Intents in the Age of Onchain Agents

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:09

   Okay, so let's get started. It's been a wonderful conference. I'm actually quite amazed by how much energy is here in a bare market, right? I mean, we just heard stories of some of the other conferences being relatively sad. This this seems amazing. I'm extremely bullish on NE. Uh, and I wasn't paid to say that. So it's not, you know, it's it's my privilege to host Adrien Brink from Anoma, old friend. Uh Gregory May from Blog Dam, new friend.

2. 0:36

   Uh Zaki Manion, well, we go back a very long way. And uh Michael Moin from Swapkit. Thank you for joining. And today we're going to talk about some buzzy topic called intents in the age of onchain agents. Right, I'm going to skip the singularity part because it's a that word confuses me and you guys find me at a very psychosis time in my life when you know claude code has taken uh a lot

3. 0:59

   of jobs away and the rest of claude keeps taking jobs and yeah so I'm a bit uh and agents are hiring humans and agents are talking to agents so I'm I'm actually very confused about what's going on but we're going to try and uh focus on what's uh in front of us which is intents Adrian you guys started started talking about agents before almost anyone else, right? Uh and I I think so. Yeah.

4. 1:22

   Yeah. Yeah. And you guys had this phenomenal paper with zero knowledge technology applied in very de in a very deep way uh to intent. So let's start with Eli 5. What are intents? Um intents at a very high level. It's just you describe a state change, right?

5. 1:37

   rather than a transaction which executes a spec like which runs a computation somewhere an intent just specifies I have A and I want B and you don't specify how the compute actually runs and this just lends itself very well to very scalable sort of distributed database applications one of the first ones being sort of finance and tokens um because executing a direct unis swap like a an transaction on unis swap is

6. 2:03

   just like it's not very composable whereas if you just specify I have some Ether, I'd like some Bitcoin. A solver can actually go and just figure out the correct path like what's the most efficient thing? What's the quickest thing? So intense are just I think a better abstraction here for what normal users really have. This is kind of how uh okay, so you have transactions where I actually format a message and then I

7. 2:26

   tell a smart contract, you know, to invoke this function and do something, right? Whereas with an intent I'm saying I I just want this to happen you know build a VC startup that becomes a unicorn send me the money uh that sort of and okay so in the age of LLMs right large language models this becomes particularly relevant uh yeah I mean so it is because it's way more composable

8. 2:50

   like LLMs are going to allow us to sort of have hundreds of thousands of interactions and intents don't necess have to be settled on chain and so I think like we'll end up speaking a lot to LLMs and LM output structure data i.e intents to coordinate and negotiate with each other, right? So you just tell uh this god in the GPU what you want and then it translates it into how to get there,

9. 3:13

   right? Kind of like a cloud code which brings us to agents, right? Uh so agents have a goal and if you you know and and you essentially orchestrate your agents in a way that until they just keep going until the goal is hit or you hit a failure scenario, right? So so so kind of this whole thing very neatly fits together. So this idea was kind of ahead of its time. So uh Gregory uh where does

10. 3:38

    block demon come into all this? What's what's your role in the world of agents? How do you see your infrastructure role changing uh from as we go from transactions to intents? Yeah. So um you know I think the Asente side of it is still very new obviously you know for from an institutional point of view. Um the just base layer intents though you know there's a lot of value that that brings. Um there's still

11. 4:06

    challenges on the institutional side. I think you know they want to see uh you know controls and uh policy enforcement around these kind of operations. So um that will lend itself well to the agents right you need to be able to have uh through policy you know defined limits around your slippage uh you know deadlines timeouts that type of thing which uh you know your your routes potentially certain assets over others and so so that that's definitely a

12. 4:39

    requirement. There's also, you know, there's the operational control side. So, you know, where they want to institutions want to see that these contracts are audited. Um, they want to see post trade uh reporting. Um, and all of that, you know, lends itself well to to agents.

13. 4:59

    When this is happening at scale on a high volume, yeah, you need that that order. Yeah. I went through this uh back in 2018 when we were trying to get compliance at RBS to use machine learning and they were really concerned about interpretability and you know our ability to explain why the models were doing what they were doing. So so you're saying institutions will still have some

14. 5:20

    concerns about but on the consumer side you know uh intents are doing very well and Zaki you've been working in this space for a long time. So, uh, intents were a buzzword, you know, our friend from was the CEO of Cosmos. Uh, yeah.

15. 5:34

    So, I think there was a bunch of friends Larry and Mag, you know, did, uh, Skip go. I don't know who all I'm I'll try and who who you're thinking about. Yeah. Then intense kind of died for a bit, you know, there was, uh, and now we are coming back big time and I think I when I started talking about intense, I didn't really understand what he was trying to do, right? And then he finally when Adrian started talking to me about intents, I didn't understand what he was trying.

16. 5:58

    So how did you come to understand what he was trying to do? Uh I think it was like 2022 in the winter. Uh I had surgery so I was stuck in bed and I like I like I was just thinking about DeFi and like how smart contracts work and like all the frustrations of actually building DeFi protocols. Um they're hard to secure. or they're hard to they're like very user unfriendly because usually you build a D5 protocol

17. 6:24

    and it's like okay it does this like one economic function but it captures so little of like what the actual user flow is because the user is like I have money and I want yield or I have token and I want a different token and instead you're like look I have given you this like one tidy piece of it and it's like and that's very unsatisfying to the user um and I think it's one of the reasons why like most DeFi protocols have been so unsatisfactory um from a user

18. 6:49

    uh users perspective is like the the protocol itself captures so little of what the user is trying to do. So anyways at that moment I'm like oh this is what Adri was talking about like 2 years ago uh in 2021 and I'm like okay this makes sense. Um and I've helped out on your intents. I've helped out on other intent protocols. It's been amazing to see but like it's so little of like what I think intents will be. Um and I think the LM world has like

19. 7:15

    accelerated which is basically I expect intents to displace basically all use cases of smart contracts. Um there's virtually like nothing that a smart contract does cannot be done better by an intent system. Um where the worst case scenario of how much money can be stolen if something is wrong with the protocol um or if some if two AIs misunderstand each other or write bad code uh gets you know shrunk down to

20. 7:38

    like the size of the transaction rather than like all the money in the system. Right. And Michael, so swapkit, right? The first use case for intents that we have seen work at scale is crosschain swaps. Yeah, for sure. Um, and in case anyone doesn't know about swapkit, swapkit is a crosschain dex aggregator. So, we work with a lot of big wallets that you know and we help trade from one asset to another. That's our biggest use case right now. And I

21. 8:05

    think that, you know, as we move into this agentic world, uh, agents are really, really good at abstraction and creativity. Uh, crypto tends to be much more deterministic and it's much more difficult for an LLM to do something like build the exact transaction payload that it needs to do over and over again.

22. 8:24

    So that's where we see a role for swapkit to you know create the tooling for thems so that they can reliably carry out the tasks much similar to a way that you know if you ask chat gpt to multiply something it's most likely going to pull a third party calculator tool and use that instead of just u you know trying to guess what the next token is right so you know two years ago I was

23. 8:48

    telling earlier to just uh bro why are you talking about intents you know because intents are seen as only market makers essentially providing liquidity and you have such a hot AI narrative going on. So I was thinking like a Twitter marketer right which is if you want pump then you have to talk about AI if you want numbers to go up you have to talk about AI. So it turns out you know then models got a lot better right and now we can do a lot more cool things

24. 9:13

    with intent. So [snorts] uh but uh what are the use cases you guys are seeing I mean in DeFi beyond just crosschain swaps where are you seeing anything exciting going on with intents and and and LLMs? So I I think there are a couple of things to it. One is authorization flows are much simpler with intents. So you can much more easily attach ZK uh proofs to them. So you can much more easily uh do like private swaps for example or private trading in generally. Um and the

25. 9:43

    other thing is you have as Zach was mentioning you have much less counterparty risk because rather than creating like these massive honeypotss um that like if you like hack the pool all the money is gone with intents you really only have the problem of like are the it's like only the specific inflight intents would be at risk. So the overall risk and experimentation you can do goes up massively.

26. 10:06

    We have not yet really seen massive experimentation outside of financial use cases. But for example, if you wanted to build something like Bitcoin um like intents are actually very good at this at modeling like non-financial use cases which have had historically a very hard time because modeling like Bitcoin with smart contracts is it's awkward as hell.

27. 10:25

    Um, so I'm very bullish that intense will actually allow us to like sort of like escape this problem of like oh crypto is only about finance and we can rather build other systems and other kinds of applications that go beyond finance.

28. 10:37

    Okay, this this part is a little bit less clear to me because if the underlying substrate the blockchains are mostly being used for finance then how do Zaki intents help us go beyond finance? So I think the way that you could think of this is blockchains are like ledgers of of resources.

29. 10:56

    Yeah. Intense are the mechanism by which counterparties like allocate those resources. Yeah. So but like resource allocation goes beyond like I mean everything. So in one sense it's like how do you make blockchains go beyond finance but it's like everything in the internet at some level is finance. like everything like the internet. The web 2 has always p has entirely been powered by ad exchange

30. 11:19

    marketplaces that are just like intent expression engines between people who want to advertise and people who have uh uh uh you know like spa have cap have have have content who have supply of adressions. So like we this huge settlement engine that always underwied Facebook and Google uh is their entire business model. How do we bring that onchain? That's pretty much like that's

31. 11:43

    that's the the frontier we're trying to solve. Okay. And Michael, uh where do you guys come into that? I mean, uh how big is the swapkit volume today? How is it growing? Uh you know, are you guys going to disrupt Binance alongside Near Intent? What's the plan? Yeah, that's definitely the goal, right?

32. 12:01

    And it's not just to take on centralized providers like Binance, but in my mind, it's all of finance. And we need to bring them all onchain because blockchains are a better way to do finance in general. They're more secure. They're more transparent. They're less prone to corruption. We can do things like fix the supply and not print money, which I think is probably what brought a

33. 12:21

    lot of us here in the first place. So definitely still bullish on that vision. But uh you know crosschain swaps are I think the lowest hanging fruit and the easiest thing to pull off, right? Is just a spot swap executed at a market rate. But why do people still go to centralized exchanges to Binance? Well, it's because they offer different products. They have things like an order book. They have things like per which we're starting to bring on chain a little bit. But uh you know how much of that is still centralized is a bit

34. 12:48

    debatable. So for swapkit um you asked about our volume. 2025 was really our first year. We did about three and a half billion in volume. Uh we saw almost two billion of that in the fourth quarter. So definitely that activity is ramping up. But I think that getting into those other verticals. So doing things like limit orders on chain in a a scalable way, bringing pers and then doing things like real world payments, real world assets, all those verticals

35. 13:14

    are going to be really important to, you know, bring normal users into this future that we all imagine, right? And Gregory, how does this change what you're doing? I mean, you're uh you're a blockchain infrastructure company, right? I mean, does that bring you closer to the AI infrastructure companies in terms of what you do today and what you're going to be doing tomorrow?

36. 13:34

    Um yeah not really like so obviously we're aware of everything going on but um you know our customers are large institutions so I think you know uh risk is their primary concern so like if we can manage the risk appropriately then then the rest of the flow will come at the same time you know obviously we're aware and I think everyone now has a view that uh you know in the future a lot of consumers of products won't the people.

37. 14:04

    You're not going to build a risk management agent for that. Well, I mean, so you talk about that, right? Like maybe you need agents that are handling your anomaly detection of of the order flows that that that type of thing, right? So, LLMs are still LLMs are checking the agents still, right? Right. Interesting times. Yeah.

38. 14:23

    Now, so in the two weeks I've been in the US, uh, my my current trip, I I I'm going crazy, right? Everywhere I turn, I see an agent. Agents are talking to agents. Uh everything is agentic. Now the SAS companies are trying to be agentic. You know, crypto is agentic. I mean, so but do we really need when will panels be agentic?

39. 14:46

    Panels could be agentic. Why are we even here, right? We could have sent our agents to talk to each other and maybe they would, you know, even have a more interesting conversation. Who knows, right? I mean, it's a it's a very it's a very weird world. And I think cloud code definitely sparked everyone's imagination about what agents can do, right? I mean sure it's a limited use case in the sense that codes code is testable and verifiable but you know who knows where we go from there

40. 15:10

    and uh so so so which where do you see the the killer app as in how much of this stuff that we do today whether it's you know speculation gambling trading uh prediction markets uh per why do we need agents for any of this stuff why I mean where do you see the killer apps coming through I I think to me most likely the killer application is going to be me talking to my local agent and my local agent just spawning off intents and like why does he use intense because fundamentally he's going

41. 15:48

    to do hundreds of thousands of these like exploratory spawning events and see like oh what are the counterparties that could for example offer me a flight like I will sort like my a I talk not like actually I I don't like typing so I like like voicecoms So like I just speak into the mic and go like find me a flight and then the agent sort of reaches out and goes okay who has flight providers here and then but the important thing is like

42. 16:11

    how do you actually verify what I sign over and what I accept in the end it's like then agent then the other agents will return me intense and my agent will then go here's sort of like these four different things combined hotel flight uh train kind of thing um do you want to do them and then I hit yes and then it that's actually the atomic execution of all three intents or none and I think this This is really like the killer application where like I just chat into my phone and to get done in the

43. 16:37

    real world. And this is going to be very difficult to build without intents or with without agents. Fair enough. I've seen some very interesting use cases, you know, like my agent is optimizing yield in DeFi while I'm sleeping, you know? So, because I don't want to be in the business of moving money around, right, for managing my health factor in a way, right? So, there's an agent that I have instructed

44. 16:58

    to keep my health factor. Maybe here I actually have a slightly controversial opinion. I don't think agents are great at this kind of discrete optimization thing because agent like LLM specifically and like that's essentially an agent, right?

45. 17:09

    Like they're good at like this structured unstructured language into structured output or vice versa. They're not great at like optimize me the boundary conditions for like what's the most yield I can get. But I think what you actually want to do is you sort of issue the intent um that like optimize for this boundary condition and then specialize solvers go out and try to do it. I think this is sort of going to be the next unlock because that's going to be much more

46. 17:33

    efficient and probably produces better results. Um because LLMs are like I can't even get my LLM to like correctly call me every 15 minutes because they're not very good at discrete optimization. Yeah. Yeah. I I ran out of my uh cloud API credits while messing around with OpenClaw. Uh it was an expensive undertaking. Uh where else I I mean so I mean so agents right? Do we need sovereign agents for any of this? Do we need what what cryptographic properties come into making agents more secure? Why do we

47. 18:08

    need any a blockchain for any of this stuff? Why not give them give agents a credit card and let them swipe my credit card at the checkout? Uh that will definitely happen. And I think that it's easier to do things in a centralized way when you have an undo button, right? But I think that everything remains true about, you know, centralization and its ability to be uh manipulated and corrupted. And I think blockchains are still going to be, you know, that immutable layer that really we're going to need to underpin

48. 18:40

    everything because if you just take this the current system and put agents on top of it, that doesn't mean that, you know, politicians can't still do their thing. So fundamentally I think blockchains are more secure, more distributed and that's going to be really important for the long-term sustainability of anything that we build.

49. 18:57

    Okay. Uh the first time I saw intents in action was when AI6Z and all the Salana Twitter bots were going off, right? And then they were tweeting like me. There was a chain Yoda agent and there was there were all these but all they were doing is tweeting. So and and these things got really big like up to two $2 billion market cap in the in the bubble.

50. 19:17

    we forget that there used to be a bull run, right? So, so I mean is there a world where intents uh you know agents start to uh using the power of intent start to bring the onchain and off-chain world together right because Twitter is offchain agents might be doing some things offchain in the real world like with my bank account or you know or yeah I don't know so so are you seeing any applications there and what's possible I guess my hypothesis about why agents

51. 19:49

    will want to will like sort of drift towards using blockchains for more and more things is because what I think agents really want is for their like workflows to compose in a way that like humans don't naturally what humans what humans want to compose they build like a company right so that like we could like we could allide a lot of like uh uh sort of like sort of transactional friction

52. 20:14

    um by having some institutional structure that we all participate in I don't think that's the way agents will largely work I think the way every agent like people's agents will want to be like sort of free agents uh and but like be able to compose their workflows together. So these like properties of like you know immutability being able to make credible commitments all of these things are actually much more valuable I think in an agent context than they were in a human context.

53. 20:37

    Yeah. So do you think agents will have an identity and agents will have a reputation like the good agent or bad agent just yeah kill him you know maybe I I think we're still far away from this I think at scale because it is very difficult to sort of like you want to be able to spawn up many agents and you want to experiment a lot so like whether these agents will have very persistent identities right now is very questionable to me because like I'm going to spawn like a hundred tomorrow kind of thing, right? Like and they're sort of going to be very exploratory. I

54. 21:11

    think what's much more important for agents is that you have credible commitments. I that they can show, oh, I'm not just like asking whether you have flights for me, but like I actually have the ability to pay because like I can include sort of like this is a credible commitment that I will take any reasonable flight under these conditions that also you can settle and I think this is much more likely to evolve rather than sort of like a global scoring system uh that requires persistent identity. Amazing.

55. 21:38

    Any We have three. Sorry, please go ahead. No, I was just going to say that there's interesting standards emerging. Uh if we you know, I think everyone's familiar with X42, but I think the other day I saw a uh a standard emerging for HTTP message signatures, but it was reversed.

56. 21:56

    So instead of the web server uh signing over the the calls, the client signs over the calls towards the web server, which is interesting cuz if you have an agent now that has a wallet can use its private key to sign over its calls and authenticate to web servers.

57. 22:15

    So you know essentially off-chain but using onchain wallets. So interesting building blocks that are emerging and and standards around that. Oh wow. Okay. So last uh you know show of the day. Anyone in the audience has any questions? Please do ask our esteemed panelists.

58. 22:33

    Any questions in the audience? Any intents to ask a question? Okay. So, so Zaki, I think you've you've got a you know a good bird's eye view of both AI and you know I think you you're the heaviest user of cloud code in this group. So if if you you know if you had to just completely throw a fantasy right say okay this is how good it can get or this is the most powerful thing we can build what would you throw on the wall?

59. 23:06

    Um I mean what you want is like what you want is well I mean okay the most powerful fantasy is like recursively self-approving agents buying resources on a blockchain. That's what you want. I see. So, okay. I think if it also high like I have much more realistic go also just like a personal assistant that like can book me a flight in a hotel by chatting to it kind of thing. But you weren't.

60. 23:35

    Yeah. My vision is flights and hotels don't exist anymore. Paper equip them. Adrian Adrian when you guys started Anuman, [clears throat] nothing about it was realistic, right? It was like a really cool project with a lot of hard problems to solve around ZK and whatnot.

61. 23:48

    So maybe there is value in us being a bit unrealistic especially that's true I think like very fundamentally we should all be very unrealistic because like even if you don't hit your unrealistic goals like hit exactly something cool along the way.

62. 24:01

    Yeah. Yeah. Yeah. Yeah. So I think please go ahead if you can speak up a little bit. Yeah. So in the example that you were mentioning about clients signing transactions that signed by the agent itself, is it a matter of liability or agency in terms of who signed the transaction of security or in which context?

63. 24:36

    Yeah. Who goes to jail if the agent Yeah. Yeah. Sure. Great. Um, so, so the first property is you do get that uh cryptographic proof of of the agent, right? They're using their wallet, but who's controlling the agent? You know, you don't uh ideally you want to get away from KYC with uh with agents, you know, that's kind of the the purpose.

64. 24:58

    So, yeah, I mean, I guess it depends on the on the use case and the setting. Um but at least you are getting that that first property where the agent is authenticated and tied to to a wallet. That's a that's a fascinating question.

65. 25:13

    There are so many questions of security right now with I mean if if I can have an agent surely Kim Jong can have an have an agent and you know they can someone can use uh cloud security plug-in to detect issues in a D5 protocol and it's harder to ship and deploy. it's it's much easier to you know find an issue and exploit it. So yeah we are in a very interesting territory with uh with security and liability with AI because this stuff is getting way more intelligent than any

66. 25:44

    human I have seen which is a scary thought [snorts] but hopefully we can put it to good use at some point. So thank you so much to our you know audience uh I know it's the last uh panel on the day so thank you to the audience as well. Thank you to our speakers. Uh it's been a fantastic day one of Nearcon. I'm really surprised by how good and bullish it's been in a in a terrible crypto market. But, you know, that tells me how good this thing is going to get when we are on the way up.

67. 26:08

    So, I I think the future of near is the future is bright and the future is near. So, thank you so much.