Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Dev Office Hours | AI and Move on Aptos with Kana Labs and RNDM
Recorded: Dec. 16, 2024 Duration: 1:01:42Space Recording
Full Transcription
We'll give it a few minutes before I introduce R&DM and Kana Labs.
Oh, nice.
It's a really nice city.
How about yourself?
I'm currently based in Toronto,
like kind of between Toronto and New York.
But I was in Korea for a year,
so I speak Korean,
at least like the basic level.
And then before that, I was in London.
That's why I spoke of Manchester, yes.
But most current labs team is based, completely DeFi.
So we guys are everywhere.
And I think we can't hear Brian speak.
He's not in the chat anymore.
So we'll wait for him to come too.
Okay.
Vijay, is this your first Aptos AMA?
Yeah, this is my first Aptos AMA, yes.
Oh, usually it's...
Oh, Brian is here.
Hey, Brian.
It was muted or it wasn't working or something.
So I was talking to myself for a bit.
We tried to fill the gap, Brian.
Yeah, nice to finally speak to all of you.
I think, yeah, I heard Servi say
this is like the first time for us meeting.
And yeah, Vijay,
same first time speaking with you as well.
Yeah, yeah.
Let me see.
Also, Brian, kudos to you for taking the initiative of speaking,
like actually pronouncing my name correctly,
or also even calling it out,
because most times people don't do it.
Like I remember I was at a panel where Avery was there,
and he literally was so scared to speak my name.
He just said,
and the gorgeous lady here, she'll be who.
Okay, well, I'm glad I got it right.
Yeah, we can just jump right into it.
I think most of the people that we want are here,
so we can get started.
Like I'm planning to have, you know,
the first 20 to 30 minutes
just kind of discussing about Kana and RNDM
and AI agents
and how, like, you know,
Kana can better support AI agents.
So, I mean, first thing is, like,
we can go ahead and introduce Vijay
and kind of what is RNDM?
Yeah.
Maybe I'll start with saying, like,
it's actually random,
and we just...
Okay.
Yeah, yeah.
Random, random, okay.
Yeah, yeah.
So, yeah, I'm Vijay.
I'm founder of Random,
and at Random,
what we do is we try to build AI agents,
and my personal experience is, like,
I've been in the space
and building AI since, like, 2014, 2015.
I was in...
Before that, like,
it was called Neural Networks at that time.
I was, like,
trying to do it more
for the automotive industry,
building, like, more, like,
models for, you know,
quality control
or for, like, you know,
development and so on.
More in the computer vision side.
And then, like,
I moved on more towards,
you know,
analyzing it more for traders,
and that's how I got into crypto,
like, more around, like,
21, 22,
you know,
quit my job
and moved full-time into the space.
Okay, nice.
We'll start,
and then we'll just move over to Surbi
from Kana.
Awesome.
Thanks, Brian.
And, again,
lovely to meet you, Brian,
and Vijay,
you as well,
and thanks for organizing this.
I'm Surbi.
I'm one of the co-founders
and also the head of business development
for Kana Labs.
And as for me,
I think I just come from
a very different background
as compared to Vijay.
I am...
I come from traditional finance,
so I got into
trading crypto
or building anything
related to trading
on blockchain
in 21,
2022.
Before
building Kana Labs,
I was working
with a company
called GSR,
which is a market maker,
and I used to do
investments as well.
At Kana Labs,
we guys have
expertise
or have been building
specializing in
tools
which allow for
regular people like me
and many others
who are not
the perfect DeFi
builders
or, you know,
users
to be able
to interact
with blockchain
and also benefit
from the benefits
it brings along.
On the other side,
why random
and not RNDM?
And we guys have
a lot of collaborative
future,
hopefully so,
because we are building
a lot of trading
bits,
such as a perpetual future
which currently
is on Testnet
and we will be
launching soon
on Aptos.
Yeah, nice.
That's great.
I mean,
maybe this is like
a good question
for our community,
but I know
most people typically
when they first
come into crypto,
they usually start
in like,
you know,
the EVM world.
I'm kind of curious
what,
I know for Vijay,
you said your
random is launched
on Testnet,
on Movement.
Is that correct?
Yeah, that's correct.
Yeah, yeah.
What kind of,
yeah, I'm curious,
like what kind of incentives
led you to launch
on Movement
and particularly
like why did you
try to try
and use like
the Move VM
versus,
you know,
just solely sticking
to like,
you know,
SVM or EVM?
maybe like,
maybe I'll start with it.
So, initially
when I first started,
right,
that's why I was
very much
like more
ETH focused.
Like,
the thing is,
because I saw,
you know,
like one of
Vitalik's speech,
this was like
early 2019,
and so that's why
I was like,
you know,
more towards the EVM
side and I was
like working on it.
And then I noticed
like,
you know,
I saw like a few
things from Rushi
and others
or like about
the Move part
and then I thought
like I was just curious
I have to try this out.
I just had a look
and it was like,
this is really
interesting.
like actually
like actually
there is the same
like address
and data
it's all
treated the same way
and then
that sort of
like,
you know,
led me down
the rabbit hole
of trying this out
and then
building it,
I guess.
Okay.
Okay.
Serby,
do you have any
comment on like
why you decided
to go with
like the MoveVM?
I would,
I think,
I think
there's several
reasons,
right?
like I come
from BD
and not from
development
but maybe
I'll give a little
bit like
when we made
the decision
of actually
building on
Move
was,
we started,
we have experience
on EVM
as well as Rust.
from the developer
side of things
like it was quite
obvious that
a lot of developers
were already built
even us CTO
and others
like they felt
that Move
was such
a much easier
or a dev
friendly language.
On the business
side of things
we realized
that one of the
key issues
which we've seen
like when we
are running
similar products
on EVM
sides
like it could
be quite
expensive
the gas
itself,
right?
Whereas Aptos
was already
or like
resolving that
issue for us
in this sense.
we sometimes
pay as little
116th
or even like
smaller portion
even from the
if we compare
with L2s
on EVM.
That was one thing.
The other bit
is all security.
for us
like when we
are building
for the masses
or we are,
I know it's a
very,
overused term
but for us
it really
means a lot
bringing the
next billion
on VET3.
These two
things like
when it comes
to security
as well as
the cheapness
or like
not cheapness
in the cheap sense
but rather
just having it
like,
you know,
you're pocket
friendly.
Aptos
provided that.
for us
it became
a very simple
choice.
Now,
on the
BD side
of things
and the
community side,
the team
is so
well-knit.
Everybody
at Aptos
is so
helpful
that
a lot
of the
problem
that
have
might
become
more
difficult
unless
we're talking
about Ethereum
particularly
to actually
gain access
to that
knowledge,
to build
something
faster
and to see
what we can
resolve
quickly
and how we
can build
for the
people
or users.
I think
there were
several
of these
reasons
why we
started
doing
and we
have now
built
almost
everything
on Aptos
versus
anything
being on
or any
other
chain
so far.
Nice,
nice.
Yeah,
I really
appreciate
the support
and giving
Aptos
move a try.
I think
very excited
to see
what you
guys build
out in
the future.
I know
with
Kana Labs,
in addition
to scanning
through the
websites,
you guys
offer a
Perfs
exchange,
but you're
also working
on developer
tooling and
infrastructure.
Yes.
Do you
want to
help explain
the products
and the
decision to
make consumer
apps and
dev tooling?
Yes,
I think
that's the
most exciting
part and
I think
we can
be very
genuine
about it.
Thanks for
that question
because it
can be
sometimes
overwhelming
for people
who come
to the
website
and what
are we
building?
Why are
we building
so many
things?
Let me
start from
the first
part.
I think
dev tooling
is where
expertise
lie right
at the
beginning.
I think
we have
been one
of the
teams
which
also
when it
comes
Ethereum
where we
brought
account
abstraction.
Our CTO
was involved
personally in
that and
is still one
of the
co-developers
over there.
Similarly,
for us,
we want to
solve the
problem for
liquidity
itself.
When I
say liquidity,
I'm not
talking on
the trading
side of
things,
but user
friendliness,
how do you
resolve that
someone who
is on
with ETH,
how do they
interact with
something on
Aptos without
leaving their
wallet and
doing several
of those
things?
It became a
natural course
for us to
build those
tools out for
a chain of
our preferred
choice so the
user itself
does not know
which chain
they're on.
That becomes
our passion
tooling and
infrastructure which
is required.
Now on the
trading side of
things, of
course,
infrastructure is
not sexy at
all.
It requires a
long time for
anybody to
start really
understanding,
especially on a
very basic
level, how do
you use that
infrastructure?
So we've had
support from a
lot of partners
like from
SKT Telecom in
Korea to
Lotte to
others where
they've used
a Paymaster
which is a
gas-sponsored
sponsorship which
is built by
Kana Labs and
as well as
the SDK for
aggregator where
we have
cross-streamed
for over
non-EM to
non-EVM as
well as
non-EM to
EVM side of
things.
But in
addition to
everything else
we also
realized that
a lot of
our tools are
so specific
to Aptos that
we can
really use
our tools to
build the
trading tools
which everyday
people can
use and we
started building
those out.
So it was
more like a
test.
It kind
of worked
for us and
then we
got a lot
of support
that this
is something
which is
missing in
the ecosystem
and we
started building
our tools
for traders
to come
on board
and also
enjoy what
is being
offered on
Move as a
language as
Aptos and
how it can
mimic or
almost out
par in the
near future
and at
some point
replace the
centralized
nature of
trading or
rather replace
Web 2 with
Web 3.
Great.
I know you
highlighted
Paymasters for
a bit.
I think this
is an amazing
feature.
Essentially,
if it's your
first time
interacting with
Aptos on
chain, if you
want to execute a
transaction, you
wouldn't have to
pay for the
gas fees.
Traditionally, you
need some APT in
your wallet.
You have to go
through Coinbase,
you have to KYC,
you have to convert
your fiat back to
APT.
So I think
Paymasters is great
and I'm really
glad you guys
have that as a
feature in your
dev tooling.
We can also
move over to
Vijay.
I know you're
working on AI
agents.
Yeah.
I know you have
a few different
agents running.
I don't know if
you want to dive
into the different
agents that you
have and any
alpha you want
to release.
I think a
majority of the
people listening
are very interested
in the intersection
of AI and
crypto.
Yeah, and
maybe I'll
start with the
preface.
This is the
crazy part,
right?
In 2023,
when I was
pitching,
when I was
explaining about
agents,
most people
actually didn't
even know the
term agents
or they just
confused it with
bots and they
didn't even
understand it.
And it's
curious how
some narratives
evolve and now
everyone talks
about agents and
I was like,
where were you
guys when we
were building
this?
So we have
been actually
building this for
more than a
year.
agents and
the idea
basically came
from, you
know, if you
have seen in
the Web2
world, like
what is
happening in
that, all
agents are
governed by
action.
So that's
the ultimate
goal, right?
Like you want
to make sure
that you can
process actions
and do that.
And that's the
key differentiator.
Whereas compared
to a bot which
just mindlessly
performs what
you're told to,
there is some
form of
intelligence in
the automation,
so to speak.
And that
enables us to
make like unique
workflows.
And that's the
key part about
any agent,
right?
Like you want
to have unique
workflows,
something which
is like both
A, which,
you know,
relieves human
burden, and
B, which also
like, you know,
comes up with
unique ways to
attacking or
solving a
problem.
And so when we
built this,
our goal was
like, or my
vision was like
to have each
and every DeFi
primitive.
So it's not
just about like,
you know,
building for like
one specific part
of DeFi and
then like, you
know, just
leaving at that
and scaling
that, but
it's to scale
like horizontally
in the sense
like go across
each and every
DeFi primitive.
And that's the
reason why we
have like multiple
agents, because
each agent targets
like a use case.
And so with that
in mind, our
first goal was,
you know, the
harder goal,
which is Perp
Exchanges, and
that's why we
built Atlas,
our first agent.
And with that
agent, the goal
was to basically
like maximize
taker volume,
have like as
much taker volume
as possible.
and, you
know, volume
equals attention.
And that's the
way Atlas, you
know, gained.
And initially when
we were doing
this, people
actually said
like, you know,
you use, you
know, centralized
market makers for
this, only they
can do it and you
need like specialized
quants.
And none of us
were actually
quants.
I mean, I have
done trading, but
I'm not from a
Trad5 background.
But it's interesting
that, you know, we
could go shoulder
to shoulder or head
to head with, you
know, the top
guys and Hyperliquid
was our first test
case.
And by the
way, again, when
we were building
this in Jan,
Feb, 2024, like
Hyperliquid was
not as well known
as today.
So that's how we
started with that.
And now we are
just scaling it.
So we have like
more and more
agents.
We have like an
agent which is
targeting, you
know, for
example, like the
cash and carry
trade.
Then we have
like another
agent which is
focused on the
prediction markets
and another agent
which is focused
on other DeFi
walls and so on.
So that's sort
of our goal.
So whenever there
is a DeFi
permit or a DeFi
problem,
we want to
replace that or
enhance that with
an agent.
Okay, so you're
saying your
Atlas agent is
primarily like a
taker on Hyperliquid?
Yes, that's what
we built because
at that time,
and this is the
crazy part, right?
At that time,
when we are giving
the objective
functions on the,
when we are building
that out, it
actually said,
you know,
there is on the
small off chance
that the point
value is extremely
high, we should
actually sustain
some losses in
the trading so
that we maximize
the thing, right?
And we thought
like, no, that
doesn't make sense.
So we actually,
this is where the
human intervention
is sometimes not
so good, right?
So we focus more
on like PNL0
or close to
higher value.
So we built it
like that and
then in the end,
you know, like
actually, oh my
God, we should
have, we
shouldn't have
optimized it, we
should have sort
of listened to
the model.
So that's,
that's the
interesting thing.
So the AI
actually sometimes
comes up with
counterintuitive
answers.
I see, I see.
Do you want to
try and explain
to the audience
like what it
means to be like
a taker, like
of Hyperliquid?
I feel like maybe
some people who
are not too
into training.
Maybe I'll give
it a more
generalized
explanation in
any purpose
exchange, right?
So if you look
at the order
book, like let's
say you're trying
to buy a bottle
of soda, right?
You're going to
have like hundreds
of people buying
soda, you're going
to have like a
seller, let's
say a marketing
person or whoever
who's going to
like, you know,
sell that through
a distribution
outlet and you're
going to arrive
at a fair price.
So if it is a
dollar of soda,
I mean, I'm not
sure if it is a
dollar, maybe it's
two dollars.
So anyway, it's
two dollars a
soda, then you're
going to have
like, you know,
someone offering
a 2.02, 2.03
and on the other
side, you're going
to have 2.05, 2.06.
So you're going
to have like a
very liquid market.
Whereas if you
look at, let's
say a painting,
like say Mona
Lisa or something
like that, that's
going to be very
illiquid in the
sense like there
will be someone
offering, let's
say a few million
dollars and then
on the other side
someone will say
there might be
only one or two
buyers who might
say one million.
so you have
these extremes
in marketplace
and that's where
market makers come
They try to like
normalize the flow
so that you have
like intermediate
prices so that when
you go and go into
an order book and
have a look at it,
you then get to
choose whatever price
you want, select that
and buy a product
or sell a product.
So the maker side of it
is making the books
in the sense like you
are able to like set
all the different
prices and then
the taking part of it
is like people who
are on the other side
who are going to
go and buy it.
And this is where
the market making
slash taking comes
in where like you
have to set up
both sides of the
book so that you
have like a liquid
place so that you
go in there and
you're able to
select on
illiquid assets.
Right, so taking
is primarily just
like buying things
off the order book.
Yeah, buying things
or selling things
from that maker side.
Yeah, that was it.
Yeah, yeah.
So yeah, it was
interesting because
I've talked to
like market makers
like who interact
with like DeFi
applications and
they're primarily
only takers
on like A&Ms.
So it does make
sense like with
Atlas, what criteria
did you set for
like the agent
to be like, okay,
like from here
we'll take this
price.
Or like how
did you, like
what, yeah, what
set of rules did
you set for the
agent to like take
orders off
hyperliquid?
I would say it's
not rules, it's an
objective function.
It seems like a
small difference and
this is where like,
you know, things
changed since 2023
because before that
what you had to do
was you had to have
like a set of if
else conditions and
then like, you know,
you sort of like you
could still get some
form of thing like,
you know, with simple
rules, you have like
a form of emergence.
But what you had to
do was define it
more like an
objective function.
So you'll have like
a series of things,
objectives which it
has to hit.
And based on that,
it will try to make
some set of actions.
So it's not having
like intelligence or
it's not trying to be
like, you know, do
that, but you still
like, it's not very
clearly defined as
rules.
So you have like set
of fuzzy objectives.
And that's what we
define.
Yeah, yeah, yeah.
That's what makes it
more complex, right?
Otherwise you can just,
you know, anyone can
run it.
I guess I'm approaching
it from like not the
AI agent perspective.
I say like what set of
rules we're using.
In that sense, like I
know recently there's
been a lot of these
new agentic frameworks
like, I mean, the
biggest one being
Eliza.
And then, you know,
there's the one from
OpenAI called Swarm.
Like have, has your
team used like any of
these frameworks or
have, do you guys
like build this in
house?
And then kind of
which, and like a
second question is
like kind of which
foundational models
have you seen work
the best with AI
agents?
So the thing is,
right, if you use,
for example,
Elisa or any of
those other
frameworks like
Swarm and so on,
what you run into
is like, you know,
as soon as you
connect it directly
with say a wallet
address or something
like that, then you
have like a huge
risk.
You saw that happen
with some of the
virtual agents as
well, like, you
know, they had some
leakage.
So that's why DeFi
is quite hard if you
connect it directly.
Second thing is you
can go try this and,
you know, even now
with the most advanced
models, you can go
ask ChatGPT, for
example, like, you
know, what will
happen with, you
know, the Bitcoin
price or what will
happen with this and
so on.
And it will fail
because inherently
LLMs are quite
imprecise.
So that's why you
need to actually
have more like a
sort of like one
kind of like a
quantitative focused
model and then you
pair that with an
LLM to get like a
unique, what we call
as like a knowledge
graph.
So just imagine
it like, I mean,
knowledge graph is
sort of like it might
seem like a fancy
term, but basically
all you're trying to
do is you're trying
to codify the
actual part of
like, you know,
let's say an
order book, what
happens in an
order book, combine
that with what
happens like if you
use like an LLM
which has like
already a lot of
information about
say Twitter data,
sentiment data and
so on, combine that
to build like
different ways in
which you can like
use this for the
agent.
So that's what
we do.
So we have
existing, so we
use part of their
framework, for
example, for the
social side and
then part of it
is our own
framework.
I see, I see.
Yeah, I know
like kind of, you
know, AI agents
are great and
these like
foundational models
can, you know,
you can chat with
it and they can
come up with
responses that
surprise you and
they're very
convincing, but
I think what
some people don't
realize is sometimes
these models can
like hallucinate
and they could
very easily
like, you
know, liquidate
all your funds
or, you
know, transfer
to like a, you
know, a wallet
address that's
off by one
character or
like how, yeah,
how from when
people say AI
agents and
trading, I get a
little bit
skeptical because
from like an
engineering
perspective, you
know, these
models can
hallucinate.
Like how do
you prevent
agents from
going rogue and
like, you
know, just
sending money?
Yeah.
Exactly.
And that's
why we kept
it the
Firewallet in
the sense
like you
have one, the
agentic framework
should have
like these two
different flows.
The LLM flow
is mainly for
getting sentiment
more like a
user-like data
and whereas
the other
part.
So it's
almost like,
you know, having
a person like
you, for
example, maybe
like, I
don't know,
like, you
know, 15,
20 years back
or maybe,
sorry, maybe
even before that
when people
add like, you
know, like when
you are given a
problem, they
try to use it
using a
calculator or
try to, you
know, do it
differently and
then the other
part of the
thing, like when
you have like a
social question,
then you answer
it, right?
So imagine it
like that.
So it's more
like right half,
left half kind
of a thing.
So that's what
we try to do.
So we keep the
quantitative part
separate from
the sentiment
part.
I see.
Nice.
Okay.
Super cool.
Do you want
to try and talk
about like any
benchmarks you have
with your agents?
I know you
have, you
released some on
the Twitter
account.
Maybe you can
share it with
here as well and
you know, promote
some of your
agent products.
Definitely,
definitely.
Yeah.
So our first
agent with
Atlas, we
reached I think
close to like
200 million
overall volume
over this year.
And that has
been achieved
with like a
really low TVL
value, like
close to like
250k.
And that has
been achieved
by basically
like using this
agent, using
this unique,
our own model
where like we
are just, you
know, making
sure that the,
you know, the
objective functions
are PNN equal
to zero and
so on.
So now what
we are doing
with the next
set of models
is like doing
portfolio management,
doing like a
delta neutral
trades and so
So for that,
we are sort
of customizing
these models
and the
benchmarks for
that is more
like around
like, you
know, capital
efficiency.
So keeping TVL
as low as
possible and at
the same time
having this.
Yeah.
Nice.
Okay.
Great.
Cool.
Yeah, that's
awesome.
I'm glad, like
I think you're
one of the only
projects kind of
really working on
AI agents on
move.
So, so very
excited to see
what happens in
the future.
Servi, I don't
want to leave you
out here, but
I know you have
like a middleware
SDK.
I'm wondering,
have you had any
developers try and
integrate AI agents
or do you think
there's any, yeah,
specific part of
your SDK that
could help with
this?
I think before
I answer that
question, may I
just add something
on the taker and
maker side, like
which is just an
addition.
I would like for
in a very lay
language for
people who, who
don't have as
high IQ as
Brian and Vijay
and who are more
compatible with me
in a way like
taker also allows
that's a real
volume.
like even in
traditional finance
wherever, taker
side is what
constitutes the
actual volume, not
the market making
side, because
that's the flow
which you would
want on an
order book or
trading, which
allows for retail
traders as well
as others to get
the prices which
they can actually
buy and which
they see on
news or which
they see on
like, you know,
like how they see
the real price
discovery taking
place.
And this is, so
it's very
interesting that
Vijay, you guys
start from the
taker side, because
most people even
on crypto right
now, most
market makers
they've not, they
still try to do
a lot of the
making bit because
that's where they
find money.
But for real
users to come in
taking is which
allows for a very
healthy order book
or liquidity to be
present in this
case.
I hope I made
sense.
I'm sorry.
I was, I really
enjoyed what you
guys shared and I
just thought like
like my
perspective, how I
would interpret
it and how this
would benefit people
who are over
here, like I
see from Mohan
to Dave to
Wajid to a lot
of people, Raj
also I've seen
like a lot of
people trade and
how that would
allow for them to
actually trade and
use this.
So I think it's a
very big part or
missing part of the
puzzle, which was
not there.
software and it
does allow, especially
in support for
users.
So not to be
taking everything
from institutional
trading bit.
Now coming back to
the SDK, sorry
Brian, I know
that's what you
had asked me, but
I got really
excited.
That's like the
first part is what
really excites me.
But yeah, from the
SDK side, yes, we
are trying to see how
we can actually use
it because as you
rightly mentioned, we
have an SDK
middleware where
people can, where
projects can use it
to swap between
different chains and
other things.
However, with what
Vijay is building
like on the
preliminary looks and
most of our team is
devs, we were
looking at how it
could actually allow
for a lot of people
to simply be able
to like in the
future, possibly
like simply, you
know, give a
command that they
would want to swap
from one place to
another or similarly
find like decking
DEXes, since we
aggregate a lot of
DEXes, which have
the best liquidity
and still find the
cheapest or the
lowest slippage in
order to get that
swap done.
I hope that was what
the question was and
that's how we see
like basically random
being integrated or in
some way like AI
models being used in
the cross-chain bit.
However, I'm very
happy to explore on
other sides like the
actual trading, not
just the middleware
part, like how we
see this being much
more effective,
particularly built for
the users or the
retail trader that we
see on a platform.
Nice.
Cool.
Okay.
Yeah, I mean, yeah,
I mean, this AI agent
narrative is like very
hot, right?
And so like if we can
in some ways like
support tooling to
allow people to go
out and build like
their own and
especially since
Kana already has
their own purpose
exchange as well,
like how can we make
it like very easy
for like let's say
Vijay, like what
would it take Vijay
to put like Atlas
on Kana's perfect
exchange?
So let me, Vijay,
before you answer,
maybe like let me,
so one thing which
like we are looking
at or we've been
seeing is that how
can like most
traders benefit from
some trading
strategies which are
present and they
would need to do a
lot of manual work
to figure those
out and I think
that is one place
where we see AI
agents to primarily
be able to, you
know, produce that
kind of a strategy
and maybe also use
the liquidity part
of it which
incentivizes the
retail to engage
in this form via
Kana and how we
can always like
integrate random
itself and Vijay,
maybe I'll let you
build on it and
then add further
if that's okay.
Yeah, I mean,
I would say the
thing is like if
you look at it
right now, right,
like Hyperliquid and
the other exchanges
they are like really
big and one of the
reasons is like,
you know,
easiness of like
integration.
So for us to go
from zero to one
was like, you know,
less than a week,
like I would say
like two, three days
and then to go from
one to ten was
again like really
fast.
So it's these fast
iterative cycles
because like our
focus should be more
on like, you know,
how do we attract
like, you know,
how do we make
sure we get the
liquidity and how
do we like, you
know, use our
agents, how do we
make sure that we
don't have execution
speed which is
quick and all those
kind of things,
right, that's what
we want to be
focusing on, not
on the actual
like, you know,
integration with the
middleware which
takes forever or
stuff like that.
So I think that's the
key part, like, you
know, really simple
things like making
sure the web
sockets are good,
making sure that
there is like a
really good to use
Python SDK.
I mean, an SDK
which doesn't fail
under like heavy
load and so on.
So I think that's
where like I would
say a lot of the
perp exchanges fail
and that's the key
part.
So I think,
thanks Vijaya for
that.
Like, firstly, the
second part of it,
this is where we
saw on the perp
deck side.
So our middleware
and the perp decks
are completely two
different products for
the listeners,
which I know this,
just to clarify.
On the perp deck
side, that's how we
see integrating
random, right?
Like, first of all,
just testing like how
it would allow for
the load to be
taken.
Secondly, also, I
also see, I don't
know, Vijaya, how
do you see this?
But I also see like
a lot of like, I
know you said the
taker side, but also
the market making
side, right?
Providing liquidity
where there might be
strategic bids,
especially in low
liquidity areas where
you can pull in the
liquidity.
So I see two parts
of it.
One is the
infrastructure support
itself where the
SDK or rather it
can be integrated
with other platforms
or rather other
devs or traders
bids.
bids, whereas
and the other
side being the
low liquidity
areas of token
which might, not
token rather, like
perpetual futures
which can, you
know, utilize where
there's nobody else
willing to provide
liquidity or also
take, but, you
know, retail has a
demand, basically.
So that also being
in the future, like
some form of
integration or
collaboration.
I don't know how
you see that, but
that's something which
would be really
exciting if that's
possible.
Yeah, I mean,
definitely, like, you
know, it just
depends on, like, the
SDK and, like, how
easy it is.
It's always like, you
know, the 1 to 10
stage where it is
get things get
harder.
The 0 to 1 is
almost always the
same.
But, yeah, that's
where we found the
issue.
So just answer to
Brian's question.
Great, cool.
Yeah, I mean, switching
topics a little bit
before we open it up
to community guests,
I want to ask, you
know, one final
question to Serby and
Vijay.
Kind of give us a
teaser for, like,
what's next for your
projects?
Do you have any
launches, partnerships,
milestones, like
anything we should
keep an eye out for?
You know, give our
audience a chance
to speculate.
Vijay, do you
want to go first
or should I?
No, I can go first.
No problem.
Yeah, I would say,
like, what we are
trying to do right
now is, like, you've
seen with virtuals and
you've seen with
others that the
fundamental tokenomics
has changed with
agents because before
this, it used to be,
like, projects would
have, like, a token and
then they would launch
a token and that's how
users would get access
to it and build on
that and so on.
Or your projects would
be, like, you know,
layer 1s and so on
where, like, you know,
there are multiple
projects building on top of
that everything was
more, like,
infrastructure.
I think with AI
agents, you see the
first whiff of, let's
say, consumer agents
or consumer-facing
apps where, like, each
agent is going to have
its own unique
identity and also its
unique thing and, you
know, one agent might
be in one chain whereas
another agent will be in
another chain and it
will then have its own
token but all of them
will be bound together
by the same kind of,
like, one single model
framework.
So I think this is going
to evolve and if you
were to ask, like,
what's going to happen
over the next year or
two years, I think this
is sort of, like, my
bold prediction.
Maybe I'm completely
wrong but I think two,
three years from now it
will be the agent
frameworks and not L1s
people look at.
Wow, did we just,
hold on, did Vijay just
say AI agents over L1s?
Yeah, I mean, I'm just
provocative, I know.
I didn't want to tweet
about it but I nearly
tweeted about it.
Brian, how do you see
that?
How do you see that?
I think let's switch the
question over to you.
I love it, I love it.
It's a huge claim, I love
it, I mean, especially
coming from Aftus Labs
as a L1, so I, yeah,
made sure I caught that
but, yeah, I like that,
Vijay, you should, you
should tweet that.
You should tweet that.
Oh, man.
And I think maybe, Brian,
like, before, after I
speak, maybe we can,
like, then speak to you,
ask you the question,
how do you see the
future, like, from the
L1 perspective, right?
I mean, the future, the
future is AI agents on
move, like,
AI agents just move
better on Aftus, so
that's the future I
believe in.
But, yeah, Servi, I'll
let you, I'll let you
go next.
Thanks, Brian.
And, and, I, like,
leaving aside the last
statement, made the
vision, like, I do, you
know, Kartik, when we had
the Aptos in there, we
made the announcement
that we will be launching
intents on Aptos, which
is very much, like, from a
technical perspective,
intents would require
solve words, which I
feel is what Vijay
described, so which
would mean that, like,
on Aptos, there'll be
seamless, like, liquidity
accessible to people from
across DeFi suites, let
it be Ethereum to
Arbitrum to Sui to
anywhere else.
So I think, and I don't
know, Vijay, if you've
spoken about this, but I
think maybe offline, that
would be one way how we
could even use the
framework of random
itself as one of the
solvers for intent.
And for people who speak
English, like, intents is
nothing, but similarly, we
are building it on Aptos,
which is simply trying to
solve the problem of
liquidity itself.
So which would allow you
guys to access any, let's
say you want to access
Kana perpetual futures with
ETH, and you'll simply be
able to do so, and
hopefully random will be
one of the key
infrastructure which we
could use to perhaps
access it.
So that will be more
like agents on Aptos
solving for liquidity
across globally.
On coming back to what
we are actually hoping and
which we are seeing right
now, is that we'll be
launching a perpetual
futures about, I think,
in the next month, and
we have a test night going
And coming back to the
very first question that
Brian had asked, like,
we're also hoping with
random, we expand a
cross-chain, you know,
middleware using an AI
agent-friendly tool, so
that allows for an easier
setup for users and also
like a much easier way to
access that bit of
liquidity, which is
separate from what we
intend to do with
intents itself, the
double word.
But yeah, this is how we
see it, and we're happy to
answer or also ask, like,
listen or answer any
questions from the
community, otherwise as
well, what we should be
building and what is it
that you're looking on
Aptos to be built by
Khan Lab, which we can
build for you.
Cool, cool.
Great.
Yeah, I think from now we
can open it up to any
community members that had
any questions, and we'll
leave, yeah, we'll leave
like the next 10, 15
minutes for anyone to
ask questions.
Okay, we've got one.
All right, I'm inviting
Baba.
Okay, inviting Greg.
Hey, guys, can you hear
Yep.
Yes.
Awesome, great space
today.
Thank you for putting it
So I just wanted to ask a
couple of questions.
So we've seen a huge
influx of interest in AI
agents lately in the
crypto space and outside
of it.
Shout out to the team
behind the Aptos plugin,
by the way, for the
ELISA framework.
It's really cool to see
developers enabled to
build AI agents that can
manage Aptos wallets.
And shout out to you,
Brian, for the text and
video tutorials you put
together on building AI
agents on Aptos.
So with that, I wanted to
ask, are there any
particular types of AI
agent projects or
functionalities that you'd
love to see come to life
on Aptos, something beyond
the current crop of tools
and experiments?
And from your perspective,
how do you see the future
of AI agents evolving,
both within the crypto
ecosystem and more broadly?
Are we heading towards
like a landscape where AI
agents become more
autonomous market
participants or even
entirely new classes of
decentralized applications?
And once again, thank you
very much for the space
today.
Yeah, I can take a stab
from the Aptos side.
I think
for AI agents, we're
really happy to, like,
it's like an entirely
new use case.
So we're very open
minded to, like, you
know, anything new that
developers want to make.
Like, if I were to, if I
were to say, like, what
exactly to build, I don't
think it'd be, like,
entirely accurate, right?
Like, you know, one
example is, like, pump,
pump dot fund, right?
That was a very, like,
Web3 native type of
application that I don't
think many people could
have predicted.
So for AI agents, I think
it's going to play out
the same way.
I think, you know, right
now people are trying kind
of, like, the pump dot
fund for AI agents, but I
think, you know, in a
couple months, we'll see
this, like, new use case
that, like, that is very,
you know, authentic to
Web3 and will take off.
Maybe Vijay can, like, add
some better light to it.
Yeah, definitely.
Yeah, I mean, really
curious with the way it's
evolving and, like, another
interesting thing is the
agent-to-agent payment
model and where, like, you
know, instead of having,
like, you know, user
paying, user interacting
with an agent and then the
agent then goes out there
and then, like, you know,
does some action, you could
have, like, a user delegating
a task to an agent and then
the agent has, like, a
sub-agent.
We already have, like,
multi-agent pipeline in our own
project anyway and imagine,
like, how this could evolve.
Like, you could have, like,
you know, a network of agents
and then several levels down
and all you see as a user is
just at the top level and this
is quite similar to what we
see in the real world, right,
in terms of, like,
distribution networks.
Like, if you go to buy
produce in a supermarket,
you don't know, like, how many
levels of things are going on
and to someone, like, say,
40, 50 years back, like, when
I asked, like, my grandma,
grandpa, like, they would
be, like, you know, like,
you know, even a supermarket
is, like, a marvel and that's
how the pace is going to be
with agents.
Like, we are going to see
totally unique use cases
simply because of multi-level
network of agents.
Oh, and I think, if I may,
I feel like, if this is how
I imagine and, like, Vijay and
Brian, please do correct me.
Like, I feel like if it does
grow in this way, then I do see
that there will be a lot of
competition or we'll see a lot
of the agents competing for
trades, arbitrage, like, more on
the trading side of things and
arbitrage opportunity as a
liquidity.
So it'll be more like a great,
I don't know if I could say
digital arms race, but instead
of humans, just be, like,
agents, like, basically
outsmarting each other in
milliseconds.
So that is where we could have
very hyper-efficient markets
and primarily, like, something
which we've seen, like, HFTs,
which were there built out in a
different way.
And I know there can be
conversations regarding, like,
you know, high-frequency
trading within crypto, like, in
blockchain as well, how that
would go.
But one way from the agent
side, the AI agent side, I see
that could also be something
which you could see in the
future, which would
inevitably make, like, trading on
blockchain, like, much more
efficient or maybe hyper-efficient
if that's a possibility.
That's actually a really good
analogy.
I didn't really think of that.
But, like, you're essentially
stating how, like, current market
making is really PvP, kind of like
it's, you know, Citadel versus
HRT or, like, HRT versus
Tower.
And you're saying, like, in the
future, it'd be, like, AI agents
versus other AI agents.
That's how, yes, I imagine it.
I feel like that is, like, the,
you know, like, I think Vijay
mentioned two things.
One is the tech part and the other
is the emotion part of it, right?
So, simply working with these
limitations and that would, they
could exhibit, like, the similar
trading patterns which we see in
these.
Like, if you've seen, like, Citadel
versus Jump, maybe because I come
from the trading world, like, there
was not a lot of strategy-based
trading but rather speed-based,
right?
Now, AI agents could actually allow
for the strategy as well as the
speed, which would completely
revolutionize.
And that is one of the reasons,
coming back to your very first
question, Brian.
Why did we choose Aptos?
That is one of the reasons why,
you know, Aptos VM is something
which we saw could, in the future,
build, allow, or make it possible
using the agents, like, for that
kind of thing in the future.
I don't see any other L1, you know,
allowing for that and that's why
we're hoping for intent framework
and seeing some collaboration with
random, like, allowing for this.
Right, that is true.
I mean, our time to finality, I
think, is, like, sub-second.
Yes, in reality.
Normally, like, 200 milliseconds, I think.
Right.
Greg would know the exact number.
So, glad Greg is here.
And I'd also like to add one more
thing.
Like, when I say hyper-efficient,
for everybody here, like,
HFT and traditional finance
allowed for a seclusion for people
who have studied trading, who are from
Wall Street, to be able to participate
in the market, Robinhood kind of
chained that for people like us
to be able to participate who are
not professional traders.
So, when I say hyper-efficient on
blockchain and particularly
with AI agents and
including Aptos here,
it means that
it will be easier for you guys,
for us to trade,
without being professional traders
and getting the right price in the
market, which does not allow us to
feel manipulated by it or being
controlled by a few select players.
So, it will become much more
inclusive, which allows more price
efficiency, even on the price
discovery level.
And also, hey, Greg!
Sorry.
Yeah, for those that don't know,
I've met Serbia in, like, three
different continents at this point.
That is true.
Wow, did not know that.
Serbia, I will have to meet you one day.
I don't know.
Where are you based, Brian?
Are you east or west?
I'm in the U.S.
I used to sit with Greg.
It's like, yeah.
But you're in the west coast, right?
I still haven't.
Oh, okay.
So, I worked with Greg in Aptos
off in New York, too.
Yeah, Greg recently went out there.
We have to meet.
So, we got, like,
10 minutes left.
Let's get Baba and Harsh
to be able to get a little chat.
Hey, GM, GM.
Shout out, Greg.
Shout out, Brian.
I'm getting really used to Serbi talking,
and I like it.
I like it.
I was speaking to her the other day,
and she had the most amazing insight,
and I see why you would have met her
in three different continents.
I think she's been all over the place recently.
Yeah, so, I like the,
I like what Canal Labs is doing,
especially with their pep decks right now.
I think it's really awesome.
But, yeah, I'm now going to talk from where,
would I say from the consumer side of things,
where everybody's really excited on.
Is there, okay,
so now one issue memes have is liquidity.
Like, that's why it takes forever
before you can have memes in pep deckses.
Serbi, would you say there is a point
we can get to that almost from the get-go,
memes would be on pep deckses?
Oh, yes.
I think, let me put it this way,
Baba, first of all,
hi, and I'm so sorry for speaking so much.
So, I do think there is,
and I think that point can be very much now.
So, in simple terms,
there's a price on chain,
which, what that means would be
if it's on peps, chain link,
or even on a dex,
or emoji coin dot fund.
If there is a price on chain,
there can be a perp around it,
simply put, from day one,
very much so.
Now, the second part comes
to providing liquidity itself.
Market makers is what,
who need to take the risk
of being able to provide that liquidity.
That's where there's still
a lot of inefficiency
because there's not enough incentive
for market makers
if, unless there are enough retail
already participating on that token.
So, it becomes a chicken and egg problem.
However, this is where I see
which is random,
being able to solve
for that inefficiency
in the day one launch of a perp,
like before even, let's say,
you have the spot
with the perp itself,
like launching,
because the AI agents,
if they are built in this form
and they're able to solve
these problems,
they're able to already
provide the pricing,
which is efficient enough
for the user,
like you and I, Baba,
like to be able to purchase
at not a very high price,
but rather like discovering price
from the very get-go.
I don't know what Brian
or Vijay have to say,
and Baba,
I don't know if that answers
your question,
but the answer is yes,
we're getting to that point.
We just need to make it
more efficient.
Yeah, awesome.
Thanks, Debbie.
Yeah, sorry about that.
Yeah, I think that pretty much
explained it,
and yeah,
I think it's optimistic now
and just looking into the future.
I think the space
will be ending soon,
so sorry I have to just shield this.
We are bringing all of Aptos
together tomorrow
on Spookspace.
I believe everybody
is going to be there.
We reached out to basically
all the projects
we could get our hands on
on Aptos.
You guys should check it out.
It's going to be awesome.
You don't want to miss it.
Okay,
it's my turn right now.
Yep.
Hello.
Yeah.
Yeah.
Hi, Brent.
So yeah,
I'm basically from India
and I'm a trader.
I used to trade
on many things
like Solana,
Sui,
Aptos.
On Aptos,
I use like Michael Trade.
There is a Michael Trade,
MKL token is there.
So I used to there
and the liquidity
they provide,
we get on that
is very good.
The leverage we get
is very good.
The,
what we say,
the cryptos
and the forex
and the community,
commodity side,
we get so much
of this perpetual
on this Aptos chain.
You know,
so second thing is
I am also using
and following
Kana Labs
from last one year.
I am also,
I have tried
perps there
and I have tried
swapping also there.
It's no doubt
the cross chain
is also good
and the same chain
is also good.
The thing is
the competition
is very high
on this market.
Same as
on Elvance also,
if you go on
Solana chain also,
they are also
finding very difficulty
on getting liquidity
like the project
like Drift
and Zeta market.
So they are also
getting very liquid,
much liquidity
problem on that side.
And same with
happening with
this SWE also.
I have tried
many DEX
perpetuals.
So how
Kana Labs
gonna solve
this problem?
Thank you so much.
I could not see
Haklu speaking
so I'm sorry
I can't say
thanks for
a good question.
Your name
basically.
I think
you've addressed
a very important
question.
I think this is
where Brian
had asked
a very important
thing like
we have the
infra tooling
side as well
as the trading.
So infra tooling
is where
we are hoping
not to just
be competing
for the same
pie which has
existed on DeFi
for a long
time, right?
Right now
whether it's
SOE,
Aptos,
ETH,
it's the same
number of
people who
are trading
and we're
trying to
snatch that
entire bit
from each
other.
Or snatch
might be a
more negative
word but
in a very
positive way
like trying
to entice
the users
to try out
the products.
However,
we at
Kana Labs
we are actually
trying to see
how we can
onboard people
who don't
already,
who are
already not
there on
Web3.
However,
they're smart
intelligent
used to
trading online
not on
chain maybe
and how we
can make
the user
a bit
very smooth
right for
them.
For example
like Aptos
Connect.
I don't know
how many of
you have
already tried
it so we
have been
using it
for a few
of our
products
where users
don't have
to actually
have a
wallet but
they can
connect using
their Google
sign-in
and others
which allows
a very smooth
transition,
right?
Of course,
that still
does require
some sort
of crypto
to be in
your wallet
to be able
to engage.
However,
there's a step
one from
a Metamask
versus an
Aptos Connect
in this sense.
And even
for Merkle
Trade,
they're doing
a direct
deposit which
is again
very user
friendly.
So to
answer your
question,
one way
which we're
trying is
not just
to fight
for that
same bit
but rather
allow
educate users
who could
really
were trading
otherwise
to come
what the
benefits
of DeFi
are,
like how
an FTX
like situation
cannot happen
on blockchain
in the
DeFi
manner
if you're
decentralized
and how
you are
in charge
of your
finances,
of your
wallet
and of
your
trading
when it's
smart
contract
based.
So that
is how
we are
doing
on a
very
high
level
happy
take
offline
explain
further.
On the
other
side
what
Vijay
doing
what
Random
doing
actually
also
trying
solve
that
liquidity
problem.
liquidity
comes
from
users
itself
onboard
those
users
other
part
actually
aggregate
that
liquidity
which
fragmented
across.
This
is where
we see
Random
also
hopefully
having
synergies
solve
last
part
which
infraside
which
trying
liquidity
building
intents
that
Aptos
because
feel
that
should
home
where
build
most
efficient
protocols
this
business
speedy
pitch
rather
from
tech
like
from
feedback
where
feel
that
liquidity
could
efficient
protocols
built
building
addition
Aptos
Connect
solving
connectivity
similarly
bridge
across
liquidity
know
this
specific
question
Kana
Labs
don't
know
Vijay
how do
this
it's
very
important
question
general
liquidity
part
similarly
Brian
don't
know
would
love
hear
what
have
even
Greg
yeah
mean
well
like
liquidity
of course
is like
an easy
part to
like
just as
previous
question
asked
about
meme
coins
agents
can only
solve
a certain
point
example
like
someone
wants
like
meme
coin
which
like
don't
know
5 million
market
100k
liquidity
agent
only
much
like
know
developer
decides
that
particular
token
then
obviously
agent
cannot
anticipate
that
only
look
order
book
know
number
sales
buys
match
that
same
they
garner
attention
look
current
right
like
example
hype
like
know
dominant
kaito.ai
then
look
other
parts
a lot
agents
doing
best
influencer
right
like
least
opinion
like
look
KOLs
just
AXBT
data
that
like
it's
more
like
more
nuanced
like
capture
attention
attention
then
capture
liquidity
yeah
I think
we ended
on a
great
question
I think
Vijay
made
a good
point
I mean
part
about
attention
right
then
intention
incentives
that
then
think
also
liquidity
Brian
that
note
still
have
points
program
running
Kana
Labs
until
30th
December
since
forgot
attention
anyone
wants
participate
please
yeah
exactly
it's
very
important
yeah
have
this
incentive
structure
think
that
wrap
want
huge
thank
Vijay
Serby
coming
discussing
Kana
Labs
Random
huge
thank
Aptus
Labs
Camille
who's
helping
host
this
Twitter
space
Greg
thank
yeah
hope
everyone
has a
really
good
holiday
really
good
year
go to
spook
space
tomorrow
spook
space
right
I know
Baba
and Yar
want me
to see
that
right
also
think
you're
looking
charts
like
happy
Bitcoin
all-time
highs
well
thank
everyone
thank
thank
guys
happy
holidays
