[Music] We are going to start our first session for this afternoon and for this talk I'd like to bring on deer to tell you about the future of consumer crypto with AI. Please give a big round of applause.
Thanks so much Cartik. Hello. Hello. Hello. Is everyone all a little groggy?
full of lunch now. Yeah. Well, I'll try and be exciting. I was I was going to be the uh most exciting talk of the day and then I looked and saw I think Vitalik's coming later. Is that that right?
That is that Vitalic?
Oh. Oh, sorry. Um so I'll do my best. I'll do my best. Although, you know, it's hard to beat the goat.
Um, so I want to talk a little bit about some concrete things we're doing on Flow this year, um, to make interacting with crypto, DeFi, um, using AI way more powerful. Um, and it's, um, yeah, it's very exciting. So, first, here I am. Um, the green face is probably better known than my human face. Um, I created ERC 721, the standard for NFTTS.
Um, I was on the team that built CryptoKitties. Um, and I was a chief architect for the Flow blockchain and NBA Topsh Shot, which was the first, uh, big application built on Flow. Um, these days I'm full-time on Flow, having left Dapper Labs, the creator of CryptoKitties and Topsh Shot. Um, and I have a great team at the Flow Foundation, and we work um, all of our time on Flow uh, trying to enable amazing application development on top of Flow. And as someone who came into um uh blockchain development as an application developer first rather than as a protocol developer first um I think we have a pretty great uh perspective on how to build a chain so that it can be used in really interesting ways by developers.
So, I took this picture back in 2019 um and and and it really resonates to me um because I feel like the real power of crypto is the thing that's come to be known as composability. And it's this idea that we can have permissionless interoperability between all these different smart contracts and all these different protocols created by all these different people. And there doesn't need to be any, you know, um governance bodies. There doesn't need to be any trust. we can just keep aggregating and building up new things and and the ultimate promise just like Legos is that you as an end user can plug these things together and create what you need rather than just you know having somebody else create an application and you're you're hopeful that it'll fit your needs.
The reality is that we're actually mostly in the Home Depot era. Um, yes, there are all of these pieces, but unless you're a bit of an expert, putting them together, um, you know, figuring out how to plug a and unis swap together is probably something you need to be a smart contract developer in order to figure out. Now, the good news is is that we're starting to see um an effort to improve this with things like um hooks now so that a and unis swap can talk directly together. But at best, this is IKEA, right? where somebody else is deciding how to plug things together and yeah, you get to plug those things together, too, but only in the ways that they've sort of pre-imagined.
So, where is that Lego promise? Part of the problem is that today the applications get built and they're designed for a certain set of smart contracts and as new smart contracts come along, we have to build new applications in order to work with those smart contracts. So what if we could introduce a layer between those things where we could create an abstraction so that the most common things that you do in DeFi can be represented by simple actions that are the same regardless of which protocol is serving those things underneath. And that's what we're introducing this fall on flow. So what are actions?
They're standard interfaces that enable effortless composability of D5 protocols. Um, and that's all very abstract. So, fortunately, we have a video here. So, I'll just talk over this. So, uh, what you're seeing here is is a a recorded session, um, using AI tools.
Um, and we're starting by installing DeFi actions into this project. So, so this is cursor. Um, and we've downloaded the version of D5 actions that is currently on mainet right now. It's in a it's in a um pre-release state. Um, and you see that this particular user has they've been farming uh tokens um here.
And so they've got uh a small amount of flow tokens that they've already earned and they would really love to take those tokens and uh put them back in as new LP tokens so that they're farming even more. Right? This is if you've ever done DeFi farming, this is something you do all the time. You go and you you capture those things and and put them back in. So now um the user is asking um the AI, can you create a transaction that'll harvest those rewards and just stick them back in as new LP tokens?
And as you can see, the AI here is has uh has been working with the flow blockchain, looked at all the documentation on flow. Um, and it's going to create a transaction um that does what we just said. Now, if you've never looked at cadence transactions before, you might be a little surprised that what I'm what is about to come up on on screen is a transaction and not a smart contract. Um, cadence transactions are incredibly sophisticated. um they are basically programs in their own right.
They just happen to be one-offs and they they run with the authority of the user who signed them. And so instead of a transaction just calling a function on a smart contract, a transaction can be a self-contained program that can can do all sorts of steps together. And so what we'll see here is is that the uh transaction text is there. The user is about to run that transaction. And then when we come back here, the the money has been um has been taken out, the rewards have been taken out and it's been put in as additional um uh additional LP tokens.
But um so you can see this is the trans transaction here, single transaction, three different um token transfers, and you can see all of the events that happened, right? We pulled tokens out of the pool. We staked new tokens. We swapped tokens. Um, and we put them back in as new LP tokens.
And we took those additional LP tokens and put them in the farm. So what does that code look like? Um, so the way the transaction is broken up in flow is is that we have three sections. We have a prepare statement which is um the part of the transaction that has access to your account. The rest of the transaction doesn't actually have access to your account.
We have the post condition which is run at the end of the transaction and if the the ability the the checks done in the post condition fail the whole transaction reverts. It's like it never happened. And then we have the execute statement that I'll talk about in a minute. But the prepare statement here is the only privileged part of the transaction and it has access to all of the assets in your account. And what you can see here is it's grabbing a certificate from my account.
Um, it is figuring out which pair of of tokens that the this particular farm represents. Um, it's grabbing a token source, which is a way to get those tokens out of my farming pool. Um, it's creating a zapper object that will let me convert those rewards directly into new LP tokens. and um and then it's it's storing this token source as a local variable to the transaction and then it's computing how much how many new LP tokens we should expect to get from that number of rewards and critically when this prepare statement is over the rest of the transaction only has access to those local variables right the staker which is just my account address the ID of the pool the a reference to the pool itself the amount of tokens I started with, the amount of tokens we expect to get, and the only thing that has any access on my account that can adjust any of my account balances at all is that one token source there, which is um taking tokens out of the swapper um uh sorry, out of the zapper, which is um which it's converting one token into another set of tokens. And so what this means is that this transaction um as as as after we've reviewed that prepare statement, nothing else that the transaction um does, no matter how complicated, no matter how inscrutable it is, um can touch any of my NFTs.
It can't touch any of my USD. It can't touch um any of the the other assets I might have in this account. Um and and then in the post condition here I am explicitly saying look if I don't have as many tokens if my current staking amount isn't greater than equal to what I started with plus the amount I expect to get then this whole transaction is going to abort. Now think about this. So I just asked an AI to write this transaction for me.
Knowing that my that this transaction can't succeed unless I end up with more than I started with in terms of my LP tokens is an a very strong assurance that this AI hasn't screwed up or been poisoned in some way to attack me. So we will look at the execute statement here because I think it is interesting um to see how DeFi actions work. So we create this notion of a sync and a sync is a place you can put tokens into and in our case the token the token sync that we're working with is um increment fi staking. So that's the farm uh functionality on increment fi we're taking those tokens out of the token source and remember the token source earlier was a um wasn't just taking the tokens out that were the reward from my farm but it was also swapping them into new LP tokens. So, what I'm getting out there, that vault that I'm getting out is is new LP tokens all in that one co line of code.
And then I can take those and deposit them into the sync. And what is the sync? That's the the farm. And so I could just put those tokens back in and they're being farmed. And then we make sure that all of the tokens ended up uh being deposited and there's nothing left in that container, the vault.
And then we can get rid of that temporary vault. And and again those post condition make sure that everything that I expected to happen happened. And so when I think about using an AI to build DeFi transactions I get really nervous about asking an AI to do a bunch of stuff um um using EVM transactions because who knows what it's doing. Whereas in this system once you understand what the you know prepare and post statements do um then you don't have to worry about what the rest of the code is doing because the only thing that this this code can touch is my LP tokens and the only way this transaction can um can complete is if I end up with more LP tokens farmed than I started with. So actions are this this new way of thinking about doing um activities in DeFi that AI is really really good at reasoning about that AI was able to understand increment fi but it also can understand kitty punch.
It also can understand trado it can also understand uh more markets. These are all defi protocols onflow. Um, and once we have those adapters that that make it work with the source sync and swap uh interfaces, the AI can reason about those and create very complex workflows that allow you in a single transaction to do some very sophisticated thing that would take you three, four, five transactions if you were to use the EVM. But what if you didn't have to run that transaction? What if you could schedule that transaction to be run for you and you didn't have to depend on anyone else to do it?
What if you could tell the blockchain to run that action for you? And that's what we're calling agents. And so starting in September on Flow, any smart contract on Flow can be set to be run at a time in the future. And of course at that time in the future, it could schedule itself to be run again. And so once we have actions on on chain, I could set up uh sorry once we have agents on chain, I could have taken that workflow that I had there.
Um and I think we have the code here. Um I could have taken a very similar workflow to that. I could put it on chain. I can tell the callback scheduler, I want you to call this again tomorrow. And then it could schedule itself to call again the next day.
And so every single day I could be going about and compounding those things all while I'm waiting in Toronto because Air Canada rerouted my flight, which is true story unfortunately. Um, and so these callbacks are available for any code onchain, but in particular, it works super well with actions and allows you to create onchain agents that engage in complex DeFi action that's specific to you, that makes sense to your use case, that makes sense to the the protocols and the tokens that you have, and that runs without any additional input from you once you've got it set up. And of course, you can cancel at any time because everything that's happening, you set up. It wasn't some other application developer. It wasn't some external protocol.
So, if you're thinking that this sounds a lot like sort of, you know, some sort of um uh uh custom service that you could write yourself or or chain link actions, of course. Um if you're willing to trust somebody else to call these transactions on your behalf, there are other ways to schedule this stuff. But this is all native to the chain. It's all completely um it's non-custodial because it's your account. It's your keys.
It's MEV resistant because the um these these transactions are executed by the network. They can't be frontr run um because they're all run by the system at the beginning of every block. So actions are ready to be um to be worked with today. So we've got a beta versions of the action smart contracts available on flow mainet today. Um and that demo that you saw was run against Flowmainet.
the Forte network upgrade that is coming in October um we will have a final version of actions available on mainet at that time and because it requires protocol changes that's when the agents capability is going to come online but we should have agents on testn net um sometime in September and the um and the the scheduled callbacks are available in a preview form right now using our emulator which is a local version of the network that you can run locally. So, we expect to um get a bunch of people building on this uh tomorrow at the hackathon. If you're one of the hackers um I hope you'll come and take a look at what Flow is doing. Uh we'll have some prizes associated with it. Um over the last year, Flow has been one of has been at most ETH global hackathons has been the number one chain that people have looked at.
If you're an EVM developer, we have a fully compliant EVM uh implementation that you can use. Um, if you want to take a look at some of these advanced features and there are more that I haven't talked about today, those are available through cadence. The nice thing about cadence is it has full access to the EVM. So, so long as you're writing to little write a little so long as you're willing to write a little bit of cadence code, um, you can have your uh, EVM smart contracts triggered on the schedule as well, you can have um, you can build connectors for DeFi actions to any EVM um, uh, protocols. Um, most of those protocols I listed, uh, Kitty Punch, uh, and more markets, um, are actually EVM protocols.
They're forks of of, uh, other protocols that you may know. Um, and so, um, uh, we're really excited to have these these great new features on board. I think that actions make it so much easier to for developers to reason about, um, building DeFi things. I think in conjunction with AI, I think we're going to be able to see end users using these this functionality to build complex DeFi flows for themselves and with agents having that run onchain completely autonomously, completely decentralized, I think is a real gamecher for certain uh activities. So, um I hope you'll all check out Flow and um uh uh thank you very much for listening to me today.
[Applause] [Music]
Automatic transcript — names and jargon may be misspelled.