Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
🤖 Agents Unplugged: Your Bi-Weekly AI Deep Dive #8
Recorded: May 7, 2025 Duration: 1:10:13Space Recording
Short Summary
In a dynamic discussion, industry leaders from Swarm Network and Oort explore the latest trends in decentralized AI, including significant project and token launches, the importance of partnerships, and the growing need for standardization in agent frameworks. The conversation highlights the increasing interest in agent-to-agent collaboration and the pivotal role of DAOs in shaping the future of decentralized governance.
Full Transcription
Thank you. Thank you. Hello, guys.
Just doing a quick mic check on two.
What's up?
What's up?
Howdy.
Hey, guys.
Glad to have you.
I think we're waiting on one more person, which is Francesco.
Is that correct?
No, Francesco will be joining halfway.
Oh, he will be.
Great.
Awesome.
Then without further ado, let's get right into it.
Welcome, ladies and gentlemen, to our biweekly AI agent deep dive called Agents Unplugged.
This is the eighth edition.
And today we're gonna go with the flow.
A lot has happened.
A lot of the guys here actually were at the token event in Dubai.
I guess we can speak about this.
I wasn't there myself.
However, Morty, our CTO, and Gus, our lead blockchain engineer, were there. And I was properly briefed. But yeah, my name is Yannick. I'm the co-founder, sorry, founder and CEO of Swarm Network.
and what we do is very simple.
We make large-scale swarm-driven businesses available to users,
and we also have a really great product that deals with arbitrary information
and brings that on-chain,
and yeah, I'm very happy to be here today together with my co-host Marco,
and then Marco, maybe you can do
a introduction and maybe you can introduce the other guests too
yes gladly yeah we definitely missed you in dubai i was hanging out quite a bit with morty
but yeah i'm sure we will speak about it in a second for those that don't know i'm marco i do
all the fun ai stuff on oasis and this has actually been
quite fun recently since i have to build quite a few agents for the events that are coming up
oasis is a privacy focused l1 it's te based and that's why we also built the framework for
utilizing these tes for off-chain compute it just sense. Lots of the AI stuff will be happening off-chain
for quite a while.
So we built a framework
so that you can build trustless agents
and kind of have verifiable off-chain compute.
But yeah, with us, we have Ort.
Speaker is Makosi.
Please correct me if I mispronounce it.
Sorry for that.
But Ort has been a great partner for a while.
Oh, hey, Marco.
Yeah, actually, it's Marcos.
What I answer to, Marcos, Marcus, Mark.
I'm pretty flexible, so yeah.
Yeah, first of all, guys, thanks for having me here.
I'll just kind of give a little snippet about Oort.
So Oort essentially is a decentralized AI startup
based in New York,
and we are essentially a decentralized cloud for AI.
And I'll just unpack it just a little bit.
So we are a decentralized, verifiable cloud computing platform
where we leverage blockchain technology
and connecting our data centers to smartphones
where you can store your data,
do data labeling, data collections.
And also we're trying to build kind of like an Expedia for copy providers.
So that's what is essentially in a nutshell.
So, yeah, I don't want to go too verbose because I'm kind of still recovering from token as
I'm based here in Sunnyside, Dubai. So yeah, just heading the main conference and so many pretty cool side events in all areas.
But definitely, I think at least AI is quite interesting.
And there's a lot of interesting stuff happening in the AI space, whether you're talking about storage know storage validation data and of course yes
compute so yeah um yeah so that's uh that's about myself and or so I'll hand it over to the host for
the most yeah I mean that's a great segue I to be fair I was really surprised how much RWA discussion
there was I don't know if this is like Dubai specific if if there's still like a massive market for this.
There's lots of local players that actually want to support this.
But I talked to tons of projects that are doing like tokenization of real estate and they're building all of these new marketplaces.
So this was definitely unexpected for me.
And likewise, the agent space is bubbling. Like I talked to at least 10 teams.
I still have to go through my Telegram messages that are doing something in the space
of agent to agent collaboration,
agent to agent negotiation, communication,
any kind of method to kind of have agents interact with.
So I think it was great for Morty to be at our event.
So Morty was actually speaking there.
And I have to say his uh keynote was
one of the busiest we had like everyone was really interested in seeing finally an implementation of
all of this agent to agent collaboration and specifically also the truth uh use case it's
just so tangible it's a problem that most people have everyone's struggling with it so it was very
cool to kind of see how many people paid attention were very curious about it i would love to actually hear
uh at one point also from morty how he kind of saw all of this yeah i can i'm sure yeah i can
actually tell you because he was um he was really pleased with the the whole like the whole setup
that you guys arranged first of all thank you for. That was really nice. And I think that he was probably was kind of nervous going into it
because I had a few calls with him to kind of like nail down
how we should present like ourselves.
And eventually we realized that the best way to present ourselves
is just to like, you know, to show what is possible, right?
And I think that actually attracted a lot of conversations post his keynote.
And he spun up a few groups and he's had a few talks with very interesting people.
So yeah, it was really great.
And it actually got a few balls rolling for us.
So that's really cool.
Yeah, shame that I couldn't come,
but you know, with the newborn,
I just wanted to make sure that I spend enough time
in the early days here.
But yeah, I guess Morty was, you know,
very much capable to deliver that message.
Yeah, that's what I figured.
But yeah, I mean, how lucky can you be
if your CTO can do sales and BD?
That's kind of best case.
You know what's funny?
Like, you know how we filter people in our company?
And did Morty tell you?
No, I have not asked.
Okay, you must be able to produce music as a hobby.
Oh man, okay. be able to produce music as a hobby oh man okay so morty is a producer or like chief blockchain engineer is a producer our uh our marketing guy knows a bit about production and uh and and music
as well that was it's not a it's not a true filter it's's more so a coincidence. But this kind of profile usually works out good in public.
You know what I mean?
The singing count?
Singing counts.
Yeah, yeah, for sure.
Singing counts.
Can you sing?
All right, good, good.
I love it.
I can do AI music very well well i wonder if that's enough
you know have you seen oh so okay let's dive into this like like side quest for a little bit
i've been getting these twitter uh these sorry x sorry you know my apologies i've been getting
these x posts all the time about like how people are inventing new ridiculous ways to use ai to
generate like fucking garbage and then like monetize it like and ai music is one of the
biggest ones man i used to be in the music scene so i naturally sometimes my time because i follow a bunch of like music related uh profiles too um and uh yeah so i've
been getting these like hey do you want to know how i'm making like three thousand dollars per month
uh generating ai music and then the catch-all is that they're literally generating like
60 000 or 70 000 different songs each month and dumping it on like spotify and
soundcloud and stuff yeah like volumes man like volumes and they're teaching that to other people
you know like oh my god that like imagine you get like a thousand people in the world doing that
and they all generate like 60 to 70 000 tracks per month like that must that must be such a burden on the provider no it's like
that's like a lot of storage capacity that they're like shitty ai generated music is is is yeah and
also lots of cues yeah hey i'm happy to listen to some of that uh volumes of shitty music you know
hey you know uh exactly the influence would have to be super
expensive no i think it's actually really cheap i think those um those music generation models
are super super cheap because like there's a bunch of them which you can literally use for
free you don't even have to make an account so i assume that it's really really cheap
interesting i mean i've only used zuno and i have to say it's really, really cheap. Interesting.
I mean, I've only used Zuno.
And I have to say, it's definitely not kind of for the general audience
to build, like, these hits that are going to be listened to all over the world.
But what it works really well with, oh, I think we lost odds.
I don't know.
Of course.
But what it works really well with well for is actually very customized songs.
I think I've shared this before, but I've gotten many people in my surrounding,
like friends, family, really emotional with these songs.
Like thank you songs that kind of mention very personal things between us
that obviously no one else can know.
It's just like songs for my parents
in their native language and they really cried just because it was so personal and customized
really that's interesting because the first time that i really like the first time that i really
got into uh listening to a full track that was ai produced was because um one of our one of our community members and junior
moderators he made like a few tracks about swarm he made like a techno track about swarm and i was
listening to it like just a few days ago so i hadn't touched it for a while but it's really good
man it's scary it's like the lyrics all make sense you know like the composition of the track is like
It's like the lyrics all make sense.
You know, like the composition of the track is like really proper.
It sounds, it actually sounds like a, like a song, like actually sounds like a song by
sung by a person.
And, and, you know, the production grade is pretty high because so, okay.
For, you know, for those who don't know music production, music production is not easy because
it's all about the quality of your recordings and the
quality of your sounds and like then once you have good quality sounds you're not done yet
because there's this whole like mixing process and you need expensive equipment um like so i
used to have a music studio i used to have a recording studio actually so we had to invest
like like a lot of money into like a
mixing desk and speakers and we had to like soundproof the whole thing and we had to have
like floating floors and floating ceilings and all of this stuff that went into it just to get
the sound right like and then after that mixing process you also have mastering and that's like
another kind of like engineer level job which requires like 20 years of experience and so on to get like a production grade song, which can go to the radio.
You'll probably spend like anywhere from like $5,000 up to like $100,000 post producing the song.
So once you've already have the song and then you want to get it production ready, you have to spend a bunch of money.
But now with AI, you know, like whatever it spits out is like production grade because it's trained on radio hits you know
it's pretty wild but is it actually do you not notice a difference as an expert no i of course
i can tell that it's ai but i can just tell that i can just tell that the sound the quality of the
sound and how you know the balance of all the instruments and the loudness and the equal like the like like the eq in terms of like how much low-end
frequencies and how many high-end frequencies i'm i know i'm talking like a nerd right now
but it's it's really i mean like you know it sounds really good to be honest it sounds really
really good and that's why people are upset you know because
you see this trend like segueing segueing away from ai like i i think that the threat of like ai
agents um you know like taking over a lot of industries and jobs and especially now that
you have all these kinds of different models which are not just you know they're not just large language models anymore right like you have like voice models you have all of these kinds of different models, which are not just, you know, they're not just large language models anymore, right?
Like you have like voice models, you have like now apparently indeed these music models, but you have like whatever, like specialized models for like recognizing objects in the real world that are used for robotics and self-driving cars and so on. And it doesn't stop there, right?
And machine learning models have been out there for a very long time already.
So you have all of these industries which have essentially already been run by agents,
like by software, by models for a long time. I just read a paper from Andrew Ng,
one of those guys,
one of those professors
that is really pushing the boundaries of AI.
I think he's at Stanford,
or it might be Harvard, I forgot.
But he was talking about models
that can improve the crop yield
and help farmers you know like visually identify when was the best time to to you know like um
for example like harvest the crop or something like that so once all of these models start to
become integrated and accessible to executors which are agents like holy shit you know like
all of a sudden i kind
of get it i kind of get why the writers are angry i kind of get why the musicians are getting angry
because it's a threat to them like really like especially those creative jobs like we talked
about before man it's like i see it i see it coming and it's coming faster than people realize
i believe yes i mean we keep talking about and i don't want to be the duma again but also they I see it coming and it's coming faster than people realize, I believe. Yes.
I mean, we keep talking about it and I don't want to be the Duma again,
but also they are pissed because it's trained on past things they have done.
So it's using their creativity and they get nothing in return.
I don't know if you guys saw it,
but there was like this TED talk with Sam Altman,
the founder of OpenAI, obviously.
And they were kind of showing this Snoopy
comic which was kind of fun
and he was just talking about the content
and the moderator was like yeah
but isn't this pretty much IP theft
yes an AI model generated this Snoopy
comic but you didn't ask Snoopy
the creators, you didn't get any of these
kind of rights to utilize
this and the audience was kind
of clapping and he was like super
ice cold like yeah yeah clap as long as you can that was such a weird like him not kind of
addressing this at all but just like hey this is publicly out there we will use whatever we can
and you can't stop us yeah and you see those companies actually championing championing for the abolishment of ip right like um like ip
rights uh i think it's like elon is a proponent of that sam is a proponent of that and there's
a bunch of other i think i'm jack dorsey yeah yeah jack dorsey and then i believe like jensen
jensen huang and a bunch of guys they're like in order and i get the state i get the i get the the vision or the idea so the
idea is like okay you know like if you abolish like ip rights um then that means that you can
train a model on anything like freely and that means that it's a lot a lot easier for them to
to train right because now they start to have like you know
certain lawsuits and they get certain especially in the united states they get certain you know
like like agencies on their ass they want to know how they handle this kind of um this kind of like
like like proprietary data and images and ip and so on so yeah it's an interesting one it's an interesting one because i don't think
it's going to happen because it's been uh it's like patents you know like patents and like
intellectual property rights there it's such a big business for companies um and it's such it's
such a cornerstone of like international business and litigation and so on there's so much money
involved in it that i don't think that that's gonna happen but it's interesting if you think about it you know
what's crazy um i was reading this book about it's well it's a paul sayer book it's about a graphic
design actually so he's like uh one of the american great graphic designers um that kind of came up pre pre pre pre Mac Macintosh pre pre
computer and then you know made that transition over to digital anyway in his
book he describes a few of his favorite rock band LP covers like the image like the the depiction of the cover so it was like
like uh like acdc and then he would describe like what was on the cover and the story was
interesting because uh his best friend's mom found the stash of like death metal and like rock and so on and they were a very christian
family so she was he was in big trouble but he described those um those covers really well but
he didn't mention the names of the bands in the description what i did was i took the description
so imagine he's a designer he's visually explaining in proper jargon, what exactly is depicted in imagery on the cover of this album. no reference whatsoever to the to the to the uh to the band name or the singer or anything it's just
a it's just a visual description of what's on the cover and where it is so like on the top right
corner there's like the band title and then there's like a jeans the background is like the
lower part of a male body you can see the genes and then you know like his
his gun was relatively big and bulging out of the genes and so on and so on and i took that
description i put it into uh sora into chat gpt the image creation model and i got the
with the description not referencing the band i got an exact replica of the cover of that band
but it didn't recognize the actual band no it did it did it it also made the title it it was
literally like a one-on-one i was actually going to publish it so i think it was uh it was like uh
i think it was acdc or something like that and it even it was just like
it was not exactly the same but like it was like acdc and then you had the genes and then you had
like you know like it was almost the same so if you can visually describe it really accurately
it will just like pull up like the closest reference apparently and it'll just reprint it
so that's really that's really the underlying issue
right so that puts it into context so if you can describe something really well and you see it with
like the studio gibili right like that that that whole thing that went on like a few weeks ago
where everybody changed their profile pictures to the studio gibili anime style it's like yeah that's
the problem right like i'm an artist, okay?
I'm making this, like, you know,
imagery, I'm selling it and so on.
And then all of a sudden, you're like,
you don't need me anymore.
Anybody can do it, you know, like,
just cut every equation.
Let's just have machines make art now.
Yeah, I mean, I have artist friends
and they are always really scared
of now putting their stuff
out there like i always love playing around with these new tools so there's for example flora where
you can build these super cool flows to get from text prompt to image to then video then kind of
adjust the video based on some other images or tags like a very fascinating tool that gives you
access to lots of different models actually also quite a few free credits so i was asking him like hey can you send me like actual nice image
or high quality images of one of your art pieces so that i just want to animate it and do something
funny like no actually please don't upload it anywhere this one i haven't put on instagram
it's no not online anywhere and i kind of want to avoid it being now used by AI models.
So it's a real problem, and I'm still surprised
that it's not even close to being solved.
Yeah, but it's also a really complicated question, right?
It's a difficult thing to solve,
because I think it's one of those scenarios where you might have to sacrifice something for something greater.
So you might have to sacrifice some smaller people.
I don't mean to neglect anybody or talk down on anybody but
like let's say like if you're not a top name in industry like it's really hard for you to champion
for your rights you know like this is just a fact especially if you're talking about artistry and so
on you you'll you'll um you'll be sacrificed but what are what is the sacrifice well the sacrifice is for
the models to become much better right and what is the upside for those models to become much
better well they can solve more of our problems in the real world and okay what does that give us
well that gives us more prosperity probably if if we deal with it correctly and if we, you know, democratize it and commoditize it and
make it accessible, then it could significantly elevate the entire world's population in terms of,
in terms of, you know, what we have at our disposal. And probably it will be able to help
us lift a bunch of people even further out of poverty
though if you look at the poverty numbers they're like they're way way better than let's say 20 30
years ago but I mean like these models like obviously they can accelerate this this development
of the entire world very rapidly and I'm really excited about one thing which is that you know
like now agents as perceived in the
web3 space they're still purely involved in the informational kind of like world right like they're
still somewhat boxed in um to a certain degree right like they are not they're like i don't see
a web3 company yet that has extension into the real world, as in, you know, there's like some kind of like deep in aspect to the agent or there's some kind of maybe embodiment.
Right. So robotics aspect of the agent, you know, you don't see the self-driving Bitcoin taxis yet.
Right. Like we're not we're not there yet.
But that's that's definitely possible if these models become much better than they are right now
right like right now you wouldn't it would still be hard to have like an let's say an autonomous
agent controlling an autonomous vehicle i think that's still a little bit wonky and probably
there's still a lot that can go wrong especially on ethical levels um but um yeah so so my notion is like
maybe this is a scenario where the industry is going to or the entire world is going to sacrifice
you know like some people's livelihood some people's um business and that is going to become something else something new right and there's a there's a
there is a in my opinion there is a upside to all of this which is that any job that disappears
will re-manifest itself somehow in a different form right so? So you say like, okay, well, you know,
like a taxi driver might become a autonomous taxi,
a vehicle safety inspector or something like that, right?
Like there's always something,
there's always something that is gonna be created
when a new industry is introduced
because there's many problems that we haven't thought about
and that creates a lot
of business opportunities right so let's say you know you're a taxi driver you see this uh you see
this you know this waymo revolution and in china and in hong kong you have a bunch of these startups
that are really really really on the edge and they're pushing things really far they have like
functioning self-driving taxi cabs in beij Shanghai, for example, that are just fully autonomous.
Way better than Waymo.
But yeah, if you're the taxi driver and you see this change happen, then you can, you know, find a new niche and probably make much, much more money with that new thing.
But obviously that takes a bit of thought and a bit of entrepreneurial mindset.
But yeah, it's not all bad you know i don't
know how you how do you guys think about this i mean yeah of course but good luck giving this
altruistic pitch today to an artist i mean they have to think about their livelihood today so of
course they will prefer to actually have some control over their ip to preferably monetize it whenever it's
trained and modeled so i get it but also i don't think it's very fair now to just make it a free
fall and if it's out there you can use it oh we don't hear you marcos oh yeah oh you hear me oh great oh that's all right no i'm just no
just reflecting uh what you guys are saying earlier so i mean i i i i guess if you look
at history and stuff you know uh some of the old previous jobs i I know, like in the Victorian age, you know, like the chimney
cleaning thing or whatever, you know, now you don't have like folks doing that. There's going
to be some shifts in the economy. I think that's, that's going to happen. I guess when it comes to
like, I think the challenges when it comes to like, you know, intellectual property in terms
of, you know, creative works and, you know, artists and stuff, that's going to be a challenge. And I guess also still got to figure out what are the actual outputs and stuff for people
to buy into this future that we're talking about in terms of AI agents and stuff.
So I'm with you, Marco.
I guess we do have, you know, obviously we have
some friends who are in that space, you know, so I think we still have a long way to go in terms of
how to, you know, reward people for their autistic contributions. I think it's a matter of figuring out how to incentivize
human contribution to AI. So especially in the AI space, a lot of the AI models,
it's all built on millions of humans' contributions, right? So I mean, think of your
data collection, data labeling, it's all, a lot of it is done in the global south.
And, you know, for example, Scale AI, a couple of months ago, they got hit with, you know,
some, I guess, the Department of Labor or something for human rights violation in terms
of payments and all that kind of stuff.
Really?
So, yeah, yeah.
I think there's some just a report there's some
articles about scale ai i think they are valued at like over one billion dollars so which is not
something new i mean especially he's got like a lot of these companies do use third-party companies
in the in the global south uh particularly like say like india kenya nigeria and what these
economies have in common is that the economy is bad, but they do have highly talented people.
So unfortunately, they get paid like two dollars an hour.
They work like 12 hours a day.
So you could kind of think of it as sort of like, you know, digital slavery, you know, in a sense.
So if you don't believe me, there is a show called 60 Minutes by Barbara.
I can't remember her last name.
She's an old white lady, and she interviewed a lot of these students from Kenya.
And, you know, they all have worked for these companies,
and they've gone through a lot of human rights issues and stuff.
So I would recommend people is there on YouTube,
which is kind of interesting to see that particular show, which has been running
for more than 30 years, cover this area of AI. So, and I guess it kind of makes sense because
not many people visualize that, hey, you know, there are millions of people who are actually,
you know, contributing their data. They're doing the most laborious part of AI,
which is, you know, data collection and data labeling.
So, and I guess we'll see more of that in the future.
So I guess when it comes to, you know, some of the creative stuff,
I mean, there's going to be challenges with, you know,
with the general population.
Then definitely we're going to have some challenges
from bureaucrats and politicians, you know,
they still want to figure out some way
how they can get a piece of the action.
You know, that's, I can see that coming.
I don't know how they're going to do it,
but I'm sure there's going to be some way
they're going to do it.
So that'll be interesting.
But yeah, so anyways, sorry, apologies.
I don't want to go on a long tangent,
but just want to share a little bit about i guess maybe what i'm seeing so
hope that made sense
yes it does marco are you still with us oh yeah of course uh i just want to ask marcos i mean
you're the one closest to us i mean i know the old data hub and like not shilling the product or anything but you guys have built the system how to reward data sharing participation in all kinds of data sets and i'm
sure some part of it actually i'm not sure but i would guess that there is also some form of
creative contribution and there are also artists part of the platform trying to get paid. But correct me if I'm wrong.
We haven't hit the, I guess, the creative side. I think at the moment we are more focused on, what do you say,
collecting data from, I guess, AI-based companies.
In terms of from the art world or from the creative industries, not yet, not yet.
So this is more like, you know, traditional AI companies.
I mean, so we basically, you know, people, you know, say, for example, if it is in construction or something, so, you know, we ask people to, you know, like, you know, collect, you know, images or even audio for, you know, like in the construction space and later get it labeled.
And then, you know, like send it off to these AI companies who will use it to train their specific models and stuff.
So the creative space, I hope so.
I hope we could maybe find some clients,
but I'm sure, you know,
I guess the art world or the creative space,
I guess maybe they haven't jumped on the AI bandwagon
for a lot of good reasons.
I guess, you know, a couple of months ago,
you had the, what do you call the writer's guild
and the actor's guild in Hollywood. Iwood i mean yeah i didn't even know that yeah i didn't even know
that they still use the word guilt i was like holy shit what are we in the middle ages
i don't sound like a warhammer or like a warcraft kind of a thing right so yeah but but can i ask a question marcus so so um essentially how
you guys would work or how your platform works is that you you pay out uh fees to
so your clients are the ai companies that want to train data correct yeah essentially yeah so yeah and then and then
and then the providers are just people that have the capacity the capacity to fulfill whatever
kind of like data set or requirements for a data set that's out there right so you gave the example
for like um construction maybe it would be like floor plans right for large buildings or
something like that right so you go out and source that material which is not readily readily
available and accessible and then provide that to the to the ai companies is that how i have to
understand it yeah i let me um because i'm because the challenge is that like part of these AI companies,
they have quite a complicated, I guess, requirements. So let's take an example of like, say, okay, say that, you know,
you want to start your own kind of like a fashion app. So, you know,
they need, you know, like raw data of people from different parts of
the world wearing different clothes, you know, like people in China, they wear like certain
kinds of t-shirts or pants, same thing in India or say in Dubai. So you try to, so the idea is
that, you know, we try to get like, you to get different images collected from different parts of the world.
So it's kind of like geographically distributed.
So the more distributed, the more voluminous it is.
And also, of course, not just collected, but also labeled as well.
So what I mean by labeled is that, yeah, you take a picture of me with a polo shirt and
shorts but you also need some context you know like you need someone to label my shirt as blue
it has a fashion logo so these things the more detailed it is the better the you know the the ai
can be in terms of you know training the models and stuff so um so yeah so i hope that kind of made made sense so yeah no it does it does make sense um
yeah and those data sets they are usually they're quite large right yeah yeah it can be yeah um
Yeah, it can be. It can be quite a lot because I don't remember the exact size and stuff, but there is like, what an update, you still need raw data, you know, you still need like from actual human beings, you know, because that's the
way like, you know, the AI gets better and stuff. So. Right, right, right, right, right. It's like a,
how do you say it? Like a, not a moving target, but like it'll evolving, you know, because, you know,
definitely the need for data is not going to stop.
It's going to keep growing and growing, which is good for us,
especially if we are in the AI space.
So I think the challenge is trying to get, for us, the challenge for us
is trying to get high quality data.
I'm sure, look, we can scrape all kinds of data
from the internet,
but the challenge is where can we get high quality data
which will make our AI agents much more better,
perform better, evolve better.
So that's going to be the challenge, I guess.
Especially as we move forward.
So yeah.
I see, I see, I see.
Thank you for sharing. And now we just
by Francesco.
Hey man.
What's up everyone? Hey bro.
Yeah.
Maybe just do a quick intro
and then you can jump right into the conversation.
Yeah, for sure.
I like what I'm listening right now.
What's up everyone?
I'm leading developer relations at ConsenSys.
So I'm playing quite a bit with agents.
We are currently not doing something specific
on the wallet side, on the Metamask side,
but I'm very enthusiastic around developer onboarding
and also like contributing
to a DAO called Modern DAO
where we are basically like building
specific like agent frameworks
and standards for agents,
basically building social graph for agents.
And yeah, we're just very excited
to ship this stuff.
Token just happened.
I hope everybody got a little bit of rest during the weekend.
But yeah, and for everyone listening, happy Pectra Day.
And yeah, thanks so much for having me here.
Yes, I was going to say, I saw you today on my feed like 20 times.
Tell us about Pectra. How did it go?
Yeah, I mean was a lot of things
happened but i think it was very smooth i mean uh compare like some of the those other upgrades um
i mean this was probably like the biggest one post emerge uh is including also like different uh
different eips i also tried to actually like you know, break down different technicalities on, you know, what
actually was happening from the developer side, from the user side.
Obviously, there is a lot of like kind of like nitty gritty account destruction.
I would say inclusion with this upgrade, but probably like it went well.
I mean, the team, the team team foundationals the research team was also monitoring
the first like 50 minutes around the finality so there was a lot of like different things but uh
he actually went quite uh quite smooth and uh yeah i mean it's if you think about the complexity
how long also like he talks to actually like uh involving those uh eips a on on mainnet is
involving those EIPs on mainnet is a very complex work.
And the fact that it went also so smooth
is showing also the caliber on this researcher side.
And the interesting thing is that this, for me,
I mean, translating this for everyone,
is this upgrade also bring really interesting points around usability around specific action that also
new dApps developer can do on Ethereum specifically, right? So it's not just improvements on the
protocol level side, but also like we are enabling i assume it's a very like kind of like
facilitating say it's like we are enabling new new kind of like function new let's say skills
to actually be built on top of ethereum so i think this is a big step for for going into different target developers.
Because I think sometimes, you know, like DApps developers,
they are more high level.
They don't care about kind of like low level development sites.
I think it's a big kind of like milestone.
So I'm very excited what happened.
Kind of like, and obviously there is a lot of like things
that now will progress on a waterfall basis, you know, new IP is coming, there
is already people like seeing okay what's happening next, but yeah very
exciting thing and yeah but I think like there is, I don't know if you guys also
like spoke about kind of like the lesson learned around Dubai, but I think a lot
of like things are happening on the on the builder sentiment side they are very positive around the ecosystem so I'm
definitely very excited also like to to see this sentiment going right because
we are wearing this space you know for for also like you know building better
you know infrastructure better also like tooling, better kind of like, let's say, privacy compliance systems
and, you know, minimizing our goal with doing stuff like,
you know, specific like meme coin trading
is maybe one or 2% of what actually this industry
is capable to evolve in, right?
So, yeah, but going back to this space title,
there was a lot of discussion around token.
I don't know what you said before in the space,
but I hear a lot around the agency, agentic systems,
social coordination around agents.
What do you guys heard about from token?
I don't know if you guys discussed about that
or kind of the main highlights.
Yeah, we did. Go ahead. I've't know if you guys discussed about that or kind of the main highlights. Yeah, we did.
Go ahead. I've seen a
meme about it. I wasn't a token myself,
but now there were some memes
saying that
people were kind of
tired of people
saying, well, what are you doing? Yeah, I'm building
an AI agent.
Yeah.
I think, yeah, there was some discussion a lot around, you know,
what agency kind of like are on production, right?
And people were speaking about, okay, even if you're on production,
there were some names, I don't want to name it,
but, you know, they're being tested and on production
for maybe a month and a half, right?
So that was a little bit kind of like funny.
But I mean, it's also like,
it's also, I would say like,
we are so like early also,
like in terms of like building a genetic system.
I think like unleashing the power,
it's also like a strategy,
meaning right now, like in specific on token,
I see two main use case, right?
On the social side, so on the social coordination, on the waterfall kind of like content distribution.
And then you have on the DeFi side, obviously maximized yielding.
But also like on the DeFi side, you know, there are so many more use cases that we can just do that is not just kind of like maximized maximized yields right on the agency side
so that's why i'm saying it super early but but i wonder what what you guys heard and what's your
feedbacks yes i said in the beginning i was really impressed how much the space was bubbling i i had
at some point like seven people in a round and just everyone was working on agent
to agent collaboration. Somehow agent to agent negotiation, coordination, all of these systems.
And it was like, what the fuck? How are you all working on the same thing? But it's such a massive
topic to solve. And there's so many different approaches. And I mean, even the big web two
players, they are working on this. There always these new swarm sdks from open ai
google released their a2a framework we have still mcp servers even though they're getting somehow
less popular it's really incredible how many are working on this so this was really bullish for me
for the space because previously i had this feeling like decentralized ai was slowing down
like people were realizing hey this is actually difficult to do.
Specifically also just like DeFi agents.
Like I know you say,
hey, yield optimizations,
it's kind of trivial,
but still there's so much like security to be done,
like to build actual trustless agents,
which is what we at Oasis keep on kind of pushing.
It's still very, very difficult.
Like how do you show your users,
hey, I don't control this agent. I don't have the access to the private keys i can't just rug you so it's
it's super difficult as soon as you move away from these fun twitter bots uh but yeah it was
exciting to see that many projects working on it and you know what what what what what signals this
well what what is the reason for that focus, what is the reason for that focus?
And what is the reason for that shift?
And why all of a sudden people are working on the bigger problems?
It's because I feel and I think that the meme super cycle is finally coming to an end.
And everybody has been like ultra rugged.
People don't have money anymore.
And then now it's time for us to focus on the real utility again, right?
Like where last cycle was also a lot of that bullshit going on, but there was still much
more people that were actually trying to solve fundamental issues and build good tech.
And I think that people are trying to revert a lot of the attention now back to, you know,
like projects that actually have product and actually really do something, which is really cool to see.
And I think that also, you know, regarding the agent collaboration and the agent negotiation, agent collaboration, like all of these frameworks, they're all going to achieve a certain level, I believe,
a certain size.
And the reason for that is because there's no one solution fits all.
There's different types of orchestrations,
different types of collaborations.
You have whatever.
You have waterfalls.
You have more kind of like
these free-for-all structures you have
very well defined like abc the efg kind of workflows so there's so many different types of
kind of like collaborative models and different types of things you can focus on within the larger
notion of agent-to-agent collaboration right so for example what we are focusing on is not
necessarily agent-to-ag collaboration, but we're more focusing
on cluster to cluster collaboration,
and then also like the formation of larger swarms, right?
Really ramping up, like instead of we're focusing
on like five to 10 agents working together,
we've kind of solved that puzzle and we're now moving
on to let's say, okay, what if you have a thousand?
What if you have 3000?
What if you have 5,000?
What if you have 10,000? What if you have a thousand what if you have three thousand what if you have five thousand what if you have ten thousand what if you have a hundred thousand like
how do you how do you make a system like that like behave well and become more powerful than let's
say like just a few agents uh that are working together so there's you know like this issue is
starting small like there's a small problem that is to be solved, but the problem is going to grow and grow and grow and grow and grow.
And there's going to be so many niches to be served, right? of these kind of like frameworks and models is that if it's too general, you will then again
require your users to understand too much and do too much work themselves, right? So if you
make it too general, then you are putting the, let's say the requirement on the user of really
understanding what an agent is capable of and so on. So I think
that there's going to be these kind of like agent collaboration frameworks specifically for DeFi.
And then for us right now, it looks like it's agent collaboration framework specifically for
handling arbitrary data. Like there's like very different kinds of things that are going to pop
off, I believe. And it's really exciting to see that so many people are working on it.
And I'm also happy that it's starting now because we're already so far ahead.
I don't want to boast, but yeah, we're kind of like early to the space.
And that makes this position relatively favorable.
I just have a quick question on the, I 100% on the social cooperation side that there is a lot of also discussion, people already implementing different things. But I want to say like two things on what you said at the beginning, beginning, right? proudest part of this cycle specifically was seeing kind of like web 2 phd coming and funding
and building agents from scratch like this is true smart capital injection and the reason why people
are coming is obviously the web 3 rail you have on-chain data feedable under like 24 7 you actually
can speak with different protocol and fetch their their. So the reason Web3Rail is really providing any value to those people,
and we see kind of like, you know, agent builder flourishing.
And probably my proudest time was actually, you know,
interviewing and learning from this kind of like agent builder
and understanding really why they came on kind of on Web3Rail.
And I think that was one point.
And my second point is agent-to-agent collaboration.
One of the, probably the most important things
I want to say is that we need also like some standardization
or some specs around agents,
because I feel like that was the reason
why actually like MotherDAO started in the first place
is that we didn't want to kind of like see
what kind of like happened with DeFi Summer that people forked Dex's, like thousand Dex's and user at the end of the day got kind of like ragged because it was the fork of the fork of the fork, right? social coordination layer and social reputation that is on-chain for agents,
we will not be able to actually surround ourselves
who are the agents really providing value
and users will not be able to actually find these out
from the website-specific.
But also, I'm happy that, for example,
you guys are pushing also hard on specs and standards
because I think we don't want to silo
more
kind of like agent builders
and having standards
and specific specs on how two new
builders come to build
is very crucial
that is the whole ideology
behind the truth protocol that
we're developing
and truth for not just for finding the truth
in let's say the context of misinformation and so on but also in verifiability and reputation for
for the units or the nodes that operate on the network and so i did a very interesting exercise where i said like
okay what if such a protocol of like reputation essentially right because uh let's say like if you
want to be able to rely on a certain output of a certain cluster of agents or an individual agent
right like the historical track record of that agent
is really crucial.
The person who builds that agent
or who governs that agent, essentially,
if it's a group of people or if it's, you know,
if it's anonymous, it doesn't really matter.
But who is the controlling entity of this agent,
for example, like all of these different pieces
of information, they are really important.
It's very hard to find, you know, like find out on an individual basis, like, okay, who made this? What did it do in the past? Can I trust it and so on. So for that, you need some kind of like standardization, indeed. And so the exercise that I did was I imagined that the that at the start internet, right, like what would such a protocol have looked like first for humans?
And then I went through five different phases and I eventually kind of like imaginarily evolved it up until like a blockchain standard, like a unified protocol essentially and what comes back all
the time is like okay like this this reputation aspect is just super crucial like because you
want to be able to rely on on a service but you don't want to necessarily have to do the research
on on the service right so if i go to like a famous company i go to that company because
i know that they do their job well because they've you know they have a like a very high market cap
high valuation and they've surfaced a lot of you know other companies and i'm i'm willing to pay a
premium for that but that's like kind of like a social construct right but the problem with agents
is that there's going to be so many of them right and they're going to be so like specific specifically solving like certain problems so
you need to have some kind of like a unified and standardized model for this because otherwise
people are just going to be so confused and they're just going to get rocked all the time and
then party's over for everybody i agree where where does people can see these standards in general
right is there some kind of like i remember back in the days with uh with the discussion around the tino magician with
the nft standards that you know people came out with rpc library specs do you see these standards
or kind of like some uh um some kind of like uh standard specs libraries somewhere where people can actually see is there some resources out there
um not not that i know of but we are going to publish something similar probably over the
course of the next few months um and and but you know like the problem with with this is that
it has to become a thing that everybody uses right right? And I don't, do you remember like how that evolution
came to be like for the NFTs?
Because at one point everybody was just like settled
kind of like on, that was just like an Ethereum thing, no?
That was just like-
Of course, of course.
And I remember, yeah, the debate started
because like a lot of people were saying,
okay, I wanna move from like, you know, open source to variable, but there was no specs on the ERC.
So like I was like, OK, that's super interesting.
And then I started to see the debate.
Hey, we need to standardize specific kind of attribution levels.
If not, not just marketplaces, but let's say also like NFT consumers and dApps builders.
Obviously, this is more kind of like scoped, verticalized.
And I think this will happen the same with like agents, because people will build siloed agents, especially on DeFi.
And it would be like this kind of like social to agent to agent collaboration would be more important.
And if we don't have standards that someone and I don't know, like the question is also whose responsibility is to build those standards.
Right. I saw it already kind of like popping up um i'm not sure if those kind of like standard can
be driven by uh the registries um or like by certain foundations but we need that something
something like to kind of like clear out the the space because it will it will come
um sooner rather than later right yeah yeah And if you really think deeply about it,
like the reason why I did that exercise
to try and imagine what that would look like
at the beginning of the engine ad
and then like propagate it through time
as new other, you know, like sub-protocols came to be,
like HTTPS and whatever,
like all the other kind of like security related stuff.
They were just mainly focused on the verification
of the source, the validity of the message and so on, right?
And then a couple of like very simple
security features built in.
But what I'm getting at is that
it's not a super complicated problem, actually, right?
Like, developing the standards and then kind of, like, making sure that they are fit with the evolution of the space in general in terms of what the agents are required to do.
But, like, obviously, there can become, like like a scenario where there will, you know, a
standard could become a bottleneck. But then, you know, like, if it's a standard, then it shouldn't
be that hard if everybody has consensus around that standard to then also have some kind of like
move forward and evolution in the standard. But it's a really interesting one, because it's kind
of like a chicken and egg problem, right? Like, what do you need first? Do you need too many people
that want to do more with AI agents, but they hampered because there are no standards and everybody's just like fighting
each other or do you start with the standards like probably it's i don't know actually
yeah it's a good point i just want to say that you need to also like at the end someone need
to drive those kind of like standard conversation out.
And I remember back in Denver that I see like Jake is also in the audience, but I was quite impressed by the guys from Story, for example, they push this kind of like standard specs
around IPs.
So at least like, you know, like you have, for example, companies saying, okay, we're
pushing kind of like this standard specs out so that actually people
can follow and the expectation on the kind of like on the agent side is that builders
will start using some of those specs so that they know that an agent will be unified and I'm assuming
the same conversation maybe will happen on the on the side maybe on the registry they will stand
they will actually say hey this is the registry specs
if you kind of like want to align with other agency uh a jantic kind of like a like agents
in the space this is something that you can follow um so assuming something like uh something
like that would happen um but yeah it's definitely exciting yeah but another thing to think about
though is that agents are so versatile right like they don't have to adhere here
to any specific thing it's very easy for them to adapt to like a multitude of different types of
like interpretation of that standard and maybe across different protocols too so it's also yeah
it's an interesting one it's very it's very interesting indeed um yeah but I mean, they're, yeah, so like the problems will arise when these agents, let's say like you build an agent on, let's keep it really simple. You build an agent on Ethereum, I build an agent on Sui. And now we want to have some kind of settlement take place, you know, like that is standardized across Ethereum and SUI, right? Like as if maybe like
as you could, you could, for users, for our listeners, let me try and make it as simple as
possible. You probably think of like a bridging function, right? So I want to send tokens from
one chain to another. And there's like, there's like a bridge that facilitates that, right? So
how that can work is that, you know, like one side of the chain that sends the tokens to the
other side, the tokens are locked and unlocked on the other end and vice versa in really simple
terms. It's obviously a little bit more complicated, but like, you know, like now if you want settlement to happen between agents cross chains, you're going to have an issue, right?
Because like there's, that's going to be very complicated.
So now there's also this battle going on.
It's like, okay, where should the agents be built?
Which I think it's a really ridiculous conversation to have.
have it's like obviously they should be built everywhere but the more important question is
It's like, obviously they should be built everywhere.
like eventually how can you make sure that they um they can do business with each other across
different ecosystems yeah i really hope like we're we're not getting too down to the conversation
like also which uh you know which registry or which kind of like uh um framework has more
distribution to kind of like agent consumers, right?
Because one thing I'm asking myself,
especially post Dubai,
is that I'm thinking also like
if there will not be a big problem solution fit
for independent agent consumers,
we will probably see more kind of like agent
being kind of more like a B2B kind of like service side.
So agents will actually servicing protocols and protocols,
DeFi specific like governance protocol will be the one actually up,
I would say up value proposing like the value to the users, right?
So basically like you are abstracting the service to the users
and the users is getting kind of like
the benefit of this kind of like up up time and uh but the service that the agent does is to the
protocol itself yeah yeah yeah yes correct but yeah but the the most let's say like if you
if you want to imagine a future where the agent space has flourished into, you know, like, like it's,
it's near maximum state, then you would have to have proper, proper, proper standards,
that and proper standards, proper foundations, proper systems, proper technical solutions,
that are able to handle all different kinds of very complicated things so what i see
where i see a lot of potential too is in things for example like dispute settlement right
so what if you know like you you get these like very large different systems and hence we also
think that's part of the truth protocol that we're and it's an important part of the truth
protocol that we're that we're developing is also, it's like,
okay, so what if there is a dispute among agents
where, you know, like, it's gonna be impossible
for a human to solve this situation,
but it can also not be that, you know,
like your agent can just go around
and like scam other agents, for example, right?
That's gonna be like mayhem.
So there are so many very interesting problems that will have to be solved,
which are really complicated problems beyond just standards.
Agree, agree.
Now, I really like this conversation, how the flow is going.
We should definitely do a part two here.
Yeah, yeah, yeah, definitely.
Yeah, yeah, definitely.
And another thing which I wanted to touch on, which you just mentioned earlier, definitely do a part two here yeah yeah yeah definitely yeah yeah definitely and um another
thing which i wanted to touch on which you just mentioned earlier i know we're a bit over time
guys but let's let's just spend another five minutes if you have time um what i wanted to say
is that um yeah you're right about it should not just be a solution for businesses right because
then like what is important for us and where I see a lot of potential is,
I want to be able to facilitate individuals to be able to adapt quickly with times and
create opportunities for these people.
And the way I see the world in the next five to 10 years is that
I will have a business and my business is probably like 90% made up out of agents or
things that look like agents and maybe a couple of like tools which are not agentic but that the
agents might use and that I might use to coordinate,
you know, the business and its functions and so on.
So what is really important in that context
is that you abstract the complexity away
from the user that wants to build this business, right?
So what that means is that like,
you need to reach a certain level of standardization
of a whole slate of different
separate like like um like workflows if you will and that are part of building an agent or
your larger agent systems such as like a swarm for example so kind of thinking and thinking about
it's like a 360 life cycle management for agent to swarm, but it has to be abstracted and it
has to be simplified in such a manner that, you know, like I can do it without having like deep
technical understanding of like, you know, like how do I write the Python script or how do I
implement an MCP server or how do I do this? How do I that, right? Like you want to make it
accessible to users. And the problem with agents right now is that yes agents are user facing but they're user facing in the way that hey I built
this agent and now you can you can come and use it but the real the real problem is like okay
build the tool that allows people to build the agents that other people can then use right and that's where you really have this snowball effect um where you know like now all of a sudden everybody
can build agents that actually are business grade agents they can become part of larger let's say
subsystems and maybe even larger systems than those subsystems so i'm talking about clusters
and swarms and then you know like we can launch a token for swarm and we can govern this
thing together and becomes like an entity and can service other different endpoints and other
businesses and so on and then you really get this whole you know economy of agent ownership
and swarm ownership kick-started but um yeah there's man there's so many things that have to
really happen before that can become like a...
Yeah, exactly.
And also it's fine if it happens in parallel.
Like we don't need to be pushing
just this kind of enablement of non-tech users
and build these no-code agent platforms.
It's fine if it's sophisticated agents
that are actually made into a business.
Like for example, I can build an agent
that does yield optimization for myself
but arma the team uh geez where geezer is the team behind it they have built a very cool one
they guarantee or currently it has like 8.5 percent yield on stables i'm like this is fine for me like
i don't need to build something uh myself i do trust them i kind of check how they use session
keys and perm keys and in parallel many are actually now building these no-code platforms and i'm definitely with you i think we will move towards the future where
everyone is their own entrepreneur you can have kind of these agent swarms doing things for you
and you are kind of the owner of those best case you're even running them locally but
there's quite a few steps until we get there
few steps until we get there yeah and you know what is also a funny uh topic is that i think
daos like the dow structure is going to become more and more important and people will like
once and for all really have to solidify the tech stack for for daos uh because i think like like
for for daos uh because i think like like the dao system is going to be the vehicle for
not just like the autonomous organization but the autonomous organization that controls the agent
swarm cool um can you still hear me, Marco?
Yes.
OK, OK.
All right.
It's such a shame that we're already way over time.
Yeah.
We should have to do a part two.
It's really interesting also flowing these discussions.
Yes, yes.
It was really nice to talk to you guys.
This was really, really nice.
Let's do a follow-up with this panel, and then we can go a
little bit deeper into the nitty-gritty. Thank you so much for joining us today, everybody,
more than a thousand people listening right now. It was a pleasure to host with Marco once again,
and we'll see you on the next one. Thank you, guys.
Thanks, Francesco, for joining. Thanks so much, everyone, for people taking the time.
Cheers.
See you next time.
Bye-bye.
Bye-bye. Thank you.
