a price of 4,000 US dollars. You don't need a smart contract for that what do expect explicitly this thing. You can compute your intent in your framework library what is deployed on a block blockchain. And a strategy what can execute it.

What change when the liquidity is non-custody, shared, and programmable? Everything shifts. One balance backs and strategies You have zero custody risk. Of course, you have smart contract risk when you approve the specific strategy is allowed to move your assets. You have infinite strategies per balance.

With every strategy, your capital efficiency is growing infinite. And you don't need any token transfers. You just cancel your strategy without moving any assets and paying gas costs. Better architecture solves supply, but who generates the demand? My bet is that the next billion users won't be human.

AI agent are the fastest growing consumer of on-chain infrastructure. Programmable, the deterministic and composable interfaces are exactly what they need, right now. And agents are already there. The most profitable polymarket wallets are bots. They have access to analytics data on on-chain data, on on-chain events and trends.

They can act much, much faster than the humans. We have more than 500 AI agent crypto projects out there already. And we see more than 1 billion US dollar volume from AI agents. So, we have of course some standards like ERC 8000 4 for on-chain agent identity standards. Something is already going on.

The X402 standard for the payments. And we see the shift from having cloud AI to personal agent. So, it's the reference to my first question when I came here. Who runs own agent on Mac mini or Mac Studio? And I feel very very familiar with this approach that you you have your own agent running on your own hardware what you install by yourself and you secure it by yourself.

You have in your own custody. Self-custody. It's very near to our our approach with Web3. We are here with Web3 for self-custody and ownership. So, nowadays, you can just easily buy Mac mini, Mac Studio.

You can install Open Claw or you code your own project what I did [snorts] for myself to ensure that I'm safe. Um my personal AI agent can even do Web3 transactions swap for me. Provide liquidity into Uniswap because I teach them to use Web3 library and Uniswap and 1inch. And you can automate a lot of things with your agent, right? You can automate reports, trends.

I get Slack summary. What's going on in the company? I get news of DeFi. Who was hacked tonight? And we faced again to another 70 million US dollar hack uh on Solana, unfortunately.

So, you can automate a lot of things, especially when you are a person like me who is very, very lazy in manual work. So, and this is happening now. It's not just like Silicon Valley self-driving cars only. Um are available in Silicon Valley. Um AI agents are already everywhere in a lot of homes.

And machine-to-machine economy is emerging. Agents paying agents via stablecoins. Agents paying humans to do some real-world thing already. So, and it's growing and growing. So, in terms of information advantage, a human with a dashboard is very limited.

All the information are built for humans. It's limited in in accessibility. It's limited in information amount. An AI agent can just make a Dune query and get the right data. Can subscribe to Web3 events by using 1inch API, for example.

Can understand what's going on. Can predict the the next trend. Can suggest you to rebalance your portfolio. And you still decide by using maybe wallet connect in your agent and a wallet what allows clear signing like 1inch wallet and Ledger. You can swap.

And AI agent works 24/7, 365. So, we have three forces here. Shared non-custody liquidity, programmable strategy execution. AI native demand. Capital efficiency solves the supply side.

AI agents solve the demand side or increase it a lot. And programmable strategies are the interface between them. Agents don't need front-ends. And people should not discover new DApp and try to understand how I connect wallet. How I swap on this interface?

I don't understand how I select a network. The human don't need any of these interfaces. They need just an AI agent what understand them. I record a voice message to my agent. It decodes and understand my my demand and swap for myself.

So, I don't have to do anything except like looking into my Telegram and seeing the progress of of of the execution. So, UI and UX will dramatically change how we interact with Web3 very, very soon. Stop locking capital. Start unlocking it. That was from my side.

And if I have I think we have still time, 3 minutes. If someone has questions, go ahead. Any any questions? AI uh or um our shared liquidity approach, how we addressed the idle capital in the AMMs. I think I think I can hear you.

Just say. Okay. But how how do you secure your investments? You you were talking about your Apple mini Mac and your Mhm. agents that are working for you.

How does your agent make sure that the uh liquidity provision that they are doing is kind of in a risk Mhm. uh matter that you want Yeah. So, the question is how I I secure that the AI agent will not mess up with my funds in my Web3 wallet. So, what I mentioned before, my personal AI agent what I wrote, I call it Cortex. So, um have specific security and safety measure measurements and layers.

I connect my wallet via wallet connect. So, I tell them I want to connect my new wallet. They show me QR code and the link. So, I can select my 1inch wallet or MetaMask. Click to it.

My my wallet opens. I accept the connection. And then I I question some some topics like, for example, okay, I would like to provide liquidity into Ethereum USDC liquidity pool on on on um on Uniswap for now since our Aqua protocol is not yet for production, but for development. So, um and I'm asking potentially if to, okay, let's look into historical data. Go to Dune.

Make a your query. Get the data. Get the price ranges and suggest based on the current trend in what price range I should provide my 100,000 US dollars. So, based on the information from the blockchain, based on also standard um algorithms, I get the prediction for the next 3 months uh where I want to sit in the liquidity pool. I I can ask them to create the transaction through wallet connect and 1inch wallet or Ledger.

I can see what I'm signing. So, 1inch wallet and Ledger supports clear signing. I'm not I'm not sure about the MetaMask. Maybe we already um this part I I don't know. I'm not using MetaMask or maybe Uni wallet.

So, through this clear signing, I verify what's going on. This even a simulation what's what's going to happen. So, I am I will get a LP NFT LP token, for example, and I see that my funds are moving there. And this legitimate because one issue is a scan for threats. There are some service provider who also scan the transactions if it's malicious or not.

So, based on that, I can verify that I'm doing the right thing and the AI is not cheating on me. And especially for me, I build it by myself because then I can verify that they are follow the process and don't mess up with MD files and and and and do it in different manner. So, I have specific programmable flow for such swaps. Yeah. Of course, wipe code it.

Um Uh I was surprised and had to rethink how to how to test now wipe code it application. So, I told it to invent like a flow to properly uh review and test everything. Yeah. But just just just the flow what I'm using right now. Thanks.

Yeah, I I see some data from AI. Yeah. So, of course, I could trust it or I could maybe go and self verify verify, but I'm convinced enough that I get the right data. And I can again verify in the wallet that I want to really provide into liquidity pool in the price range between 2,800 US dollars for Ethereum and 3,500 US dollars for Ethereum. And it's if it's valid for me, I provide and I accept the transaction through wallet connect.

So, um I think we are done. Yeah. So, um thank you very much. It was a pleasure to be here. Um very very very proud to stay on this stage, famous stage, and uh happy that uh this amount of people came to uh see my presentation.

Thank you very much.