The ONLY Practical Crypto+AI Use Case | Kartik Bhat - Sei Labs

[music] Cool. Thanks for coming out today. My name is Cartik. I'm a founding engineer at SE and today I'm going to be talking about X42 or as I like to call it the only practical use case in crypto and AI. So brief background, I used to work at NVIDIA.

I was on the self-driving team, was at Carnegie Melon after that doing a lot of AI research and also worked at Google. So, I come from a more traditional AI background and I care a lot about practicality, not just ideology, right? Um, and today we'll first go over the crypto and AI landscape. Then we'll talk about sort of X42 and why it actually brings practical benefits and then what we're doing here at say to improve the X42 ecosystem. So I think there's a fundamental disconnect in crypto and AI.

Whenever you ask someone in crypto, they always say AI needs crypto. But then when you ask anyone in AI, they say definitely not. Everyone in crypto thinks that you need decentralization, tokenization, onchain agents. But in reality, you just need the normal centralized APIs, traditional a infra, and anything out of the box right now, right? And so the challenge that I'm really interested in is how can we bridge this practicality gap?

What actually is a net new use case that crypto provides that normal users can benefit from? And I think that's payments. So I'm going to run through some examples of these crypto and AI projects that I think are, you know, interesting to say the least. There's a new chain called Sentient out there which is, you know, supposed to help incentivize people collect data, run emissions. you get a token if you, you know, contribute to the model.

None of this is actually useful and and I'm willing to bet that these new models communityowned don't actually provide a benefit to users. It's just decentralization theater. Story protocol is another one. You know, they raised so much money to build this like IP layer to collect data. You know, they they're they're clearing the IP that normally you can't access data.

Give it a second. So story is another one where you know they raise a lot of money. There's a lot of incentive to say that IP is something that blockchains can enable. But in reality blockchains are unnecessary here. Data gets collected fine through normal data collection schemes openai the big labs etc are collecting data fine.

There's no net new addition that's practical here. And I think the chief problem with the way that we in crypto think about these problems is that we start with this one broad problems. Maybe I don't trust open AI. I don't trust these frontier labs. And then they go to an extreme.

Oh, I need decentralized private inference that's communityowned. In reality, it's more like something in the middle, right? If you don't trust OpenAI, you don't trust the big labs, you might go to something like open source or something like a hosted open source. You don't need the blockchain or decentralization just for the sake of it. And another example is ERC80004.

Again, I personally think that it's interesting, but it's mostly overkill. Reputation and ID does not need to be on chain. You could probably just use DNS and signed headers. So, the takeaway for me is I think again blockchains are made for financial transactions and there's one key thing that causes issue right now for agents and that's card networks. So, let's do a brief refresher of card networks.

Right now, they have a minimum fee. So they're 30 cents usually or even higher on normal card networks where you cannot do a microtransaction. Settlement is T+1 or T+2. So I don't have instant settlement. I have to wait a day or two.

There's always a chargeback risk. So you know if I buy something there's a potential that uh it it gets reverted. Right now there's no global access, right? It's very like country by country and depends on the card network that you use. And another important case is that it's not 24/7.

and it's only on business hours and depending on the rail you use it's it's not consistent and so I think stable coins have four clear properties that are practical for normal users Ignore the decentralization they're micro so you can send micro payments which are less than a cent you can instantly settle them right so on normal fast finality chains like say they settle in like 400 milliseconds as opposed to t+1 or t+2 they're global by default So anybody who trusts the network by definition has access to these stable coins and they're crossber operating 24/7. And they also have some cool programmable properties, right? Like they have guardrails, spend limits, whatever you need you can program in. And so I want to walk you guys through this how you open claw and this whole explosion of AI agents, how you normally connect them to these MCP tools. And this will kind of paint a picture for why I think stable coins make perfect sense for payments of these agents.

Right now, if I have an openclaw agent or whatever agent I have right now, it first discovers these tools. Like let's say I want to find the weather in San Francisco or I want to buy uh specific data about uh you know a specific trade. It has to discover each of these tools separately. It has to manually create an API key for each of these tools. Now again that's a security risk, right?

I have to manage those keys somewhere. I have to store them properly and make sure they don't get leaked. And then I have to pay a subscription in fiat at the end of each month. Right? So for every tool I have to set up an account, I have to pay it monthly and I cannot get charged per use.

It's a system fundamentally designed for humans and not for agents, right? It's very high friction. And when I'm vibe coding or I'm just having my agent do something, I don't want to have to manually come in and set up an account for each of these specific tools. it should be able to discover it on the fly and do what it needs to do. And so the crucial unlock that a lot in the industry are noticing um even outside of crypto is give it a wallet and give it stable coins, right?

So if my openclaw agent has a wallet, it can discover each of these tools like getting the weather in San Francisco or whatever other instant tool that it needs and then it can instantly access it by paying in a microtransaction rather than needing to sign up or manage an API key. Right? So again, no management of API keys. You only pay for what you use. And because micro payments are allowed, these APIs become allowed to uh pay on the fly.

So again, like if I request the weather in San Francisco, you're talking about like a hundth of a cent, right? And that's just not possible when card network minimums are 30 cents, right? So this sort of flow makes a lot of sense for normal users. You don't have to care about decentralization. You don't have to care about censorship resistance.

You just have to care about what's practical. So at say where are we focused? X42 still is very early and and candidly there's a lot of missing pieces but we feel like this protocol has a lot of potential. So let's start with the wedge. I think one huge issue that people notice for agentic commerce is that they jump to the top left quadrant.

So you know this is the transaction value mapped against the transaction frequency. Stable coins do not make sense for high frequency low value transactions or highv value low frequency transactions sorry. So for shopping and commerce where I pay like $10 very infrequently stable coins make no sense. Chargeback actually is needed because if I don't receive let's say the jacket I need to claim fraud. I only do it maybe once every week or once every day.

It's not very frequent and car networks sort of fulfill their purpose here, right? It's way above the 30-cent minimum. X42 doesn't matter. So when Stripe talks about, you know, ACP or all these other like, you know, agentic commerce protocols, shopping is not that interesting to us. I think where stable coins and XRO2 actually win and where the wedge that we are focused is the bottom left, bottom right quad quadrant.

That's very high frequency but low value transactions. So these are APIs in particular, right? So you're talking about, you know, just like the getting the weather from San Francisco, you're talking about a thousandth of a cent. You're making a lot of these calls. Stable coins make sense because they're micro and they instantly settle.

So I don't have to deal with any chargeback risks and I can settle each payment independently. That is the wedge in my opinion. So how can we take this wedge for APIs and actually productionize it? We've been looking a lot into these proxy servers. So there's a really interesting business model that pops up with stable points.

I can buy APIs in bulk and I can resell them per call in stable coins. So think about it like in Costco I buy a bunch of toilet paper. I can then sell you individual toilet paper rolls at my deli or wherever that is per use. And so that's what's happening right now in the X42 ecosystem as well. There's all these upstream APIs like OpenAI's API, you know, a weather API that maybe one organization can set up a fiat API account and then I set up a layer which is like this proxy server layer and anybody can pay per use against that API that I'll proxy, you know, to that upstream API.

They can pay per use in stable coins. So any agent can maybe pay me a scent. they can it will go through to my API to let's say open LLM but they don't have to worry about setting an API they don't have to worry about paying in fiat just what they use right and so you can even charge a margin on top of that but this sort of buy API in bulk and resell per call is in my opinion an important wedge right another huge problem for you know X42 is that fundamentally stable coins are not widely adopted right like in crypto and you know in in this conference People are excited about stable coins. Normal people like, you know, throughout like the US, they they they could care less, right? And what are they used to?

They're used to credit cards, right? So, if I want something, I get it. I buy it immediately. There's no friction and it works, right? But with stable coins, I have to deal with this whole on-ramp mess, right?

And this has not been solved cleanly. So, stable coins operate like debt fundamentally, right? You have to fund it first. Um, so if I don't have USDC, I have to go through this hurdle of like going through a crypto exchange, doing KYC verification, waiting for the bank transfer, like it's just such a high friction that normal users are not going to do this, right? And if I had to do this every single time, like I just drop off.

And so we're thinking a lot about these preunding hurdles that we can get over with things like an embedded on-ramp. And you know, this is common in the industry. There's, let's say, Coinbase Pay or Apple Pay. uh inside the app you can sort of do Apple Pay and maybe get $20 in USDC. But the problem with this is of course they hit limits, right?

The the maximum is usually like $20 or $25. The more interesting case is the credit layer. Um and so in this case, I'm not going to go into a ton of detail, but you can think about having this basic risk engine that fronts you the stable coins and then you pay in fiat maybe later on at the end of the month, etc. Um, and so the risk engine can be based on the agents onchain history, uh, your repayment behavior in your identity, and it can front you, the stable coin so that you don't have to do that on-ramping itself. And there's a bunch of protocols coming out here, but this avoids the UX problem of needing to on-ramp.

We've looked a lot into like facilitator fees and so for those who are familiar with X42, we actually have a proposal out now. Uh, facilitators are the gateway to X42, right? So when I send a transaction, it lands on the blockchain via facilitator. It's almost like an RPC or a paymaster. But the problem right now is that facilitators are charging fees.

There's no consistent standard to disclose those fees. So am I charging BIPS on those transactions or am I charging a flat fee? You don't know. Um so we're trying to standardize that and make it really easy to route dependent on which specific fee model you prefer. Um and then we we we care a lot about privacy.

Yeah, I I won't spend a lot of time here, but the interesting one that even at the L1 layer that, you know, we're thinking a lot about is payment lanes. So, fundamentally for payments, you care a lot about two things, inclusion and latency, right? Trading is a very different case, right? You care a lot about ordering, but when there's high congestion, I could care less that my transaction sending money to this API or this server uh comes before someone else transaction. Like there's no price difference, but I care that it's included immediately, right?

And so you can create this like sort of payment lane that validators uh can can sort of stake to be a part of and you can guarantee inclusion and maybe take like a BIP fee on top of it, right? And so this makes payments guaranteed for folks who need like guaranteed access to like these APIs and congestion that is caused by let's say trading or other activity on the chain doesn't influence X42 volumes. Um yeah, so that's it for us. Like again like I'd really love for you guys to reach out to me. This is my Twitter.

This is my email. We're thinking deeply about Oh, it's again. We think deeply about how we can bring X42 to the masses, not just is this like an interesting nerd type. So, I'm really interested in the sort of product research that can actually bring this to users outside of crypto, not just people who are at this conference. Appreciate it.

Thank you. [applause]