Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
The Role of AI and AI Agents in Blockchain
Recorded: March 27, 2025 Duration: 0:53:51Space Recording
Short Summary
Exibits is launching a groundbreaking platform that transforms GPU power into enterprise-grade AI compute, enabling faster access for developers and researchers. The initiative includes a token launch, creating a self-sustaining ecosystem for AI compute, while emphasizing the importance of owning AI infrastructure in the evolving landscape.
Full Transcription
Thank you. I'm going to go ahead and do it. Thank you. Oh Thank you. Hello everyone! I'm going to go to the next video. I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video.
I'm going to go to the next video. I'm going to go to the next video. Thank you. At Exibits, we understand that AI is changing everything.
From self-driving cars to medical breakthroughs, the future is built on artificial intelligence.
But AI runs on compute power.
And right now, there's a problem.
There isn't enough of it.
GPUs are the engines of AI.
But on their own, they're just raw power,
unrefined, fragmented, and inefficiently allocated.
It's like crude oil before it's turned into fuel.
There's potential, but it's not usable yet.
Right now, AI companies, researchers, and developers
struggle to access compute power. Traditional cloud providers are centralized, researchers, and developers struggle to access compute power.
Traditional cloud providers are centralized, expensive, and exclusive.
Meanwhile, decentralized GPU marketplaces are unreliable and inefficient.
This bottleneck is slowing down AI innovation.
At a time when AI needs to move faster than ever, that's where Exibits comes in.
Exibits acts as a refinery for AI compute,
taking raw GPU power and transforming it into enterprise-grade AI-ready compute clusters that are optimized, scalable, and ready for deployment. Think of Exibits as the base layer of AI compute.
We don't just provide GPU time. We process, optimize, and distribute
high-performance AI compute to power the entire industry. This means AI companies, researchers,
and developers get faster, more efficient access to compute without dealing with fragmented supply
chains or unreliable resources. But XBits isn't just about using compute, it's about owning it. With XBit,
compute becomes an asset, not just a service. Holding XBit gives you a stake in the AI compute
economy, earning EXA rewards as demand grows. By staking XBit, participants earn EXA. EXA
tokens is used for incentives, governance, and liquidity. This creates a self-sustaining ecosystem where GPU providers, AI developers, and investors all benefit from AI compute growth.
Exibits isn't just another GPU marketplace.
We are the foundation that powers them.
AI research, cloud computing, or Web3 applications, Exibits provides the optimized enterprise-grade
compute power that industries need to scale. That's why GPU marketplaces and AI enterprises
rely on Exibits to supply their compute needs. We are positioned at the base layer of this industry.
That means as AI demand grows, Exibits grows with it.
The future of AI isn't just about using compute.
It's about owning the infrastructure that powers it.
Exibits is financializing AI compute,
turning GPU power into an investable, yield-generating asset.
Whether you're a developer, investor, or enterprise,
Exibits gives you a way to participate in the AI revolution.
Not just as a user, but as an owner of the most valuable resource in the digital world, compute power.
Join Exibits today and be part of the next evolution of AI compute. Hello, hello, good morning, good afternoon, good evening, everyone, wherever you are.
So thank you again for joining us for this amazing AMA.
My name is Miel, I'm Trivially in our community. And as always, this is going to be in a question and answer format.
So we have pre-prepared questions for everyone.
And we highly encourage all of you to engage as well
by sending in your questions
through our Telegram community, Discord community,
through Binance chat for those watching via Binance Live
or through our chat here on our Xpaces Live.
Without further ado, let me turn you over to our host, Mark.
Mark, you're probably muted.
Mark is not a speaker yet. He's still listed as just a listener.
he's still listed as just a listener oh okay hold on
Oh, okay. Hold on.
my bad i thought i i already sent the speaker
there you go the speaker on this one there we go all right welcome to today's ama session where
we delve into the transformative intersection of AI, AI agents, and blockchain technology.
My name is Mark Fidelman. Thank you, Miel, for introducing me.
CMO at Exibits, and I'm going to be your host today.
At Exibits, we're pioneering the financialization of AI compute by tokenizing enterprise-grade GPUs,
enabling individuals and businesses to become stakeholders in high-performance
AI infrastructure.
This approach empowers participants to contribute to AI's growth while earning substantial
rewards from compute resources they help build.
Go to our website to find out more.
But today, joining us are two distinguished guests, Andrew Smith, founder of Formation
Cloud, and Russ Ventrano, Arch Personal Chief Sales Officer at Exibus.
Andrew leads Formation Cloud,
a company at the forefront of fog computing,
combining the scale of cloud computing with
the responsiveness of edge computing to support
autonomous applications.
Russ, of course, brings extensive experience in AI sales and
GPU computing, driving Exibus-Bits' mission to provide
affordable and reliable Cloud GPU compute solutions.
Today, we're going to talk about AI and AI agents,
one of my favorite subjects and how
they're revolutionizing the blockchain landscape.
To kick things off, we are going to get started.
But first, let's have
Russ then Andrew introduce themselves.
Thanks, Mark.
Russ Petrano, I am our Chief Sales Officer.
I've been with Exhibit for almost a year.
Prior to that, I spent about 15 years working with and for AI companies, including some
including some time at Microsoft Nuance Communications.
time at Microsoft Nuance Communications.
And during that time, it became clear to me that GPU compute was going to fuel the next evolution of AI.
So I was excited to join Exibits and become an enabler by providing high-quality, affordable compute
so that companies of all sizes, not just those the biggest in
the world, but have an equal opportunity to participate in this evolution.
So happy to be here.
Andrew.
I am Andrew Smith.
I'm the founder of Formation.
As Mark mentioned, Formation is really the first public fog compute platform. The big distinction between fog and cloud
and fog and edge is kind of what it sounds like, right?
Fog sits in the middle.
It's like cloud on the edge, right?
So you're closer to the end user.
You're closer to the devices that may need inference.
And really, we built this from the ground up
with the focus on autonomous services and devices.
So our platform is completely built to optimize agentic workflows and to optimize what I believe is coming very, very near in the future, autonomous devices, which I think is oftentimes a little bit overlooked in the agentic world. We have a lot of focus on kind of the software layer,
but we shouldn't forget that robots are coming.
Some of them are already here.
They're more and more going to come.
And when robots are everywhere, they need to be able to, you know,
run inference against cloud scale, but do so with really, really low latency.
If you're going to rely on robots and
autonomous devices to make decisions, you can't wait a couple hundred milliseconds for them to
make a decision. They may make the wrong decision or they may freeze. It's sort of anecdotal,
but I've seen this happen with some of these delivery robots. I live in Miami in the city,
and these delivery robots are very
common here. Oftentimes, yeah, they're everywhere. And oftentimes, you'll just see them stuck at a
intersection. And what's happening there is that the latency is too high for them to make a decision.
By the time they make that decision, information is changed. They're sending off a signal to,
you know, some cloud provider that's in Ohio and it takes 300,
400, 500 milliseconds to get back.
And the light changed or cars started moving or people are walking.
So you end up with this problem where these robots can't make a decision.
That's really kind of where fog comes in and sits in the middle between cloud and edge.
It allows devices and services that need really fast response times,
but need the scale of a cloud data center in order to make those decisions.
And as we move towards more and more reasoning models,
where they really do need that powerful GPU compute,
we're going to need something like what Formation offers,
where you have lots of compute close to
the edge, but in a cloud-scale data center. So that's what we're building. And we built this
entire platform from the ground up with that particular use case in mind. We believe that
this is going to be a trillion-dollar-plus opportunity over the next decade. The entire
world is moving towards AI and agentic workflows and autonomous devices. Yeah, I posted
this little meme here in the space. There's no turning back. It's here. We're in the autonomous
age. We're in the very early innings of the autonomous age. This is the fifth industrial
revolution. There's no better time to get involved. And I think the work that Exhibits
is doing is extremely important. Mark and Russ heard me talk a little bit at EAP Denver
about what I think the real moats in this space are. And I think we'll get into some of that here
as we talk about AI and AI agents. Yeah. I mean, you've hit the nail on the head and it just
blows me away that so many people are not really paying attention
to this i mean they see chat gpt they see other llms but they're not really understanding how
all of this is going to change especially when we get into agi and then when you add them to
robots and devices i didn't even think devices i knew about robots but yeah you can have autonomous
devices now it's going to be crazy.
And I've said this, you know, within three to five years, 80% of us will lose our current jobs. I think they'll be replaced with better jobs, more strategic and more value add to whatever business you have or humanity.
But we're going to be sounding the alarm until it happens and maybe more and more people start to pay attention.
So let me start with my first question, Andrew.
How does Formation Cloud leverage AI within its Fog Computing framework to enhance blockchain
applications?
So our focus is really more on Web2 applications.
The way that we use blockchain is for coordinating incentives, for trustlessness, for verification, verifiability.
One of the main focuses that we have is auto redundancy in these applications. And this
becomes really, really important as we rely on agents more and more in our life. If you have an
outage and your aid agent is only hosted in a single region because it's a pain in the butt
to back it up to multiple regions.
You need DevOps team.
You need, you know, you got to set up cron jobs.
You got to make sure that those cron jobs
actually execute on time.
Like the modern DevOps is built for humans.
It's built for human intervention,
stepping in, taking over and taking control of that process
and making sure that it works properly.
So our entire system automates that process. We kind of get rid of all the DevOps headaches that
you're going to run into. And as a result of that, we guarantee SLA, obviously, not necessarily that
we can physically guarantee it, but we do ensure against any outages. So our SLA guarantees 100% uptime and any damage that's caused to our customers
is automatically paid out via the blockchain. So that's really kind of where blockchain comes
into our protocol. The way that we use AI and agents within our workflow is we enable agents
to actually manage the workloads. And so you can imagine a scenario where you have an agent that has a team
of agents underneath that before you would need a human to actually manage that, that entire
process on our platform. We actually enable the, the agents to manage that directly. So we have
an MCP server that enables agents to actually work within our protocol, deploy additional agents if they need them, scale them back if they don't need them, manage the entire lifecycle of instances of agents underneath them.
And so that's a really powerful tool because now you can actually build truly autonomous systems that don't rely on human intervention, where you get auto-redundance.
You automatically get a high availability cluster out of the box at no extra cost. And as a result of that, you can actually rely on these systems. If you didn't have
that and you ran into an outage, well, you're right back to requiring human intervention.
You need a human there to go and redeploy things, move it to a different location,
move it to a different region. With our system, because you have that automatic failover,
region with our system because you have that automatic failover in the event of an of an
outage the agents can still continue to operate there's still a backup of that of that agent that
can pick up where the uh one that suffered an outage left off and can continue to work this
is extremely important if you think about something like uh you know an agent that controls the power
grid right uh which we're going there.
It's not whether or not this will happen, it's when.
Yeah, this wins.
And so, you know, if you have an agent that controls the power grid and that agent is offline,
well, what happens if it needs to throttle power?
And as a result of not being able to throttle power
because it's offline, you end up with a blackout.
Like there's a lot of things that we just don't think about in between here and when agents come online. We just think about it as like, hey,
we'll deploy the agents and they'll do their job. Well, you got to have infrastructure that's set
up for the agentic era. And right now our entire infrastructure is set up for human intervention.
And so when we were building formation, one of the main goals we had
was to design this from the ground up for what I commonly refer to as the autonomous age.
Got it. Wow. Okay. Well, you put a lot of thought into this. I want to move to Russ real quick. And
Russ, can you elaborate on how Exibit's GPU services support that computational demand of AI-driven projects,
and specifically when we get into, you know, eventually when we get into AGI and powering
those robots, and also what Andrew was talking about with the AI agents, how do we kind of talk,
how do you summarize how Exibits does that, and what are we doing to make sure it runs as seamlessly as possible?
Yeah, sure, Mark.
I mean, that computational demand, it needs GPUs.
That's the lifeblood of how we power them.
And Exibits does this by, we work with data center partners around the world to provide
high-quality, low-cost, distributed computing.
So today, we're working with about 30 data centers around the world to bring compute to the edge.
And we're continually adding more providers so that we can offer things, you know, better pricing, better terms, more flexible than the hyperscalers.
Still the same GPUs.
We call ourselves the base layer compute
because we don't rent from community.
And Andrew and I and Mark had a great conversation
in Denver how there's a lot of noise out there
renting from people's laptops, crypto miners,
A100 in their basement, And that's not very reliable.
So we're at the base layer.
We source all of our GPUs from data centers.
And we physically operate, manage, and maintain every GPU that we rent to our customers.
And, you know, we're leading edge. We just rolled out thousands of H200s and we're currently testing and benchmarking the Blackwell chip, the RTX 5090s.
And we're leading edge.
We just rolled out thousands of H200s.
So we're going to be rolling out, you know, the flagship GPUs of NVIDIA just differently than hyperscalers do.
I think if I could jump in here, why this is so important. So I think there's two pieces of this.
One is the financial piece.
Enabling anybody and everybody to own this base layer compute is so vitally important.
Because right now you see it.
If you follow this space on Twitter, you see the most powerful people in the AI world going
and asking for government privileges, going and asking for government to protect
them against competition.
And the worst thing that could possibly happen to humanity is that we end up in a world where
only open AI, where only X and Grok, where only these few handful of organizations have
enough compute to run these models.
That is the worst possible outcome. That
is dystopia. That is, they become our oligarchs. They control everything. All the money flows to
them. We have to enable individuals and smaller organizations to be able to compete in this space
and own some of the base layer compute that powers these systems. If we don't do that,
then we're going to be at the mercy of these monopolists. So this is why- You're right. I mean, I had mentioned earlier, that's why I
specifically came to Exhibit because I felt like I wanted to give an equal opportunity to all
companies to participate in it. So 100% agree with what you're saying. And then the other piece of
this is having distribution. Having global distribution is so
important for the same reason that you want a blockchain to have lots of nodes around the globe.
You want to protect against nation state attacks. Not every country on earth is going to have a AI
friendly posture and those postures are going to change over time. And so you need to have access
to this compute in different places so that we
can protect against some of the potential harms that may come this industry's way as it becomes
more and more important. And so having the combination of global distribution and enabling
anybody and everybody to actually own the physical hardware that fuels AI is extremely important.
AI is extremely important. This is like, you know, GPUs are the oil of this era. And if you don't
have access to GPUs, you can't run your factory. I was at GTC last week. I think some of the people
from your team were there as well. You know, and the way that Jensen presents NVIDIA is that it's an AI factory, right?
That it's this factory that enables you to produce AI a lot faster and a lot better and, you know, in sort of an assembly line manner where you have all these open source tools that you can combine together in your assembly line and you can pump out AI on the other end. Well, all of that's great, but if you don't have the fuel to then power those models, if you don't have the GPUs,
then they're useless. And you can imagine a world where you have these powerful AI agents,
but you just simply don't have enough compute underlying them for them to actually do their
job and to work. And what good is a really powerful AI agent if it only works three or
four hours a day? What makes it to where we're going to have this much more efficient and abundant
world is where those agents can work 24-7, where humans can't, where they can do more than humans
can and they can do it around the clock. And if we don't have enough compute to power that,
then it's all for naught. So having companies like Exibits that enable anybody and everybody to finance these GPUs,
it allows you guys to scale in a way that the hyperscalers never will be able to, because
they have to make these investments off a balance sheet.
They have to either go to private equity and raise capital, go to the stock market and
raise capital, issue bonds.
That takes six, nine months, maybe longer.
They have to register with the SEC.
They have to do all these things for their securities offerings just to go build another
data center and acquire more GPUs, or they have to have the cash on balance sheet to
do so.
Well, with a project like Exibits, now anybody and everybody can use crypto to form that
capital to acquire the GPUs that are necessary to power the autonomous age is vitally
important. I cannot like the big bottleneck, the big bottle. It's not software. You see it every
single day. Some new agent gets released. Some new model gets released. It keeps getting better
and better and faster and faster. And a lot of this stuff is open source. The big bottlenecks
are the access to the compute and then the energy that powers
that compute. And if we don't solve those problems and make the ownership of those resources
distributed, then we will be at the mercy of monopolists, which is something that you don't
want to, we don't want to go back to the feudal era. Yeah. I mean, Andrew, you've been echoing
exactly what I've been saying. Anyone following us from AMA, I say exactly the same thing. Couldn't agree with you more.
And I'm wondering if you could comment on that about how we address security within these AI agents, especially when there's going to be an interoperability layer that allows AI agents to talk to other AI agents.
Yeah, talk to and pay other AI agents, right?
So I think this is one of the roles of blockchain.
You need some form of verifiability.
And there's nowhere better than a blockchain to prove the computation.
You guys just rolled out a bunch of H200s.
H200s have TEs.
This allows you to actually verify the computation, allows you to verify that the model that you're inferring against is the model that you're expecting it to.
that you're inferring against is the model that you're expecting it to. This enables us to,
you know, load up programs, agentic programs onto computers, onto servers, and for those agents to,
for their computations to be verifiable. Well, where do you put that proof? Where do you put
the TEE's proof? If you just put it in a database somewhere, no one can access it. No one can actually audit it. There's no way to actually verify it. So you need some form of verifiability
and this is going to become more and more obvious and more and more important as we go down this
path. That is one of the roles that I think blockchain serves. So verifiability is a big one. I do think that in many cases, you can solve a lot of the security concerns around AI agents
by separating the data layer from the compute layer.
Um, you know, a lot of people will look at like, uh, you know, you saw this going around
when DeepSeek was released as like, oh, but we're sending all our data to China.
data to China. Not if you run it locally, not if you run it on your own machine. The issue is if
Not if you run it locally, not if you run it on your own machine.
you're using DeepSeek's app, where they're actually storing your data. They'll say they're
storing it for context to be able to reuse later. Same thing with OpenAI, same thing with Twitter,
et cetera. I think Twitter just, X just gave us $150 in credits if you're willing to share your data
with them.
So, you know, these companies are going to come up with ways to like convince you to
give up your data.
And I think we have to be cautious of that.
And so separating the data from the compute is one way that we can really kind of, you
know, nip security issues in the bud, so to speak.
We want data to remain mostly local or
under the control of the owner of the data. And we want the compute to be as open source as possible.
And so the model should be open source, the data should be closed source. And if you're just using
the models, then you don't really have too much to to worry about um there and i think there's two
different concerns here right there's like rogue agents what happens when an agent goes rogue
versus security concerns and and and data privacy the data privacy part of it is largely a solved
problem i think the rogue agent piece is one of those things that we're going to kind of have to run into it before we come up with solutions.
I'm not too concerned at this stage.
I, you know, I say this sort of tongue in cheek, but there's some truth to it.
us how to build like self-replicating nanobots that could go and mine the silicon and like build
the semiconductors and build the the lithography machines and and so that they can build their own
compute i'm not too concerned about rogue agents pour some water on the server unplug it like you
know there are going to be solutions to this that humans will still have a lot of control
now when agents start teaching us how to build self
replicating nanobots, then, you know, my, my alarm bells will start ringing.
Well, we're not that far away from it, from what I'm seeing out there.
Maybe not, maybe not, but we're not there quite yet.
No, we're not, we're not, but these things happen pretty quickly as we've seen how fast
people are innovating in this space. Let me move to my next question and this one is for
russ and for andrew and it comes from one of our community members named smith and he or she asks
what potential benefits can ai agents bring to blockchain technology
technology? Yeah, so the AI agents can bring a ton of things to blockchain technology.
So some of the things they can do, you know, they can do things like automated training bots,
network optimization. They can help with fraud detection and security,
optimization. They can help with fraud detection and security, you know, data verification,
verifiability. There's a ton of things they can be used to do. Smart contracts, dispute resolution,
predictive execution. So there's a lot, you know, there's so many ways that I see how
the AI agents can help in this space. We wouldn't have enough time to discuss all the things here
today. Yeah, I mean, it's literally anything. And when you talk about smart contracts,
imagine when you're throwing AI in with these smart contracts, so they become a lot smarter,
a lot more adaptive, and talk about cross-chain trading and cross-chain communication
uh it's i mean if you can think it you could probably do it with an ai agent what do you
think everything smarter faster more secure more reliable so uh you know it's it's multi-factor
how they're going to improve uh everything that we're doing. Yeah, I agree with all of that.
I think one of the obvious ones is making it a lot easier to onboard people with a better user experience.
I am very crypto native.
I bought my first Bitcoin in 2013.
I've been using blockchains for a really long time, been using wallets for a really long time.
for a really long time, I still hold my breath every time I send a transaction.
I still hold my breath every time I send a transaction.
And so, you know, increasing the confidence, delivering a better user experience, you know,
about a year and a half ago before kind of the agentic wave, one of the things that a
lot of people were talking about were like solver networks and intents.
And I think that AI agents can really deliver on the promise that solver networks and intents. And I think that AI agents can really deliver on the
promise that solver networks and intents kind of just kind of fell short on, which is, hey,
I want to do these four things, orchestrate a transaction for me and or orchestrate multiple
transactions for me to accomplish these goals. You know, I want to purchase some soul. I want to stake liquid stake that soul.
I then want to find a, a, you know, LP pool that I can LP into. That's going to earn me an additional
300 basis points or better. And then I want to loop that. Okay. Well, if I wanted to do that on
my own, that's like seven, eight, nine steps, different transactions all along the way,
clicking links, holding my breath, hoping that I didn't get click on a link that was, nine steps, different transactions all along the way, clicking links, holding my
breath, hoping that I didn't get click on a link that was, you know, an injection from some,
you know, North Korean hacker. You know, the entire process of, you know, executing DeFi
transactions is very cumbersome. It's very intimidating. It's very difficult for somebody
to become an expert at and to onboard.
And so AI agents can solve a lot of that for us.
And I think that that's probably the lowest hanging fruit and the most obvious.
You definitely have some teams out there working on it.
I wouldn't be surprised if you had entire blockchains that came out that were focused on this type of use case.
on this type of use case. So that's one. I think the security stuff is very obvious.
So that's one.
I think the security stuff is very obvious.
Fraud detection, tracing when there are hacks, tracing hacks, learning from that.
Another area where I see agents really solving and my friends at Eigenlair and Shriram, the
founder in particular, talks a lot about like intersubjective, intersubjective consensus and like the difference
between objective, subjective and intersubjective consensus. So an intersubjective thing would be
like, you know, LeBron James is the all time leading point scorer in the NBA. Now, all of us
in here can go and look that up and see that it's true, right? We can even see the date when he surpassed Hakeem Olajuwon and all that. There's no way to actually cryptographically or mathematically
prove that. That is a subjective truth or an inner subjective truth. It's provable,
but it's only provable through social consensus. That's something that could be provable through
agent consensus. And so you can
open up the world of blockchain to lots of new data types where you can have intersubjective
proof around data that is not mathematically or cryptographically provable, but is provable via
social consensus. The problem with social consensus is that you got to get people to
participate in it. Well, if you had a team of agents that that was their job, now you can kind of get rid
of that sort of lazy validator problem or lazy human problem where people don't really
want to participate in this inner subjective consensus.
So opens up the world of data that can be proven on chain dramatically to basically
anything that's that's provable off chain can now be provable on chain. to basically anything that's, that's provable
off chain can now be provable on chain. You have different solutions to this. Some people are using
things like ZKTLS and other, other technology, but I think agents can step in here and really
solve this problem. Uh, and then on the flip side, I think, you know, the ability to, uh, to have agents commit to their purpose on chain is a very powerful concept. And what I mean by that
is I build an agent whose purpose is to go build an e-commerce platform. Well, I don't necessarily
want that agent to all of a sudden start trading, right? I want that agent to just be focused on building this
e-commerce platform. Well, how can I do that? Well, I can use slashing to require that agent
to stick to the thing that it's supposed to do. And so by enabling agents to make commitments
to purposes on chain and having penalties when they go off their commitment, that is a really powerful
sort of opposite side of the spectrum of how blockchains can kind of rein in agents and make
sure that they're only doing the thing they're supposed to do and prevent sort of rogue agents
from going off and doing things. And then the last piece to it is verifiability between agent
to agent payments. We are going to live in a world where
my agent needs to pay Mark's agent for some service that's being rendered. Well, how do we
verify that that payment actually occurred? Use smart contracts for that. Now, other agents can
actually look at that smart contract and say, yes, Andrew's agent paid Mark's agent. Mark's agent
rendered the job for Andrew's agent.
That money's sitting in an escrow somewhere, release it from escrow to Mark.
These are like very obvious places where the traditional financial world doesn't work very
well for these types of use cases.
It's not built for these types of use cases.
And we don't want to just have to keep introducing humans in the loop to solve a lot of this stuff.
We want autonomous systems.
So that's a very key thing.
So I was going to come hit on that.
We hear AI agents building better AI agents.
And Jensen talked about it in his keynote, getting to a point where we remove human in the loop so that we can let AI train itself faster. So autonomous AI training, agents doing data labeling, tuning models, exploration, continuous learning.
So getting AI agents to train and build better AI agents and start to remove humans from the loop will make that go exponentially faster.
Well, each of these have been like billion dollar ideas.
So if you're listening to this and want to know what your next startup is, it should
be exactly what those guys are talking about.
I'm only going to throw in two more because like Andrew has said and Russ has said, it's
endless.
But the big one I see is having AI agents making getting into crypto, getting into tokenization
a lot easier. Because
right now, a lot of people aren't doing this because A, they're afraid of it, or B, it's too
hard to set all these things up and to transfer money back and forth. But if you had an AI agent
that you just instructed, hey, set up a crypto wallet for me and fund it with $2,000 and then
go buy Bitcoin, imagine that happening that fast,
that would bring a lot more people into this industry. So if you're listening now,
that's more than a billion dollar opportunity. Go out and build it for us so that we can all use it.
Secondly, much more complicated, but just as important is creating a supply chain,
just having supply chain management
and having AI agents able to track all these things
on the blockchain.
Now that requires a little more infrastructure changes,
like people actually reporting into it.
But imagine the AI agent being able to track
anything anywhere,
kind of like Amazon or FedEx does,
but for any type of supply chain.
That's going to come a little bit longer, a little bit further out there, but it's going
to come, and governments and big businesses are going to demand to know where all these
things are coming from and whether they're behind or low in shipments or what have you.
It makes the whole system a lot more efficient.
But let's now jump, and that was a great topic, by the way.
Let's now jump, and that was a great topic, by the way.
Let's now jump to some of the other things from our audience that is asking questions.
What do you think, Andrew, on the privacy implications of these AI agents?
We talked about security.
We talked about fraud.
How much of this, how much do we need to worry about privacy?
You need to worry about privacy a lot if you're using, if you're using closed source models and you're using closed source applications, you need to be very concerned with privacy.
You know, the old saying is, you know, if, you know, if the product is free, then you're the product.
Right. Um, and, and, and I think that that is something that we should be very, very concerned
about. Uh, your data is going into chat GPT it's being stored. It's being used to train their next
model. You get nothing in return for it other than getting to use the product. And if you want to use the API, you're going to pay $600 per million tokens output.
You know, so I think that, you know, there's kind of two different ways that you can avoid
that.
One, focus on open source projects where you can actually, whether or not you can read
the code, you can, you can have cursor
audit the code, see if there's any data storage that's occurring.
You can, you know, rely on open source community audits, or if you're actually able to read
the code, you can go in and look at the code and see whether or not they're storing your
data.
I think, so I think open source is really important and open source projects are much
less likely to actually store your data because they know they'll get called out for it.
So that's one thing.
So wherever you can try to use open source.
And that's on the application layer as well.
That's not just the model.
The application layer is typically where the data is being stored.
The models don't actually store data.
The models just do prediction.
So you want to be very conscious of the applications you're using and what they're doing with your data.
The other piece to this is, you know, this is an area where I'll give a little bit of a shameless plug.
Use a platform like Formation and use dedicated instances that you have control over.
We don't store any of your data.
Our job is not data storage.
If you want to host a database with us, you can. Fully encrypted. Only you can access that. So that's a way to solve some
of these problems. Use platforms that allow you to have dedicated hosted databases that are encrypted.
Even the platform itself cannot access because only you hold that encryption key. Use platforms that enable TEEs and secure enclaves so that
your encryption keys are secure and even the people that have control over the machines that
this software is running on don't have access to your encryption keys. You're going to have to be
a little bit more conscious of what platforms you're using in the AI era and in the agentic era, because, yeah, data is at a premium and everybody wants to collect your data and use your data.
And so it really at the end of the day, it's kind of on you as the consumer to make sure that you're not giving up your data to platforms that you don't want to give
your data up to. I can tell you, I know a lot of, you know, we're in crypto. There's a lot of Gen Z
in crypto. Privacy seems to be a lot less of a concern for Gen Z. And I think it's one of those
things that until it's required, it's not really thought about. And until you start realizing that this data is being used to maliciously target you
or manipulate you, or in some cases attempt to brainwash you,
you're not gonna care that much about it.
But one day you will be red-pilled
and you'll understand that, wow,
like this data is being used to try
to manipulate my behavior.
And you'll wake up and you'll be like, I don't like this.
I want my data to be private, it's my data. So the best thing to do is to use platforms that don't store your data,
store your data locally, use it as context into your models and into your agents.
If you need more scale, use platforms that offer you databases that are encrypted,
that use secure enclaves for storing those encryption keys and use platforms where data storage is not the thing
that they do. They're focused on compute and serving models and hosting models and doing
inference. They're not focused on storing your data. Right now, all the major closed source
providers and some of the open source providers through their application layer are collecting
and storing your data and using your data yeah also on that too
andrew you know we talk about the data privacy and data security not all we talk about exhibits
being a little bit different being gpu at the source so when we look at working with data
centers we're looking at tier three and tier four data centers data centers that are iso compliant
data centers that have sock compliance so data centers that have SOC compliance.
So we're very serious about protecting your data. And again, a lot of people just, they don't,
oh, a GPU is a GPU and it's not that way. So we want to make sure that we're working with
data centers that, you know, treat your data securely. So that's, I think, another thing that separates Exhibit from other providers
and taking that attention, even like, you know, we were a big AI company.
So there's health care, AI models, and there's a whole nother
of compliancy with HIPAA and everything else.
And to be a really big player and to support all the companies in this space,
you have to work within some of those
standards to do that. Exibit absolutely is doing that today. Okay. We got time for one more question
and I want to phrase it in a way I think that'll really draw out a lot of ideas from you guys.
So some of the smaller organizations that are here and they're doing information products or maybe they're even in manufacturing, what are some ways that they can start exploring AI agents and maybe with blockchain technologies?
But let's start with AI agents without huge, massive investments.
How do you do it at Formation?
Yeah, I literally is a perfect, perfect question.
I just got off a call with
a manufacturing company that's based in Cleveland, Ohio. They do about $50 million a year in revenue.
So it's a good sized small business, but it's by no means a publicly traded enterprise. And the boss,
the owner of this company is very, very interested in automating and creating
autonomous services around a lot of their workflows, but has no clue where to even get
started.
And so we kind of talked about what are the things that take up the most amount of time,
but are very monotonous within your organization.
And originally, he was kind of thinking HR, payroll.
He was thinking about accounting and bookkeeping. He was thinking about customer service. Well, what came up when I asked
that question was, well, you know, we get about a hundred receiving documents a day. We're bringing
in all of this raw material. We get all of these receiving documents. We have 35 different suppliers.
is receiving documents. We have 35 different suppliers. Every supplier's document,
each document that a given supplier gives us is exactly the same, but they're different from
provider to provider. So if we wanted to automate this where we scanned a document and we used some
sort of like OCR mechanism to extract the data and put it in a database, we'd have to have 35 different programs
to parse this. Well, not anymore, right? This is where an AI agent can come in. You give it some
OCR tools, you give it some context around what it's looking for. Maybe you fine tune a model
around that particular type of data. And the next thing you know, you have an end-to-end autonomous
system where you
don't even need to take those receiving documents anymore. Maybe at the receiving bay, there's a
scanner that they feed the document through, or maybe they send them to you over email,
and an agent listens for them and pulls those documents in and parses them and puts them in
the database. I learned something about manufacturing that I had no clue that, you know, there's something called a heat and the heat is basically the batch that the steel was made in.
And same company heat to heat, the actual metal bins differently.
So the machine that bins the metal needs to know the heat, the batch that needs to be optimized for that particular batch. Well, right now they've got humans that are entering in that information
every time they run a batch. Well, there's a way that you can automate it from receiving,
understanding the heat number in the database. Another agent that lives on the computer that
controls the machine can look at that heat number and can optimize it for that heat.
And so this is a manufacturing company that coming into it was thinking, okay, maybe we use agents for customer service. Maybe we use agency,
even for bookkeeping, it'll, it'll make bookkeeping a little bit cleaner and easier,
reduce our costs, increase our efficiency. Well, now the first thing that they want to build
is a receiving agent. They can go from reception of, of raw materials to optimizing their machines to actually bin the metal in
a proper way.
You know, that's from one conversation with them.
You know, a year from now, there's probably seven or eight different agents that they
have deployed on our network in order to enable autonomous workflow across the organization
and increase the efficiency and the productivity of
that organization allows them to grow, maybe allows them to lower their costs, win more business,
increase their revenue and grow their business. And this is a manufacturing company in Cleveland,
Ohio, that does under $50 million a year in revenue. And if we can make that company a
little bit more profitable so that they can grow better and be more efficient, then that's fantastic. And we can handle that end to end. They can keep their
data local. They have some on-prem servers. They can use our inference engine to run inference
against their fine-tuned models. And they can host their agentic workflows on our platform.
And we can make sure that that's hosted local, close to them,
but also has backups and redundancies. So in the event that there's an outage at the data center
where their primary hosting is, they have automated failover and they don't get interrupted
because they're relying on this autonomous technology. Okay. I have two, and then Russ
will pass it to you if you've got one. The first one every company should be doing, and that is from a marketing perspective.
Of course, I'm in marketing.
You should have an AI agent that's going out, looking across the internet, diving in deep
to all your competitors, and trying to see what they're doing and shapes the narrative
and reports back into you what's going on every single day.
There should be a report report and if you're hyper
interested in it maybe it's twice a day what have you just looking at what the current market's doing
from a larger perspective uh then you know doing sentiment analysis on that and then as well as
going in and look at every competitor their home page their blogs any and everything that they
publish out there and so you have a synopsis
of what's going on. That's an easy one. The second one is more sales related. And that is where you
instruct an AI agent to kind of go out and look for potential customers or current customers,
anyone that could potentially buy your product and service and looking for opportunities based
on what it's seeing out there. A little bit harder because a lot of these companies aren't
telegraphing what they're doing, but you'd be surprised to see if somebody opens up a new division or somebody
publicly, one of their executives talks about something that they're interested in getting
into, especially in our field with AI. We've got that AI agents reporting back to the sales team
saying, hey, you guys better take a look at this. This is what's going on. And that way, that
salesperson's armed with intelligence so that when they pick up the call,
the phone or email or however they're communicating with with that potential customer,
they already know what's going on. So so those are two I would set up right away. Russ, do you have anything to add? Not a whole lot to add. The one thing I would add is I come from a
background where we used to use data to train models and we were limited.
So this idea of AI driven, you know, agents to agents to improve your long tail of your models.
I'd like to see more companies doing that.
I think your model is only as good as the data and the training.
as the data and the training.
So I think more investment in there
to make those models better, more foolproof
is something I'd also invest in too.
Okay.
All right.
I would echo both of those.
I just, since you brought up manufacturing,
I wanted to give that example since it was fresh.
There are applications for AI agents
in places that you wouldn't even know.
And unless you're an expert in that particular field, you wouldn't understand how or where to
even get started. And so finding people, like if you're thinking about building an agent company,
finding people that are in that space, talking to them, asking them their pain points.
This is the most important thing you can do as a founder.
Like you have to understand who your customer is
and you need to talk to your customer
and ask them what they need.
Don't just guess at it.
And now they may give you some,
but you need to go and ask what they need.
If you're not doing that,
then you're doing a disservice to your investors
and to yourself
and to your time. Yep. Right. All right. Well, with that, we've got to wrap things up. Andrew,
where can people find you? I'm here on the X platform at CryptoNamiCon. It's con with a K-H-A-N,
CryptoNamiCon. You know, that's really kind of my primary place to connect with people.
You can visit our website at formation.cloud.
We're launching our beta on April 14th.
So be on the lookout for that.
Maybe follow us here at FormTheFog.
That's F-O-R-M-T-H-E-e-f-o-g form the fog here on on twitter turn notifications on
uh we will have a discord and other things like that we don't yet um but you know we're very very
product focused so you'll keep an eye out for all that stuff but yeah those twitter's the best or
the x platform is the best place to follow us great thank you all uh look forward to the next one
