Smart Money, Smart Machines: How Rollups Power Stablecoins & Onchain AI!

zeeve

🎙 AMA with Karthik Senthil (Venture Partner, Lattice Fund | ex-ZKSync, Circle, Amazon) 📍Hosted by Karthik MS, Digital Media Manager, Zeeve  What happens when stablecoins meet rollups and AI agents go fully onchain?  In this high-signal AMA, Karthik Senthil dives into how the convergence of rollups, AI, and stablecoins is transforming the next generation of crypto infrastructure.  With experience building at ZKSync, Circle, and Amazon—and now serving as a Venture Partner at Lattice Fund—Karthik shares sharp insights from the frontier of Web3 infra.  We cover: 🧪 Karthik’s transition from Web2 giants to Web3 innovation ⚙️ Why rollups are the key to unlocking scalability and programmability for stablecoins 🤖 How AI agents are beginning to interact directly with smart contracts 💸 The potential of programmable stablecoins and onchain monetary systems 🔮 The emergence of app-specific rollups designed for AI-driven economies  Featured Questions: 🔹What experiments is Karthik working on right now? 🔹Why rollups are foundational infra for stablecoins & AI agents 🔹Can AI agents manage treasuries or autonomously operate onchain economies? 🔹What might agent-native rollups look like in the near future?  📌 Whether you're a DeFi degen, infra geek, AI enthusiast, or just exploring the intersection of money and machines — this is an AMA you won’t want to miss.  👉 Don’t forget to like, subscribe, and share your thoughts in the comments!

Transcript

00:00hi everyone uh welcome to with three voice by c uh we talk about all things i've changed roll-ups

00:06and d5 uh ai gaming and similar interests and industries of course like you all know

00:11uh i am karthik as well uh funnily uh apparently when i told uh ama today will be between uh

00:19kathit central and me uh who's who's karthik works at who works as deep uh our vp made a

00:24very uh funny joke he asked me if the ama will be called kathik calling kath which was funny so

00:31without wasting a lot of time uh today's discussion will be on the topic of uh smart money smart

00:37machines uh how develops power stable coins and on-chain ai so our guest today uh is uh mr kathik

00:44fendel who's uh uh who's building something very exciting at the moment uh we'd love to know what

00:50he's building he's a venture partner at lattice one he's a former uh dev at uh getting circle and

00:58amazon he is he's contributed heavily in building these ventures uh welcome to the zv3 voice uh

01:05karthik so would you like to introduce yourself for our listeners no i appreciate that karthik um yeah

01:11i actually haven't seen karthik calling karthik i've been meaning to i know the movie is probably at

01:16least like 10 or 15 years old at this point but i've been meaning to to see it at some point so maybe

01:20it'll be another inspiration for me to see it um no thank you so much for having me um yeah um just

01:26a little bit about myself i like many people i i first got into crypto about hearing about bitcoin

01:33uh it's kind of at least for some of the the ogs or the older kind of part of the industry it's kind

01:39of the gateway drug i know different things have uh have inspired people more recently i first read

01:44the the white paper like kind of in the late 2016 early 2017 time frame and i think for me as

01:50as an engineer what really stood out to me was this notion of like guaranteed behavior like the fact

02:00that you know the the consensus algorithm worked uh the way it worked there would be emissions the

02:08way it was scheduled and it was not impossible of course you could have a fork and everything but

02:15basically um you couldn't change the algorithm right you couldn't change the way it worked uh

02:21there was this notion of kind of assured assured guarantees which is so different from software

02:26right in fact software is is very much uh grounded on the opposite thesis that you can go iterate and

02:34change and fix bugs and do this you know at a moment's notice it's actually a been a key feature of

02:40software that people can ship stuff really quickly it's not working just go fix it and redeploy right

02:45you don't have to get it right at the first time so that's kind of what uh what um initially drew me

02:49to to crypto and of course like through the 2017 bull market you know when prices go up it's it's

02:55kind of more interesting and fun to get into the space um kind of uh you know i was at amazon at the

03:01time leading engineering teams um and um spent a little time spent some time following crypto but

03:07didn't get really um deep back into it until 2020 uh and so just a little bit about kind of my story

03:13i um left amazon in early 2021 to go be a founder full-time in crypto i was focused more um initially

03:20on crypto education and then eventually trying to build um a decentralized version of like a glg and

03:29for those who aren't familiar with that it's basically like a knowledge network um today you have

03:34all these companies like glg and there's others around the world that basically curate experts and

03:38then companies go pay them to have access to they basically they basically are like kind of an

03:44aggregation of of knowledge and so i wanted to see if there's a way we can uh invert that through

03:49crypto rails of course like so i did that for about a year uh you know obviously 2022 is a crazy time

03:56with all the different blobs and stuff so yeah it wasn't as fun uh to be a founder then and i ended up

04:01winding that down that summer which probably retrospective is a good thing because chat gpt

04:06came out like six months later um and then yeah i've i've been a venture partner at lattice since

04:11the fall of 2022 it's been awesome just um really going deep uh with the team on early stage deals

04:17and i've done a bunch of stuff uh there and then i spent a year about a year at circle uh leading

04:23engineering teams um on a bunch of stablecoin stuff which we can get into and then over the past a

04:30little bit under a year and a half i was at zk sync um as the you know the leader for zk sync era uh

04:36rl2 so thought deeply about product strategy uh ecosystem founders tech um kind of you name it and

04:45i'm obviously uh the going to our you know tg and airdrop was a you know definitely a once lifetime

04:52experience so definitely have some some stories from there and then now i'm working on um a new

05:00a fund focused on ai agents and with the thesis that crypto infrared plays a really key role

05:07in helping agents scale their models move money and eventually access uh internet capital capital

05:12market so it's a little bit about myself um but i'll maybe i'll pause there for for more questions

05:19okay sounds brilliant uh kathik so you mentioned that you started your journey with uh amazon and

05:26followed by circle and then vk sync red so uh could you touch upon i mean before we start uh start our

05:33journey because the dk sync is one of our protocol partners for rollups and we've been uh teaming

05:40with them for quite a few years now since you can just touch upon your roles at dk sync and uh what you

05:45have built at dk think and we can probably follow follow through with that question with a little bit of

05:51inside questions and a lot of questions after that you know absolutely so yeah the vk sync team is

05:56amazing um and really what i uh led there was being the the the single thread of the owner for

06:03zk sync era which is our canonical l2 so zk sync protocol it's a couple different things we obviously

06:10have our own l2 but really the our end vision for zk sync is is uh or the the elastic chain and i keep

06:17saying r because i i was working there i left a call months ago but the uh the envision is elastic

06:22uh elastic network right uh elastic chain and so really what that is is it's an interconnected network

06:30of blockchains um and they all use the same underlying zk stack as a result of doing that

06:37you're we're able to facilitate native interrupt because they can trustlessly interact with each

06:43other because using uh zk proofs um so that's that's the end vision but you know zk sync era is

06:48l2 uh that was launched on mainnet in early like maybe spring of 2024 uh i'm sorry spring of 2023 uh

07:00that's right figure 2023 and so when i joined it was about really thinking about what we want to do with

07:05the chain and so obviously we went through um through tge and airdrop which was a you know really

07:10uh kind of watershed moment for the ecosystem uh and obviously the token as well on era and then

07:17we spent a bunch of time thinking about really just you know it's obviously i think one of the

07:22things that is a common theme around roll-ups is that being a general purpose roll-up is really

07:27difficult i think uh in today's kind of ecosystem that was something that was very popular you know

07:33when armature and optimism first launched in 2019 but there's now a lot of different general

07:38purpose roll-ups and it's very difficult to differentiate yourself to builders if you're

07:43just you know by and large offering the same your block space isn't differentiated and so we're we

07:50spent a lot of time thinking about what we want to do with era um is there a certain um uh kind of

07:57domain or or kind of that of builders we want to focus on what did our tech uniquely enable

08:02so we you know we did a bunch of uh work uh there uh focused on that um and then so that was kind

08:09of what i love at ck sync okay that's brilliant context so uh it's always fun to uh speak to

08:18somebody as we can see that is uh i have listened to vesselists speak quite a few times on his youtube

08:24videos and you wouldn't want them to stop so it's it's brilliant when he when he gets into the

08:29group of talking about era and elastic chain which is of course uh changing uh rollups and the infra as

08:36we know it so uh my first question to you karthika is uh how do you see rollups fundamentally changing

08:43the infralayer of stable coins and ai agents operating on chain if you can just elaborate on

08:49that yeah i think um there's there's kind of two answers the first is kind of the obvious answer which

08:56is like you know rollups by and large like the the design and the reason that they're invented right

09:02is that they have um they can optimize for different things and so most rollups um are you know faster

09:10have much higher throughput than than uh ethereum mainnet l1 and they're much much cheaper so that's

09:16kind of like the obvious answer of yeah like if you are whether you're a human user or an ai agent um

09:23you'll probably want to access um you know a roll up if you want to do something on a blockchain versus

09:29ethereum l1 um i think what is going to be really interesting for rollups as maybe you as maybe you

09:36know as they compete with some of the other uh block spaces out there that's not that's not evm based

09:41right so solana um you know sui some of the other kind of blockchains that are kind of competing for the

09:47same kind of uh usage is really around figuring out um the type of use case that they can be designed

09:55for and uh being a much much more like user focused right so historically blockchains have

10:04kind of been like hey um and i'm painting a broad brush so i know not every blockchain operates like

10:10this but they've kind of been like hey we have this cool new blockchain um it's like you should come

10:16use it um it can do a lot of things but it's not very what i would say like you know customer obsessed

10:23like it's one of the things that you know was a big thing on amazon's like you need to deeply understand

10:27who your user is um understand their pain points and then go build a solution for those pain points

10:33and typically historically in crypto it's been it hasn't really that mindset really hasn't been very

10:39pervasive right it's been mostly like we're going to build this cool technology and then uh people

10:44will use it and i think one advantage rollups has uh if done well is they can find a use case so like

10:52you know the robin hood uh l2 that they will eventually build i know right now they're on arbitrum

10:58they can begin to think about okay if we want to you know uh support tokenized equities what are the

11:04trade-offs do we need to make on scalability are there other specific features whether it's kyc

11:10on ramps connection to settlement layers pricing what are the things that we need to kind of go

11:17offer and build and modify to make this use case exceptional same thing for ai agents ai agents are

11:23going to likely require a other things that are going to be important for them for example

11:33you know i think aliens will be different there'll be some ai agents that'll be

11:36you know purely off-chain and all they'll use for on-chain is is payments and value exchange um

11:43so i think that's like something that may need to be specifically supported um for some of the on-chain

11:49agents it's like okay well how do you go get compute to do inference ideally a roll-up that offers that

11:55natively or makes it really simple for developers is going to be much more useful than uh someone who

12:02uh it's like hey you need to kind of go figure that out on your own if verifiability and the notion

12:09of having a te is important to you uh especially if you're doing on-chain inference and that might be

12:14really important if you're doing things like uh like ai-powered defy where you need uh model inference

12:20as part of you know a default defy protocol uh implementation or kind of risk assessment that's

12:28dynamic you'll probably want to have some notion of uh te and so you want to kind of build that

12:33natively so anyway basically what i'm trying to convey is that the power of roll-ups is you can be

12:38more user obsessed and design your system to for the use case you're trying to to tackle

12:47of course brilliant so uh that's the entire uh debate as of now right the l1 and l2 debate is

12:54for quite significant and it's been stronger today than any any any time else in the past so uh my

13:00next question to you is uh what are the biggest limitations today for stable coins when it comes

13:06to scalability and programming that roll-ups are specifically solving in your opinion yeah i would say

13:15the biggest limitation that i've seen today with kind of blockchains in general is confidentiality

13:21the ability to for a user or for a wallet or for for whomever to send money to somebody else

13:32especially as you get into kind of trading use cases like a robin hood you know if they want to do

13:38settlement there's ways to kind of obfuscate it that but you want to ensure that uh these things are

13:45confidential to the user and to the company and that they don't want all these things to be publicly

13:50available and so i think things like roll-ups um can solve this uh you know to to talk about ck sync

13:58ck sync recently launched the providium and that's effectively um a validium which allows which enables

14:07the blockchain to verify um that the transactions had had the right um state transitions and that nothing

14:16funky happened but it doesn't reveal the actual uh contents of the transactions who sent to who

14:23how much they send etc so this is a uh a really good or sorry a good example rather of a roll-up if

14:32they use a providium for example they use a ck stack of saying hey look this is something that we want to

14:37enable and we're making a trade-off uh of of doing this because the use case that we uh want to tackle

14:43demands so i think confidentiality at least for stable coins is one the other one that i haven't

14:49seen tackled too much but i think will need to be tackled and maybe this is tackled more at the app

14:53layer than at the blockchain layer is last mile so stable coins you know when i was at circle right

15:00the the big challenge of stable coin payments is all it hasn't been the fact that you know it hasn't

15:05really been throughput or scale i know that's kind of a convenient thing for people to talk about

15:08a lot on crypto twitter it's really been around the fact that um if you get a stable coin payment

15:15let's say you're a merchant you're a person like what do you do with that money how do you actually

15:20take that money and take it off of crypto rails to you know do something with uh it's fine if you

15:27know if i pay you in usdc and you want to then go yolo into a meme coin or or buy some nft or something

15:34else like that's that totally works but for most people they want to take that that stable coin and

15:39go buy something in the real world and that last mile piece has been really challenging historically

15:45i think still continues to be a challenge even for merchants things like that

15:49and so um if there is a and i think this is what plasma is trying to solve as well but like

15:55if there is a roll up that decides to take more of a full stack approach and says hey look beyond

16:00confidentiality we're also going to go partner with a bunch of last mile i don't know i'm not

16:04an expert in the space but like last mile banks last mile um settlement centers such that you can

16:13once you have your money in your wallet if you want to go ach it to your bank at any location

16:18any country you can do that like that would be powerful for a lot of users to kind of create more

16:23of an end-to-end solution um then like either making the app do that or something else so i think

16:29those two things like last confidentiality and last mile are things that you know roll-ups can

16:34offer that not because they're roll-ups but because the people building these things can take a more

16:40specialized viewpoint on the info that they're building

16:45okay brilliant uh so i think can we touch upon ai agents a little bit here so with ai agents

16:53increasingly interacting with smart contractors autonomously uh how do you think roll-ups

16:59helps help in ensuring these transactions are efficient cheap and of course trust minimize

17:07yeah i would say that

17:12i'd say a couple things right i'd say number one you know one of the things roll-ups are just good at in

17:17general is making sure that costs stay low and so with ai agents um being able to uh use these roll-ups

17:29like i think uh you know pricing is gonna be key the the where i think roll-ups can really help with ai

17:36agents is really around um identity and onboarding um you know as agents mature they'll need to think about

17:49you know what kind of flavor are they are they going to be if the flavor were they're kind of fully

17:56sovereign and someone creates a wallet for them and then somehow they this this key is managed um by them

18:04and they have full control to the wallet and in that scenario uh you'll want to have probably some

18:10sort of like mpc sharded solution that again the role of a fresher should provide such that the agent

18:16doesn't have to that like never has access to the full key because we've seen with things like freza

18:20that you know very quickly i think what roll-ups can begin to think about is offering uh more the

18:26identity and kind of um the identity layer for agents so if an agent is fully sovereign you don't

18:33want the agent to have uh access to the to the key because as we've seen with things like freza

18:39these agents can be jailbroken so you'll probably want to have some kind of mpc like solution where

18:45there's key sharding and um there's a validator set that can kind of keep um these keys protected if

18:53the flavor of the of the agent is more like a companion where it's like hey the agent has access

18:57to um a user funds to go invest on their behalf but there's some guardrails and limits on which

19:03funds how much things like that you'll want to have some technology like account abstraction smart

19:10accounts um that can enable um these kind of these interactions in the more programmable space so

19:16to me roll-ups can really help agents in my view again it's really this theme i want to hammer home

19:22is um you want to create differentiated infrastructure for the use case you're trying to

19:27solve if you're trying to solve a ai agent use case you should deeply understand what is the pain

19:32point that you're trying to solve for and then build infra for that same on the stable coin side i think

19:37uh team is trying to go build infra first and then figure backing into use cases is just going

19:42to be really difficult today in crypto oh absolutely context so uh my next question to you is let me

19:50just uh jump right to it so uh as you know a roll-ups enable app specific scaling so uh could we see

19:57specialized roll-ups optimized purely for a native scaling points or agent economies in your opinion

20:04what would you think it would look like yeah 100 like i think that is probably the end game the

20:10end game is to have hyper specialized blockchains that are optimized for a specific thing and like

20:18that's the best uh option for a blockchain and there's a whole different debate on like is that even

20:24the right path or um is the app who's basically building on top of this should they actually

20:33vertically on the stack downwards and create the blockchain themselves versus a

20:38infant provider that's building a hyper specialized use case that's a different debate but 100 agree

20:44i do think i think i certainly think you'll see experiments on um ai you know ai kind of

20:55managed hedge funds or ai managed traders that people can lp into like i think that's you've already

21:01seen examples of that um there's some good ones on base and a couple other places so i think

21:06that that you'll definitely see ai native stale stable coins seems like that there'll be shots on

21:12goal i'm less uh i'm more skeptical on that actually playing out just because i don't think that a

21:20stable coin needs to be ai managed unless you're trying to optimize for yield but i think you could

21:27do that a different layer than kind of building in the protocol what i will what i do think you'll see

21:31is more like ai native d5 where today d5 um is you know to some degree uh static about the like the

21:40risk parameters for ltv and for some of these amms sometimes they're like just configured at a static

21:47config file and then you have to have a governance vote to change that i think things like that make

21:51sense for ai agents or ai models at the least to uh be able to to modify those based on what the

21:58context they're seeing so 100 agree that you'll see that and you know as we see more agents come

22:04on chain it is inevitable that there'll be agent economies that that proliferate where agents will

22:10begin to think about um you know paying each other or that you know if they're fully sovereign accessing

22:17capital markets if they decide to tokenize themselves and then do things with that money like you know

22:24acquire human labor or invest in other things etc okay brilliant that that sounds pretty promising and

22:33hope that is that comes to fruition sooner than later so uh my last question uh for the session today uh

22:41carthag is uh what's your long-term vision on a vision for on-chain ai and rollups uh do you see a

22:49world where ai agents are a major economic actors transacting in table coins powered entirely by

22:55the role of native infra in europe yeah it's a good question um i absolutely think that

23:05there is a world where ai agents are major economic actors on chain my view on this is that

23:13uh there needs to be a good reason for agents to come on chain and in my view payments and capital

23:21markets isn't the isn't the right wedge if you talk to most ai agent builders like there are payment

23:28solutions out there um like in kind of traditional quote-unquote web 2 that work perfectly fine for

23:33agents and there are other ways to get money for capital markets ie fundraising ie getting access to

23:40liquidity that are again all perfectly functional in traditional environments so my view on this has

23:47always been that uh maybe a little bit contrarian and this is a big thesis of of the fund that i'm

23:53building is that model scaling is going to be the right wedge or is going to be the wedge for for

23:58agents and what i mean by that is that in my view we've like i think we're kind of ending the

24:08era of what and like we're towards the end of an era of model scaling which is this era where hey if

24:14you you need a ton of money and you just need to pump that into data and compute and human labelers and

24:23even like test time compute it's kind of all part of the same era and um my view on this is that the

24:30next era of model scaling is going to come from a lot of it's a little bit more of an unknown but it's going to come from

24:38more experiential data it's going to come from a bunch of people trying different techniques i think

24:43the days of just going and hiring like you know these hundred million dollar i know it's just happening

24:48now but these crazy huge ai labs um to me is um i think well time will tell is it going to be a bit

24:56outdated and so in that world i believe crowdsource strategies um similar to open source software right

25:04where a bunch of people on the internet thinking about a specific use case again when i say next

25:10year of model scaling i should kind of uh preface this i don't necessarily mean frontier general

25:16purpose llms i think you know frontier labs that do this i think company formation is still likely the

25:21best way to do this i'm talking about for more specialized use cases you're building a model for an agent or an

25:26app that's really good at x-ray detection or you know doing your taxes or whatever um for those kind

25:33of things i do think crowdsource strategies will over time uh yield better and cheaper results than

25:39trying to go stand up a whole ai staff and paying a bunch of people a bunch of data labelers and all that

25:44stuff um and these models will complement frontier models but they'll kind of be the secret sauce for

25:50these agents and crowdsource strategies i think will benefit most from um from the fact that you can

25:58pay people to contribute or incentivize people to contribute uh on chain so i think if model scaling

26:04is the wedge that brings people on chain then over time agents will and teams will find it more natural

26:10to um to do payments uh on through stable coins uh access capital markets and i think roll-ups will

26:18will be an opportunity for them or the infrastructure for them to go do that i'll the last thing i'll say

26:23is you know no i know um this chat is a bit more roll-up centric but again this comes back to the

26:30infrastructure piece like the infrastructure has to be great for that use case you know if the roll-up can

26:35be specialized for a specific use case i think it can make it can make sense if it's not and it's this

26:40general purpose thing or it's not as performant or effective as some other l1 or some other blockchain

26:46it's going to lose i i think humans are much more emotional about oh i'm a ethereum or a solana or

26:54whatever ecosystem player i love evm i love the ethos here most of these teams coming from ai or

27:00certainly agents are not going to care at all about this stuff right they're going to pick the best

27:04solution uh for them and so the solution has to win out not the ethos

27:09oh absolutely we uh agree completely with that so the communities that we work with uh hold the

27:19hold the particular stack that they follow very close to their heart so they can vouch for that

27:23in more ways than one cardiac so uh we're at the end of the session uh as of now so do you have any

27:30last words for our listeners no uh i think you appreciate you having me on i would say again for

27:39any listeners out here building infrastructure um just to kind of keep hammering at home um make sure

27:46you deeply deeply understand the user and the user in this case is probably a developer that you're

27:52building for understand their pain point um and and make and understand and ideally um identify

28:03how you can measure that point point that way when you offer your solution it can be very oriented to

28:09their pain point and you can verifiably demonstrate that your solution makes their pain point better

28:16because it's measurable so if you can think about it from that frame i think you'll be better off and

28:21it's it's a it's a harder frame to think about it it's not easy but i think that you'll be better

28:25off as a as a builder brilliant okay uh thanks thanks for your work karthik so we will have uh the

28:34recording of this session on our brand new youtube uh podcast channel if anyone anybody who is uh who

28:41wasn't able to money so they slim for us so uh we usually have quite a more quite a few more listeners on

28:49us so uh yeah you can go and listen to this a very recording on youtube uh after the fact so uh by

28:57saying that i would i would i want to thank karthik once again for patiently accepting our invitation and

29:02coming and discussing on on topics of stablecoin and on chain ai and these chances were absolutely

29:09brilliant and we will be posting and you know publishing the same uh transcript of this recording on

29:16on our website as well and for everything else uh you can uh our listeners can visit zeep.io uh

29:24zeep.io slash rollup if you want to build rollup with us and zeep.io slash app chains uh to build our

29:32app chains and of course computers.io so uh thanks again karthik for joining us and uh we have another ama

29:40on 24th of uh this month and our ceo dr ravi will be watching that and that's with shadium and that's

29:46what's coming next so that's that katik so thanks again

Category

Transcript

Be the first to comment

Recommended