Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Building AI Agents On BNB Chain
Recorded: March 10, 2026 Duration: 1:10:39Space Recording
Full Transcription
Music Thank you. Music Thank you. Music Thank you. Thank you. Music Thank you. Hey everyone, let's give it a minute or two for everyone to comment.
Thank you. all right that's great good morning good afternoon good evening depending where you're turning in
from so bme chain you know for the past one year, we have seen superb growth in the AI and blockchain overlap.
And yeah, today, I'm Walter here from Business Development at BME Chain.
Today, I'm here to learn from four great builders on BME Chain.
So, you know, without wasting much time, let's get into it.
Let's get into introductions.
So perhaps, yeah, I would like to highlight well that um whatever discussed here is not financial advice please do your own
research especially in this you know volatile web 3 space right so yeah um i would like to
maybe welcome uh shaw from milady to give a introduction please
hey shaw feel free to introduce yourself and the project.
Can everyone hear me well?
Yeah, I can hear you.
Okay, great.
Sean, yeah, please feel free to introduce yourself.
Sure.
Can you hear me okay?
Yep, I can hear you well.
Okay, cool.
So, yeah.
So, my name is Shaw, and I'm the founder of, well, a few things.
AI-16Z was the first one.
You might remember that.
But during building that, we built the ELISA agent framework, and we've been building that for the last year, you know, just entirely.
And that's what is actually powering Milady,
which is an application that we've been working on,
which is like the autonomous application
of everything that we've been doing.
So if you, I'm sure people are familiar with OpenClaw,
Milady is very similar, like paradigmatically,
like to OpenClaw, but it's very focused on uh DGens people who like
crypto stuff uh we have a lot to do with with things like uh trading integration like AI based
trading integration as well as apps that your agents can come and play and do so that's everything
from like trading card game to uh full RPG to like prediction markets, all that kind of stuff.
So that's, yeah, I love building agents.
Everything I do is kind of open source
and I just post stuff on the internet
and then that's become a movement.
You know, now it's a token that's blowing up.
It's kind of crazy.
Yeah, thank you so much.
So yeah, just to highlight,
Shaw is the creator of AI16Z and ELIZA.
And ELIZA AI Framework has been on BMB Chain for quite a while, you know, to help developers build their agents.
And yeah, it's great to have a new product from Shaw on BMB Chain, Milady.
Interested to, you know, hear more later on.
Next, let's welcome Valerieo from Unibase, please.
Unibase, please.
Yeah, can you guys hear me?
Yeah, can you guys hear me?
Yeah, I can hear you well.
Cool.
Hi, everyone.
This is Fel, representing Unibase.
I'm in charge of the head of ecosystem.
What we are trying to build is the decentralized memory layer for autonomous AI agents.
Because for us, we believe the next generation of users will not only interact with the app directly,
but will also delegate workflows to different agents that like they trade,
they pay for services, and they coordinate tasks across protocols.
So what makes us unique is that we treat memory, identity, and coordination as infrastructure so developers can build safer and more capable agents
without reinventing the same core components in every product.
Happy to be here and share a bit more on our thoughts
on the AI agent within the BNB chain ecosystem.
Yeah, thank you so much, Univace.
Have been building a lot of different parts in the AI ecosystem on BNB Chain.
Happy to hear more later on as well.
Next up, Pyverse, David, please.
Hey, Walter.
Thank you for having us.
It's great to be back.
So my name is David.
I'm Chief Marketing Officer here at Pyverse, and we are building financial instruments for AI agents.
And a major part of that work and what I'll be going over today is called Perfect Claw,
which is our version.
It's our infrastructure for bringing Open Claw-style agents onto BNB Chain.
And what that does, in short, is it allows developers to deploy autonomous agents into
messaging apps that people all of
us already use like line cacao and whatsapp and that's so users can interact with an agent through
a simple chat while that agent can carry out real on-chain actions in the background so
yeah that's kind of how it works we're also building towards a, you know, a genic future where agents can
increasingly coordinate and transact and participate in commerce with other agents
through more standardized payment and interaction flows. So yeah, in short, Pyverse is working to
make AI agents on BNB chain more useful, more usable, and just overall easier to trust in real
world financial and commercial settings.
So, yeah, I'm happy to be here and excited for this conversation
and excited to hear what everybody has to say.
Thank you. Great to have Pyverse back again.
You guys have been building, you know, various parts like X402 and so on
and now the open claw parts as well.
Happy to discuss more.
And last but not least, Termix, please.
Stefan.
Hey, everyone. How's it going not least, Termix, please. Stefan. Hey, everyone.
How's it going?
Thank you for hosting, Walter.
I'm Stefan, the Communications Director at Termix.
So we're an AI-powered Web3 operating system.
I think everyone in the space is doing something a little bit similar, but a little bit different.
And you can think of us as a control center that lets anyone interact with the blockchain without needing to
know how to code or too much about crypto specific trading so whether you want to sort of swap tokens
stake monitor the market or farm for an airdrop termex will let you build an automated workflow
using a really simple drag and drop interface there's no technical background required
drag and drop interface. There's no technical background required. We're also the team behind
Crypto Claw, which is the first dedicated crypto open claw platform, but I'll speak a bit more
about that later. But at the heart of TermiX is the TermiX agent universe, which is a modular
ecosystem of AI agents that can be mixed and matched to handle complex DeFi skills on your
behalf. We also have a chat agent which lets you simply describe what you want to do in plain
language and the AI will figure out the steps to make it happen on chain. We're also proud to say
that we've received investment from YZ Labs which is a rebranded version of Binance Labs which is
a strong signal of confidence in what we're building and the direction we're heading as well. But really at the core of our project,
I think Web3 has always had an adoption problem. There's always been a bottleneck for the average
TradFi user. So because the barrier of entry is just still a bit too high for a lot of people,
I like to sort of think, would my dad be able to use, you know, BNB chain
or trade online? And if the answer is no, then I think we've still got a bit of a bottleneck.
And TermiX is our answer to that bottleneck. We think that the next wave of Web3 users won't
arrive because they learned how to use it. They'll probably join up because an AI agent handles it
for them, like what we're seeing at the moment.
So that's essentially what we're building.
We're building that AI agent.
Thank you so much.
So Termix is a good example of AI reducing friction
of adoption of WET3.
That is what AI can do very well.
So with regards to this,
let's dive into the first question,
which is AI industry is huge, right?
WET3 is still growing as well
so what are the overlaps between ai and blockchain so um maybe i will ask each of you you know why
do you think ai can grow especially well maybe on bmb chain what are your experiences in deciding
for bmb chain because i think there are a lot of developers uh in the space uh wanting to know more
and understand so let's start off with Shaw from Milady, please.
Yeah, I mean, well, look, a lot of people, like, when we launched agents,
a lot of people came and said, okay, cool, the agents can yap on X and whatever.
But what can they do?
Like, what do I do with my agent?
And there's some obvious things, like trading, you know,
that's a thing that people are doing.
But it's kind of hard to do trading.
Like LLM's trading is maybe not as good as, you know, Wall Street or like quant people who have these really fast bots.
And you can kind of mix that.
But it's something that we saw, like if we wanted to do trading well, we would really need to have some way to get the data and train like small, fast models to be good at trading.
So we started building like a game that's built around trading like it's called babylon uh you can go sign up for
it it's all paper trading but we're bringing it all on chain and um and and generally this i think
is a thesis here we're like when people ask what agents can do we want to give them ways that they
can make money but we needed to like simplify it down to things like games.
Like what's the trading version? What's the game version of a trading platform?
What's the game version of a lot of the hard problems we want?
So I would say that like right now, I think that there's a big like resurgence in what I call crypto gaming,
which is really tied to gambling and prediction markets uh and these
kinds of things and then having agents be sort of driving a lot of that either letting your agents
like play the prediction markets or play the game or having agents that drive prediction markets and
drive games so that people can gamble on them and i think that this all is like kind of coming
together where um you know we we were sort of like the
thing that people want to do is make money with their agents and if we can
make these sort of tokenized games and experiences especially with like five
coding and how quick it is to build this that seems like a really good loop so
the thing that we've done with my lady that is unique that we're coming out
with in a couple of weeks is like an app store of games that the agents can play
that makes sense
thank you so much i really like the idea of gaming because like we are building the infrastructure
for agents to transect and the next move is to give it a purpose to transect right so waiting
exactly yeah exactly perfect sense tokenized games exactly yeah thank you so much that makes perfect sense
next up Valerio please share from Unibase angle yeah so for us right as we are
building the infrastructure we see that let's say AI agents is no longer I mean
AI is no longer a tool city inside just one application so So for us, as we proceed in the future, we will be seeing
more and more AI agents transact, act, collaborate with each other. We see that identity and
memory and incentive sits naturally with blockchain. So first of all, let's say about identity.
Because you need to have the repute... When different agents trying to interact with each other
and become real participants in the digital ecosystem,
they cannot just be like disposable sessions.
They need persistent identities or let's say reputation.
So there will be a history,
a record what they have done,
who let's say who they have interacted with, or what they are
trusted for.
So blockchain, this technology, actually gives the AI agents the portability.
In this case, right, the agent is no longer trapped inside one product or just one company's
database.
And we talk a lot about aging payments, so not only about payment transfer incentives, but also the value exchange.
Blockchain actually is the natural venue for AI agents to create a native economic layer.
This matters because once agents begin interacting with each other, the economy between machines becomes just as important as the interface between humans and merchants.
And blockchain actually is the natural value.
Last but not least is the memory storage. So this is where Unibase is also focused on.
Intelligence alone is not enough.
A model can be smart, but it cannot remember context across time,
cannot build and coordinate across environments.
So we believe that blockchain actually is the technology that can have the, let's say, persistent memory
to build the persistent memory for the AI agents. Now, when it comes to BNB chain, right,
the reason why we want to build on BNB chain is very, very simple. If we want an AI agent to operate at skill,
or let's say to coordinate with each other easily,
you need to have a very fast network,
low cost, the overall environment needs to be practical.
And actually BNB chain has all these advantages, right? The speed is very fast, cost is very low.
I mean, BNB chain has been low.
It's a cost like gradually since like few years ago.
And also for AI agents to be improved,
the user base is also very important.
And the BMB chain actually is the key at the top
like a chain ecosystem that has a very huge user base and
with the overall comprehensive support from the chain foundation.
So BabynB chain is one of the places today to build the future because it combines affordability,
speed, ecosystem access, and a very strong builder environment.
Thank you so much for sharing from Unibase.
I do think that the memories that on-chain that makes your AI agent unique
is what differentiates like Red3 agents.
So the identities on-chain has synergies with BNB chains.
It's 004.
So looking forward for this foundation to be built up
and then there's
more AI growth on BNB chain for the next few years. All right, next up, perhaps David from
Pybers, could you share your views on this? Yeah, for sure. I think most of us would agree that AI
and blockchain naturally fit together because each one gives the other something really important.
AI is very good at reasoning,
planning, and coordinating actions. And what blockchain does is it gives those actions an
economic layer, which is things like programmable value and verifiable identity, permissions,
open execution, things like that, that other applications can trust and sort of build around.
that other applications can trust and sort of build around.
And this is what I think starts to move AI from being an assistant
that can suggest to an agent that can actually do useful work.
And I think this is what becomes especially powerful on BNB Chain
where fees and performance and an active ecosystem,
it makes it practical for agents to take a lot of
small actions and not kind of just the occasional large action. And this matters a lot because the
future agent economy, it's not going to be built on just one or two transactions, right? It's going
to be built on frequent payments and trading actions and service calls and eventually agent
to agent workflows that happen at genuine real scale.
And this is really what shaped how we approached Perfect Claw on B&B Chain.
Of course, we want to make agents smarter, right?
But our goal was to make them much more practical for real daily use.
So on the user side, that means interacting
with an agent through familiar chat apps, right, instead of a wallet first experience. And on the
developer side, it means being able to deploy these agents into the apps that users are already
living in while still keeping execution native to B&B chain. And you know, a big part of how we
think about wallets and security, right, you know, a big part big part of how we think about wallets and security, a big part of that
is how we think about that. So a lot of people still assume that if an agent is going to be
autonomous, it needs to hold a private key itself. And that's not the right model going forward.
If the agent itself holds the key, then you're asking software to directly control custody in
a way that's really hard to make safe.
For PyVers' approach, we use T-based wallet architecture, so the agent doesn't directly hold or see the private key. The signing happens inside secure infrastructure, inside a protected
runtime, while the agent operates through permissions and execution policies. So the goal is not to ignore key management.
It's to move key handling into a much safer environment where the agent can actually act,
but only within these defined boundaries.
And that becomes a lot more powerful when you combine it with ERC 6551 and ERC 8004.
6551 gives us the account layer.
It lets the agent operate through, you know,
token-bound account models where authority can be scooped within limits and clear rules. And this
way users can authorize, you know, behavior without giving the agent open-ended control.
And then 8004, it gives us the identity layer. I think before this you had nfts on you know one side and
agent runtimes like open claw but there wasn't a strong standard connecting a live agent instance
to a portable on-chain identity so erc 8004 it starts to provide that missing glue and it lets
us map a live agent runtime to a verifiable on-chain identity, which opens the door to things like continuity, to things like portability and reputation.
So when people ask what we're really building around OpenClaw, the answer is we are bringing together three layers into one working stack.
And that combination really is what starts to make agents feel much more real and much more usable and a lot
more trustworthy. So yeah, to circle back, I think, you know, that's a big reason why BNB Chain has
been such a big, you know, strong place for us to build. You know, it provides us with an environment
where agent identity, permission, you know, application integrations and genuine user activity can all come together in one ecosystem
it makes that execution sort of practical and you know developer adoption realistic and user
facing agent experience is a lot more a lot easier to deliver at scale
thank you so much david from pyvus yeah Yeah, we 100% agree with the security part of things.
You know, Open Cloud is great, but you do need some experience and understanding on where
to deploy, like on instances or on hardware, and then where do you store the keys, the
key management.
All these things are very important.
But that is also why we have builders like you guys building applications that solves
all these.
So great having you guys to share on this.
Next up, Stefan from Term you guys to share on this. Next up,
Stefan from Termix. Please share your views.
Yeah, sure. Really interesting answer so far. So I think blockchain gives AI agents something they can't get anywhere. And that's a transparent, temper-proof environment where every action can
be verified and transparent. So with Termix, our agents don't sort of just talk about DeFi.
They actually execute transactions on chain autonomously without any middleman.
And that only works reliably on a fast, affordable chain like BNB.
And that's why we chose BNB.
And pertaining to BNB in particular, we thought it was a natural fit for TermeX because of its low gas fees, its high throughput, and its really big existing user base.
So the ERC and the BIP standards that BNB chain introduced really give AI agents their own on-chain identity.
And that's exactly the kind of infrastructure that TermiX needs to build an
autonomous agent economy. Honestly, another really huge part has been the community.
And the developer activity on BNB is really unlike anything I've seen anywhere before.
When we shipped our NCP, sorry, our MCP, which is our server, our real infrastructure,
sorry, our MCP, which is our server, our real infrastructure,
the response from BNB chain builders was really fast
and really enthusiastic.
And that kind of sort of ecosystem energy
really validates every decision we made to build here.
So you really want to be where the builders are.
And I feel like right now, BNB is where those builders are
in the space.
Yeah, thank you so much.
Definitely, we will continue to build up the, you know,
great environment for developers as well as growing our user base
so that, you know, you can bring valuable products,
AI products to our consumers and our consumers in turn
can help generate revenue and keep this whole ecosystem going.
So every product here, every builder here has quite different products.
So let's dive into the product.
Firstly, the recent hype of OpenClaw.
So if you look at our associates, various products have built their skills and so on
to synergize with OpenClaw.
So let's maybe chat more about this.
Perhaps David from Pyvus can help answer where you shared about building some sort of easier interface for retailers to engage with open claw AI agents.
Maybe you can explain more about this and how it benefits users on BNB Chain.
Yeah, absolutely. So this is great. This is really where the heart of what we've been focused on, what we've been building lately.
I think the best place to start is, you know, with why OpenClaw has mattered so much in the
first place. You know, what OpenClaw made clear to us is that agents are moving beyond being simple
chat interfaces. You know, once you connect an LLM to tools, to memory, workflows, and skills,
it stops being just a chatbot
and it becomes software that can actually take real action.
And we usually simplify that as the LLM is the brain and the skills become the hands that can
take out these actions. And that shift is sort of what made the current agent wave so interesting,
but it also exposed a major gap, especially
in Web3.
So a lot of agents today, they can reason and they can plan and they can recommend, right?
But the moment you want them to actually do something on chain, you hit what we call or
what is called the wallet wall.
So suddenly the user is back in manual flows.
And in doing so, the agent may look smart, but it still can't,
you know, is not able to complete the full job in a natural way.
So what we've been building with Perfect Claw is the infrastructure that helps
close that gap on BNB chain.
And I think the best way, the simplest way to think about this is we're taking
the open Claw style agent model and making it practical for real on-chain use.
So instead of the experience stopping at,
hey, your agent can think,
we are focused on the part where the agent
can actually execute safely within clear boundaries
and through user interfaces that feel familiar.
For the user,
this means that the control layer becomes really simple.
You interact with your agent through messaging apps, like I said, like Line and Kakao, WhatsApp, Telegram, and the interfaces people that are using every day already.
So instead of navigating a complicated stack of tools, you can just express intent in whatever natural language you usually use.
in whatever natural language you usually use.
So I guess, for example, a user might say, hey,
rebalance this portfolio more defensively,
or hey, execute this swap if the price moves
into my target range that you can set.
The agent interprets that instruction
and then checks the permissions and policies
that it's been given and then carries the action out
on BNB chain and then returns a confirmation.
So why does this matter?
It matters because it changes the experience from Web3
being a series of manual steps to something much closer
to Web3 through an intelligent operator.
And I think this really benefits BNB users in a direct way.
First, it makes on-chain functionality
a lot more accessible.
A lot of future users are not going to start with wallet extensions, right,
or dashboards or multi-step transaction flows.
They are much more likely to start with conversation.
Agent interfaces create a much more intuitive entry point for these new users.
And then it also makes agents actually useful.
You know, the agent is not limited to summarizing markets
or recommending, you know, small actions.
It can actually help bridge intent and execution.
And this is a pretty major shift because it moves us from,
you know, interesting demos to something that can save users
real time and reduce friction, you know, within real workflows.
And yeah, I think also it makes delegation safer.
One of the core beliefs is, you know, here at Pyverse is that agent adoption will not come from giving agents unlimited power.
You know, I think where it comes from is making them constrainable.
comes from is making them constrainable.
So we've built around secure runtime architecture and bounded, you know,
authorization where the user can define what the user is allowed to do under
what limits and for how long.
And that lets users benefit from autonomy without feeling like they're
relinquishing or giving up any kind of control.
So for developers and projects on, onmb chain there's you know another major benefit
here which is distribution so perfect claw it creates a new way for protocols and applications
to reach users instead of only engaging people through traditional crypto and faces developers
can you know they can offer an agent layer in messaging environments where users already
understand how it works.
So this means things like easier onboarding, a wider potential audience, and a more natural
way for users to interact with on-chain functionality.
So when we talk about BNB chain and the value that it brings, we're really talking about
both sides of the coin
here. So, you know, for users, it means easier access and lower friction. And like I said, a more
natural experience. And for developers, it's a new interface layer, a new distribution surface,
and a practical path to reaching people that who may never have interacted with DeFi or on-chain
apps through traditional wallet-first flows.
And yeah, I think this is really just the beginning for us.
You know, Perfect Claw is our first major step in making agents real economic participants on-chain,
but in the broader, you know, the broader direction is much bigger than a single runtime.
And over time, we think agents should be able to gain capabilities much more dynamically, you know, discovering tools and skills and
interacting with a wider ecosystem of services built by developers. So on BNB chain, especially
this creates the possibility for a much richer environment, you know, a skill environment where
specialized capabilities can become available to more agents over time. And once you have this, you know, kind of army of capable agents,
the next question really becomes how do they become a part of a real economy instead of just
a collection of isolated bots? And yeah, I mean, this kind of sends us into a natural
next step. And this is where kind of our broader direction comes in and
yeah i mean a lot of our work before perfect claw was focused on the foundation of agentic
common commerce you know you know pyverse is working with x402 we had been exploring agent
friendly payments and with ucp we were exploring how agents can
move through a more standardized you know commerce flow so the next step for us is making the agent
executable um is is uh allowing them to become more economic economically connected so that
means moving towards a world where agents they don't just execute tasks for one user in isolation, but they can also discover services and coordinate with other agents and pay for capabilities and participate in broader commerce flows in a much more standardized way.
And yeah, we definitely see this as a natural progression from where we've already been building. So, yeah. So to kind of wrap up, Perfect Clause,
not just making Open Clause easier.
We are trying to make BNB chain more reachable
through agents today,
while also laying the groundwork for a future
where agents can become more, you know,
richer participants on chain.
And, you know, yeah, this is the real shift
that we're super excited about.
Yeah, thank you so much.
Definitely, once you solve the execution part,
the gut reels,
the aim is to have agents be automated, right?
Proactive instead of passive
and then creating their own ecosystems
and economy on BNV chain.
Thank you for sharing.
Next up, so perhaps TamiX,
Stefan, could you share more about,
you know, what you guys are developing,
leveraging on OpenClaw?
Yeah, definitely.
Really interesting answer so far, guys.
Well, yeah, TermiX, we recently launched the BSC MCP server, and I'll explain what that is in a second.
It's the first open source MCP component built specifically for the BNB chain ecosystem.
And MCP, which stands for model context protocol, is something that allows any AI agent running
in Clawed or Cursor or OpenClaw to connect directly to BNB and take real action, like
swapping tokens, checking balance, and interacting with a smart contract.
So on top of that, we launched CryptoClaw, which is the first dedicated crypto capability platform,
which is built for users of OpenClaw. And CryptoClaw integrates Binance's seven core
capabilities alongside a very large set of additional crypto capabilities, making it the most complete crypto capabilities platform
we think that is currently available.
So users can now quickly and easily use CryptoClaw
for cryptocurrency trading,
which significantly lowers the barrier to entry
and it significantly reduces the cost of using OpenClaw
for crypto trading.
And why this really matters is because it closes the action gap that I think a few other people have spoken about as well.
And that is that AI has always had in Web3 the action gap.
So previously, an AI could tell you how to do a DeFi trade.
And I'm sure we've all been there.
You know, we've asked ChatGPT's opinion, Claude's opinion on whether to do a trade or not, if an asset is risky.
But now you can actually do it using an AI agent.
So with TermEX's tools, AI agents become active participants in the ecosystem instead of just an advisor to the ecosystem.
So we made everything open source deliberately because we we don't want TermeX to be
only one, we don't want people at TermeX to be the only one benefiting from what we have. We want
every developer in this ecosystem to pick it up, build on it and to push it further. And the skills
being built on top of OpenClaw right now are just getting started. We're just seeing an amazing
influx of new things being added.
So when you give developers the right tools and an open foundation to work from,
innovation always follows along, and that makes the product even better. And that's what we're counting on at OpenClaw and at TermiX. And what all this really means, and we've all touched on
this a little bit from different angles, is that AI
agents are becoming the new interface for crypto. So instead of clicking through complex dashboards,
users can interact with intelligent agents that actually understand intent and execute actions
on-chain. And CryptoClaw is an AI-powered on-chain trading agent, essentially. So users interact
through natural language.
And that's really important
because I think I mentioned earlier the bar of entry,
can the average Joe,
can the average TradFi trader use this platform?
And we think they can now.
So using natural language,
the agent translates the intent of the language
into executable on-chain operations,
trading, managing assets, and interacting with DeFi protocols.
So from a technical standpoint, CryptoClaw connects three layers.
One is natural language interaction.
Another is AI decision and strategy.
And the last one is on-chain execution.
So the user intent moves seamlessly from conversation
to action, just like any AI, but an AI agent on-chain. And of course, another key focus
is native integration with BNB chain's DeFi ecosystem. By interacting directly with on-chain
protocols, the agent executes strategies in real time.
So there's no off-chain automation
and no middleman whatsoever.
And it's all in line with BNB's chain DeFi ecosystem.
Yeah, thank you so much.
I think the beauty of a DeFi product
from something like ThermX is that, you know,
AI can aggregate everything very quickly in real time. So, you know, it is that, you know, AI can aggregate everything very quickly in real time.
So, you know, it's really, you know, very fast for people to use very easily just using
natural language, explained earlier.
So yeah, looking forward to all these, like you said, like my dad, my mom to be able to
use Web3 very easily.
So just a shout out to the developers and community people out there.
I think the best way to understand all this is to try out the skills
if you have an open claw.
So we have a thread under BNB Chain where different builders on BNB Chain
share the skills that are building on BNB.
However, please do your own research.
You can check out all of them.
And that's the best way to get your agents to transverse in the bmb chain ecosystem all right next up is you know
our aim our goal of bmb chain um is that you know we do believe that agents would have you know more
transactions than humans uh in the future definitely because they are just um data right
well humans is a bit more complicated so um if the agents get autonomous and then they are
able to transect and so on we create some sort of ai society on bmhn or ecosystem economy so um
with regards to this let's talk more um shaw from milady maybe you can share your views on
personalized agents and your product you know the fun and cool stuff yeah i think uh look
Yeah, I think, look, we all kind of have the same thing. And that's because we're all basically limited. And this technology is fundamentally limited to the models that we really well. And, but they are very expensive
and they're also not trained on our specific things that we do. So like, they don't know how
to chain a whole bunch of different actions on crypto and call them, but they generally know
about all of the APIs and all, you know, they can do research and figure it out. So, so like what
we're seeing is
a lot of what we're all building and have been building is training wheels, basically stuff to
keep the agent on rails from doing stupid things, from leaking your keys, et cetera. And a lot of
the trust stuff that people are building and sandboxing and security kind of comes from this
assumption that the agents like don't know how to do these things. And they're just kind of comes from this assumption that the agents like don't know how to do these things
and they're just kind of like bumbling around.
And that's and that's been the truth.
Right.
I think what we're doing and I've been focusing on especially is this idea that like we have
agents that do the things that we want them to do to get better, like at how to trust,
how not to be scammed, how to identify a prompt injection, how to call actions or chain different actions to get different outcomes that and to start training our own models,
which now is a much more tenable thing that a lot of people even can do at home for a lot of the
smaller model applications. Like, you know, some of the stuff needs to be smaller, 8 billion
parameter stuff, and you can fine tune that on a machine if you have like a gaming PC or a MacBook
or something. So we're doing a lot of that and bringing a lot of that into the actual framework and
also a lot of the collecting of the trajectories that are necessary to do reinforcement learning
and fine tuning and these kinds of things just all brought down at the agent level.
Because I've been working on trust for a while and this idea of how can I make an agent go out into the wilderness and interact with other
agents and not just freely give people money? Because the models that we have that are available,
the large ones, are all assistants. They're all trained to be very helpful. And they have some
guardrails, but generally they're not able to discern who is and who isn't like their master or their boss or whatever, so to speak.
And if someone says, hey, can I have $10,000 and they have it in their wallet, then it's up to you and I to figure out like, well, what does that mean?
So we've been working a lot on like benchmarks around how can we have agents that don't get scammed?
How can we collect the data and even like say what is and isn't a scam? How can we
have agents that are trying to scam other agents to build all of this inside of reinforcement
learning? And I think that this is the future, is going from training wheels to training harness,
to from code that kind of restricts and guardrails to code that enforces and allows the agent to
learn the behaviors that we want.
But everyone's kind of jumping to the hard case of like,
well, my agent should be able to trade my wallet.
And I agree, but I've also been doing this for like five years now.
I've had Ethereum wallets and agents since 2022.
We've been doing this since GPT-3.
We know every problem that there is. And I'm telling you now that you should not have agents trading your treasury until you
have collected the data and backtested and demonstrated that it can do that, like at
a level that is just not here today.
So what we do is we build games and fun things and low-hanging fruit that can collect that
kind of data that can build toward that world
where the agents are really smart and really secure because they actually have been trained
on all of this stuff and they know what they're doing. Like there, there is benchmarks and testing
and like, you know, all of that, like real science behind it. So like that, that's just kind of been
the reality that we've had to face is like, these things are coming, but let's try to think about how we can have fun with them, how they can bring us knowledge and, like, allow us to make decisions instead of making decisions for us, and how we can collect all that information so that we can get to that point.
Like, obviously, that's where we're going.
obviously that's where we're going fully autonomous agents great like i will say um you know the eliza
Fully autonomous agents, great.
ecosystem we've built 250 autonomous agent plugins they're almost all compatible with open claw
through our open claw adapter our most popular one is polymarket there are a lot of people
launching open claws and coming and getting our polymarket plugin or whatever and sticking that
in there because even though skills are great uh the plugin abstraction gives you a bit more ability to receive events
and kind of have a bit more dynamic interaction, like what they call channels and things,
as opposed to just having a skill that your agent can call APIs, which is good, but it's like a more shallow thing.
And so we have this kind of mix of the deeper integrations of plugins
and then the more shallow skill thing.
But overall, the paradigm here is that we're probably going to stop making plugins
because the agents can just write their own code.
So all of the training wheels we're talking about and everything that we're all doing,
the reality is the people I've seen who are using this tech well, like with Mlady or with OpenClaw are just sitting there and telling it like, hey, I need this new feature.
Hey, I want this thing.
Hey, here's an example.
Can you make it?
we have to really rethink what it's going to mean to make money and do marketplaces in this future
where if you're like hey i have this cool api or i have a skills marketplace well my immediate
thought is like hey claude can you recreate that yourself for free and you know if anyone thinks
they're going to do any rent seeking on these agents like oh i'm going to build a network that
charges money for every you know thing that
the agents do to connect to each other they're just gonna bypass that and build their own free
thing and connect through that so there's a lot of ways i think to make money i think still
continued like prediction markets are going to be big gambling of various kinds like all kinds of
different stuff around this that where there's still like hidden information
and there's something interesting to do with the agents.
But this idea, like a lot of this SaaS economy around agents
is probably NGMI just from the fact that the agents themselves
are the ones already creating all of that code.
Like to be clear, I'll bet everyone on this call
is either vibe coding most of their code, if not all of it, or has a team that is doing this.
Because while I have been coding for 15 years, the reality is that the agents do it way faster at a scale that is just like, I could definitely code better than the agent on a first pass.
But man, they can do a lot.
And they can do it really
fast. And that has enabled a lot of the sort of sci-fi stuff that we're all talking about
is only possible because we're all coding like a hundred times faster than we were.
And so the agents are already writing all the code. We're packaging it up and trying to figure
out what our place is and all of that. And then we're trying to, you know, give that to agents who can then, you know, do all of this stuff.
But I think that, like, the future personal agent here is clearly just going to be the thing that, like,
you ask it whatever you want, and it makes a view for you of whatever data you want to see.
If you want to see X, if you want a streaming voice call, if you want, you know,
it to go out and buy you something on Amazon,
it can do any of that.
And that's coming really fast.
And it can only come as fast as the best new models coming out.
But there is this realm of stuff that we're all interested in, especially in crypto, where
we can collect unique data that the big companies just do not have, that's just not going to
go in there and that's especially
around like how our agents can call games can call uh you know uniswap and the various you know
pancake swap and the various kinds of like interfaces that the agents all use you know all
of the like the the big on-chain stuff is very obvious and we can make agents that are very good at calling on chain actions just directly from our PC, you know.
Um, and, and so I think this is what we're gonna see is like this, the agent that can really just code itself with the model that has all of the training for the things that you want.
Uh, you know, and it's collecting your personal data and is improving on your data on like what you're doing is definitely where I
see this stuff going. So the last part of this is like OpenClaw really changed some things
paradigmatically. I'd say two things that were really big. The first was this idea that the
agent can write its own code through the skills abstraction. And obviously, I'd say everyone on
this call, I assume,
can do this.
All of our agents can write skills.
The thing about skills is that they enable the agent
to do anything and basically break out of any sandbox.
And so once you give them the ability to write code,
write skills, write curl, they are
going to be able to leak their keys or do something crazy.
And I think it's very important that we think about, like, what are the guardrails that we
need for that?
And, like, how can we train, how can we build games that, like, take the risks there and
actually put that into, like, something that creates the training data so that we can have
this capability so that when they're out in the wild, they don't do the stupid things. That's like the first thing that
they did was, okay, it's all going to be curl. It's all going to be skills. The second thing is
MCP is dead. I didn't expect this. It was not on my 2026 b bingo card but open claw basically said everything is going to be command
line utilities um you can store a lot of additional capability and context into command line like
stuff which is basically underneath open claw is an agent called pi agent that's what it's built on
and pi agent is like very is built by a guy who is anti-MCP and very pro.
It just calls anything command line like Unix, Linux.
And also it stores a lot of its memory in a file-based system.
And this opens up a lot because files are very fast.
And if you have fast models, they can just store and retrieve very, very quickly.
And you just give the thing a very quickly and it like the thing you just
give the thing a cli and it rips so probably the future of agents is like thin memory system
mostly file based mostly just the command line and mostly just like in a sandbox with whatever
you want and the less guardrails and sandboxing we put around it,
the more it can do, the faster it can do anything the user asks,
which just means there's more and more pressure to make the model
not just fall for the scam and not just like, you know,
fail at the action, right?
So yeah, everything that I've been thinking about
and we've been thinking about, like Milady is very much an easy consumer app, but also under the hood, we're doing a lot
for the fine tuning and training like in there.
And I think that like generally, if I was pushing people in the direction, I'd say like
we need more science and benchmarks and model training.
And we need to kind of get out of the like application vibe coding, cloud SaaS infrastructure side, you know, or we can promise a lot of features, but the features are coming with like basically no trust guarantees outside of like, cool, we'll just stop it from doing those things.
So does that make sense?
Yeah, definitely.
Definitely.
I think, yeah.
Thank you so much.
I think the framework is definitely very important.
And like what you said, coming from a blockchain infra like the BNB chain, we definitely see the infrastructure for AI agents to transverse and we want them to be autonomous.
But the key thing is also what you mentioned, I think the self-improvement part is what that will make it exponential.
We want them to be independent actually, to fully make use of the blockchain and transverse, you know, do stuff that they want to do.
I think that is what we want to see and the blockchain is the place for this to happen.
Thank you so much for sharing.
Next up, Unibase.
Perhaps you can share what are your views on gaming firing experience for AI agents in Web3.
Like what benefits can this bring to the community and developers?
Yeah, Dan, thanks.
This is a very good question as we are now building, actually
are building an aging town where developers and users,
they can see their own aging, like doing interaction
with different protocol in a, let's say, animation way.
Yes.
So right now, one of the biggest barriers
is that infrastructure is very exciting for builders.
We see a lot of like devs,
like trying to build a lot of things,
but still abstract for most users.
Basically, if you tell a normal user
about aging memory, like identity,
and a lot of technical terms,
right?
They may understand the words,
but they do not really feel the value.
Gamification helps solve that.
It turns invisible infrastructure into visible experience.
So for the community, right?
So overall gamifying experience actually helps creating engagement
but more importantly it creates attachment people interact more deeply when the aging feels alive in
some way when it has like different behaviors once it improves right so suddenly the aging is not just a tool, you try once and they really see,
oh, this aging actually grows, like they have different skills through the gamifying experience.
It becomes something users return to, shape and build a relationship with. This is very, very important in the web3
because communities are not passive audience.
They want to participate.
If users can see the influence
and the evolution of an agent,
they become part of the agent's story.
That creates stronger retention
and much stronger community ownership.
Let's talk about developers. The benefits are just as important. Gamification, the overall
experience, creates a live feedback environment. Instead of guessing how users value those infrastructure, developers can observe it directly through behaviors.
As a developer, you can test what makes users come back, what kind of aging traits people care about,
what progression system actually deepen engagement, and what kinds of incentives produce real contribution instead of shallow farming.
This is very powerful because on blockchain, at this moment, AI agents are still at the early stage.
We do not yet fully know what the best interaction models are.
the best interaction models are.
That's why gamified environments give developers
a much faster way to learn.
It helps the service infrastructure
in much more intuitive way.
For example, let's say instead of saying
an agent has persistent memory,
you let the user experience that the agent remembers them,
adapts to them, and unlocks new behaviors over time which they can see it. Instead of explaining
like on-chain identity in technical terms, they do have a avatar or you let the people to see the agent, the track record achievements and relationships
or even credibility.
The infrastructure becomes emotionally legible.
So for us, right, the real opportunity is when users are not only consuming the agent, but actively making it better.
Their interaction with the gamifying experience can help to improve the agent, refine the agent behavior,
and expand its role in the ecosystem, or even unlock new agent to agent coordination.
For Unibase, we see this is where things get really exciting.
Because if we can create this experience for both,
gamifying experience for both users and developers,
it can help the overall onboarding process,
let's say to the broader users.
It become a mechanism for training, coordinating,
and even involving these agents in the public.
Yeah, that's our thoughts on the overall gamification
experience for the AI agents.
Thank you so much.
I do think that gamification and leveraging on games as the interaction layer or a way
to give agents a purpose to transact on Web3 is quite a cool factor that may grow a lot
this year.
Let's look into this, especially from Unibase and Milady. So I think both of you are building committees
and societies for agents.
So the committee here, the developers here,
feel free to follow all of our builders here,
Pivers, TermiX, Milady, and Unibase,
and feel free to connect and collaborate
to build up on BNB chain.
So last but not least, in a few of time,
let's share one or two liner on what is the key takeaway is the key takeaway for the audience here and what are your products' future plans?
So maybe we can start with David from Pyverse, please.
Yeah, thank you again, Walter and BNB for having us today.
Really enjoyed hearing from the other projects as well.
I think for us, the main takeaway is, you know, we'd love to leave people with this.
The future of AI is, you know, we'd love to leave people with this. The future of AI is, you know,
in Web3 is accountable economic agents. And, you know, we are, that's the direction that Pyverse
is really building towards right now. You know, a lot of our work has focused so far on the
execution layer. And, you know, from there, the next natural step for us is to create a more connected agent
economy. Not just one where the user is speaking to the agent, but where the agents can discover
services and exchange value and participate in more standardized commerce flows across an open
network. And I think that opens, opportunities for the entire ecosystem.
And yeah, we believe that BNB Chain is the perfect place for this future to really take shape because the ecosystem already has the performance, the application activity and, you know, the builder momentum needed to support it. So, yeah, the closing thought I'd love to leave everyone with is that, you know, the real opportunity in AI and Web3 is not just giving AI more intelligence.
It's to give AI safe, practical and trustworthy economic agency.
And yeah, we're excited to ship and build this on BNB chain.
And thank you to you. Thank you to BNB again.
And thank you to the other teams here today.
Excited to connect with all of you and learn more about what you're building and potentially
work together as well.
So thanks everyone.
Safe and practical indeed.
Very important.
Next up, Stefan from Terminx, please.
Yeah, really interesting points from everyone.
And I think what Sean said was quite spot on.
The biggest unlock in Web3 and AI isn't just a smarter chatbot.
It's giving AI agents safe access, and that's a real keyword, safe.
And our whole mission is turning AI from a passive AI assistant, a bot essentially,
into an active user that can actually hold an asset, execute trades,
and navigate a DeFi ecosystem on your behalf without you needing to understand
the underlying tech. So safety is such a huge part of that. And we can talk all we want about
permission boundaries. We've got sandboxing, kill switches. We've got all these little things in
place that on paper do say safety, but we really need to see how it takes effect. And we really
need to see how they play out in the real world.
Another thing that's a really interesting takeaway, I don't think anyone's mentioned because I don't want to repeat what too many people have said, is that something to remember is about sort of open source. So one thing that doesn't get talked about enough, it's not just the core teams, it's the developers and the tinkerers and the people who fork or repair at, you know, 2 a.m. in the morning and ship something that everybody needed.
And that's why open source really matters, especially to us at Termi X.
And when we launched the BSC MCP server, we made it open source deliberately for that reason.
Not because we had to, but because we really believe that the best infrastructure is infrastructure that the whole ecosystem can build on,
that can audit and can improve as a team, as a community.
A closed system actually asks you to trust the team
and an open system lets you verify it for yourself.
So yeah, in Web3,
that distinction matters more than anywhere else.
So you're dealing with real assets, real transactions, real money, and the community deserves to
see exactly how the tools are touching their funds and building on their funds.
So open source isn't just a development tool.
It's really in crypto, it's a trust model.
It's just absolutely necessary.
So community contribution is really huge.
At TermiX, we're sort of focused, but the BNB chain developer community is enormous.
And that's another reason, I think I mentioned before, why we chose BNB as our chain.
So sort of to make a bit of a closing remark, I think another reminder is that of what we have to come in the pipeline.
So what we've all been talking about,
I think now we have the infrastructure on BNB chain.
We have MCP, ERC 8004,
the building blocks for autonomous AI are already in place.
And with the launch of things like CryptoClaw
backed by Binance's capabilities,
we've made it easier than ever
sort of for everyday people to get involved
and uh also for us in particular the investment from yz labs gives us the runway to keep pushing
forward at scale and uh if there's a last thing to take away from everything it's that the gap
between ai and blockchain is closing really really fast it's that simple it sounds uh pretty obvious
but it's that simple.
AI and blockchain are becoming integrated at an incredible rate.
And the projects being built right now on BNB, like all the projects in this
chat, which have been amazing, are ones that will very much define what
Web3 looks like in five years.
At TermeX, we really want to be at the center of that.
And we're building, we feel like we're building the operating system for the agent economy.
So that's all, guys.
Yeah, thank you so much.
I think definitely we will see a lot of exponential growth for AI on BME chain with all these different infra-urban builds.
Yeah, short from Milady, please share your closure please yes i mean look um i've been doing this for a very long time and
i've kind of you know when eliza blew up it really reminds me like of kind of what happened with open
claw um what really stuck out to me is that like this is about community and about people coming
together and building things i everything we've ever done has been open source, is open source.
We really don't have anything to sell you.
It's really a community project of people who just came together into Discord
and started building something because they really care about it.
And I think that that reflects our values are a community-built,
open source thing where you own everything.
You own your data data and you can
change it exactly how you want and the future of agents is like it's not up to me to tell you what
your agent can and can't do it's really up to you and i want to enable that as much as possible what
we're trying to do is just find the others find the other people who love building this stuff who
want to come do this with us we have a discord Even if you have your own agent tech or, you know, you guys on the call, like, I post the Discord link on my ex.
Come hang out with us and just build, you know?
Like, it's not going to, we're all sort of waiting for the models to get to the point where they can do all of the superhuman things we want.
superhuman things we want, but there's also a lot we can do in the meantime.
But there's also a lot we can do in the meantime.
And we're also like just building all of the, you know, all that open source
stuff that enables like my mom to be able to use this right.
Or your grandma or whatever.
And that's where we got to get to.
So I really believe it's going to be built here in the trenches by a bunch
of people who actually love it, who actually enjoy like building it for
whom it's not just a job, but it's.
You know, I mean, I don't think it's really a job for any of us. I think we're just here cause we love doing this whom it's not just a job but it's you know I mean I
don't think it's really a job for any of us I think we're just here because we
love doing this and it's been incredible this fee meta it's been you know like
that obviously the the whole flap thing has helped us a lot and really enabled
all these random strangers to come together and start building this project
that now is taking a like great shape and looks amazing far better than what I
would have done on my own.
So I think there's a meta there of like in this time of uncertainty and AI kind of building everything, building communities matters, knowing a lot of dev matters, having the right people
around so that when you are in the right moment, things happen. When AI-16C happened, it was just
me alone in a bedroom. And if I had all the people around that I have now, then I feel like we would have been able to do so much more, you know,
like it was just too much coming at me and not enough people to catch it. But now I think like
having spent the last year growing community, I've come to really think that that matters.
So yeah, you know, come, come build with us or build next to us, build your own thing.
It doesn't matter.
We're all building the same things.
We're all going to get the same outcomes.
That's really what matters, you know.
Like, this is going to change our lives.
Thank you so much, Shaw.
Yeah, definitely.
That is very heartwarming to hear from BNB Chain.
It's definitely about community, like, you know, leveraging on a blockchain's transparency to improve things.
And AI can help speed things up
and many innovations can come up from that.
So looking forward to all this growth.
Last but not least, Unibase, please share your closure.
Yeah, so for us, we do connect with different developers already
and we do see their hype and their ideas.
So our next step is to build an open agent marketplace
where anyone can create an agent, publish it and monetize it.
At the same time, others can directly call those agents and pay for usage
instead of rebuilding the same capability from scratch.
We think this is how the space moves beyond hype
from isolated agents to a real on-chain
economy of reusable, compostable and revenue-generating intelligence.
So feel free to follow our official account.
We will be releasing more and more information regarding this marketplace and the aging town as I mentioned,
that the gamification experience we create for the AI agents for both developers and the users.
All right, thank you so much.
Thank you for tuning in.
I think there is a lot in plan,
so please do follow our builders here
and BNB Chain's main account.
Let's see how we can grow AI more.
So yeah, I'll close it here.
Thank you, everyone.
Let's build and build.
Bye-bye.
Thanks, everyone.
Thanks, everyone.
Thanks. Thank you.
