Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
SKALE School: ERC-8004 with @VittoStack
Recorded: March 10, 2026 Duration: 0:57:26Space Recording
Full Transcription
It's scale school time ladies and gentlemen boys and girls we're gonna be diving into some very very interesting fun topics today
with
Mr. Vito from the aetherium foundation. Hello, sir. Hi everyone
To be here. Thank you for having me. No worries at all pleasure for you pleasure
Yeah, yeah, yeah, yeah, yeah, yeah. It's a pleasure for you to join us. But no, thank you so much for taking time out of your busy day, your busy schedule for jumping in and, you know, having a grand old conversation. So yeah, how you doing, man?
Better than you, apparently apparently you're having trouble today
yeah it happens so uh normally this show is actually an hour later so i'd be like
an extra coffee deep um but americans decided to change their clocks like
a month earlier than everybody else so uh yeah we're starting the show a little bit earlier.
Thanks, Sawyer.
It's literally the same exact time for me.
Yeah, it is.
Not for everyone else, though.
Not for the rest of the world.
That's tough.
That's tough.
Well, I'm excited for this one.
I think there's a lot of traction in the agent economy, AI agents, etc.
So pretty pumped.
This one's going to be a little bit less agentic commerce-y and maybe a little bit more broad on AI, but we'll see.
But yeah, I'm pretty pumped.
Hell yeah, hell yeah.
So Vito, why don't you give us a little intro about yourself
and then we can kick things off. Yeah, 100%. So hi, everyone. I'm Vito, why don't you give us a little intro about yourself and then we can kick things off.
Yeah, 100%.
So hi everyone, I'm Vito.
I work on the DAI team of the Ethereum Foundation.
The DAI team of the Ethereum Foundation does a bunch of things, but mainly is focused on
AI security, privacy, and genetic economy as a whole.
Plus we do, of course, other things always aligned to the Ethereum Foundation mandate. Me uh me personally i'm the ai coordinator the team is very small and it's kind of new
david the crap is the the head of ai kind of founded it back in august last year uh but
we worked on things like 2004 that is one of the topics of today's conversation oh yeah
hell yeah and soya you know you've been playing around with 8004 a lot um and actually
you know what we'll do um before i go over to soya vito why don't you give the people a little
bit of an introduction into exactly what 8004 is absolutely uh this is going to be a long one
no i will try to make it as short as possible um but it doesn't for so it comes from kind of a few problems that
davide marco and and other folks started noticing in the industry so taking a step back is a new
standard right it was released the 29th of january uh is written by folks from coinbase eric rappel
from jordan and this from google marco de rossi from metamask head of ai metamask and david the
crappies head of ai at the ethereum foundation, and David Krapis, head of AI at the Ethereum Foundation.
I know the official story, and it's a fun story.
David was already thinking about agentic commerce and how agents talk to each other, how they work to each other.
But he knew that it was like a big problem.
The first problem is one that we all experience, right?
When you want to decide which tool to use, but even just what API to use.
You only have so many ways
where you can find the right tool for you,
whether it's asking your friends
that already are using that tool,
or it's like going on chat GPT,
so using an agent to tell you which one is the best tool
for whatever you want to accomplish,
or you go on Google, but these are kind of the ways,
or of course, you start trying them all,
spend a bunch of money and see which one is best for you. is already a big problem and it's even a bigger problem if you're expecting
agents to find the right tools for themselves the other one is understanding the quality of
these tools right like how do you know the quality once you have understood like what is the least of
the tools that you can use well the only way you have is literally trying them or like looking at
reviews online but
with the possibility that some of these reviews might be fake or the reviews are not like looking
for the use case that you're actually trying to to use that tool for so all of these were a bunch
of problems uh on top of it there were not like real identity attached to this agency it was kind
of early so there wasn't really the concept of like the open club going out in the wild and doing things but you already could see like that that could become a problem
so they thought about 8004. 8004 at the core is a very simple standard it's probably one of the
simplest standards we have right now in amongst all the urcs is like three registries two which
are already online uh one is still not it would go go out, I think, at the end of this month,
end of March, early April.
So you have an identity registry
which contains all the information related to the agent,
whether it is the name of the agent, the description,
if it's X402 compatible or not, what are the services.
And when I say services,
I literally mean any type of service.
It can be A2A, it can be mcp it can be an endpoint but
it can even be like an email an ens a did and many different things what are the skills what is the
domain so identity registry is the identity of the the agent now we can go more technical is a 721
is an nft is connected to it but that's not really important you just understand the identity
registry is what makes the identity,
stores all the identities of these agents.
And then you have another registry
that is the reputation registry,
which contains kind of like the feedback of this agent.
Now, the way I usually explain it is
you can think of like Yelp stars
or like Google business stars.
The reality is that it is a little bit more complete
and comprehensive than that.
It's much more similar to if you have some experience on SEO, how Google ranks pages
on the search engine results page.
So you have the quality of the past operations, quality of pages, you have the latency, you
have the uptime, you have multiple different things that then get aggregated and give an
overall high level score of that page
in this case of an agent.
And then the third registry,
which is the one that is not online yet,
is a little bit more esoteric,
but it's like a way to validate the operations of this agent.
Now, the way I like to see, to think about this
is think when you go on like chat GPT
and you send a message to 5.3, right?
You don't really have a way
to understand if you're talking to 5.3 3.2 4.1 like they have all the numbers right uh you can
only trust the ui now that's not a problem if you're using chlgbt to know hey how do i cook
this steak and make it like fantastic it's much more of a problem if you're dealing with like
high sensitive financial operations
or like dealing with your health data and anything that is kind of private.
You want to make sure that the setup of that AI is kind of like trustable
without, of course, having like trust assumptions.
That is something that we don't like, right, in our industry.
So how do you do that?
You use like a validation network.
You can think of like an Oracle network, whether it's Redstonestone chailing and all together they can test the operations of this agent replicate their setup
see what is the output compare it and make sure that the what the agent is telling you is doing
is actually doing now taking a step back the 2004 i told you it was going to be long no no man you
are you are 100 nah trust me you are 100 fun Nah, trust me, you are 100% fine.
But the thing is, with 2004, we call it the trustless agent standard.
I think that name is doing more harm than good in a way
because it's literally like the trustless service standard.
It's not only for agents.
It's for any service.
It's for APIs that need to be found.
Think about an example that I always make.
Right now, when you want to build an agent, you need to put different tools. Think about a universal MCP
that gets the request from the agent, goes on the identity registry, checks the service
that does, I don't know, gives you the price of oranges and checks the reputation and gives
you the best service for the price of oranges at that price for that day. So that's what 1004 is. And these are the problems
that it's solving. Amazing. Amazing. Oh man, there's so much to kind of dive into. But I think,
first of all, thank you for the detailed breakdown. The growth of the 8004 has been really exciting. I think one of the things
that I think you guys have done really well is obviously it's an Ethereum Foundation backed
initiative, but you haven't really denied anyone who's wanted a deployment. And I think that's
really important because there's so many developers building in so many ecosystems, scale included,
who want to contribute and be a part of this agentic ecosystem and community.
My kind of first question for you, thinking about how there's all these chains and we have agents registering everywhere.
Why does someone pick one chain over the other?
Why would someone pick scale versus polygon versus Ethereum versus Solano?
Why should someone register on one versus the other?
It's a question I've gotten quite a bit and I have my own answers,
but curious from your point of view, just they're on one versus the other. It's a question I've gotten quite a bit and I have my own answers,
but curious from your point of view,
yeah, what's kind of the reasoning that Sloan may choose?
Yeah, that's actually a great question.
And I think you have like multiple separate reasons.
One is probably like a practical reason
and what is available on that chain, right?
I mean, taking a step back,
8004 kind of tends to be multi-chain by default,
because if you already look like at all these scanners,
they're already exposing like APIs,
like 8004 scan is one of them,
which everyone thinks is from the Ethereum Foundation,
it's not, it's a different company,
but they're doing an amazing job.
They have an API, you ping the API
and it gives you like a list of agents, not by chain,
if you want this by chain,
but it's usually just aggregated and run
to based on their own reputation registries.
So that's not really the reason.
I think different chains have different tooling
that is available right now, right?
One thing that I'm thinking is like X42 facilitators.
X42 facilitators are not available on every chain.
And that's a big component of this agentic economy right so you might want to pick the ones that have more facilitators whether it's
polygon whether it's base for example on mainnet we have very little right we're working now on
getting more facilitators of mainnet now why don't we have and it kind of like a trickle down problem
like why don't we have lots of facilitators well that's another thing that you want to consider when you're deploying your agent is like probably
gas fees right your agent wants to to transact a lot it will transact a lot you don't want to spend
like hundreds of dollars on on gas fees and on all these transactions so you want to pick a chain
that has probably high tps high transactions per second second and you want a chain that kind of
like is affordable when it comes to gas fees.
I think another reason is for sure like ecosystem, right?
The different chains are putting different amount of efforts in prioritizing the agentic economy differently.
And that means a lot.
There is chains that are writing to me every single day being like, hey, how can we contribute to this?
How can we work on that?
A, we want to work on a watchtower.
We can chat about watchtowers later, but we want to work on the reputation problem because it's a problem,
right? So all of these is going to be probably a differentiator. It's already a differentiator
when it comes to builders building on their own chain. And I think that the last reason is probably
just like belonging, like which chain kind of reflects your personality more. That's always
been a thing, right? In Web3. The Solana folks like Solana,
not because sometimes he has better TPS or whatever,
just because they grew up on Solana.
So they want to stay on Solana.
They know more people on Solana.
So different reasons.
Yeah, I think that's kind of my summary.
No, fair enough, fair enough.
I think the one that I personally tend to use the most, I don't think it's necessarily true yet, I think it will be, is the cost of feedback.
It's something that was raised very consistently in the Ethereum Magicians thread during the discussion around 8004 is the fact that posting feedback on chain, to your point from a gas perspective,
just doesn't really make as much sense on chains like Ethereum where, yeah, it's more expensive,
right? You're paying for a higher level of security. There's a much bigger validator
ecosystem. And, you know, for some things, it just may not really be feasible or realistic.
And so I think as we see more and more agents come online, as much as I love the
scaling Ethereum roadmap, I do think that we're always going to have a place for this kind of
growing ecosystem of chains supporting this because it's just really not feasible that
everything's going to run on one chain. It never really has been in my opinion, but hey, that's That's just my two cents. That being said, I love the call out around availability of resources.
This is something I think is really interesting and not thought about enough is how do we create
an agent? How do we get an agent up and running? How does that agent stay alive? How do we create an agent? How do we get an agent up and running?
How does that agent stay alive?
How do we know that agent's legit?
There's all these things, that terrible word,
things, there's all these things we have to think about
when it comes to an agent.
And we're obviously in the early days, right?
So we have things, again, things like the watchtower.
We have SDKs like Agent Zero. We have your CLI, the
Create Trusted Agent. We've got X402. We have all these resources. And I think what's really
great about 8004 is in general, you can kind of stand up everything you need by yourself
if you want to. But we're seeing these resources grow.
Before we kind of jump into maybe some of the new things we've seen,
what do you think is kind of the biggest missing piece right now from 8,004? We've seen a lot of agents get stood up, but what are we missing to really see these these agents uh i don't know
maybe become a bit more mainstream yeah multiple things is not one thing is a lot of things right
when we talk about agentic economy economy is a big word like he has a lot of parties involved
right even the real world economy is is millions billions of people coordinating each other to make it work.
So the same thing is for agents.
For the genting economy to exist, you need different things.
You need payment rails, whether it is X42, whether it is like something new that is going
to come up, who knows?
You need that.
Again, you need identity, but you also need, again, ways to verify the reputation, right?
Reputation right now is a problem but it's not
just a problem for 2004 if you ask the google folks hey how do you deal with the feedback on
google business they're gonna cry right away because it's a problem and that's and that's
much easier to solve right because it's like a walled garden and they have like all the moderation
tools they can decide who gets approved and who doesn't here we're talking about a completely
open standard and everyone can leave reputation feedback so how do you make sure that the reputation feedback i mean to give an
example we released the 29th the day after there were already companies selling feedback services
where they're like getting paid to leave feedback for for agents which is ridiculous but that happens
right it happens in on yelp it happens on google business it happens in web free
so how do you make sure that a feedback is legit?
How do you verify who's leaving the feedback?
How do you verify that the job really actually happened?
How do you keep track of all the agents?
So definitely there is a level of abstraction
that needs to be built and is being built.
We were setting up this thing called the Reputation Alliance,
which is just like a number of teams that have been working or are working on like reputation systems and um and
they're building like different in different ways they all have their own thesis their own testing
their their own thesis uh we have the concept of watch towers that i that we mentioned before
where you have these like you can see them as cron jobs, essentially, they are going to go agent by agent,
pinging them, making sure that the agent is live,
that the endpoint to that agent is actually reachable
and leave a feedback hint based on that.
I think these two are definitely like the biggest ones.
I think another one is making 8004 useful, right?
So what it means is you need actually useful services you need
useful agents on it uh we have plenty of agents right now i would say a good number of them is
is not useful like it's just like we have them because it's like the last shiny thing everyone
loves the word agents but we need like not only agents but agentic services that do things and because of that we
were talking to companies that have already released like redstone like chain link like we
need the price feed on 2004 because we want agents to use it so we'll say feedback and services are
the first ones and then it's like a broader adoption how do we get out of like our niche
uh talk to the actual ai teams to the actual like bees of the situation,
to the banks of the situation.
What I can tell you though is that banks are like super excited.
I've never seen banks so excited to work with like blockchains,
like since we started talking about the genetic economy,
because they're all looking at agents.
How do we make sure that agents are passing KYA,
know your agents, AML regulations?
Well, 2004 can do a bunch of things for it to do that.
Yeah, that makes a lot of sense.
I think to kind of maybe put a bow on that,
there's a lot of things that just have to be
kind of solved in tandem here.
I think it's really interesting you call out
the use of the, we need agents that actually do things properly.
I need a rubber band so I can stop saying that word.
That's terrible.
But it's funny because that was actually one of the conversations
I've been having with a lot of the teams building in X402 as well
is we see this level of, call it like spam, right? And I mean, blockchains have cult-like spam, right?
And I mean, blockchains have always had this problem, right?
Like if you look at Ethereum and base,
it's like a significant amount of the block space
is quite literally filled up by just spam
that is annoying in there.
And so we've seen this with the ERC-8004 registry.
There's all these agents that get created,
then they're just empty.
They don't do anything.
They just are there. They take
up space. It's annoying. And then unfortunately, because of the design of most blockchains, it's
not really feasible to filter them out on chain. So they filter out off chain,
but that's beside the point. But it's this question of what does the agent actually do?
And it's really interesting because I think we, as an industry, we jump toward, hey, let's build agents.
And there's this belief that, which I think kind of came from OpenClaw still,
there's this belief that every agent will be autonomous and make money.
And I personally don't agree with that. I think there will be agents that
are stood up and they're fully self-functioning. But I think the most valuable agents are ones that
are almost like your personal assistant style. I have all this information that comes to me every
day and I have all of these things I have to do, rubber band.
And it's just hard, right?
Like the more the AI world grows,
the harder it gets because we're just trying to play catch up all the time.
And so I see like the personal agent
being the number one type that gets created. And I think so far everyone feels like they want
to create autonomous agents as part of the registry, but I haven't really seen people start
to say, well, I can create a personal agent that is designed to be my public presence,
put it on the registry. And then if people need to send me an email,
right, my agent has an agent mail email and people can send it to my agent inbox or,
right, it has access to all these resources and people can ping and pay to communicate and
contact me and my agent has these payment requirements to filter out spam. I haven't
really seen that yet, but I think that could be a really big catalyst for growth because I think it's so, like to me, it feels obvious.
I don't know about you, but I think everyone kind of would love a personal assistant,
but it's not really feasible for everyone to have one. So maybe this is a way we get there.
But yeah, not sure how you feel about that. Yeah, like the yellow pages of personal assistants, right?
You want to talk to the Sawyer,
you go on the 1004 scan
and look for Sawyer personal assistant
and you have, absolutely.
I think that's the thing about the 1004, right?
It's a registry and it's a directory.
It's like a open Google, you can see as such, right?
So you can register whatever you want,
whether it is the agent
that is your personal assistant
and you want it there
because you want people to find it
and maybe becomes like part of our culture
to go on a 2004 scan
to look for your personal assistant
before trying to spam you on Twitter,
trying to reach out to you.
It could be the service
that just like wants to be found
both by user and by agents
and wants reputation and stars there.
So there is a lot that can be done.
I think I don't see if your agent doesn't fit
in any of these categories,
so it's not helpful for the public,
it's not helpful to provide a service
and putting a user in contact with you is also service, right?
It's not helpful to optimize your yield.
It's not helpful to help you do groceries, to book an Uber.
They probably shouldn't go on 2004.
Like, you can stay on your Raspberry Pi, on your Mac Mini, on your house.
And that's totally fine, right?
But everything needs to go there.
Now, there's also, like, situations where just people register on 2004 because they want the cool fine, right? But everything needs to go there. Now, there's also situations where just people register in 2004
because they want the cool number, right?
Like you can try and look at how many wanted the number one on each chain.
We literally had the bots, like, snapping that up.
That happens a lot, and it's totally fine.
It's very early, but in many ways, like, it's not just with 2004, right?
Even if you look at OpenClaw,
and that's a big discussion on twitter right now lots of people went completely crazy on open claw but very little
use cases i would say other like the personal agent i've seen people spending tens of thousands
of dollars because they wanted their agents to make money by themselves i think one succeeded
and it was because it went viral on Twitter, not because their agent did anything crazy. Yeah. It's definitely funny, I think because
there's, I get this question a lot, which is, how does blockchain truly derive value from this?
And I think a lot of it for me comes down to in the short term, I think the
commerce pieces will be the easiest because it's easiest to visualize and manifest this concept of
my agent is paying for something. But to your point, and I think a lot of people agree here,
we need agent first things. This is a totally separate topic that we won't go into a tangent
on because I could talk about it for hours, which is I think a significant amount of the world is going to, I think, maybe not a significant amount of the world.
I think a lot of the internet-based services, we will start to see new companies spin up that are just fully agent first.
Agent mail is a great example. It's an API first email.
There's a reason like, I mean, I love the Google team. I love Google as a company,
but using OAuth with email is really hard when you're trying to run multiple agents.
It's hard to bifurcate your email accounts and set that up securely and properly.
You don't want to give it full access to your email. and set that up securely and properly. You don't want to give
it full access to your email. So I think we're going to see a significant shift in that direction.
And I think that will ultimately help back with your statement and what I agree with of we need
more relevant things for the agents to consume. When that happens, then again, I think becomes
easier because to your point, maybe I've got one public agent and I've got 100 private agents that are doing all sorts of different things.
Those agents can be transacting on chain.
They can be leaving feedback and paying for things.
They don't need to be registered.
It doesn't matter.
They're consuming and using blockchain as their primary consumption of identity and payment rails, but they don't
actually need to be registered, right? The registry is, I think to your point, it's almost,
it's a great point. It's almost like the open, it's almost like the open page rank, right? It's
like the new way of thinking about how we find and discover things, but we don't need every single thing on there.
Rubber Pant.
Here's how I'm going to get myself one.
But I digress.
Anything else, Vito, there before we move on?
I wouldn't say so.
I think we kind of touched on everything.
I just want to reiterate on the fact that if you have, like, an API,
but even, like, the classic, you know, weather API,
that should probably go in 800.
That's a great use case for 8004,
because you're getting a place where it can be discovered,
that it can be used,
you can be included in all the future developments.
That's the interesting part, I would say, of 8004,
is that, really, like, if you get your API
registered there and tomorrow creates this universal MCP, then your API is probably going
to be the first ranking weather API in 2004.
Great business opportunity.
I agree.
I was actually talking with a couple of the weather teams and man, there's a number of
opportunities for people to help bring people to this because people outside of our little bubble are just,
they don't, it's hard, right? But beside the point, I actually have a question for you. And this is
something I've been wondering a lot. And I've gotten others that asked me the question. So
you'll, you'll, maybe you'll be able to help me answer this better in the future.
this better in the future. 8004 does not require X402 as a payment mechanism. X402 is, I think,
right now the most popular. And that's because I think X402 and 8004 are just very complementary.
They both have a similar, I think, thesis of very simple, right? As simple as possible, very extensible. But one thing you just said,
which is interesting, and I haven't heard others say this is you said, register your weather API
on 8004. 8004 to register is you're registering an agent identity, right? So my question for you is
Right. So my question for you is when we're thinking about onboarding people and explaining this to businesses, is it a I'm registering a weather agent who then has these APIs and I'm calling the agent and they're then like I'm saying, hey, what's the weather? And they're sourcing this information for
me? Or is it purely as part of the discovery layer, all of their resources are embedded
and 8004 is actually the secret discovery layer that we didn't realize for X402 and
agentic visas and things like that?
Absolutely the latter. So I wrote an article recently that and, you know, agentic visas and things like that. Absolutely the latter.
So I wrote an article recently that is,
you're reading 8004 wrong, which is exactly about this.
Because again, the title, Trust as Agents,
kind of puts the agent word subject in every single sentence
when you're describing it.
But if you look at the standard, if you look at the specs,
there's really nothing that separates agent
from just services, right?
You can, yes, add your agent
if he has like an A2A endpoint,
but you can just add your API.
Because if you look at the services
and the registration file, really,
you have a name, which can be weather API.
You have a description.
This API gives you weather information of things about things.
It's getting contagious, but of weather around the world.
And then you have services.
And within services, you can have HTTP endpoints.
And that's it.
You don't need anything else.
And that's a totally legit spec aligned registration.
So, yes, absolutely.
You don't need an agent.
You don't need a weather agent.
You just need your weather API.
And then that weather API is going to be consumed by agents
that are looking at the 1004 directory registry.
But that's it.
Yeah, it's not only for agents.
It's also for pure services.
Makes sense.
I feel like as a community,
we should potentially think about changing the trustless agent thing out then because what you're saying makes a ton of sense. And I agree. It kind of creates maybe an interesting point of confusion because I have a couple of X402 merchant endpoints that I've been running. Pretty lightweight and basic, but thank you for sharing that.
But it's very interesting because I didn't think about using 8,004 in this way.
So yeah, it's kind of funny.
But that being said, I want to talk a little bit about the commerce layer.
So I think this is a good segue into it.
So I've been reading 8,004 wrong, clearly.
That's okay, my bad.
Maybe more directly, right? As we start to think about what are agents going to do?
We're in this really chicken and egg situation where we need more, but in order to get more,
we also like literally need more, right? Like we need more services, we need more agents,
but to get them, we need more of them. So we have people, you know, trying and experimenting with use cases. Obviously a lot of them fail.
It's the way it goes, right? We're almost creating a dual sided economy here with totally new rails,
which is really cool. So as part of this, there's been a lot of questions around how traditional
A lot of questions around how traditional commerce flows work on chain.
We've seen this on the X402 side.
We've seen this being solved by the Z-Off team, who's introduced some refund capabilities.
If I'm not mistaken, I believe the Stripe team has as well with their new machine payments
introduced some refund stuff.
I might be wrong, though.
And we've seen a couple others focus on the refund side. There's also been some discussions
around how to properly use escrows and prepayment, almost subscription style. So we're starting to
see people thinking, in my opinion, about the right needs to actually satisfy real commerce
on chain, which is good, right? It's not, not everything in the world is
just here's, you know, here's a dollar, give me it. Be nice if it was, but, you know, it's not
realistic. You guys started working with, you know, the virtuals team. And for those that don't know
the virtuals team, I would argue right now they're probably the leading agentic protocol in the world
from an on-chain perspective.
They've got the most volume, the most agents that I can see, and just overall really, really strong team.
And you guys introduced the new ERC with them, 8183, which talks about the commerce layer.
Yeah, I would love an introduction to that, how it relates to 8004, X402. And then I have some
fun stuff with our conditional transactions I would love to throw in there. 100%. So first,
big shout out to the virtual team. We've been working with them for the last two weeks and
they're fantastic. First, they're always online, always receptive, like really great team.
they're always online, always receptive,
like really great team.
So 8183, every time I call it 8138,
I don't know why, 8183 was released yesterday.
So it's very fresh.
It's a spec that got proposed on the Ethereum Magician forum.
So you can just go there and check it out.
We can probably link it.
And essentially comes from a thesis
that is not really a thesis,
just like the reality of things, as you said,
is not every transaction is like a few cents.
Not every transaction is like,
I give you $1, you do this for me,
and then we're done, see you later.
It comes from the need to hook multiple components
to these like buy and selling kind of mechanisms.
So what it does, it creates an escrow.
So for those that don't know what is an escrow,
essentially put money somewhere,
they get locked until the job is completed.
And then if the job is successfully completed,
then these money are unlocked and sent to the seller.
Now, how the money are unlocked
is kind of like the tricky part
because you need like a three-party
slash four-party mechanism, right?
You need a buyer, you need a seller, you need like a three-party slash four-party mechanism right you need a
buyer you need a seller you need the escrow contract but you also need someone that judges
and validates whatever was the job and it was completed correctly and then potentially like
activates you know um refund mechanisms or disputes mechanisms and so on and so forth so 813 8003 8183 oh my god i will never be able to say
that essentially what it does is it creates this escrow mechanism makes it very very easy like
1004 is he creates this four party kind of way of transacting buyer calls the seller the seller
tells the price you you lock all the funds within an escrow the
job is completed the validator that is not the validation registry is like a validator whether
it's a human or an nlm validates the job is completed unlocks the money send them to the
center uh very simple uh again author david acropp is together with the virtual team
released yesterday put together in two weeks.
I think it was the fastest DRC I've ever seen
being put together.
Of course, I mean, the virtual team has been working
on something similar for like ages.
It's their ACP framework that we just like simplified
to the bone and put it under DRC.
So that's the idea.
It kind of creates an alternative to X402.
X402 is great for small payments, fantastic, super fast, very easy to use.
8183 is fantastic for, like, bigger payments.
They also need validation and might be prone to disputes and refunds.
Amazing, amazing.
This is, it's funny. Do I have this pulled up? Yes, I do. So I don't know who this individual is, but the guy's name is J.U. from Pantera Capital.
He wrote a tweet two days ago about X402 micropayments and how they won't scale in current form.
He said the average X402 transaction is $0.09.
The average Visa transaction is $50.
Visa processes 24,000 transactions per second for the same volume.
We would need a chain that has 500X throughput of 12 million TPS.
of 12 million TPS.
And then he said, we need new batching
And then he said we need new batching or designs for agent scale.
or designs for agent scale.
So it's funny because we always get into this argument
in blockchain about needing more capacity,
but the reality is we've never hit the capacity limit.
It's just the truth.
The only time we've ever come close
to hitting the capacity limit is during market meltdowns
when Ethereum gas prices spike,
which is just, it kind of is what it is. If you really want to put
your transaction through that, you're going to pay the premium. And that's just part of the deal
we've agreed to. And we saw Solana go under some stress during the meme coin crisis. Outside of
that, I've never seen a blockchain actually hit its limits. I've seen for years people try to
launch blockchains claiming TPS things, but it's been just irrelevant. Rubberband.
That being said, the part that I love about this proposal is it's not concerned about TPS. It's finally concerned about something that actually matters, which is in a scenario where we need
to move money in a specific way, this is aiming to approach that and find a solution.
aiming to approach that and find a solution. As part of that, I kind of have two items that I see.
The first is, and I need to read through this spec because it really popped up for me this morning.
I don't know why Elon is not surfacing this information for me immediately. It obviously
is what I want to read. But I want to say that it should be feasible to explore as it continues an X402 extension,
just because that is the design of X402.
So I personally am not an X402 Maxi.
I don't think everything needs to revolve around it.
I think the nice part about it is it creates a really easy standard to onboard teams and
companies into the agentic commerce world.
And so that is something that I'll be very interested to see.
And if that's not already been discussed with the virtual team,
would love to chat with you guys and work with you on that.
Because I do think that's one of the things X402 has done a great job of
is it is really easy to get a new merchant involved.
a new merchant involved. We've taken most of the headaches out.
We've taken most of the headaches out.
Additionally, and this is something I'm curious your thoughts on, is privacy.
I think the reason X402 hasn't really come under fire for this so far is because most of the
payments being made are, they're so small that I don't think people really
care. And they're primarily agent wallets being seeded with a very small amount, right? If you're
giving an agent wallet, let's just say $5 and the average transaction amount is $0.09, that is 55 and a half transactions, right? So if we look at the actual X402 consumption,
which as of, again, just like off the cuff here, pulling it up live, in the past 24 hours,
there's been 167,000 transactions, right? And there's been 4,000 buyers. These numbers are
not massive, right? We're still in the early stages. And so
these numbers are massive. And I think as we continue to grow, we see this split where
maybe people want to start to trust their agent with these larger payments. But as that occurs,
you now need to see the agent wallet with more money. The agent literally needs
more money. So as you start to put in more money, it's either sitting on more money or it's being
delegated money from an account that potentially has more. And so now instead of... It's very
common. A lot of us are developers. We all do the same thing. Binance, Coinbase, Kraken,
A lot of us are developers. We all do the same thing. Binance, Coinbase, Kraken, Bybit, whatever your preferred exchange is, you pull five or ten bucks off. That way it's fresh. You throw it into a burner. Boom, your agent's funded. But the reality is there's a lot of people, most of us included, that have funds on chain. And we'd like to be able to use those funds to deploy out to our agents. The problem is, well, now everyone knows who that money came from because it's public.
Curious if you see any issues as we start to think about these larger payments around
privacy and confidentiality of the balances of both the humans and the agents.
100%.
I think I started this conversation saying that on the AI team,
we work on three things, right?
Agentic commerce, agents' privacy, and agentic security.
Agentic commerce, we kind of touched on that
for the past, like, 30 minutes.
We're on that, different components.
Agentic privacy and AI privacy is a mess, is a big problem.
Because when you wanted to start bringing up, like, remove the part on blockchain, right?
We can work on that.
I can mention the classic things like ZK proofs.
Okay, great.
We're going to implement that.
I can mention, like, the classic mixers.
So you don't know where this money are coming from.
There is multiple things you can do.
There is, like, the guys from Builder's Garden that created Cwa, which kind of creates like a key store for your keys,
which kind of separates the wallet from the agent.
So the possibility of prompt injection
kind of drops down drastically,
which by the way, prompt injection is a big problem.
22% from the latest study
that came out yesterday from Xenolix.
22% is a wild number if you think about it.
But anyway, so multiple things that can be done there.
We need better tools for agents.
We need better tool for agents that can pay.
And it's going to be a problem also for the banks
because if you can manipulate an agent,
that agent has full control over the medium of payment.
And if the agent has full control,
that's a risk by itself, right?
So you
need two-factor authentication, you need whitelisting, you need a full framework of
tooling that needs to be created. Now, on top of it, AI per se is not even private, right? Like,
there's nothing private. Everything you tell an AI gets sent to, or if you're using Open Router,
like OpenLLMs, you get sent to open router uh which by the way
i love open router i don't know if you know the cto of open router is the cto or the previous
founder of openc so like they're super connected to to web3 and blockchain as well so that's top
of mind for them as well if you're using anthropic that's uh then everything gets sent to them if
you're using gpt everything gets sent to the open If you're using ChagibD, everything gets sent to the OpenAI. Everything?
Wait, wait, wait.
Everything?
If you send a message to your ChagibD?
Everything.
Everything and more.
You don't even want to know about the things they're collecting
that you don't know.
Cookies!
The damn cookies.
Gotta stop accepting cookies historically flagged people to the fbi because they said something that was a little bit
spooky so yeah no everything they literally went from oh my god social media know everything about
me to let me tell the darkest secrets i have to my child gpt instance at 3 a.m in the morning
and we're totally fine with it uh the problem is that the the solution to this problem is a problem
by itself because to make inferences private and maybe i'm diverging a little bit from the original
question but just because it's a systemic problem that we need to solve that you have two ways one
is fully homomorphic encryption we're
not going to get into fully homomorphic encryption on this call but just know that we've gone through
we've gone through it many times i love it okay shout out to the sama guys um so you already know
i think your listeners already know what is fully homomorphic encryption just a way to perform
operation on encrypted like strengths uh and another one is t trusted execution environments now with full of morphic
encryption if you talk to the folks at zama which by the way are the kind of the guys that invented
fully homomorphic encryption in a way they're researchers at least there's no way we're gonna
get that sorted in the next 10 years the The computational level is wide, it's like 10x.
You cannot do things, like you cannot do divisions and multiplications
on fully amorphic encryption,
which sounds like a little thing is massive if you want to run inferences.
The other one, again, is TEs.
The problem with TEs is that you can branch attack them,
so you can just go to the physical TE that is running the inferences
and kind of download all
the data so i know for sure that this is genius in my opinion there is companies that are planning
to shoot tees on space in space because that way no one can go and connect to the te so to answer
your question that's a problem uh is a problem 360 degrees with ai don't tell your darkest secrets to
open ai and anthropic
run your local model that's probably the only way to solve it right now even if it's a shitty
model of course um but but but yeah i mean if you if we want now to have an agent that has like
millions of dollars in a wallet just make sure that it has a multi-sig and you have to confirm every signature
and accept the fact that everyone will see that your agent has a ton of money in the wallet
that's another great great point as well as to why you know obviously with the the you know the
likes of open clawed and things like that the thing that everyone went to first was trading
you know and i guess like another thing
as well when using these these uh you know third party llms and providers and things
i mean they should be able to also see like like if there was a rogue employee right and
you were running everything through open ai or something, they could technically have access to your wallet, right?
If the key somehow winds up in the prompt,
then yes, that's why...
Like, in the information that gets injected.
That's why, like, most of the usage of wallets ideally comes
through a tool, right? Because the provider should not be able to see what occurs behind the scene of
the tool. However, again, it depends on how you're doing this. So like a lot of, there's a lot of
nuance here, let's put that way. Cause even like the, for example, I've seen people try to do where they have a local private key
on a, like with an agent, and then they're trying to use like cast from Foundry to execute
transactions. And the problem is, is every time they make a transaction, the private key is actually in the prompt because the prompt is the CLI request.
So it's like, my two cents there is don't store keys on the machine directly, or at
least if you are, you need to be able to rotate them constantly, like a secret, because you
just have to assume it's exposed.
If you're not assuming it's already exposed, then at some point you're probably going to lose.
Is that fair, Vito?
Yeah, 100%.
Your agent shouldn't be able to see the public.
That's a no-no.
Yeah.
Vito, I have a question.
So I want to share something with you and our audience of however many people may be watching today. And then I want to kind of just pulse check and see what you think about this. So this is my scale show for the day.
of the scale team are, you know, we've many members have been in crypto for a very long time.
You know, and so we're, no one here is against any of the privacy protocols, right? Like we
respect that people want to experiment and have the level of privacy they feel comfortable with.
However, we recognize from a business perspective, some of the protocols that exist have gone through
various levels of regulatory
sanction. They may not be compliant, et cetera, et cetera. Again, not discussing legal items.
What I want to focus on is what we can do today, right? Many of these things are not functional
for an enterprise to come and say, I want to deploy AI agents that move money on my behalf.
They just like, they can't use those, right? There's no
proof point that they can legally. And so we've been thinking a lot about that. And I want to
argue that we were one of the first teams to start talking about confidentiality. I won't say names
of who may have taken that from us, but it's all good. I like that people picked it up. That was the point.
Confidentiality is key. We call it the barista test where it's, you should be able to do actions where the action shares enough information to get executed. But the barista, the person serving the
coffee or the waiter, the person you're working with here, should not have access to your bank
account. They shouldn't be able to see your full transaction history or like all these things, right? That's that level of confidentiality that
we ideally expect in society today. If you have a credit card, when you spend your credit card,
people can't see what's in your bank account, right? It's just that simple. And so we actually
have this new, relatively similar to Zama, we can kind of call it a new token
standard, but we aim to be as close to ERC20 as possible.
We call them confidential tokens.
This is actually available right now in our sandbox.
We've got some teams we're working with that are testing it.
And what it allows you to do is actually have a shielded balance. So for example, I can take 100 USDC, I can have it be encrypted, and now I can move that 100 USDC around.
People can see that I'm sending it to someone else, but they can't see how much I'm sending.
That balance and that value stays encrypted from my wallet through the transaction to the other person's wallet.
It only gets decrypted when it's decrypted.
You can decrypt your own balances locally because obviously you need to see it.
For compliance reasons, you can decrypt your transactions,
but the actual balances stay private.
Our belief is that this is a necessity for Agenda Commerce.
And there are a lot of teams aiming to do this.
We've seen some good stuff come out for the Tempo team. We've seen some good stuff come
out from the Arc team. We've seen some cool stuff recently come out from Stellar.
And of course, the Zama team as one of the OGs in the space. But the reality is,
no one's really hit this in production yet outside of Zama and Zama is only on Ethereum right now.
I'm curious on whether that's something you believe both as part of just both transfers,
X402 payments and possibly this new 8183 standard, if you think it will be, I don't know,
maybe something that could be,
let's call it a default for agents moving forward.
I mean, I can say you should, absolutely should.
Like privacy is extremely important.
We talk about this.
If you talk with Vitalik today, like that's the first thing he's going to tell you.
It's going to be, hey, how are you doing?
Privacy is important, Like, obsessed about it.
It comes with a bunch of issues, though, because what I think,
I think it was Vitalik that said something similar is, like,
privacy is hygiene, and it requires education, right?
It requires education because people won't default to privacy automatically.
Because the problem with privacy is that it usually comes with frictions.
So I think for sure we need to do a good job
of making it as frictionless as possible,
but also we need to all together,
make sure that we educate the people, right?
About the importance of it.
Not just, of course,
like talking specifically about these tokens,
it's important that you don't have to disclose your total value or your total net worth every time you make a payment.
That's key. That's expected. That's how it should be.
It's important that I don't get all of my movements and purchases tracked because it could be a problem, even like a social engineering kind of problem
right at that point someone knows what i'm looking for what i'm purchasing let's say that i start by
i don't know pregnancy tests now everyone knows that i'm waiting for a baby like things like that
that's important to keep them private and it's the more blockchain the more this kind of like
payments will become popular and common the more we need to
make sure that this is taken care of but taking a step back is is about everything privacy should
be the default for everything it should be the default when when i'm sending a message to my
child gpt is wild that he knows everything that i'm thinking it should be the default when literally
i'm doing when i'm using my phone it shouldn't be tracking everything i'm doing
so to answer your question simply i'm very much into like this kind of privacy tokens and uh privacy overall 360 degrees i do think though that most of the times privacy teams do a no full job
at dealing with the user experience and explaining privacy privacy to the normies
because they're giga brains right like super smart people so they don't even know explaining privacy to the normies because they're gigabrains,
right?
Like super smart people.
So they don't even know to speak to the normies, not non-gigabrains people, but people that
don't think about privacy.
And developer experience needs to be very smooth as well, of course.
Yeah.
No, I fully agree with that. It's one of the things that we're definitely seeing right now is it is impossible to create a end-to-end privacy token
based on the ERC-20 standard.
It's just like, it is impossible.
So it's not, you need information additionally
that's just not feasible.
We've gotten about as close as we think we can.
And I mean, there's definitely limits to compatibility
that pop up almost immediately.
We know the Zamba team is dealing with the same issue.
It's just like nothing works out of the gate.
But definitely agree with that.
One of the things that I'm really excited about
is in the last, what, three weeks,
we saw this shift from MCP to skills and CLI. Kind of exciting because it
opens up the door for, I think, some really, a lot easier developer experiences and agentic
experiences where we no longer have to try to force everything behind the scenes through a tool.
We can now actually just say, like, run this operation when you want to encrypt something
and it just works.
So pretty excited.
We've got some really cool things coming.
Vito would definitely love to chat with you more offline
about our confidential token,
but that is something we're pretty pumped to see
right around the corner.
And I think we'll,
we haven't been doing privacy for 10 years like the
Zama guys, technically longer since a lot of those guys are like career cryptography people in
privacy. You know, this is obviously something a little bit more recent for us with our byte
protocol. So this is definitely something we're very, we're very interested in seeing on the
agentic commerce front. So if you are listening and you're interested in building, you know, agents that have a little bit more privacy by default,
things like that, definitely make sure to, you know, reach out to any one of us. We're happy
to chat with you and see how we can support and get you using our new stuff. But I'm being
screamed at. We're about at time here, unfortunately. Falalkor. I love how you tell people to contact anyone.
And then I get telegram messages like,
oh, hey, I want to know da-da-da-da.
And I'm like, okay, fine.
And I'm like, bro, I don't even know
what half of those words mean.
Contact me or Vito.
Don't contact Falkor.
If you want to contact Falkor, don't do it.
You can do it. Just don't contact falcor if you want to contact falcor don't do it you can you can do it just don't
expect a response that's going to help you with your problems you know you could just tell people
hey talk to that guy instead that would be acceptable i i do i do i do i say that well
they must be terrible following up you know and they're like oh but Soya told us to talk to you.
And I'm like, no, go talk to Soya.
And then-
I don't tell people that.
Oh, you just did like two seconds ago.
Guys, thank you so much for joining us
for another episode of Scared School.
Vito, absolute pleasure having you on, my friend.
It has been awesome.
Diving deep into ERC 8004
and everything else that is coming out of the ethereum foundation
and that really cool new thing with virtuals as well but um you know we are at time it's been
pleasure having you anything final before we kick things off no thank you for having me that was
fantastic if you want to dive deeper in 2004 go go on 8004.org. We have all the information there.
Reach out to Falcor 100%.
It will answer all of your technical questions.
And again, thank you for having me.
And coming back, Vito will not be.
Guys, thank you so much.
And we will see you next week
for another episode of Scale School.
Goodbye.
