Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Building Trust in the AI Agent Economy with ERC-8004
Recorded: Feb. 25, 2026 Duration: 0:56:28Space Recording
Full Transcription
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music Music Music Music Music Music Music Music Music Music Music Oh I'm going to go to the next video. The Oh I'm going to go to the next video. Thank you. Music I'm going to go to the next video. Music Oh Oh The Oh I'm going to go to the next video. Hi, hi, hello everyone. And thank you so much Fabian and Vito for joining. Let's wait for
a couple of minutes for more of our communities to join and then
we kick start the space awesome thank you for having us good is my audio okay yeah we Yeah, we can hear you. Fantastic. . Продолжение следует... Thank you. Það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það er það Thank you. So So I think we can kick start up for our space and we'll have like more people joining along
the way.
So welcome gentlemen.
And hello, hello everyone who joined here, all the agent economy and to see us just and
people who just started building or anywhere in your journey in web 3 so we also brought wonderful guests for today's this last last talk episode eight and.
I'm your host to do and if you wanna like send me some details about yeah, you can have my Twitter zero X do and I'm currently leading DevRel at Metis and Laz.ai is a product of
Metis.
So Metis is a parent company and Laz.ai is powered by Metis.
And in this last talk, we're going to dive deep into the infrastructure for AI agents,
which is the ERC-8004. And our mission today is we're going to unpack
how agents can discover each other,
build portable reputation, get validated,
and actually transact with each other.
And if you are building an open agent economy,
trust can't be permissioned, right?
So joining me are two absolutely heavy eaters who have been in the
space and deep in the trenches today everyone from the ethereum foundation and fabian from stateless
hey guys welcome to last talk so vito kick us off uh can. Give us a bit of introduction about yourself, what you do and what you are passionate about, especially in the agent economy.
Yeah, 100%. So thank you for having me again, as we'll say. Anoctero is a legend.
I've been in the space since 2020. It was a very different space at that time.
And I started, before that I was doing something completely
different, I was in the movie industry
and then from 2020 to
2022, 2023
I was at Alchemy
working in their dev rel team
and then I joined Cyphering with Patrick Collins
I know most of you know Patrick Collins
is the guy that creates these very
long courses where we created a security company, Cyphering.
And then since the beginning of this year, I joined the artificial intelligence team at the Ethereum Foundation,
where we're working on the agentic economy, security, privacy, and making sure that builders have all the tools that they need to be done on Ethereum Mainnet.
Both of it. and making sure that builders have all the tools that they need to be done if they remain it perfect
it's thank you so much for introducing yourself and it's it's really like wonderful to know you
you have a different background before 2020 i can't imagine you in i mean it makes more sense
you're in uh like a movie industry i can't picture yourself someone from
you know like a italian movie actor no no no i wasn't an actor i was doing animation and
mock-up so completely different stuff i was working with hundreds of other very talented
people on on games movies and yeah mostly animation you know the cool suits where they
have all the dots and then the cameras the capture movements that's what i was doing awesome but you could have had a shot as a like an actor
for sure so you'll get character actor i think i would say like underused potential
appreciate it awesome so fabian it's your turn
Awesome. So Fabian, it's your turn.
Hey, first off, thanks for the invite. It was wonderful meeting you throughout these
years and now that we've gotten a chance to talk about agents and stuff. It's really
wonderful. I would say I started my journey with Buffbri around a similar time when Vito started, around 2020, where I started
one of my B2B NFT analytics products.
It was pretty interesting.
Over the years, I've gotten to work with wonderful people in crypto.
I've worked in governance, grants, public goods space as lead engineer at Parmafrior
previously, contributing to protocols like Bitcoin, Octent, as part of a grantee accountability protocol.
In my previous years, I've also worked with Eigenlayer as one of their earliest first AI AVS
over there. But I'm mostly known for my e-global records.
Got into do a lot of hack turns over there.
Won a lot of finalist titles over the last three years.
And currently I'm building an AI call center stack
for customer experience at Tenori Labs for AI.
Based out of APAC region.
And Stateless is a community that I run
with a wonderful group of builders and engineers.
Outside of work, I love running, travel, and I'm trying to produce music.
Yeah, that's been a pretty good introduction of whatever I do here, I guess.
awesome great energy loved it so we got like the uh one of the top builders and uh people in the
Awesome. Great energy. Loved it.
space for like a really long time and then you've seen like multiple cycles so you've been in the
trenches you've seen bitcoin like going to the sky and then you also seen like bitcoin and like
few thousand dollars so you've seen it all i wouldn't say that i'm that
old but yeah i've been a little older than was he great it's great i love it so let's set the stage
with some basics so the first question for both of you would be like can you explain the erc 804 that's going to be the core topic of this
whole space in one sentence for someone who's never heard of it and you know like even make
it some uh like a meme worthy or make it some quick quality if possible yeah we don't you want
to take this first yeah 100 oh god explaining it in one sentence is difficult, but I think that one sentence that would explain it is the trust layer for the Gentic economy, which is very difficult to understand if you are not into a Gentic economy.
So if we want to go more in depth of what it actually is, we can do it. But Fabian, I'm curious to hear your one sentence.
one sentence. Oh yeah, sure. I mean, maybe I would take it a little bit more personal.
I always thought ERC-8004 was more like LinkedIn for agents. Yeah. Or more like Craigslist
or a combination of Craigslist and LinkedIn together. I love it. That's actually very
close. That's actually a very good description.
Tiro, do you want to go more into the depth of what 8004 is?
Yes, sorry, come on in.
Do you want to actually go into the depth of what 8004 is and how it works?
Yeah, the registries that it provides for the AI agents and stuff?
Yeah. Yeah, the registries that it provides for the AI agents and stuff? Mm-hmm.
Yeah.
So, yeah.
So, it does have the three registries, identity, reputation, valuation.
So, let's, like, talk more into, like, how these three registries help other agents to, like, understand, evaluate, and work with each other.
Mm-hmm.
100%. So, thank you so much for, like, putting work with each other. 100%.
So thank you so much for putting it in those words
and especially Fabian putting it like a correct list
and LinkedIn for agents.
That's something really interesting point of view.
And so this is what ERC agents report
like kind of aligns, right?
So it provides registry for other agents to validate
in such a way, either for an agent to hire another agent. So in this human world, we
look at LinkedIn and Twitter also. But for agent, it does have the ERC-AJSR4 registry.
And apart from validating, what other feature it could have so we gonna like
go all in depth in this space and before that let me put you on another question for discussion here
is in the last like one year even like one year to like one and a half year the world of ai
basically shipped from like just a chat interface for any websites
to like autonomous agents. Now we have claw bots, the open claw brought it up and there's
been like a huge revolution I would say around agent tick frameworks and agency usage. It's
just from like asking like chat GPT to give you a code to like having a cursor or
a Vinsurf to just basically build you stuff.
And now Cloud also introduced Cloud Code.
And this space has been evolved a lot in the recent space.
So what do you think the core problem ERC-AG for created to solve uh especially around trust and coordination
when we have like multi-agent systems uh for like high level perspectives so vito or phibian who
want to take it first um i could go yeah yeah i think when when x004 first came up, they addressed a really important code problem.
And that problem is, like you said, once we start moving from chatbots to autonomous agents,
there are a few questions that you need to answer for one agent to an agent.
Who are you as an agent?
What do you do? Are you the same agent as yesterday? Are you know, who are you? Like as an agent, like what do you do?
Are you the same agent as yesterday?
Are you working in a sovereign manner?
Like whatever code that you execute, is it what you were intended to execute?
Right?
All of these questions that sort of came up and these were really important questions,
but not a lot of people really understood why we even need a registry standard like this. And the true girth of the situation sort of went out when
everyone started posting their own OpenFlow agents out there. Like, you know, every other person had
an OpenFlow agent coming up on Twitter and like, you know, every other person had a open claw agent
coming up on Twitter and like, you know, talking like them.
I've personally known like at least two out of three
of my friends had an open claw agent
taking care of their socials right now.
And these agents are not just agents.
We've evolved to a point where agent to agent communication
is sort of normalized.
And every single agent has
started talking to other agents and this is where things get really interesting now this open class
spike made erc 8004 more relevant if you ask me because now you really need to tie identity
and reputation together because you've got all these different distribution rails but
there are absolutely no standard way to for you to trust and this is what erc 8004 already addressed
and it had been leading all the way up to this right you've got so many different ways you can
validate so many different ways you can put things on registry and maintain identity and reputation.
Now it all aligns really well
and people can really permissionlessly think
about letting their agents talk to other agents
and use these capabilities.
And we've also seen the darker sides of security
of vulnerabilities with Open Claw,
how people have been misusing the
skills and whatnot. But all of these problems are in a way, tying to what we're talking
today. I mean, what we're talking about today, and how this is solving a major problem, solving
shared identity, portable reputation, and all of this while not keeping it very strict
to one network, but keeping it really
composable and modular so that coordination is standardized and anyone can build on top
of this.
And I think that's pretty cool.
And this is not just coming from a random developer.
This was built by really huge people from Ethereum Foundation.
You've got people from MetaMask, Google, all of these interesting collaborations sort of gave birth to this solution. And I think that's an important problem for people to like
understand. And yeah, it's pretty cool. Yeah, Fabian is explaining 1004 better than I could
ever do. I love your explanation. I will just add another thing. And it's something that kind of
gets lost into the wording of how 2004 is written, because the name of the standard is
trustless agents. Now, the reality is that we don't need trust only for agents. We need trust
also for agentic services. When you think about machine economy, there is really two components,
right? There is the agents that are purchasing and doing actions
in a self-driven way autonomously.
And then you have services that can just sell data
or sell outputs in general.
And that's exactly what 2004 does, right?
As Fabian was saying, it's like the LinkedIn for all of these things.
You have three registries.
You have an identity registry that kind of binds
these agent or agentic service
to an
identity that doesn't change.
You have a reputation registry that
the way I also actually
usually explain it is like Yelp stores,
but in reality
is a little bit more extensive
than that. And if you're familiar with SEO,
it's like how Google kind of like ranks pages
on the search engine results page,
where you not only have like the length of the text
or the quality, but you also have the latency.
You have the overall like relevancy
of the content within the page.
So it's a lot of data points
that then need to be aggregated
and are not aggregated by the standard.
These are like abstraction layers
that are built on top of that. And then these agents and then you have a third registry that is
this validation registry that kind of like fabian was hinting to where um it the way i usually frame
it is i think if you're on charge gpt and you're writing to charge gpt i don't know what number
we are at now it's like 5.3 i think. You write to 5.3, you send a message,
and you get an answer back.
The only way you have to understand
if you're actually using 5.3 is to trust the UI.
But under the hood, you don't know
if they're using 3.2, 5.3, 4.5, and whatnot.
Which is not really a problem
if you're only dealing with chats,
like what am I eating this evening.
But it can be a huge problem
if you're dealing with high-s, like what am I eating this evening? But it can be a huge problem if you're dealing with high sensitive operations,
not just financial operations,
but even like healthcare operations and so on and so forth.
So what the validation registry does,
it kind of validates the operations of these agents.
Now, as Fabian will say, the standard is very open.
So it's not super opinionated.
If you look at the specs the specs are really
well written if you ever read a spec of a near co ips they tend to be very convoluted whereas
the 1004 specs are very easy to read we also have a website that is 8004.org that you should
definitely visit because he has a bunch of faqs and tools that you can use to start building. But if you look at the specs, you can see that other than like a name,
a profile picture, skills, and domain, then you have a big array of services
where you can literally put whatever you want.
And there is A2A support.
There is MCP support.
There is simple endpoint support, going back to the idea that is also
for agentic services and not only agents but you also have dids decentralized digital identities
enas uh email support so it's a very open standard and the goal yes is to create this
non-walled garden where agents can discover services and vice versa of course
and vice versa, of course.
Lovely.
So that's a really extensive explanation one can provide for ERC-G04.
Thank you so much, Peter and Fabian.
And it does clear everything.
And I have one personal question on the ERC-G04 stage studies.
One of them is reputation.
So this is something really hard to build even in
web2 space or web3 space and especially for an agent and i did saw some uh like post on the same
and i would love to like get your opinion on it so it could be uh played right it could be gamified
and it could be manipulated how could we like solve that how could we prevent that
sorry is the question how we can prevent the gaming reputation yep that's a great question
this is the the most like asked question i would say when it comes to the reputation registry
now the reality is that if you think about Google Business or, again, G2, all these review platforms, they have a problem with verifying the reviews.
A lot of time, also in Google Businesses, and this is a centralized service, lots of the reviews are fake.
So understanding which review is right and legitimate and which ones are like a gaming of the service is very difficult.
Now, 8004 per se doesn't solve that.
The only thing it does is to register in this feedback.
As I was saying before, though, on top of the 8004,
there is a bunch of abstraction layers that are being built by users,
one of which is like Beader's Garden, but there is like LineF,
there is a
towns protocol that is building on top of it which are creating an agent zero sorry because
it's from marco de rossi which is one of the co-authors which are building these watchtowers
and the only job of the watchtowers is to analyze this feedback and actually also leaving feedback
now how do they analyze the feedback is about provenance.
So who's leaving the feedback?
Is this like a new agent or a human that has no reputation registered in 2004,
or is a reputable agent?
What are they doing? Where do they come from?
What are their historical actions on chain?
Now, based on that, once you have all of this data,
once you have all of this data, which again is public because it's on 2004,
which, again, is public because it's on 2004,
then it's very kind of, not very easy,
but it's decently easy to understand how to aggregate this feedback.
On top of it, these watchtowers also leave feedback themselves.
So what they do is they cycle through all the agents and services
registered in 2004 and check things like,
is this service up or is offline?
What's the latency of the service?
Is the answer that I'm getting from the service the answer
that I expected?
And on their own, they're also leaving feedback.
So it becomes like about circular economy, right?
Where watchtowers leave feedback, analyze feedback,
ranks them, and then people can use it to connect
with the registry.
Yeah, this is pretty cool because um i was working in a in a dow product where we were
solving for an exact same similar problem where we have to find out, evaluate reputation for delegates in DAOs, right?
So that it was a very similar problem, but it was for humans.
And surprisingly, I mean, it's so cool that even for agents, we have to come up with similar
solutions.
And there are so many vectors out there to come up with these reputation metrics and
rubrics, right?
And there is no single way to do this, but I think pretty much out there today, like
Vito mentioned, there are different ways to solve this.
And most of them sort of tie to having multiple layers over this reputation flows, right? Like you've got people
bringing in staked reputations where if they don't really have one person or like any, not anyone can
just go and put up a fake review and, you know, make sure that an agent gets a good score where
we've got reputable reviewers where you assign weights to people who've been giving good
feedback on these agents.
There's a lot of these attestation knowledge graphs that basically work on top of this
to understand which attestations do I really give attention to, right, when I want to look
at a reputable agent.
So this problem is sort of a very open problem
that a lot of people are trying to solve and that's sort of another cool thing that i think
erch004 allows that anyone can come up with their own uh reputation flows as well like
and that's also a nice um you know spotlight point that I would like to leave with.
Oh, I think Vito got made into a listener. I need to.
Yeah, I'm sorry.
So this is a classic mistake I always get into.
So this is a fantastic take.
And yeah, so Vito is again speaker so i was looking into that
so thank you so much for adding um more things on the reputation management and uh i'm glad like
like a lot of solution being coming into the space and uh that's like really interesting point is
like anyone can bring their own um like solution in order to integrate with the ACA 04 because it's open
open standard for ethereum so that's like a really interesting part for me and on at the
same time we will be all like looking at the registries let me also like ask you this as
this been uh starting as a infra for uh agent economy I think there's me there's gonna be like a lot to come i
also uh like witness and uh been in the like x402 phase as well where erc ages or four and the x402
goes uh in hand in hand along with that um so how are we gonna add like the governance guard rails
and any other integration that we could like do with
like physical agents and even like bringing more support i think we do also saw your post of giving
like physical body for your agents so in those spaces how it could be like very handy and
what are the things we need to actually bring in to make this space better
things we need to actually bring in to make the space better.
But there is a lot we can do.
So per se, again, as I was saying,
2004 stays very flexible.
And you can hear me, by the way,
because I think I got kicked out at a certain point.
Yeah, yeah, please go ahead.
Fantastic.
So what we need to do, I'm seeing it, Yeah, yeah, services, not agents.
I think today, it doesn't afford, it's first and foremost for APIs that are difficult to discover or want extra discoverability.
It's for MCP servers that are non-discovered and you need to be good at SEO to be found,
good for all of the services that, again,
want to be used within the agentic economy.
And as an example, for example,
you're probably familiar with Redstone.
Redstone didn't register an agent on 2004.
What they registered is their price feed,
which is incredibly useful for 2004,
and their market rating API, which essentially gives agents a way to rate certain tokens.
And there is a lot of agents trading so they can use their API to evaluate what they should trade and guess what.
So I think the first step is that one.
Second step is getting into mainstream.
So one thing that we're doing,
I think, wrong in this space
is kind of like isolating ourselves.
And this is not only for 2004s in general, right?
Where we always say like web free against XYZ.
2004 is not a web free product.
It's something legitimately useful for the AI economy.
So it's getting all of these AI teams that are building
and data layer teams that are building for whatever
that is not like blockchain to get registered on 2004.
Once we have that, then it's going to become probably a virtual cycle.
Just to give an example of cool tools that are built on top of these two primitives
is one is a universal router.
And this is an example that I always bring
because it's very interesting
and kind of makes people understand
why 2004 is very useful.
Think right now you have to implement an MCP.
Actually, when you're building an agent,
you have to implement different MCPs and different
services.
Think if you only have to implement one service and that service uses the 8004 directory and
the reputation directory to understand which tools you use for whatever task your agent
asks it to perform.
So these are the three things that I think we need.
And then, of course, the the war the sky is
the limit right there is so many other things that can be done but without this these are the first
steps yeah thanks vito philman do you want to add a few things yeah sure so i think one of the other
things that uh you were talking about was physical uh agents and how this would type into the usc
and even the x4 not to payments i think it would be pretty cool to see this in real life and i
think i've seen a couple youtubers already made this in real life you know where a personal agent
uh in you know like a handy pocket agent that would go out and like make orders in a in a handy pocket agent that would go out and make orders
in a coffee shop where they have another agent
that sort of distributes coffee to people
where you can basically go there in front and order.
And this is sort of very primitive,
but I can already see how Tesla's robots
and all these newer next-gen agent solutions are going to work really well
with physical robots and it's pretty cool to see how this standard would also evolve in in the
direction right you've got device fingerprints you've got like how the boot form was are written
into these agents and and all of these critical things that sort of will again
we have uh will have to be updated but but i think we are in the right direction i also saw
that ethereum foundation is also working towards this on the road map and uh it's pretty cool to
see how all of this is going to tie in the next few years now that physical agents is no longer just something
that you see in movies,
but rather something that you can already see in life
when people are building on top of it, right?
So, and X402 and 8004 ties hand in hand
to solve for all of these things.
Of course.
So the future that we're looking for is still building and we are being a part
of it as well. Yeah. So that's, thanks so much for adding this. And I would love to
like have like a showcase of like these things. Yeah. I think we could work on it like later.
So I have a personal favorite question for me. So this is something I'm a bit curious about myself so agent can work multi-chain so how erc aj04
work on that case so if we have a we have quite few l2s so evm layer 2 let's just take very
specifically if a agent registered on chain a can of course work with like another agent on any any other chain and we
have x402 for the payment rails in the case like how erc is easier for works and is there gonna be
any changes or it either it supports or it require more uh like integrations to make it work so how it i would work in the case of multi-chain
fabian do you want to start yeah sure um i think this is pretty simple um or at least this is how
i envision it uh in this standard anything that is immutable, right? Like whatever that needs to be tamper groups
has to be on chain, no questions asked.
But there are this computation of the sticks that,
let's say things that don't really necessarily
have to be on chain.
Treat this as like your agents RAM, right?
Like you don't need RAM to persist.
And these are things like, let's say,
aggregators that would just work on some of these metrics
and then perform compute and store these metrics on chain.
These things don't necessarily have
to run on contracts or on chain.
And all these computation statistics, again,
like I mentioned, has to be done through coprocessors or things that can be scalable.
And that too, it's not just a matter of reducing cost, but a matter of efficiency.
And I think this segregation is also something that a lot of people have been thinking about when they came up with the whole standard in the first place.
And I think I would love to hear more on this, on what Vito has to say,
because I'm pretty sure he was working with some of these technologies early on.
And I've seen some of his Twitter posts as well.
I would love your take, Vito.
Yeah, 100%. as well like i would love your take peter yeah 100 so the thing is that we made it easy enough
for anyone to to interact with 1004 meaning that right now 1004 is available on i think 23 chains
already and that happened within a week literally as soon as we released 2004, all the chains were like, we want to deploy 2004 on our chain, which was fantastic.
So again, if you go on 2004, I think under the build page,
you can see all the chains where it's available
and more are coming as well.
We have a long backlog of chains.
Now, the cool thing is that on all these chains,
the contracts have the same address.
Leonard, that is my contract developer,
that worked in 2004 registries, did an amazing job
and deployed all the contracts by himself on all the chains
and made sure that all the chains share the same address.
On top of this, we already have scanners.
What the scanners are doing is kind of enabling
this multi-chain behavior.
So what they do is they take all the chains
where this is supported
and they aggregate all the data
and then they expose an API
that allows you to explore the data across chains,
chain agnostically, I would say.
So at that point, it really doesn't matter,
for example, where you deploy your agent
because your agent can interact
with any other agent on any other chain.
Now, there is one small blocker that should be solved, I think,
by the next week, I hope.
That is everything that is not a thousand and forth.
Like, for example, you want to deploy on Xchain Polygon,
but you want to pay with X402 on base.
Now, that might be a little bit of a problem
because not all the facilitators that are supporting base
support also Polygon and vice versa.
But we are working on that.
So we're working closely with PayAI, for example,
to make sure that they support all the chains
that are supported by 1004.
At that point, once that happens,
then it's multi-chain by default.
It's really not a problem.
Interesting.
I think some people are also working on agent proxies which can relay these calls on gather chains as well.
Right?
Yes, there is any source of experimentation.
Absolutely.
Now, personally, this is totally personal opinion.
I don't think we really need to have a copy of that agent and all the chains.
I think what we should focus on is make sure,
and I think this is not only for 2004,
I think it's for the overall blockchain economy,
is to try and connect all the chains together.
So at that point, who opportunities you just decide based on the ecosystem or on the tools and whatever support
you want to receive yeah that's a pretty cool thing yeah yeah thank you so much for adding that
so that's a really interesting take and then it's interesting
to see how we can able to like fix that and all other things that is ef is taking efforts to make
it uh make it better and uh at the same time we are speaking more in uh gender for uh agents
perspective and uh lot in theory so i want to uh like uh turn this this like space into a direction of DevX, how easy or how hard
it is right now for someone to integrate ERC-HC04 and how easy of it is a plug and play or how
hard it is to like integrate for any specific use cases and what are the efforts that has
been taken and just want to like have a shout on the devx especially for this
Yeah, 100% so again
It's very easy
Meaning that the contracts are very straightforward, right? Like the registration is is very straightforward
is very straightforward.
the registration files can be hosted on
The registration files can be hosted on IPFS, Base64,
so on-chain, HTTP is really up to you.
When it comes to tools,
I think it's already kind of well-supported.
So one of the most, I would say, popular tools right now
in SDKs right now is the Agent Zero SDK.
You can find it by going on ag0.xyz which is built by
marco de rossi and is one of the co-authors uh there is uh tools to scaffold your agents so
there is this tool that i built that is create a 1004 agent and you just run npx create a 1004
agent and then you go through like a wizard that asks you which chain do you want to deploy on what kind of features do you want to have uh do you want x402 or not so all of these things and then
it gives you the scaffolding of that there is scout cause chain uh there is no code tools so if
you go on any of these scanners like it doesn't enforce scanner or trust it doesn't for uh all of
these then you can just like register your agent or agent service through the UI.
You just need to connect your wallet and pay for the fees,
and the fees are very, very low.
In general, we also have guides and demos
and examples that you can follow if you go on 8004.org.
So I would say we kind of took care of the developer experience.
Now, the only
blocker that I'm noticing right now,
the only friction, not really a blocker,
the only friction that developers need to overcome
sometimes is these
kind of
scattered building
components that you have
across different chains. So the example
that I was making before, some chains support X402, some chains do not. So that you need across different chains. So the example that I was making before,
some chains support X402, some chains do not.
So that you need to kind of understand where do you want to build and what you want to build.
8004, I wouldn't say is the problem.
Again, very, very easy.
You can get your agent registered in a few minutes
and listed in even less.
The problem is building services and agents
that are actually useful and work well,
which we've seen and is proven
is not that easy, of course.
Cool.
I think you mentioned pretty much everything
that people really need to get started.
I think the best entry point for people
to get on 800 code
is to head over to the EIP itself
and understand what it's actually doing.
And of course, when you want to build using your agent,
I think Austin Griffith posted any skills
for specifically some of these agents' skills as well.
So you don't really need to even like go and read
through the tons of docs that are available out there.
I'll just go ahead and use the skill
and get started as soon as possible.
The DevEx is pretty good.
Like the DevEx is pretty much not,
I mean, I would say it's not even needed at this point.
So yeah, there's nothing stopping you from going
and downloading the skill and getting
starting to build the whole, whatever you have in mind.
Yeah.
That's really interesting.
And especially on your last take of DevEx is no more needed.
Now we need more of like agent X.
So make sure our LLMs can find the docs
and then find the workflows that needed to be done.
Definitely.
It's also good to see like the community support
in order to bring this DevX more easier
for the upcoming, all the like EAPs
or any other protocols, especially for like x4 not to when
x4 not launched we also see some really good tutorials at the same time so it's really good to
see and those like improvements in the devx especially uh yeah back in the days i only got
like videos and then uh i've read like a lot of blogs and resources and also from like before I've had this YouTube
channel.
That's what I've used back then.
And I also like referred to like a lot of people.
So now the agent X is a like must for anyone building apart from like the devex is not
matters much, but still like I've personally seen like for some things um like the the llm is not actually
like reading it well uh the human intervention is required and we have to like uh like provide
guidance or like provide directions so it's not it like completely uh like irrelevant so
but we are heading towards the direction so So we had a really good chat,
and I wanna have this question, especially to Vito.
So looking ahead, what are the biggest
remaining challenges or limitation for this pick?
And on the other hand, I also wanna throw in this question,
how do you address the agent privacy?
Because you don't wanna disclose everything, but at the same time we want to be um like transparent but i also don't want my
uh like others to know like what my agent does so how do you do that and uh how we can prevent
the misbehavior uh of with any particular uhologies. It can be like reputation penalties or slashing mechanisms.
So can I have a take on that?
Yes, absolutely.
So biggest challenge is the one we were talking about before,
is reputation that is not reputable,
meaning that is not trustworthy. And meaning that it's not trustworthy.
And this is something that we're solving actively.
We're working with multiple teams to make sure that we can build this abstraction layers
on top of the 1004 standards
to make sure that the reputation and feedback signal
is evaluated, is aggregated in a way that actually matters
and a way that is trustworthy again.
I think when it comes to privacy
more than security but privacy is a big topic that the ITM and the
Ethereum foundation is working on right now and there is a lot of different
components like AI created so many issues when it comes to privacy we went
from I don't want to give all my information to social medias, to I'm giving all of my information
to AI companies around the world,
including my darkest secrets at 2 a.m., right?
So that's the biggest problem that we're trying to solve.
We've been talking about, well, from a non-inference standpoint,
we've been talking about private api calls private subscriptions private token usage which
we're actively working on right now i think davide crappies and vitalik buterin davide is our head of
ai you know vitalik of course um he they have been working on this new research paper slash protocol that I think is going to be out pretty soon, which kind of enables these private API calls through ZK's NARCS, which is interesting because it's also kind of quantum proof.
And it will still allow service providers to rate limit callers, even if they don't know their identity.
But if you talk with providers like
dennis ai for example they're amazing they have like kind of anonymized api calls and you can pay
cryptos and stuff api calls are just like one percent of the issue that they have when it
comes to privacy the biggest problem is inferences uh when it comes to inferences
it's not easy because you could use fully homomorphic encryption.
But if you know about fully homomorphic encryption,
and here I give a big shout out to the Zama folks.
They are the highest experts when it comes to FHE.
It becomes, A, very computational expensive, very expensive,
and almost impossible in certain cases
because there are actual mathematical operations that you cannot do.
So the only solution would be
with trusted execution environments running LLMs.
There is a company that is working on that
that is called Space Computer, I think.
I just learned about them this week.
Because the problem with TEs is that they're super safe
unless you go there physically and try to, I don't know,
plug something in the in the te and at that point you can download all the data so what they're trying to do is super fun they're essentially shooting tees in space because
in space no one can go there physically and attach to them um so privacy when it comes to ai is very
complex when it comes to payments there is a number of things that you can do.
You can use Aztec, for example.
You can use, there is a number of privacy-first tokens that are being created
and privacy-first payment systems that are being created.
So yeah, these three things, I would say.
One is using lots of ZK for API calls.
Two is solving the inferences problem with probably tes because
fully homomorphic encryption is too difficult unless we have a new breakthrough in like this
year or so and and third is private payments which we kind of already have solutions we just need to
be more aware of them it's a wonderful day thanks thanks so much for addressing it and drawing the alpha about
uh the new project as well so definitely check it out and do you want to add your thoughts into it
uh not much um i'm i think this is something i am also looking forward to maybe once uh we have some more info on like what what uh
the ethereum foundation is going to do about all of the the security aspects i think that there
has been definitely a lot of learning in the last few months especially with a lot of these
ai related attacks and how personally identifiable information and prior and all these things are being masked
by the current uh agency systems and i think with all these learnings i think the whole
uh direction is going to drastically change and it's for the better and i'm looking forward to that
yeah good yeah thank you so much for adding your thoughts and then thank you so much for uh being
like accepting our invitation and being as a speaker in um class talk you usually have it
as a live stream and with uh like technical issues popping up and it's challenging sometimes working
with the x platform so we had to switch it. And then thanks so much for cooperating.
And it's been a fantastic ride
and huge thanks to Vito and Fabian.
Thank you so much for dropping in a lot of alphas
and serious thoughts on agent economy in general,
and especially on ERC-804.
And I really appreciate your honesty
about how the reputation currently working
and how we can actually improve it along the way
it's been a really knowledgeable and learning session for us and everyone been joined thank
you so much for me it's especially a lot of learning for me personally so i really loved it
so if you are watching this you are building agents go register one build reputation join
the stressless revolution guys and thank you so much for our speakers again.
And looking forward to have you in another settings
and another last talk.
And looking forward to meeting you all in IRL again.
Thank you so much.
Thank you for having me.
Thank you, Theru.
It was a wonderful session.
Glad to talk to you as well, Peter.
My pleasure.
Let's keep in touch.
Yeah.
Bye, everyone. Thank you. Bye-bye. Thank you so much touch. Yeah. Bye, everyone.
Thank you so much for joining.
Thank you so much.
Have a nice day, all.
