Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Polygon x Semantic Livestream
Recorded: March 20, 2026 Duration: 1:04:07Space Recording
Full Transcription
Hello everyone and welcome to our live stream and if you're catching this as a recording,
welcome to the recording. We're going to get going in just one moment here. We're here with
James and Dex from Polygon and Semantic Layer and we'll dive into some really cool Agent 5 stuff,
everything going on on Polygon, as well as Agent 5 coming
from us and Polygon. And we'll dive into everything in just a minute after we give people a little
bit of time to join here. Let me get a quick tweet and tag out real quick. All right.
I think we're good to go.
And I think we're ready to kick this off.
This is going to be really fun.
I've been looking forward to it, guys.
So, yeah, I guess before we really dive into all of the Xpro2, AgentFi, transaction flow,
maybe we can get a little bit of intro from both of you.
So, James, head of DevRel
at Polygon. Yeah, give us like a quick little background, how you got into everything on chain,
how you got into Polygon and like what you're doing right now and excited about.
Ooh, how I got into crypto blockchain. That's a long story. i think i'll take most of the restrain but um
you know quick synopsis i've been in crypto eight or nine years now um originally worked on a lot
of enterprise stuff and um around 2020 got really into the defy sector um worked there for a few
years and then i uh joined a company called sequence Sequence did a lot of work on the infrastructure side.
So like if you're building an indexer or wallets or interoperability, so we did a lot of work there.
And how I ended up at Polygon is Polygon actually ended up acquiring Sequence and acquiring our
infrastructure. And so we basically, the goal of Polygon now is tying a lot of those products
into what we call the open money stack.
And really the goal of Polygon
and what a core focus of ours now
is moving to this consolidated
and vertically integrated platform
that we call the open money stack,
which allows basically any businesses, users to
basically move money globally more easily. So ultimately, we're, as you kind of see from a lot
of the stats on Polygon POS, we do significant numbers of like stablecoin volume of payments
across the board and a lot of jurisdictions around the world. And so really what we want to do is just make it easier for people to move money
wherever they go.
And a lot of my focus right now
is working on the agentic side
to make that more easy to build that out
along with our developer experiences.
Epic, thank you.
Beautiful intro.
Yeah, the open money stack,
full vertical integration,
I think is what we're uh
we're going for and i think we're all on target for it with polygon it's like pretty sweet to see
uh like this you know i don't know it all coming to life here it's like been many years of uh
number one side chain and now i'm getting kind of expanding from number one side chain to number
one vertical stack like it's happening real time the stats are coming through uh it's pretty awesome Dex uh can we get a little intro from you as well
yeah hi everyone my name is Dex I'm one of the co-founders at Samantha layer um so how I got
in crypto um well I mean like a couple years ago ago, when people discussed how they got into crypto,
it sounds like everything was just a couple of years ago.
But now I think about it.
I first got into crypto in 2014.
I was a high school student at the time.
And at the time, I was living in a school dormitory.
And all of the other guys, they have super beefy gaming computers, PCs in their dorm room.
And I had one as well.
But at the time, the reason I had one is not to play games, but to mine Litecoin.
Because at the time, it was still profitable to mine Litecoin with your GPUs.
So that was how i initially got started and now
actually if we if we count the years right 2014 well that was like 12 years ago so i mean this
industry really came a long way uh since then so that's how i got started but um uh when I really started working on Web3, that was 2020, that was DeFi summer.
I was super excited and at the time I was working for MarketMaker and that's how I got
my first job in this industry. And so, you know, been working with a lot of different teams and a lot of different
people. And eventually I co-founded Samantha Lear. And what we're building from day one
and has always been is to help dApps to sequence their transactions. As we all know, in the blockchain, everything is deterministic and the ordering
of transactions actually matters a lot in terms of who gets the better execution price
and who gets the ability to extract arbitrage value. So we decided to build Semantic Layer in late 2023, and transaction sequencing has always
been our focus and our value proposition.
And right now, the reason why we are also reaching to the AI agent space is because
we realized that the majority of transaction volume might be driven by AI agents in the future.
So we're building all this AI agent adjacent infrastructure
and dApps to bootstrap the order flow from those AI agents.
Awesome, thank you.
Yeah, I can resonate with that early mining. Like, I remember my first
real gaming computer, and I was like, finally, I got a graphics card that can mine some ETH.
And I remember tracking it down, and the T-Rex Miner and all that good stuff, or Trex Miner.
Actually, that's really funny. I didn't mean to even think of that. But yeah, Trexminer, completely separate.
The T-Rex, like massive tokenization, real-world assets.
Polygon, you guys just partnered with T-Rex.
I don't know if you want to give a quick little shout-out or word on that, like, crazy big partnership that just dropped yesterday, James.
Oh, yes.
So I can definitely, on the T-Rex side, it's so funny that just dropped yesterday, James. Oh, yes.
So I can definitely, on the T-Rex side,
it's so funny that the pipeline to a lot of crypto is always gamers buying gaming PCs and using it for money.
So I think that was my initial introduction.
But yeah, on our work with the T-Rex ledger,
with Tokeny and the Apex group.
So this is something that's really exciting.
It just was announced yesterday.
And so a little bit about what it is.
So basically Apex Group,
who services over $3.5 trillion in assets,
they're going to basically tokenize
over $100 billion in tokenized assets.
And their network is called T-Rex Ledger.
And so this will basically utilize our enterprise CDK stack. And so basically
as we talk to a lot of like payment organizations and real-world asset
issuers and several others, we basically have realized that you know a lot of
enterprise focused use cases they want their own
block space um because they need to be compliant or they need to service certain um things and so
we developed this um the enterprise cdk and this also is interoperable with other chains via ag layer
and so people can easily um move assets back and forth to the chain and it's all like completely
driven by their tokenized compliance engine.
So this is really exciting.
They're moving a significant number of assets.
And yeah, we're really excited to be a part of that.
And I think this is a great first use case of what we kind of see
and what we're accelerating towards is just more money moving on chain
and more assets moving on chain.
So we see this pretty consistently through a lot of our conversations.
Yeah, and it's really cool, too, to see like these really big asset managers
and stuff like actually use some of these chains like polygons already evolved
and been working for years rather than try and start up like their own L2
specifically for their own enterprise. Like it's, you know, enterprise. We see a lot of companies try to do that and then it kind of
backfires or fails often. Why not tap the resources that have already evolved on-chain?
And so I think that's what's going on here, which is the right way to do it for the long term.
And yeah, one thing Dex mentioned too, as well as like the
sequencing and ordering, like that's always, I mean, of course, it's always been foundational
important for blockchains and DeFi and everything. But as we're getting into this
agentic era, it's slowly but surely accelerating, like the sequencing and ordering matters
even more than ever, because we're getting to that future point, like the sequencing and ordering matters even more than ever, because we're
getting to that future point, like really quickly, where all of a sudden, maybe we're
about to have a million agents, and they're all trying to call, you know, call contracts
all at the same millisecond.
And it's going to potentially like overpower some of the infra that we have on chain.
So it's like, yeah, the sequencing is going to matter more than ever
as you go into the future. And humans are now competing with a crazy amount of agents on this
block space. So we're going to see probably a lot of interesting new intents and solvers and
different systems that handle all this agentic traffic. But yeah, I guess the internet's about
to get a lot weirder. Agents are going to be calling APIs directly, paying for them, paying other agents to do work.
Probably a mix of centralized agents and non-centralized agents.
I know there's like ERC 804 trying to do a lot of standardization, but there's kind of that cold start problem.
And then we're already seeing people who people surpass that because the tech is going you know advancing so quickly in
real time that like now we're seeing open claw agents get these skills to sign transactions have
their own wallets and all this stuff and they don't really need to have that uh centralized
identity they're already acting and you know the people who who created them are their governed uh
voice in a way so it's it's like every day we're seeing the advancement, you know, kind of happen
a week or a month at a time per day. It's like the acceleration is actually insane here.
But I guess I'd like to just, you know, warm into the Asian-Fi stuff. Have either of you guys like
run any autonomous agents, play with open claw, like try to get any agents on chain yourselves
or play around or see the stats coming in like
what do you guys think um i mean i can say yeah for my side i uh i use open call agents i use um
the new the new hype one is hermes agent um which is really nice i actually prefer much more than
open claw but most of my work is also now using
agents in one way or another, whether this is using agents to build applications, deploying
OpenClaw agents to work autonomously. So whether this is like interacting with Polymarket or,
you know, looking for the best yield opportunities, or just exploring kind of what's out there and
like modifying my agents to do personal stuff.
So this is something that I love to do.
And it's like a big part of where I see
and where Polygon sees like, you know,
payment flows happening
is they'll be increasingly done by agents
on behalf of humans.
Yeah, yeah, yeah. yeah yeah i think uh i think we had semantic layer on the the same page there the the
genic overflow coming um yeah how about you dex
yeah um i've been using open call for a couple weeks right now and um i think um i think actually will draw most people
into open call what they did well is a couple usage story you know some of some of them i
assume that people really know them as a meme like oh i just gave my open call one task. By the time I came back a day later, it has built an entire tab.
Something like that.
I mean, in my actual experience, that's not true.
In my actual experience, often when I have instruction,
the agent is going to get stuck in your loop at some critical steps.
So I think obviously there are more to improve,
but I think the success of OpenCard stories
that it tells what people actually want.
People want something that can just help them autonomously? It gives this specific thing a simple task
and this thing can understand the boundary of the task
and what is the condition of the,
what is considered as finished the task
without you giving the step-by-step input.
Where so far, most of the tools right
different softwares apps they are uh they require human input a step-by-step input
um but uh what open call presented to you know to everyone is this possibility that
well maybe you can just that air agent to automate everything
yeah yeah so i guess like one big thing this is touching on is that's basically like agent 5 2.0
maybe i should back up for one quick second like agent 5 1.0 in my mind is kind of like um
Like Agent 5.1.0, in my mind, is kind of like been playing out for maybe a year and a half or so.
But Agent 5.2.0 is like really brand new and cutting edge.
So in Agent 5.1.0, it's more like agent managed strategies.
There's like a number of protocols out there where, you know, maybe you put in your USDC and it'll auto rotate through different bolts, through different like highest yield with the
risk parameters you set with these different protocols and some of them are a little more
advanced and kind of maybe using a couple different daps to do a certain type of flywheel
or strategy but uh you know it's interesting to see so much talk about like agent phi but an agent
phi 1.0 there's only maybe a hundred million total tvl between all the
different agent phi 1.0 dApps so i think there's maybe like a little bit of this trust issue still
going on where people know the agent can theoretically like source better yield
continually than them but maybe they don't trust it to execute with their own capital and then we
get into the agent phi 2.0 where we're talking about.
It's like, okay, people are gonna set up
their own autonomous agents.
And it's not just, oh, I'm gonna add money to this app
that has an agent managing my money.
I'm gonna like be the full loop myself,
send out my agent with my own governance.
And it's going to just go into DeFi on chain in the wild,
maybe with its own wallet, its own signing, you know,
ability and just YOLO and hopefully follow my instructions. And that's, you know, very, very
new. But I guess like to go into some of what we're really focused on here is like, at Semantic
Layer, we have the profit arena, which is our core DAP right now, where it's actually it's a bit of a combination now I think about it of Agentify 1.0 and 2.0, where the 2.0 layer is we run all these autonomous LLMs on Polymarket on Polygon, and they're just executing strategies with their real-time APIs on predictions.
And there's no human involved.
We hope GBT and Claude and Gemini make good decisions,
but we don't know.
They really are in control.
But then the Agent5 1.0 kind of comes in layered on top of that
where people can then have a dedicated agent that is fully on-chain as well,
and it's watching what those Lms are doing on the predictions and
then they're executing the strategy on top of that based on what the user wants so like
maybe gpt llm is making really bad predictions so you set up your your agent 5 1.0 uh strategy
manager to copy or inverse it so if gpt is doing bad your automated strategy is inversing
gpt or maybe claude's doing really good and you copy trade um long on claude's like prediction so
it's kind of this interesting combo of uh of both layers happening at once maybe uh dex you can pull
that up a little bit later we can do a little bit of a demo. But yeah, yeah, yeah. It's like, it's cool to see both of them happening at once.
2.0 kind of taking the reins here, but then being able to work together.
I guess like getting in deeper here, like I guess, how do you guys view Agenic Finance?
Like do you, are we early?
Are we late?
Is like 2.0 going to take over?
Just I guess, what are you guys' thoughts?
Like I know there
are a lot of different big companies and you know kind of big vcs and players out there that are
really um pushing the agent 2.0 and picturing like the billion agents on chain in the next couple of
years uh how do you how do you feel about it what do you think like their main real use cases will
be after kind of the hype wave of the unknown starts to turn into reality.
I think everyone wants to know the answer to that of what's going to be the biggest use case for
agents. But that's awesome on Proph Arena. But I think touching on what Dex said in the
previous question that you mentioned, the first part was people manually typing commands to their agent,
and it's very iterative.
And then the second is like, you saw examples of this,
whether it's open claw or like a route loop,
where it's acting autonomously over a like fixed length of time.
And what I found very interesting is basically adding additional,
because when I spin up these agents, Dex mentioned this,
I can give my agent a wallet, I can make it transact with something very specific,
but then when I wanted to do something else, I need to know that this skill, this API exists,
I need to add it manually to my agent.
And so to do anything useful in the real world,
like I don't know, buy something on Amazon,
or book a flight, or book a trip for myself,
or whatever the case is, I need to cobble together
like 50 different skills and different API endpoints
to actually do this.
So I think a big one, because you mentioned
like what is the next thing, I think having a opinionated discovery layer will be more and more
commonplace, which basically enriches these agents that are already quite good, but it guides them
down a path of what can this agent actually do and specialize in?
And what is the scope of what they're experts in?
So I think that's a really common one.
In regards to what an agent really is going to,
what's going to take off in the end,
I mean, prediction markets,
yield farming right now really has a lot of the most volume, as I'm sure you guys are seeing over a profit arena as well.
But I do think that gradually more and more people will be comfortable with the fact of turning over some amount of their capital to these agents for them to act more autonomously.
of their capital to these agents
for them to act more autonomously.
And we'll get more agents doing more consumer-like activities
on behalf of humans that are like predicting what you want
before you even know it yourself.
Yeah, yeah, yeah, yeah.
I see it, I see it, I see that future.
So I guess like,
I like kind of connects into like x402 like maybe not
all the d5 transactions are going to be going through x402 but like more the consumer commerce
side um is probably going to be tapping a lot more x402 as the uh the future evolves here and
we already see like x402 transactions all-time high on polygon and uh we tap x402 as well for
profit arena um maybe i can get the x402 perspective from both of you because i think
there's some like interesting things going on within that realm and uh some insight that we
can share with the public because x402 i think is still a bit un like not fully understood by everybody
yeah sounds good dex do you want to take this one first
yeah um well i think for s4 too uh the biggest uh usually solves it simplifies how ai agents can
pay since i mean obviously like if they will prefer you want to do it like they can have all
sorts of ways to make the AI agent
able to paste things, right? They can set up some temporary APIs, some stuff, but the
easiest way for an AI agent to actually be using crypto. Like thousands of users on open
call, they are already using their agents to interact with on-chain DeFi protocols. So I think what the biggest significance of X-FOR2
is that enables rapid prototyping
and enables rapid onboarding of AI agents
to pay for things on the internet.
So I think that's the biggest point of X-FOR2.
I think what I hope X-Fore 2 could add in the future
is support for different currencies.
Right now it's USDC's.
But in the future, I think we should enable AI agents
to pay with other currencies,
maybe even like natively with ease
right people have been saying okay is the ease is a asian money and i think we should be doing that
by enabling AI agents to pay for things with ethereum instead of only usdc
Yeah, so I fully agree. I think I would even extend this more like, from our side, I just
want to be able to facilitate where an agent can pay basically anything they would like.
And it settles in the preferred like merchants payment method as well. So I think that's
ultimately what we're building towards the
Polygon as well is like, whether it's an agent or a human is they can pay, you know, in a credit
card, they can pay in USDC, they can pay whether it's XFOR2 or, you know, their favorite protocols
token. And it doesn't really matter what the agent actually wants to transact in or what the
user holds. It's more about the flow of money
and then at the end of the day the merchant can be settled in whatever currency or method that
they like um and this basically just helps everyone um get more volume through their applications
and make it more like you have more universal liquidity in whatever you'd like to pay with
um of course there'll be like fees like depending on the different payment method.
But I think that's ultimately what we're seeing is,
yeah, this broadening of the different payment methods.
I will say on X402,
it's such a game changer for like,
especially if you're a developer,
is I don't even want to have to interact
with like traditional dashboards anymore.
I just want to be able to, via my agent, get an API key,
or not even get an API key, just pay for an API endpoint directly with X402.
And this is just the smoothest method,
because many times it's just kind of one-off requests,
like can you give me flight booking information for X, Y, and Z?
And X-Fore 2 is a great unlock just to enable better business models for these people that have valuable data that they want to serve via API.
Yeah, you guys both make really, really good points there.
I see a lot of perspectives like online, people kind of trying to force their own narrative or hopes on the expo too like it is very locked down into just
usdc right now and you know like eth being kind of the the most easy uh i guess easily programmable
kind of central money on chain would make sense to be added but i see a lot of people too like
saying like agents are
going to prefer uh fully decentralized stable dollars and you know trying to get it pushed so
some of these uh i guess fully decentralized stable dollars that exist currently mostly on
ethereum mainnet um want to like get their name in the bucket and and get sourced and all that
but it's like do the agents actually care if it's a decentralized stable dollar or not?
They seem to be perfectly happy with USDC for now.
And especially if it's gonna be going to merchants
and real world actors, the real world,
merchants are probably gonna be happy with USDC
over a decentralized stable
that maybe they're not able to get into their normal bank
and stuff as easily.
But it makes sense to expand to basically cover all assets and maybe include different
fee tiers depending on liquidity and all that.
So I think, yeah, I think X402 really is just at the beginning with how the rollout and
everything is going.
But I guess to slightly move on here a little bit like um i want to get into the polygon payments
and just the the vertical integration in a second before we do that maybe dex can you tell us a
little bit more about like semantic layers kind of north star here like yeah yeah um I think for the cemented layer, it's pretty clear now that the future of blockchain
will be dominated by AI agents.
Majority of the transactions will be initiated by AI agents.
Just because of the simple fact, the entire space has gone more and anyone that even even sees on DeFi users who have been using DeFi protocols
since day one just cannot keep up with everything so I'm 100% in the future 90% or even 99% of the
transactions are generated and signed by AI agents so with this assumption, where some of the protocols, some of the infrastructure that
developers might start during today so they can get prepared for tomorrow, one of them
I think very interesting would be transaction policies, for example, allowing users, the
end users, those agent owners to set spending limit or set a list of different programs that the
agents can direct. And the second interesting infrastructure I think will be allowing either
it is the apps or the agent owners to capture any of the order flow revenue from the agent initiated transaction flow.
So I think these two are some of the very much needed
infrastructures in the future where the majority of the
transactions are being driven by AI agents.
And that is what Semantar, our entire team, are working towards.
Epic, beautiful.
Yeah, thank you.
That was like really, really great overview
and explanation there of the agent North Star future.
Yeah, I guess kind of the same thing,
question for you, James.
Like what is the North Star kind of for Polygon?
I mean, we can kind of see it.
You guys have had a bunch of acquisitions
in the last year building towards this vertical payment rail integration.
And now, you know, we see T-Rex coming on and proving it just yesterday. It's like perfect
timing, literally. But like, I guess, how did this become like the core focus? Or I
guess, like, how should people really think about Polygon today?
Yeah, that's a great question.
So really our core focus,
which a lot of work is happening behind the scenes,
is the goal is to move money globally,
like full stop, that's basically it.
And we wanna make this the best developer experience possible
and the best builder experience. So us it's actually like it doesn't necessarily matter whether this is coming from ai
agents but which we of course think is going to be a massive market um or like b2b cross-border
remittance payments or invoicing and so on and so forth. Ultimately, we're building our open money stack
in order to make this easier to on-ramp, to off-ramp,
to pay people in stable coins,
to solve compliance challenges
and give people a secure wallet, as Dex mentioned,
in order to make sure they can actually move this money safely
and they can move it globally.
And so you can pay your grandma living across the world
very simply with the Polygon Open Money Stack
at the end of the day.
So really the North Star for us is
expanding the amount of partners that are building
on the Open Money Stack, that are building on enterprise CDK.
So whether they need their own private space or they want to use the public
block space for Polygon POS.
So we have multiple offerings for them and it really depends on like,
you know, kind of their,
their compliance needs and like their scalability needs.
But that's ultimately at the end of the day,
what we're trying to do is move money globally more simply.
And we have a lot of exciting stuff coming out around this on the product side.
Yeah, I can't wait to see more.
Like I saw one really cool note like just a couple of days ago.
Like safe deployments on Polygon, like just insanely
skyrocketed this month. So I don't know if you if you know, what that's all connected to.
I can't say too much about why people are deploying a lot of safes. But yeah, we got some
exciting stuff coming out. And, you know, people generally deploy saves when they want to secure a
lot of money and they want to move a lot of money. So I'll leave it at that. Yeah, yeah, yeah, that's
that's fair. I think we we lost Dex's connection, but I think we'll be back in one second. But all
good. So yeah, like one thing I kind of see from a lot of the like big funds and, you know, billion
dollar names out there that kind of help influence a lot of
what we see and the direction of on-chain finance like a lot of these big names are kind of talking
about how stable dollars make sense as a larger settlement layer like outside of just defying on
chain but like a larger kind of picture of the whole world here um Basically programmable dollars, not just programmable money. And it kind of,
it's funny because if we go back, you know, five plus years here, I think the U.S. government
was, oh, Dex just came back. Let me get Dex back added in here. There we go. Got Dex back on online.
But like, I think the government seemed to be kind of fearful of stable dollars like what
it would mean for destabilization of uh the us dollar kind of dominance and all that but like
if we've seen anything it's really been the opposite like um i know swiss francs and the euro
um have been you know tokenized to an extent but it's like the dollar really dominates um all on
chain finance between usdc
usdt and a few other like top stable coins like they're all very much dollar focused
so like i think this programmable dollar idea is is kind of coming to fruition um
but i guess like uh what do you guys like are you guys focused pretty much on usdc are you open to
all stable dollars like how do you see the pretty much on USDC? Are you open to all stable dollars?
Like how do you see the stable dollar wars in a sense like coming together?
Or like are you, you know, in the more open spirit that whatever people want to process in, we're happy to use USDT as well.
Like do you guys have any internal thoughts on specific stable dollars and like the programmability of them?
The stable dollar wars um you know from our side like obviously like there are a few um stable coins that have a
significant share of the market um i think what's interesting though and what we see at polygon
because we have a lot of usage
in like, especially non-US jurisdictions on Polygon.
And so we see a lot of traction on like actually using non-USDC payments.
And so, and we also see a lot of like, whether it's an individual protocol or a government
or a country, they want to issue their own stablecoin.
And so we do see this accelerating pretty rapidly.
And I would expect that you'll have a tokenized version of most kind of government issued fiat currencies in the not so distant future.
Because people see that like it's a massive way to basically get more influx of revenue into their treasury.
And so this is ongoing.
People see it on the US dollar.
And I wouldn't be...
It's going to happen where people want to tokenize more of their dollar assets.
So I think that's also a big focus of ours is not only being focused on the USD stable coins, but the non-USD ones, because even though they've been lagging a bit in terms of adoption, there are significant portions of the world where this makes a massive amount of like benefits in terms of usability and transferability. So I'd expect more over, you know, the near future,
people moving more money into dollarized stable coins or tokenized stable coins.
Yeah, yeah, yeah.
I like I I'm on the same page.
It's like maybe some governments that were a little more hesitant now
are kind of feeling like the fear of being left behind if they don't pursue,
pursue their own stable dollars and catch up and have like some competitor in this race um i think a lot
of people too like on crypto twitter and who are just like fully on chain are a little too zoomed
in sometimes like polygons not just you know it's like we said it's not just the on-chain only
transactions it's like connecting to the actual full real world physical uh stores
and merchants and everything and this is like a much much larger pie um than our current kind of
small closed on-chain ecosystems um normally view but like as polygon here is is really expanding
basically the payment layer for i wouldn't say the whole internet not just
quote on chain like over the next year I guess what are people going to kind of see like what's
going to be the first thing people think about as like this changes from just on chain to the
internet and then kind of maybe the more important kind of secret question here is like in the
background maybe some hidden stuff or some stuff that's like not so easily present to people what's
like the the real underlying change that's like gonna be setting new
precedent for you know maybe the next hundred years or so that you think might
be a might be coming up so you mean on Polygon specifically yeah well Polygon
specifically but I guess just like stable dollars as well, just like as this whole the internet settlement layer, not just like the on-chain settlement layer.
I mean, this definitely ties back to the agentic side of things where like right now, if you look at kind of what is the total transaction volume in terms of like US dollars transacted on chain via agents, as well as like the number of transactions, they're still relatively like small versus what's happening versus like what's hyped up and also what's actually happening like in just traditional like consumer payment space.
so what's actually happening like in just traditional like consumer payment space but
the thing with like agents is even if you're if you're gone for a week you're behind and this is
like crypto was like this but like crypto and ai is accelerating at a extremely rapid pace and so
i would expect whenever that um agents start gaining volume,
it will be exponential and we'll quickly see a massive rise
of people catering specifically to agents.
I kind of touched on this earlier,
but like a lot of the internet has been curated
for like humans to basically utilize.
And, you know, of course there are API endpoints
that BOTH use and, you know,
there are ways that have been done in the past,
but I suspect that we'll have a massive transition
to people actively selling and actively moving
in order to get more agents actually using their,
whether it's software platform or their protocol or whatever
the case is, will have a much stronger focus of how do you actually market to agents?
How do you get more agents moving with your protocol versus others?
And I suspect a lot of UI, UX things will disappear or be much more malleable where
they either get created only when a human views
them or just in time and they become customized to like what a human wants. Yeah I think one thing
you kind of touched there is like the the internet isn't gonna not be just web 2 versus web 3 anymore
it's gonna you know everything going to mesh together into this one
semi-unified layer as agents use on-chain and, you know,
transact and hold their assets, but are really working in normal web 2,
you know, the normal internet that has been built for people.
So we're going to see a lot of flow come in.
Maybe Cloudflare is going to go crazy.
Agents starting to figure out how to pass their CAPTCHAs and everything.
The revolution is happening in real time.
But I guess one kind of just crazy out there question here is if every API endpoint on the internet,
Web 2.0 and 3.0, everything's merging together. What maybe are some new business models
that may become possible?
Or what are some crazy things
that maybe we don't think about right now,
but maybe agents might see as like,
oh, this is obvious that agents will be able to do
if they can start calling these different APIs
and APIs can just charge directly to these agents
without any human interaction.
Like, do you think there are any new mechanisms that are going to potentially evolve?
Yeah, I don't know.
Just any interesting brainstorming from either of you guys.
I mean, from my side, the obvious one with X402 enabling it,
it still hasn't really come to fruition, but microtransactions are a very common one.
And I think this will just accelerate
when it can be completely automated.
The second is what I've seen and experienced as well.
Agents don't have switching costs associated with it.
So typically if you're buying software from someone,
you integrate with them, it takes some time, you know, you have
a deal in place, you don't really want to move around. But agents don't have this structure.
They basically just search for what is the lowest, lowest friction in the lowest fees,
and they move around with virtually zero switching costs. And we see this in like,
when agents are interacting with DeFi protocols,
they have the same kind of like flow.
They just search for the maximum amount of yield
with the lowest loss.
And I suspect like,
this is not really a business model that you asked about,
but I suspect that businesses will focus
and compete much less on this like switching cost aspect.
And this will have massive impact for existing businesses,
but allow a lot of up and comers to have their foot in the market.
And then basically what I touched on last time in the previous question is,
how do you market agents and how do you optimize your software
or whatever you're building
in order to make it agentic friendly?
And I suspect a lot of optimizations
will be built around this.
And how do you get first into LLMs
and how do you outrank others
will be really common to make sure
agents are building on you versus someone else.
Yeah, yeah. I think like something you touched on there, which has not really been discussed much
like openly, at least like on Twitter, you know, maybe it's we need to get more agents online and
get some of like the laws and everything in place about them and their identity. But like a lot
of that comes into you're saying like agents optimizing for you know, the quickest path,
the quickest transaction or return or lowest fees. But like sometimes that quickest path is
not necessarily the safest path. So I have like something in my mind a lot of the time is like
thinking about like governance in the trust layer. So like, I wonder
if there will be some type of unified trust layer in the future, that agents will be able to tap to
say, Oh, this is the fastest path, but it's not as trustworthy or theoretically as reliable as
another path. So maybe some of the trust aspect will come in that'll help be the second variable
in their decision making. I don't know, like, I think that's still really evolving.
I don't know if either of you guys have any perspective
on any of that.
But we can-
Well, I was gonna let Dex take this one.
Yeah, Dex, what do you think?
I mean, I think that kinda,
some of that comes into like the sequencing and ordering especially like uh as or ace like application
controlled execution like do you think that's going to play a part in uh the trust layer of
agents ordering transactions and final execution um or you think that's going to be kind of like
a separate layer from the intent and solver kind of side of things.
Right now, the biggest thing about sequencing is that from the user's perspective,
each transaction maybe only carry a few cents.
So it's not worth the while for human users
to learn how to use different solutions.
But that might not be the case for AI agents.
It's relatively costless for AI agents to try to optimize the trade execution as much
as possible.
So yeah, I agree.
I think a lot of the old business model that rely on human being um being too
oh no we we lost you dex for the last 10 seconds no right at the precipice oh no we're getting to the juicy bits dex can you hear? Oh no.
Dex, Dex, I think he fully lost connection. I think he'll come back here in a second.
Well, I think some of what he was saying is kind of leading into like
more specialized um rpc's med protection and some of this tech that we really need to advance that's kind of built for humans right now that needs to adopt so we can manage the agent especially if
there are going to be like these billion agents online one or two years from now um just like
more of a fun interesting question here
James like what do you what do you think maybe like I mean this is just you know
theoretical maybe like one or two years from now what do you think is like maybe
a weird job that AI agents might be doing that's like gonna be totally
normal in the future but like today we don't even think about the possibilities
of it or not assigning that risk or execution to agents.
Ooh, the agents will be doing or the humans will be doing?
No, agents. Maybe agents like governed by, you know, a human's initial prompt or something like
that. But or like maybe if some agents are acting even more autonomously in the future, like someone
just sets up their open claw, gives their agent a little bit of capital and like agent controlled
wallet and just lets them loose and says, like, agent a little bit of capital and like agent controlled wallet
and just lets them loose and says like, have fun.
Like what's some weirdness you think
like an agent might go down
because, you know, if they don't have a human brain,
they have a very different program thinking pattern.
Yeah, so I mean, this is, it's getting pretty weird.
So like I'm, you know, I,
in addition to like working at Polygon Labs,
like I also run my agents extremely often, whether this is Hermes or something else.
And it basically has a lot of context about me personally.
So it knows my calendar and it knows like, you know, my family and it knows my pets and it knows like my birthday and like these kind of things.
pets and it was like my birthday and like these kind of things and i suspect and like now i have
this set up like extremely detailed where you know every every day it'll look for things that i'm
interested in and then actually surface that information to me instead of like me going out
and finding it and so i think that we'll get into this world of agents are much more proactive and much more personalized to
the human and i can already see like i'm not that far from being able to just like oh here's um you
know a thousand bucks and just take care of any like uh anticipatory purchases that i need so like someone mentioned um like oh if your
kid is going back to school um previously you'd have to go and like oh i need to get them a
backpack and a pencil case and you know all their school supplies and stuff but if an agent knows
when when these different events are happening and they can predict it in the future and it can anticipate your needs
and it can just autonomously go and buy this for you.
I think that will be a really weird world
where whether this is groceries
or buying supplies for school
or going on a camping trip
where agents are then anticipating
what you need in advance
and managing this aspect for you.
And I suspect that this will become much more commonplace
in the next year or two as they become better
and they learn what you need more
in order to anticipate this.
Yeah, yeah, yeah, I agree.
I think one thing there is it really comes down to uh memory to
a bit there like you know your agent knows about you um through these these uh like the memory
saving files md files the skill files and all of this the soul uh dot md so to speak yeah but like
um i i don't know like do you think there's any? Well, there's like, I guess my mind
goes to two things. Like the security side of it is like your agent knows all this stuff about you.
Now it's open on chain. Like, well, you know, what if it leaks information, like private personal
information about your family or something like that's, I think, a big scare a lot of people have.
And then the other side of it is just like on the tech side is like the one kind of big limiting factor right now is memory.
Like everybody's trying to get their agents more memory, which is like I think Dex also has some ideas on memory.
But like what do you think about memory, memory expansion and like memory safety?
safety?
I mean, I would say on the memory stuff, I would say a lot of this is actually, as long
as you kind of have like a well structure, that's the thing is like a lot of this is
extremely manual and tedious now.
And you have to know what you're doing and you have to know where to search to find these
things.
But as these different agent harnesses become more opinionated and we kind of find what resonates the most in the market, I suspect it'll be like, you know, one click installs or you just, you know, you authenticate people, whatever data you're willing to give over to this hosted, yeah, cloud instance you give over.
And it kind of just orchestrates a lot of this on your behalf.
On the memory side, I have a pretty good structure.
I think the key is a lot of people just throw like 10,000 line MD
into an MD file and call it a day.
And a lot of times you can use kind of like almost like a table of contents
for like the top level of your agent.
And then whenever it runs into different things,
it links to different MD files of where to find them.
And then of course, just using a model
that has a big context window usually helps.
Yeah, yeah.
I see the context windows expanding,
you know, 100K to 200K,
like doubling and stuff overnight um yeah oh dex we
finally we got you back on a stable connection you were just getting to the juicy part of your
your last answer there i wonder if maybe you could uh re-say it so we can uh all here because the
connection was a little shaky there and then maybe like expand into like um you know maybe your thoughts on like net protection and
customize rpcs for agents and like that whole realm and route yeah i mean like right now a
lot of business bottle if it's our travel agents like travel aggregators uh stuff like hotels.com
right speedia um their copies of oh yeah or i mean you in web3 uh you know different
uh d5 profile you swap or switch swap um most of their business model are relying on the fact that
human users are preferred conveniences right they prefer to just do everything on one stop
uh without you know spending too much time and effort to compare, you know,
different prices, different execution,
finding the real, real like optimal strategy.
So this is a lot of business have been building their
business models, basically being the middleman,
being like introducing, artificially introducing
some inefficiencies into into systems such that they can
extract money from consumers and such were the longer be the case in the age of ai agents
because for the ai agents it's relatively cost uh cost free it doesn't cost anything for the
ai agents to look up multiple multiple different places for the best execution for the best most optimal prices so I
think that in the you know in era of AI agents probably most of those business
models that rely on being middlemen they're going to go away yeah yeah yeah that's actually a great point something i haven't thought a huge
amount about but it makes total sense is like um you know uniswap front end they're getting a little
bit of fee there and it's like these agents aren't going to be using the front end they can just go
and call contracts directly um it's definitely yeah it's going to shake up a lot of business
models going to be very interesting to see play out in real time.
I know we've been, gosh, this has been so fun.
We've been almost live for an hour here already, which is kind of crazy.
So we'll try and not take too much more time.
We've got a couple more questions, but really quick before we get to the end,
maybe I can do like a super quick one or two minute demo of Profit Arena and show like what's actually live on a polygon right now um
with us with agent 5 1.0 and 2.0 um so here we are and we can kind of we've been running this for
about three months um so it's it's perfect data we basically started the big five llms with like
ten thousand dollars hooked up to some real time financial APIs,
some basic, very basic guardrails, and then just let them loose. So we can see here, you know,
the positive and the negative, this is like the inverse GPT versus the direct copy GPT. And like
one of the most interesting things is GPT just got destroyed early on the first month and a half. It had like 10 losing trades in a row, but now it's on a 13 win streak.
So it's still at an L, but it's like it seems to have changed its behavior a bit,
which is like very, very interesting.
It's like seems to be taking less risk in closing positions earlier.
But yeah, users, you know, can copy trade or inverse trade in any of these
within essentially AgentFi 1.0 copy trading managed strategy, whereas these LLMs themselves
are kind of acting on the AgentFi 2.0 where they're controlling their own wallet and executing
autonomously and no humans are telling them what to what markets to go in or when to close
their trades. So like over here on the
side, we can see that maybe not every single agent is in a position on all these markets, but they're
continually analyzing them. So, you know, Clarity Act right here, we can see only Claude and Grok
actually have positions in this. But all of these different agents are like continuously looking at
this specific prediction market and putting their input in on
it. So even if you're not using this as a copy trade, maybe you just want to see what the LLMs
are saying. So like every, you know, every little hour here, all the agents are updating their
perspective on all these different markets. So this is kind of interesting to see, like,
from a non-human perspective, what do agents think about this? And I guess something, something
really interesting that Dex has brought up in the past is, like
right now, we're kind of we kind of run on benchmarks for all of
these different models, like how fast can they code this specific
project? Can they like, understand, you know, healthcare,
doctors, surgical tasks, like how well do they
answer these high level questions, like astrophysics, like all these different, like, you know,
hardcore specialized like PhD tests, like how well do they perform?
And that's great as like, maybe a layer one benchmark model.
But I really like this idea that Dex has mentioned before that a kind of new version of benchmarking
is like, well, if you give all
these LLMs the exact same kind of directory, the same real funds, and you set them loose on a much
more complicated market than just trying to answer these research questions, like, you know, I think
a prediction market is kind of the max pain point, in a sense, for humans and for agents, like how
well can they actually perform in the real world with unknowns and like, you know, unlimited variables.
And I think the profit arena is kind of doing this case study in a way of this new type of like indexing or comparing of these LLMs.
So that's really interesting to see play out in like how they actually perform in the real world and not just answering test questions.
I don't know if maybe Dex, if you want to expand
or have a little more like idea or perspective on that.
Yeah, I think what AI is really good at
are compression and high dimensional information.
For example, especially in the case of production markets, there are so many real events that are
going on and the real-time updates that are going to affect the outcome of specific events.
And eventually, I think it just becomes impossible
for human analysts to analyze all the information.
And this is perfect use case for AI agents.
Yeah, yeah, thank you.
Yeah, yeah, that's like, I mean, it's on point.
The future is coming here.
I guess we're gonna like wrap it up here in a second.
But yeah, James, like, I guess, what's the final takeaway we should all know?
We can already see Polygon is doing the vertical open money stack.
It's happening real time.
X4O2 payments all-time high.
All these safes coming out, which may be a little bit secret,
but I'm sure some news will come out soon.
The tokenization in T-Rex rails as well.
I guess what are your final thoughts and views for the future?
Yeah, yeah.
So the last thing, a nice little call to action is we recently released a Polygon Agent CLI.
So I definitely recommend to check that out as well if you haven't already.
I can put a link here shortly.
But basically, it allows you, you get a wallet with your agent, you get access
to like a RPC, to an indexer, you can fund your agent via various on-ramps and
like crypto transfers which is really nice. Yes beautiful, you're already
shilling my my X thread. And yeah and it allows you like a really streamlined path to, you know, deposit
to DeFi, you can register your agent with 8004, but this is really the way to get started with
using your agent on Polygon. So this is great. And it has some nice like features where it
provides some safeguards for your agent. So just, you know, is not sending money to random people.
It has some boundaries there.
So that's what we're focused on right now.
And, you know, into the future, we plan to expand that more,
add some more services, discovery layers,
and ultimately make it more useful,
as well as integrate it more with OpenMoneyStack.
Beautiful.
I'm going to have to take a much deeper look at the CLI there.
Yeah, yeah, yeah.
It was perfect timing.
I just pulled that tweet up.
Yeah, and Dex, I guess the same question to you,
final thoughts and closing words,
but also maybe you have a little bit of alpha or special thoughts you haven't shared before
on some of specialized RPCs and info that's agent specific here.
I think right now the biggest challenge for people to really give their AI agents some money or give some real meaningful money is that you never expect if your agent is going to read something random on Twitter.
People are going to poison the prompt and your AI agent will send the entire money to others.
So this is what's stopping most people to give their AI agents some serious money.
Most people, even for me, when I use open call, I only give the AI agents less than $100
because that is the amount I'm able to
to lose but more than that I'm not comfortable than any agents to allocate the entire of my
portfolio so actually we're building a product that allows people to set a spending limit for
their AI agents and yeah just stay tuned
epic yeah the guardrail is coming in because i think that really is like the kind of key concern
of people is that trust issue so um i think there are going to be a lot of different ways to tackle
that trust issue and and one of them is for sure guardrails on spending and more hard rail controls around what the agent's doing besides these soft MD files.
I think a lot of people overemphasize the MD file is going to keep your agent on track, but an agent can still drift from that versus these hard guardrails that can be in place on top or underneath the soft rails in a sense.
So yeah, it's all coming to fruition it's like crazy
time to be alive like every day I'm literally seeing more and more agent 5
like brand new things getting pushed open source just so many amazing repos
getting shared on Twitter it's yeah it's an amazing amazing century we're we're
in right now like I don't know we. We're becoming the aliens in a sense.
The tech is going.
So yeah, everybody, this has been an amazing hour.
Make sure to throw everybody a follow.
Throw James a follow.
Throw OX Polygon a follow.
We're at at semantic layer.
Check us out.
Throw us a follow.
So yeah, everybody, let's keep building around agents, payments, info, everything else.
It's going to be a very beautiful future.
We got to guide it in the direction that we want.
No Terminator vibes.
We can keep it clean.
So yeah, I guess any final words or any outros, guys?
All good from my side.
So yeah, download the CLI, let us know what you think.
Any feedback?
Happy to add it.
Awesome.
Check that CLI out, guys.
It's linked directly on James' Twitter and retweeted from Polygon very recently as well.
Beautiful.
Well, thank you guys for coming on today, and I think we'll end the stream here.
So anybody watching, drop comments and questions,
and we'll get a bunch of clips out in the future.
This has been super fun and stoked.
Thank you, James, for coming on.
Thank you, Dex, as well. I know it's late for you.
We're all over the world right now.
So until next time. Bye, guys.
Awesome. Thank you all. Bye-bye.
