Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Agents Unplugged: Your Bi-Weekly AI Deep Dive #14
Recorded: July 30, 2025 Duration: 1:05:01Space Recording
Short Summary
In a dynamic discussion, industry leaders from Fetch.ai, DAP Radar, and Intuition explored the burgeoning agentic economy, projected to grow from $2 billion to $28 billion, and the critical role of decentralized technologies in shaping the future of finance and data privacy.
Full Transcription
Thank you. Hello, hello.
Give us a second to sort the speakers and then we'll jump into it
oh never mind i think everyone is here except for one person yanik who is usually my co-host in these this is session number 14 so he
managed to do 14 sessions without missing and now is the first one i heard he got food poisoning
i'm very sorry to hear yannick we're thinking of you this shit is horrible man i've heard so many
horror stories about food poisoning so i'm really not jealous but uh good luck and the show must go on and we have an amazing lineup of speakers again
and i think we can jump right into it um maybe let's start with who am i in
hey hey guys pleasure to be here nice to meet you all
Pleasure to be here. Nice to meet you all.
Hello. Maybe tell us, who are you, what do you do, and why are you excited about Agents?
Well, we came up with the whole term. I run a company called Fetch.ai, a project called Fetch.ai.
We've been in this space since 2018, and we started this whole movement of AI and AI agents.
And, yeah, we have a whole infrastructure for agentic systems, and we're progressing, and we're seeing a huge uptake coming in the sector.
Perfect. Thank you.
I'm sure we will discuss quite a bit more about agents.
Nathan, what about you?
Better with me.
I was just taking a moment to post that we are live with agents unplugged uh i had to
make sure that the the people on the timeline knew what was up really a pleasure to be here
thank you so much marco for the invitation glad to be here with such esteemed guests and projects
my name is nathan van dee i'm head of dowAP Radar, the world's DAP store where we are building the trust and credit layer for the agent economy.
We're really, really excited about this really pivotal scoring layer, which is needed for not just discoverability,
but also coordination of AI agent swarms and really exciting financial applications that will come
from this eventual trillion dollar economy. So super excited about all of the different things
that are happening in the AI and agent space. The fact that it's slowly starting to mature,
we're finally seeing DeFi agents that are actually usable and even more profitable than your typical vaults and algorithms
and super bullish for the agentic future thank you so much once again i'm happy to dive in
with you all thanks for the reminder to do the post i do encourage everyone to just quickly
share it does help if it's on the timeline and i see after nathan posted we already got
tons of more listeners i i think it was you uh billy matthew i don't know i guess you both want to introduce yourselves
i don't know who pitches intuition the best go ahead i guess i can go first and i'll hand it off
to matt to fill in the gaps um everyone i am billy been in crypto since 2012 and involved with Ethereum since very beginning, 2014.
Did a bunch of stuff over those years and most recently started something called Intuition,
where we are building kind of this shared memory, shared context layer for kind of the future of the Internet.
So Intuition is this big decentralized knowledge graph composed of
verifiable claims slash attestations about things. So anything can claim anything about anything,
and it just lives in this nice big knowledge graph that's very easily traversable. And so how
that kind of plays into the AI agent spaces, you know, we're working on AI agent reputation,
we're working on personal portable context, we're working on platform
reputation and discoverability, and many other things. But yeah, you can just kind of think of
us as like the intuition module of the collective conscious. And that's kind of where the name
intuition came from. And where this whole thing sprang from was, you know, the AI is this kind of,
it is it is the aggregate of the collective human conscious and we think that that thing
needs an intuition module
and the way that you get good intuition about
things is you get more better data
about things so that if you boil it down
that's our goal is get data
to where it needs to be
at the point of interaction
from the sources that you trust
and with that I will hand it off to
Matt I have to give Billy some props from the sources that you trust. And with that, I will hand it off to Matt.
I have to give Billy some props.
The explanation of intuition can be a rabbit hole,
but I actually have no gaps to fill.
That was bang on.
My name is Matt,
and I head of strategy and business development at Intuition.
I got into the space full-time end of 2016.
I actually ran a hedge fund in the space for a little over five years. And intuition was a portfolio investment. And I
got closer and closer and closer to Billy and the vision that he had for the future of the web,
for decentralized data, and for user experience and application development. And I sort of got
intuition pilled and asked myself, why should intuition not be bigger than Google with its
approach to structured data, not at the application layer, but at the actual layer of the internet?
So to go back to your question of what we're most excited for about the agentic future, I think it's
becoming more and more obvious that a mouse and keyboard and sitting down
and a screen to touch is actually a very poor interface.
And speaking and gestures and having things completed
for you in a way in which you can trust
is more and more likely to be the future
of how we interact with technology.
And we hope to be part of powering that with our decentralized knowledge graph.
I was going to say, yeah, this was a really good pitch just based on the website.
I was struggling a bit to get it, but this made lots of sense.
But might be my bad, just not digging too deep on the web.
No, we'll fix that soon it's a work in
progress everyone says the same thing yeah sometimes it's a part of being in crypto like
you're not supposed to understand at first glance it should just sound exciting and then at some
point you get into it uh before we dive into some discussions uh some housekeeping so we generally
keep the discussion very open
I'm not gonna pick everyone one by one like we just did in the intros if you have something to say just unmute yourself no need to raise hands or anything you can bring up new topics we do try
to keep it as shill-free as possible but of course the projects you guys work for or are kind of
building are super exciting and happy to learn more. And I do have
quite a bit of questions. But before Yannick fell sick, he did gave us some pointers that he wanted
to discuss. So thinking of him, I think we can start with this. And one point that he left was
that apparently the Argentic market has reached more than $2 billion in value and is supposed to grow to 28 billion.
Honestly, if you ask me, these numbers seem very small, but okay.
So his burning question was, where is this growth going to come from?
I do have some ideas, but happy to hear yours.
Trillions.
Yeah, definitely think that the agentic economy is going to trillions in the future
um and so does pwc and a bunch of other players in the game i think when he was looking at the
numbers he was like focused on the current like crypto market but when we're thinking about the
vision of the economy we're seeing a convergence between
the traditional capital markets and the crypto markets and sometimes it's hard to see a forest
in front of the trees i think one really good um anecdote to help you realize how early we are in the crypto market is during Donald Trump's trade wars in Q2
the crypto and also the stock market had crashed and a couple weeks in someone messaged an
influential news reporter message that Donald Trump was going to pause the tariffs for six weeks.
And in a matter of minutes, $2.4 trillion had flooded into the stock market. And that's really
mind-blowing for two reasons. The first reason is that at that point in time, the crypto market was around or even less than $2.4 trillion. So that shows you how small the crypto market was around or even less than 2.4 trillion.
So that shows you how small the crypto market actually is right now.
And then second of all, like moving 2.4 trillion isn't done by humans. It's done by algorithms and by sentimental,
mainly sentimental slash fundamental trading algorithms.
And it was exciting to see those two things
in understanding how, one, early we are,
and two, how the future economy will be managed
by algorithms and agents in the future.
All that money will be moved by agents and agentic vaults.
And we haven't even touched how much it's going to be managed by them
because we've only started to build the regulatory rails for this institutional money to come
in like the genius acts like the meeker like vana which is going to allow institutions
to be more confident moving forward and we're starting to see some of them owning crypto
in their their spreadsheets a
lot more now and so when i saw that number i was like not to be disrespectful it's cute but i
definitely like see us slowly bringing in like the capital market which is in the hundreds of
trillions right um rather than um the the ones or tens of trillions and crypto is just a more efficient way
to do that we just need to build the trust and credit rails for the institutions to feel confident
enough to do so we need to make sure that there's privacy um on chain so that they don't get front
run by b gens we need to make sure that we can build
institutional grade technology which is robust as well and then eventually
there's going to be a ton more agentic and traditional financial players on
chain and that's all that the ladies sang
might be to follow that up um yeah i think i think that the numbers being thrown out are like
for for predictions are kind of a gross underestimate of where we're going. And I think it's due to the fact that, you know,
we're not really programmed to see things exponentially
and we just have all of these overlapping exponential curves.
So like supply of money going exponential.
So even if the denominator just changes,
that number seems small
because they're just going to keep printing money
across the board.
Then you got like overlapping acceleration, exponential acceleration curves of just software
development generally, and then AI development and all of these things.
And then the AI are going to start programming themselves.
And so I think they take over way sooner than we expect.
And I think they capture way larger of a share of the market than than we expect and i think that this is just this is just an inevitability and it's the way that it's going
to go and so it's kind of up to the intersect i don't it sounds a little bit i don't know like
egotistical but it's kind of like up to i think the intersection of crypto and ai to make sure
that that goes well because we could very easily enter some crazy dystopia where you have these centralized actors who control the entire agentic
flow. And maybe you can't get access to the best agents. Or maybe you only see what they want you
to see and you don't see what they don't want you to see. And they only allow you to interact
with certain platforms because you're interacting with the platforms through the agents now.
And if the agents are picking which platforms you get to interact with,
then they control the flow of commerce.
And so, yeah, I think the wave is here.
It's coming.
It's an inevitability.
And then we kind of have a duty to do our best to make sure that this future
upholds the values of the crypto ecosystem,
which I think are just kind of generally good moral values that will lead us
into the direction of a better future. But yeah, I think that we don't want to spend
our lives behind devices, tinkering, context switching between different platforms, doing
research. I think the human experience just generally gets better as we
identify it more and more um I hope so it could go it could go poorly or it could go really really
well and we can just enter this potentially like abundant society where everyone's actually getting
to live their lives and pursue their dreams and do the things that make them happy and so I think
we're gonna see like, even though the amount of
technology and the advancement of technology is going to increase, I think we will be interfacing
with it much more frictionlessly and fluidly so that it kind of falls into the background more
and more and we can actually hopefully get to live our lives. But we'll see.
Yeah, there's a really important, I think, risk that Billy's pointing out, which is the risk of these centralized LLM or AI providers posed to sort of like the future of information flows, right?
And to control information is to kind of control the world.
control the world. To date, we've seen these technologic centralizing effects, and they've
been very good, and they're very efficient. But if you ask people maybe five years ago,
six years ago, about the browser wars, this was in Delphi Intelligence's most recent,
or their first AI report, which is a recommended read, You would say five years ago that the browser wars are
between a handful of pretty established players and they're using their devices and all of these
things to push. And that model went on for a couple decades of devices to distribute software
to get ad revenue dollars. Out of seemingly nowhere, in pretty much less than 24 months, you now have Comet Browser, Dia Browser.
So Perplexity has a browser.
The browser company is putting out a new agentic browser.
OpenAI is putting out a new agentic browser.
And a browser as the gateway to AI is like the honeypot of personal information, right? Like all that stuff that you
maybe were smart, maybe you use the VPN, maybe you're blocking advertisers, maybe you're blocking
cookies, like maybe you're doing all these reasonable things. Like, well, that's going
to go away when you want your agent to be like of high utility to you. And it's sort of up to the decentralized agentic movement
to use verifiable programmatic assurances
because without those, they just simply won't come.
So whether it's private data, public data, semi-private data,
whether it's verified inference,
like whatever it is in the stack,
like it must be as easy to use as these browsers.
You could just download, link an account and get going.
So like, I hope that we've learned enough from like the DAP boom to finally know that
like it's not enough to have all this friction in terms of using these quote decentralized
applications.
But what we need to do is for the first time
and hopefully with the agentic wave
compete with these amazing products
like the new wave of browser agents
or agent browsers that are coming.
I think there's a slight overlap
in what people think agentic systems are and how we're going to evolve into this new economy.
I think the numbers which are thrown around, I mean, it's like saying what's the value of all the websites in the world, because that's how I see agents.
So all these websites will be, at so all these websites will be at least at least all
these websites will be agents and the e-commerce will be carried out in in an agentic way not
focusing on crypto as such but just generally on the agents and agent economy. I think that's the scale of change that is coming.
And I think as somebody pointed out, these LLMs which are centralized and this is the new way
of browsing agents effectively. And we need to see how that evolves because nobody knows what's going to take the trophy here,
but I think there will be multiple winners.
Where we have seen true growth, I think, is within organizations.
So entities like who are writing this software, which was monolithic or very archaic effectively
because it's not making live decisions.
That's going to be the first change because that will enable centralized entities
to actually convert their software stack and make it more organic.
So I think the number is very very
conservative that was thrown around of course I don't want to be overly
enthusiastic about it because I think there are definitely hurdles to getting
there because you know legacy systems don't adapt to new systems that quickly
so it will take some time but there's definite you know change that is coming
it's pretty obvious the question is you know who's going to take it first how how does that manifest
then in everybody's life which we can see chat gbd has made an impact and i think that's the kind of the beginning of this
this whole change which is coming agents will become part of it
change is definitely coming i am still wondering where this growth will come from i don't think
it's going to come from crypto um normally when
i think about anything driving industry it is coming from web to corporates them finally
implementing some new tooling and driving adoption of this implementing because those are the ones
bringing actual financial value to users yes yes or no, because I think if you look at, okay,
where did the new things come from?
And if I give you an example, it's not always the corporates.
Google was not a corporate in the beginning.
Corporates came because the consumer came.
Airbnb was not a corporate in the beginning.
Corporates came because the users came,
and individuals came. The same applied to Amazon, because all it was, was individual, the consumer
using that capability. So, and I think the reason why that happens is because an individual,
a consumer, where there's no high risk of change is much more acceptable to change
and would take on the change a lot quicker.
And if that scales well, then you can trust that the enterprise will definitely join in.
I mean, we've been working with the automotive sector,
with the agentic systems for five, six years now.
They're not very receptive to change till the consumer comes and says,
hey, I want to use this. And that's when people start to use change. So to bring a new technology,
I think you could just turn it on its head and say, for example, gig economy could use agents
very easily. Now, I agree with you that crypto is not the main one, but if you think about gig economy
using agentic systems, you could easily add digital payments and crypto payments or utility
of some kind of crypto stuff into it. So I think crypto is actually quite lucky that the timing is
working out in the right way way but AI and agentic systems
can actually bring a lot of momentum to crypto itself the problem is that the ethos in crypto
is probably not the best still because we're always chasing these get rich quick kind of schemes
these meme coin schemes we kind of feel that any serious project is takes too long to make you
millions so I think that's that's the balance we need to strike interesting so you would say that
this the adoption will be consumer driven and not productivity and cost efficiency driven
well if you look at chat GPT um look at what happened there. I mean, I was part of DeepMind
right from the founding team. I was the first investor in DeepMind. The reason, and we had
LLMs, everybody, most big companies, most big corporates who were researching AI had LLMs. We
were part of OpenAI initially because it was meant to be open. Of course,
I mean, there's a whole other drama which I think somebody touched on the decentralization part.
But where did this scale and where did this adoption come from? Adoption came from when
students started using it, individuals started using it. Suddenly, AI become known to the world. It wasn't the corporates who did that. Corporates always had this robotic process automations. RPAs have been existed. You had chatbots and you had all of that. I agree the technology is much better, much more kind of intelligent. You say intelligent, but let's call it intelligent for now.
intelligent well you say intelligent but but let's call it intelligent for now
but the adoption came from individuals right that's what brought the scale
okay and i mean i think all of you mentioned at some point the role or importance of crypto xai
like we can't let the centralized players go as they did in traditional software
but where specifically how does crypto actually help us and there's a question for everyone of
course well we launched a decentralized llm we trained it in a decentralized way um asi one ASI-1 which Fetch launched is the only non-corporate run, non-one-entity run LLM and that's available
and it requires you to lock up crypto, it doesn't charge you anything, you lock it up and you can
use the LLM. It's at par with any other big LLM. So projects using could choose a decentralized option.
The point for me is that decentralization is not the point.
True utility is the point.
Are you delivering to the community?
Are you delivering to the user something they can actually use?
Just coming up with X y and z and just
attaching the word decentralization doesn't bring utility to it so the first thing is we should
create utility can you just quickly explain the difference between a decentralized llm and an open
source one well it's the data which has been used uh has been trained on, how it has been trained on,
multiple nodes which have used to train the LLM, and the ability, and the guardrails.
The monitoring of the guardrails is done in a, I mean, that stage is coming very soon in our case,
because we're now going to give community the ability to control the guardrails, for example.
And the whole point is that you're making it more
controlled by the community in a decentralized fashion. I have a bit of a different take,
although I think there's plenty of overlap. And I think like the killer use case, I mean,
maybe we can call it utility, but I think it might be better characterized as convenience.
Like what chat GPT brought to the world was like the ability to get questions
answered more quickly. Like what email brought to the world was like the ability to, you know,
send a message in real time without the person needing to be available. What the phone brought
to the world was like the ability to send your message instantly, you know, without having to handwrite it.
Like with each iterative step,
it's become more and more convenient
for us to do the things that we do.
So I think like decentralization,
like actually isn't the core feature.
Like I think decentralization is an input
to these outputs that ideally we want to provide that create convenience.
Now, we kind of talk about this like potentially dystopian future, which doesn't necessarily
exist in this very moment for everybody, but for some people it does.
And this brings me back to like the early 2018 to 2020 conversations around Bitcoin. It was this thing will never become big because we don't
need this like permissionless money because we trust our banks and these institutions and gold
has been around for thousands of years. But that was only a fraction of the global population for
many other people through centuries, like they've woken up and their savings are gone,
or capital's been seized,
or capital controls prevent them from transacting freely. And we're starting to see these same types
of curbs now with these centralized LLMs. Grok recently had an update and started spewing some
pretty out there ideas and thoughts, like OpenAI will decide that some questions aren't answerable for you, right? Like that knowledge or that information is off limits. And what we know is a guarantee is that with enough time, the censorship of information will be abused for control and power. like there's pretty much not a single example in history where that isn't true.
When you look at like the progress of an imperial force,
whatever that might be from a news media to a government to a television news,
whatever you want to call it.
So like I think that the convenient factor that crypto slash AI has the potential to capture is, you know, uncensored
LLMs, LLMs, which help you, uh, or don't censor you or give you, I think, further control around
how it, um, interacts with you or responds to you or the tone it uses so that like you truly have
what feels like a personalized, customizable experience that doesn't have these friction points of, oh, I'm sorry, I'm not a lawyer.
I can't answer that.
It's like, I know.
For the 35 other times I've asked you a question like this, we established, I know you're not
a lawyer.
Stop having a disclaimer, right?
These little small friction points are an opportunity for decentralized AI to deliver
for decentralized AI to deliver a more convenient experience.
a more convenient experience.
And it starts there because I don't think we have both feet in this dystopian future.
But you start there so that you have a moat and a core user base and, quite frankly, start
to create this counterforce to ulterior motives.
counterforce to ulterior motives, right?
Like right now, globally, you know,
fiat currencies kind of have this counterbalance
to like gold and Bitcoin,
meaning that like we're seeing central banks
de-risk from over-leveraged paper debt
and start to increase like their gold reserves, right?
So that is a raw power force that they're able to do when they feel a little
less comfortable with one system and a little more comfortable with another system. And I think
decentralized AI has the ability to also provide that counterforce, which says, hey, open AI or
Anthropik or anybody else, like if you guys get a little too greedy with your data or your censorship,
or you're working with certain governments, like we will, you know, we, the users,
will leave you this, your LLM offering
and move to this other LLM offering one,
which like doesn't have these tentacles in it,
for lack of a better term.
But I'm rambling a bit.
I'll stop there.
I can pick up.
So I really like framing the importance of decentralization with a very simple flow,
where let's start at kind of what you're trying to do. So let's say you're trying to do a swap,
or you're trying to book a flight, or you're trying to book a restaurant. I like the restaurant
flow. So let's use that. So, you know, the way that you're going to book a restaurant. I like the restaurant flow. So let's use that. So, you know, the way
that you're going to book a restaurant reservation in the future is you're just going to say to your
agent, book me a dinner reservation for tonight. You're not going to need a huge prompt. You're
not going to need to provide a ton of context. Like we want to remove all of the friction from
this process of just booking a dinner reservation. So you say to the agent, book me a dinner reservation for tonight. And the first step of that process is you want the agent to
know things about you. And you don't want the context to live inside of the walled gardens
of open AI. You want personal self-sovereign identity and context and preferences and attributes and social graphs.
So that you can bring those preferences, you can bring your data with you to any model,
any platform. You don't have this vendor lock-in where they're just stealing your data and, you
know, doing God knows what with it. So, and then you can't bring it to you know, you can't take your open air context and bring it to anthropic. So all right, first step is, you know, you say, book me a dinner
reservation. And the agent says, Okay, you know, I know that you like sushi, here's this guy that
you trust in the context of sushi. And he likes this sushi restaurant in your area. You know,
maybe you're with someone and they got this allergy, whatever. And it's like, okay, now here is a very personalized response. And fun fact, like, we actually have that right
now. So like, as you interact with intuition integrated things, you're building out your kind
of like your intuition, your second brain. And then we have this open source MCP server,
where we put an AI chat on front of it. And so now if you just ask AI questions, like I asked it the
other week, like, hey, what book should I read next? It's like, oh, based on all of this stuff,
you know, you should read Project Hail Mary next. Hadn't heard of it, read it, great book. Like,
holy shit, it's actually working. This is crazy. So the first step is personalized responses from
the agents. The next step is, which agent does this request get routed to? Like what's going to
evolve is this coordinating agentic swarm
where you can think of each agent
as like a little neuron in the hive mind
of the collective conscious.
And it's like, okay, you know,
which agent is good at booking
dinner reservations in my area or whatever.
And so now you need agent reputation
and context and metadata.
Where does that live? Like, does it live in
the walled garden of open AI? Like, like who controls the like the the context and the
reputation and the metadata layer for like the discovery of good agents for the right job.
And the answer is no one should own it. And even Google agrees, which is kind of crazy. They put out this A&S spec recently that is a call to action to have a decentralized registry for agents and MCP servers and all of these things where it's like they are even recognizing the need that no one should own that layer.
Because whoever owns it would control which agents the requests get routed to and they get put in a really strong position of power.
So we don't want that. So now you need decentralized, you know,
agentic reputation, like first you had decentralized people reputation and context and metadata. Now you got agent reputation, context, metadata. Then the third step of the flow is
okay, which platform does the agent use to book the dinner reservation? Does it have a whitelisted address,
like a whitelist of URLs?
Does it do a Google search and then choose the top result
and get phished and lose your crypto
or your credit card info?
That's probably not so good.
So now we need platform, decentralized platform,
reputation, and context and metadata.
So in order to just fulfill that very simplistic flow where it's like, hey, I just just booked me a book me a dinner reservation,
like at each hop of that flow, you need a bunch of things that I think would benefit
pretty significantly from decentralization. And I just mentioned kind of the ones that
we're playing at the intersection of, but you also need like payments, and you need a bunch
of other things. So like, that's the whole flow. If you just think about it, just think where are the trust
assumptions that we're taking on by interacting with a centralized stack when going through this
flow. And then think about where, you know, those trust assumptions could lead to the centralization
of power. And those are the areas in which we need to kind of focus our efforts.
But at the same time,
like what's been pointed out,
I think is super valid
where it's like no one,
the average everyday person
does not care about this stuff.
I think that probably the people in this space do,
but like 99% of the world does not give a shit.
And so what we actually really need to do
is just offer better
more frictionless experiences for end users and if we can use crypto to to do that then like that's how we're going to win and we're not going to win by like preaching this philosophical gospel of
like the world's going to enter dystopia if we if we don't do these things like that will that will
never work but that's exactly the thing i mean i think all of us
mentioned now users don't care about decentralization and it shouldn't be the end
goal it's not a feature you want decentralization for things like resilience security for example
censorship resistance not that trivial to have in a centralized system uh But what you mentioned is privacy. So again, there's always a price for
privacy. So it's very difficult to get the same value with privacy compared to without privacy.
So how are users ever going to be convinced to use something non-centralized? Because OpenAIR
has already done very smart with this like cross
chain cross chat memory pitch that i think is just a hoax i don't feel a better ux or
results from just by sharing more infos i think it was just a ploy uh yeah hashtag conspiracy
theory to get more users to comfortably share data because you have the feeling like hey the more i
share the better it's going to be i don't think it's the case but still like in your example of a sushi place
recommendation or general restaurant recommendation you want the best restaurant so i think most users
would be willing to share this with open ai to share as many points as necessary as long as they
get the best result until the point where open ai gets hacked and then
suddenly there's this massive leak of exposed prompts etc but until then i don't know how any
decentralized setup can provide the same quality result yeah it's a really it's a really good
question um the way that we're approaching it is, I think one reason why,
and maybe this is happening behind the scenes and I'm just ignorant,
but I think one reason kind of the shared context across chats
and OpenAI doesn't maybe feel so good is because it's like,
the context windows are like pretty small right now.
So you can't just like feed it
you can't feed these these agents or these models you know infinite context they they got a limit
you can you can train it on infinite data but like can't give it infinite context so i think
you know if if you're taking the raw slop that you're entering into open AI and trying to use that as context.
I don't think it's so good. And so what we do is we have, this is a little bit technical,
so I'll try to keep it short. But basically, we have like, under the hood in our knowledge graph,
like everything is like, very structured, and symbolic and semantic. And so that's kind of
like where I think the original kind of AI efforts were focused was on like kind of symbolic AI.
And then we kind of moved into machine learning or deep learning land.
And I think we're going to have a reversion back to the mean where we're going to kind of have like a neurosymbolic AI,
where if you kind of combine the best of like very structured semantic explicit
data where you can get, you know, deterministic results with kind of the non-deterministic
aspects of like deep learning and things like that, then I think you can get like a much
higher percent accuracy.
So all that is to say, basically, I think what we can do is we can condense down the slop into kind of very structured, simple, semantic statements about things.
And with that, we can kind of like we're already doing it.
You can already get like very, very personalized response responses with what we've built.
And, you know, it's still the super early stages and you get better responses than you can get with open AI because we just kind of like condense and consolidate the data into these semantic triples.
So I think that we can, there's also this, to your point, it's like, okay, like step one is consolidate the data.
Then step two is like get the data and maintain privacy.
get the data and maintain privacy. And the big issue is we have the infrastructure for
having self-sovereign private data. That's kind of the field that I've been working in for the
past 10 years at the intersection of crypto. So we have the ability for people to own their data
and use it for different things. It's really hard to your point, though, where it's like the more private the data, the lower the utility. And I actually think that it's not really crypto, but it's like
cryptography and the intersection of like our industry with that world where it's like,
there are these mechanisms that are coming into play, wherein you don't lose the utility
while you maintain the privacy, or you always lose some utility, or you lose some kind of,
you know, efficiency, but you can still you can have self
sovereign private data that can actually be used. And you can
like derive value from it beyond just, hey, let me pass you this
credential. So you authenticate me into the system. And where the
it was kind of a
weave, but where the, the intersection of the both of the two come into play is the issue with
the verifiable data set right now is it's not traversable. So think about, you know, you have
all of this private data, maybe hosted in your personal identity hub, which might be your phone. It's like, how does an external agent,
without seeing everything,
effectively traverse this set of private data?
And maybe it's not even just yours that it needs to traverse.
Maybe it's like, it wants to scan the whole ecosystem
of verifiable data that is all stored on private devices. Like how can it do that? And the answer is we need like a discovery protocol for self sovereign private verifiable data. And that's, that's kind of started keep ranting about intuition. But like, that's kind of what we built is like this big map that just like points to data, regardless of where it lives so that you can find it and you can request it when you need it.
So, you know, now the agent comes to you and it's like, oh, like, I want to know about your, you know, food preferences.
It doesn't need to comb through all the slop.
It doesn't need to, you know, do some crazy, like, I don't even know how you would access it in traditional kind of verifiable claim world.
But like now it can request the explicit data that it needs to fulfill your request in like a very meaningful way.
And so like the discoverability of data, I think, is also super important to kind of increasing the utility of private data, which I think is a huge goal.
Okay, let me play once more devil advocate, and then we can move on.
Even if you have full control over your own data, you have never shared it anywhere, you
have a complete profile about you, there is some reputation and credentials attached to
How do you share this with the compute provider with the actual agent because the agent
is not autonomous itself like it's still running on some gpus someone is executing it someone wrote
the code for it so i mean this is my usual pitch like currently i only see three options it's
confidential compute which is the biggest mind to me because nvidia has had confidential
compute on h100 200 even on blackwells for a while now like i think two years so it would
honestly just be a matter of open ai activating in a sense um confidential compute to be private
and actually fully private and it is verifiable because it's a TE. So you get remote at the stations,
you could verify this, et cetera.
Second option is something like a mixer.
I think that's what Venice AI tried or is doing
where you route multiple user requests
through one API, for example, to OpenAI.
So they don't have a complete profile about you,
but they still collect all of the data.
The third one,
and I think that might actually be the biggest driver,
even though it's a bit technical,
but I think that's a UX thing, is local hosting.
So if NVIDIA or anyone manages to provide AI-grade hardware
that we can run at home that's like less than 1K
compared to now I would have to buy like four Mac Mini M4s
for me to have like a home server
where I can run all of the models I think that would be a big game changer because then suddenly
everyone can be just going into an agent marketplace fetching the ones they want and they just execute
them locally so they don't have to think about any of the privacy things and decentralization things
that we're discussing because everything is happening
on your device best case of course still it should be open source and the code should be audited but
it solves lots of the problems i think that's like the holy grail of where folks are building towards
i'll preface this by saying i'm i'm not technical and i'm not an academic in the world of large language models,
but I did set out to better understand ZKs at ECC this year
and had some really great conversations with a few builders who are experts in that realm.
And the high level, as I understood it, was this pursuit of these ZK data cleanrooms.
Essentially, you get a certain number of interested parties in a sort of TEE of sorts.
It might be multiple agents plus some data or something like that.
And then you use ZK either, and I don't know the implementation details of what he was explaining, but you use ZK to essentially allow for the ensembling of these parties, the inference, the output, and then have this programmatic assurance that after that output is delivered, the inputs don't ever leave
with the parties who are privy to it.
It also sounds like that technology
is relatively in its infancy,
but is working on much simpler use cases.
So I think it's a holy grail problem
that will inevitably be solved.
I think a more interesting question is, will people care? Because we're decades into personal data not being very personal,
and the abuses and mismanagement of personal data are abundant.
And no one really seems to care, quite frankly, right?
Like pull up any password manager and like look how many of your average person's passwords
are compromised and leaked and they don't care, right?
And as long as that utility continues to be good
and they get recommended the products
that they actually want to buy,
it kind of is this like,
I don't know if you can call it an opiate of the masses, but like, I guess the individual perceives
the utility from the lack of privacy as greater than the lack of privacy to the point where I've
even seen some folks, I've younger nieces and nephews, and I've even seen arguments coming from them that the expectation of privacy is an irrational expectation, and that privacy is really only a more modern concept. necessarily ascribe to that explanation but i definitely do wonder like or rather do see a
potential outcome where maybe privacy isn't valued uh at scale i don't agree with that but um i think
it's it's worth at least bringing up the the counter i think they um the technology is almost
there i mean we trialed out um Secret Networks, which is another project.
We've tried agents running in the TEs.
That works fine in terms of bringing context.
I mean, actually, agents really help in the core principle of decentralization
because all these agents could be run by different entities.
And when you exchange services,
exchange data information,
it's based on completely under your control.
So nobody else controls it.
So in a way, actually,
agentic systems enable decentralization
if dealt with correctly.
But I think the point you just made was quite valid, which is, do people care?
Well, I mean, you're absolutely right.
I don't think people do care because, I mean, actually, if you look at our social media,
people forget about privacy.
They want to share more and more.
So if I look at the next generation, they don't like privacy.
They like to share as much as they can.
They share their food.
They share where they're going on holidays.
They share everything.
So in that kind of social media environment,
I don't feel anybody apart from the older generation cares about their privacy.
That's not to say, again, that I don't agree with it,
but it just shows you the trend.
And also I'm not saying that it won't change because something might change it.
But the technology is actually not that far.
It's available.
We've tested it.
We've trialed it.
But to the point, nobody cares.
So privacy is a lot less efficient than a centralized kind of process. So in that case,
as I said, it will come if there is a need and if you can provide the utility without
without compromise, then I think there might be a chance.
Yeah, I think there's two things following up. And I think first and foremost, we only see a subsect of society on online, like mapping
what we see on the internet to real life is a fallacy, but it's not one to one.
So there are a lot of people that still do care about privacy,
and we don't see them talking about it because they're not online as well
because they are private, which is funny.
But I do understand that there is a trend of people sharing more
and being okay with sharing more.
I guess everyone in crypto, to an extent, more or less cares
but doesn't care about privacy because
all of our financial information is on chain as well. And that isn't going to change for
a long period of time. And just as a devil's advocate or a little counter or maybe additional context like I do think that agents are going to be
helpful for increasing the amount of privacy that we do have but there's two areas that
I'm interested in getting or there being more information about before that happens.
First is what type of agent that people are dealing with because
if you're dealing with an agent that is managed by a team that has access to all the private keys of the data and the whole data pipeline they have access to then that agent even if it is running in a TEE where there's an algorithm that processes that data in a way that means that that agent doesn't need to have access to that private information, that information is still going to be accessible to that team until we get more autonomous agents, which don't have human control over the data pipelines,
that there's always gonna be that balance
between perceived privacy and confidentiality
and then accessibility as well.
And I guess that's also a question about
trusting autonomous agents to process
that information as well and
this is a conversation that's going to be developing further and further as we
get a better understanding of how performant agents are as well most of
the agents that we have right now that are processing either our money or
processing our data are more deterministic, right? They have really strong
guardrails and those guardrails mean that there's always a human in a loop. And the question is
whether it will ever have autonomous agents which are able to manage our personal data without
having access to them or having a backdoor to them either.
Because right now it's just a spectrum, even like the most private preserving or seemingly
privacy preserving social media apps, like the likes of Telegram, they have backdoors in them as
well. And that is the direction that we're going with agents, even if they have TEEs as well.
Yeah, for sure.
I mean, that's why I said it should definitely always be open source.
It should be code that's audited.
And there's so many things that go into it.
It's not just like running something in a TE and suddenly you can trust everything.
The TE just kind of verifies the execution and makes sure that what is coming as an output is
coming from the agent but still the code is the thing that rules it and the privacy discussion
always makes me so sad because i don't think that people don't care it's just that they don't
realize the extent of it it's the same as like 10 years ago with google and facebook of course we
care but we didn't know what's actually happening with our data we didn't know that facebook has a shadow profile about my mom even though she's not on facebook that they're
using all of the data to manipulate me with algorithms to get as much attention from me as
possible and similarly here i mean i think now it's way worse with the world of big ai because
we're sharing very sensitive data with them and And again, it can be very easily used to manipulate us
because the profiles they have about us
are way more complete than before.
And as I mentioned, like if there would be a chat GPT hack,
I think people would suddenly care
because everyone would be able to see what they prompted.
I mean, assuming that they used
publicly recognizable email like i actually
do i would be in lots of problems with my girlfriend if chat gpt chats were leaked suddenly
now i do discuss still like lots of private stuff and i would love to use a privacy preserving
solution but i think as we discussed now quite a bit there is always a price for privacy and
currently it's a bit too high i want to do a bit more of a
lightweight question because i am really curious where we're currently at so what kind of agents
do you guys actually use uh right now i mainly use defy agents i I also use the more operational agents.
So I use Giza, I use Fungi as well, and also Mamu.
So I focus on like stablecoin yield bearing agents
at the moment, but willing to move along the risk cycle,
um, risk spectrum later on as well.
Um, and then I also obviously use like your chat GPT's your deep seeks.
And, um, I'm starting to explore the OLAFs agents as well now, just, uh, to see what,
what's being built.
So yeah, those are, those those are main ones that i use i'm just playing around with them at the moment and some of the agents have been
printing for me as well shout out like fungi and also geezer uh my kind of agent usage is slightly different.
I mean, we have on Fetch Agentverse around 2.7 million agents.
So any query that goes into a Psy1, our own LLM, uses multiple agents to actually give whatever is needed.
So it's not about naming those agents because they're not separate projects.
They're all just agents which live in there and come together.
When you kind of ask for a query, it kind of comes together to deliver the task.
So, yeah, I generally use, every query uses three or four agents for me.
Every query uses three or four agents for me.
Marco, I would love to talk about the agents I'm using,
but unfortunately I'm going to have to hop out of the space here for another call.
But before I did, I wanted to thank you for having me on.
And the conversation was amazing.
And Huma Yin, I would love to have a conversation offline.
And same to you, Nathan, and find ways to keep building together.
And for the audience, you are in great hands with Billy still on stage.
So thank you again, Marco.
Thank you for coming.
I think we will wrap it up in a second.
I just want to hear Billy's agents, and I can share mine and then we move on.
Yeah, I kind of, my agentic experience, I guess it's kind of both perspectives.
Like one, I kind of have my own agents for kind of like my own personal tasks. And that concluded everything from, you know, just general,
it's like got a CEO agent and a CMO agent and a legal agent and a coding agent.
But, and then also from like the builder's perspective,
kind of the direction that we're building in is the vision described of just
the agents being behind the scenes and the request
is getting routed to the right agents for the job. So from like a development perspective,
that's kind of the path that we're going down, which is the agents are, you don't even potentially
know which ones you're interacting with. It just kind of, they're able to fulfill specific requests
better than a single agent or model ever could
better than a single agent or model ever could.
the personal agent ceo cmo whatever have you built those yourself
yeah i mean i'm not like building models myself but i mean for for the most part it's just kind of
like the the the context that you provide those things and the the system prompts and things like
that but still
they're running locally and you use something like the agent SDK from open AI or how can I imagine it
yeah exactly um it's not like like I don't I I've been meaning to build like a good AI machine
because you know like can't get access to like the the bigger models and it's relatively slow but
it's pretty cool like it's just it's just kind of crazy to have like little intelligences on
on your device locally
nice okay uh so actually before we wrap it up so mine i do use giza for yield optimization i tried
mamu i tried it in the base app but I really didn't like the experience.
I'm definitely more of a desktop guy.
And I do build lots of my own agents.
So for example, over the weekend, I was really curious about VideoGen and where we're currently
So I built like this multi-agent system where I had this content strategist.
I had the video producer agent calling all the relevant apis and evaluator agent
then checking the results but i mean agents is a big word for those because i mean some use llm
apis but most of them were just like python scripts or one is a telegram bot so i don't know
and what i just realized during this discussion was i didn't even think about looking at agent markets or shops to see if someone has already built it.
Because it would take me more time to research the team behind it.
Can I trust it?
How do I actually integrate it with other agents than me just building this myself with Cursor?
So definitely one of my to-dos after this session to check it out
a bit more i think you should also list this on agentverse so other people can actually use it
because if you already built it you can attach the trust and you can actually make it discoverable
for llms quite easily within within like three of code. You can list it and other people can actually
discover it and interact with it. So try out Agentverse on our fetch site and you'll find it
quite easy because the whole objective here is to create that marketplace for agents. So if you can't
find it, you can find it over there. All you need to do is go go to llm and ask the question and it'll list all the
agents that can do the task for you nice i'll even host it on the oasis te cloud with decentralized
key management etc so that people don't need to trust anyone executing the agent but it happens
from the te app yeah so you if you just list it on agentverse.ai,
and it will become discoverable on the fetch LLM,
and there's plenty of people
starting to use agents through there.
Very cool, thanks.
I will do that.
Any closing thoughts, Nathan, Billy, Humayun,
and then we close it up here.
Oh, Axel wants to say something from the Swarm account.
Okay.
Just a thank you for the invite.
It was by super knowledgeable people here
building really cool stuff.
And if you're excited about building agents
that actually work,
DappRadar also just built something called Hivemind where we utilize our data lake of DAP data to create a really, really nice copilot for DAPs.
Basically like a FAQ, super powered FAQ for any single DAP information with rolling out with pixels.
And so if you wanna understand pixels, the game,
we aggregate data from YouTube, the Discord, Telegram,
and it works really, really smoothly.
So definitely check it out.
It's worth seeing that there are real agent builders out here,
not just tokens but building real um agentic workflows
and um once again thank you looking forward to the next one marco and cheers billy ma and obviously
um oris um was really great chilling with y'all
yeah thank you guys so much for having us.
And I guess like my closing statement is
it's still the beginning.
The wave is coming fast.
I think that, you know, for everyone in the audience,
like it is not too late to kind of get your hands dirty
and start just experimenting and doing crazy stuff.
And then I think kind of the superpower of
our ecosystem is really the composability and the coordination. And for any one of us to go
up against kind of the tech giants of the world is going to be relatively difficult, but together
we might actually be able to make a big difference here. So yeah, we'd love to stay in touch. We'd
love to figure out how our individual pieces of the puzzle come together to to compose the the ultimate
puzzle that's actually gonna be able to change the world here it's definitely worth the fight
and we can't stop uh thanks a lot everyone uh for all the listeners please follow the speakers learn more via
their twitter feeds
thanks a lot again we will
close it here and see you in
two weeks again
