Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
BNB Builders Talk: AI Agents – The Next Wave of Web3 Apps
Recorded: March 25, 2026 Duration: 0:43:22Space Recording
Full Transcription
Welcome guys, can you hear me? Can you hear me guys?
Yeah. Yes. Hello.
Perfect. Perfect. Perfect. So yeah, everyone, welcome to another episode of BNB Builder Talk.
And today's episode, we're going to discuss about AI agent. Basically, the topic is AI agent, the next wave of Web3 apps,
where we have speakers from Coscopy on Intelligence
Cubed and Smart Sentinels.
So I'm going to give you guys the stage
to go with your introduction for our audience
before starting the .
You can guys go one by one.
So let's start with Clark, then Andre, then Florence.
Hey, everyone. Can you hear me okay?
Yeah, absolutely.
Yeah, brilliant. Great to be here. So my name is Clark. I'm one of the co-founders of Coscription.
I'm one of the co-founders of Coscription.
So Coscription is a decentralized AI platform where users can create,
customize, and monetize AI models, starting with image generation
and expanding into agents and more advanced use cases.
We're building the infrastructure for user-owned AI,
where models are deployed with built-in monetization through smart contracts.
Awesome. Awesome, awesome.
Andrew?
Yeah, hi, everyone.
My name is Andre, and I'm founder of Smart Sentinels.
We are building a decentralized network
where AI agents perform real-world useful work,
starting with automated smart contract auditing
and agents in healthcare already.
This will allow to mint new tokens and reward the community.
And I'm excited to be here. Thank you.
Awesome. Florence.
you can hear me clearly? No, I'm good. Hello?
Yeah, I can hear you now.
Oh, you can hear me now.
Okay, sorry about that.
Technical issues.
Okay, so I'm Florence, CTO of Intelligence Cubed.
We're building the modelverse, a decentralized marketplace
that turns AI models into programmable on-chain assets
with pay-per-use access, royalties,
and transparent benchmark. So we're just creating a decentralized model, AI model NASDAQ. Yeah.
Thank you. Awesome. Awesome. Welcome you guys to this session. So yeah, thanks for your
introduction. Now we're going to start the AMA.
So basically this is going to be have four segment.
The first segment will be starting
with understanding AI agents.
So I'm going to ask you questions one by one.
So same question to Clark, then Andre, then Florence.
So the first question is, as of now,
AI agents still very new for lots of audience.
So let's start with the simple questions, like for beginners, what exactly is an AI agent in Web3?
Can you break it down and simply so any layman can understand?
Thanks, Jutendra.
Yeah, so the easiest way to think about an AI agent in Web3 is that it's AI that can actually take actions and not just give you answers.
So instead of just chatting with a model, an AI agent can do things like interact with smart contracts, move assets or execute tasks on your behalf.
The Web3 part is what gives it that ability to operate because it can plug into wallets protocols and on-chain systems and i think where this gets really interesting is when you can combine that
with ownership so agents aren't just tools but something you can actually control and monetize
which is a big part of what we're building at coscription
interesting and uh and, as per you?
Yeah, great question. Great question. So, in Web3, an AI agent is more than just a chatbot.
It's an autonomous on-chain worker. It can take action, it can send transactions, it can
interact with smart contracts, it can execute professional tasks independently, it can own assets.
But most importantly, it can verify work.
So through our proof of useful work, the agent's output, like a smart contract audit, it's cryptographically proven on the blockchain.
cryptographically proven on the blockchain at smart sentinels our agent aren't just software
they are decentralized employees basically that perform real business values
real business services like in a medical office for example
it's an ai with a job and with a wallet, basically, in my opinion.
Yes, new building.
So, Florence?
Yeah, so in my view, an AI agent is best understood not a conversational interface,
but as an autonomous or semi-autonomous software entity capable of global directed behavior it can
interpret context reason over objectives invoke tools access external system and execute actions
with a degree of persistence and adaptability so what makes the web street context distinctive is
that agents are not merely computational actors they can also become economic actors they can
control wallets interact with smart contracts, pay for services, receive revenue, enforce rules, and participate in open coordination systems without relying on a centralized intermediary.
The reason this category is accelerating now is the convergence of two maturing technology stacks.
convergence of two maturing technology stacks.
So on one side, model capabilities have improved substantially
while inference costs have declined
and model specialization has become more viable.
On the other side, blockchain infrastructure
has matured enough to support low-cost program
coordination around ownership, settlement, incentives,
and access control.
So what the market is reacting to is not just AI plus crypto,
it is a realization that intelligence and economic agency can now be coupled in a native way.
Yeah, that's my answer for the first question.
Yep, yep. And for the next one, I would like to combine two questions because I believe
most of the answer you have provided in the first one is I would like to combine two questions because I believe most of the answer you have provided
in the first one is completely covering also
the second question too.
So that's why I would like to only understand
how AI agents are different from smart contracts
and from the traditional boards.
And according to you guys, what are the recent changes
or improvement that AI agents are doing
actually viable now in Web3 particularly?
Yeah, great question. So I think the main difference is flexibility and intelligence.
Smart contracts are very rigid. They just do what they're programmed to do, nothing more.
And traditional bots are also similar and follow fixed rules.
So if X happens, then do Y.
What makes AI agents different is that they can actually make decisions based on context.
They can learn over time and adapt to new situations.
So instead of just following instructions, they can figure out what to do
and then use things like smart contracts to actually execute, excuse me, actually execute it.
And I think I think what's changed recently is probably a combination of a few things coming together at the same time that sort of makes AI agents viable now.
First, the big one is that AI models have improved a lot, especially LLMs.
So now they can actually reason, handle multi-step tasks and adapt instead of just giving
simple outputs. And secondly, the tooling has matured. So it's much easier to connect AI to
workflows, APIs, different systems, which makes building agents way more practical now.
And then on the Web3 side, everything is already fully digital and on-chain, so agents can
plug directly into that environment and act without needing any human intervention.
So I guess it's really that convergence, better AI, better tooling, and the Web3 infrastructure
that's turned agents into something
you can actually build and use today.
And I think that's why we're starting to see
real use cases emerge and not just experiments.
Yeah.
Interesting, interesting.
Andrew?
Yeah, great question.
In my opinion, the difference between a smart contract and an AI agent,
the smart contract, it's basically rigid.
There you have if and then and the logic.
And the AI agent, it's proactive and it's adaptive.
They have a specific goal.
They need to find vulnerabilities in a code.
They need to have a task to go for the medical office or whatever the business can be.
They observe data, they plan the steps and they decide the course of action.
It's like having an expert employee with their own office keys.
In this case, the office keys would be the crypto wallet.
And in terms of improvements,
I think having the on-chain identity,
the agents now have a verifiable reputation
and you can see the agent's success rate.
And before basically copying the agent or deploying this agent,
make a copy of this agent for yourself
and to trade for you or to do various tasks for you,
the on-chain identity is a recent improvement
that's on-chain.
Okay. Awesome. All right.
Okay, so to me, a smart contract and AI agent are both forms of software automation, but
they operate very differently.
A smart contract is deterministic.
It follows predefined rules written into code and execute exactly as specified once certain conditions are met.
It does not interpret contacts, adapt its behavior, or make judgment calls.
If you define that, so the smart contract will do that.
Yeah, and an AI agent is different because it is probabilistic and goal-oriented.
It can interpret inputs, reason over context, choose between actions, use tools, and adapt based on new information.
Instead of simply following one fixed path, it can decide on how to achieve an objective that makes it more flexible,
but also less deterministic than a smart contract.
And what, according to you,
what are the improvements AI agents are doing right now
and that actually we have now?
Florence?
Oh, sorry?
And one part of that same question is, what are the changes recently that made AI agents
recently that made AI agents actually viable now?
actually viable now?
You mean, what are some viral AI agents currently?
Yeah, implement or changes.
Okay, so recently we created on our intelligence cubed
and auto routerouter agent. So we have like more than 500 our intelligence cubed models.
And if a user's like prompt or like create a task and which the user does not know like which specific model they want to use.
model they want to use. So we use this like all router agents that defines from the our
our democratic dynamic benchmark based on the proof of intelligence score and other like
latency and stuff. It will eventually retrieve the best model that that match with the user's task.
And so that's what we're currently building on.
And we receive a lot of like on-chain usage recently
that currently a lot of users are actually using
our Aurora agent, trying out our Aurora agent, yeah.
Interesting. And one more more thing which is also i feel like
agents can communicate to each other but boards can't but what can't bots new uh and
bots bots basically uh bot can cannot interact each other, but AI agents can.
Yeah.
And now let's go to the segment two, where again, I'm combining two questions.
So what does a real AI agent stack look like on BNP chain? And how do you design an agent that can act autonomously but still remain trustless?
agent that can act autonomously but still remain trustless?
Yes, I think in practice, it usually looks like a hybrid stack. So the AI layer handles the
reasoning, decision making and understanding of what the user wants. Then you've got the
infrastructure layer, which is things like the backend, APIs, wallets, data sources that help the agent actually function smoothly.
And then you've got BNB chain,
which sits on the execution side
where smart contracts handle things like payments,
ownership, permissions, and any on-chain actions.
So the simple way to think about it
is the agent does the thinking off-chain,
and then BNmb chain handles the trust
and execution on chain um that's that's probably well it's really the model that makes the most
sense today because it keeps things scalable while still using blockchain where it adds the most value
got it got it yep andry Got it.
Got it.
Yep.
Andrei?
This is also a great question.
On BNB chain nowadays, an AI agent task is not just a script.
It's like a verifiable lifecycle.
It begins with the identity and the trigger.
It follows with the execution.
It goes further with the on-chain settlement.
And in the final result, it would be reputation update.
In short, a task on BNB chain, it's a transparent loop of identity. It's an off-chain work.
It's an on-chain proof and it's an automatic reward.
For example, it's efficient.
It is scalable.
It is completely trustless.
Okay.
All right.
Sorry, can you like repeat the question again?
I didn't quite catch it. Like, yeah. Yeah. Okay. Okay. Let me repeat the question again? I didn't quite catch it.
Yeah.
Yeah.
Okay.
Let me repeat the question.
The question is, what does a real AI agent stack look like on BNB chain?
And how do you design agents that can act autonomously but still remain trustless?
So you mean like what stack that we prefer what users use to build an agent?
Yep.
It can be AI infra smart contracts,
that request will be on BNB chain.
And how do you design an agent that can act autonomously but still remain trustless on BNB chain?
So I think it'll eventually, like...
eventually like um so eventually the fallback to reliability, observability, fallback behavior,
latency, discipline, and a clear scope matter more than um maximum ultimate autonomy in the early
stage um so a constrained agent that executed more workflow exceptionally well is usually more viable than a high ambitious but unstable system.
Hello, can you hear me? Yes, we can hear you.
I think Florence complete her questions, her answer, right?
I think we can move to the next question because as of now, you guys know this is a session for builders who are building
on a bnb chain and we are educating them and motivating them to build more agents on bnb chain
so just to help them the question is what tools sdks or in in front bNB chains would developers explore first
when they are planning to build a agent on BNB chain?
And what are the biggest mistake a builder make
when trying to build AI agents?
Yeah, great question.
So I'd say for the first question,
I'd say start with the basics first and keep the stack very practical.
On the BNB side, the main things are solidity for contracts, a good RPC connection, and the official BNB chain docks for BSC or OPBMP, depending on where you want to build.
BNB chain also has a really good developer tooling hub, so that's a good place to get
oriented quickly. And then for the agent side, use whatever LLM framework or backend stack you're
already comfortable with. I think the real value is in the workflow, not forcing a specific AI
framework. And if you want even lower cost, high frequency execution, OP BNB is especially worth exploring.
And so honestly, I'd start with one simple loop, agent logic off chain, smart contract on BNB chain, wallet and RPC connected, and then build from there.
Yeah, that's kind of the way we think about it.
And then your second question,
what are the biggest mistakes builders make?
Was that the question?
Yep, yep, yep.
So I see a few common mistakes coming up a lot.
The first one is probably overcomplicating things too early,
trying to build full multi-agent systems
before you've got one useful workflow
working. The second is focusing too much on the AI itself and not enough on the actual use case.
The agent needs to be solving an actual problem day to day. And the third is trying to put too
much on chain. In reality, most of the intelligence should probably stay off chain and only the
important parts go on chain. So I think the best approach is start simple, build something
that actually does one thing well, and then expand from there. So yeah, that's probably
the main idea.
Genuine. It sounds genuine.
So yeah, Andrej?
Yeah, so BNB Chain ecosystem has released specialized infrastructures to handle the needs
of the developers.
You do have the BNB Agent SDK.
It's an essential tool to build the agents on the BNB chain network.
You can use also the ERC 8004 protocol for this.
You can use the Binance AI agent skills.
You can also use BNB Greenfield for this and also ELISA framework you can use to build
agents on Binance blockchain. There are lots of tools that BNB chain has provided for the developers to effectively build.
And for the second question in terms of the mistakes,
I think, in my personal opinion, I think the to need to focus on one agent that
performs one task and this agent is not trying to do basically everything they
can use a multi multi-agent structure not train one model or one agent to do various tests because that will fail.
You mean basically start with something simple then once it starts working then
maybe you can add more things or maybe create more agents, right?
Yeah, I mean if you are going to to deploy an agent for now example we do
have an agent that audit smart contracts.
This is what he does. He doesn't go in the healthcare, the same agent.
So it doesn't mix the tasks that he has. So the memory, the context and so on.
So basically, if you are trying to build something, stick with it.
And if you do want to expand the services on the project or on the website or so on,
start building a new agent that is specifically meant to do that task.
Yeah, yeah, got it, got it.
It's like making experts.
So, Florence?
Yeah, so for the second question,
my advice would be to begin with a well-bounded problem definition and a modular architecture.
So first, not start with an abstraction where, like,
I want to build an agent.
Start with a concrete operational question, like,
what decision or workflow or user tasks do you want to improve?
Strongest product usually merged from a narrowly scope but highly valued use case.
And use a composable stack.
So in most cases, that means strong reasoning models,
retrieval where external context matters,
tool invocation and controlled execution layer,
and memory only where persistent is generally useful.
Yeah, so I think that's my advice for question number two.
Okay, interesting.
Yeah.
So now let's go to segmentary,
where we'll learn real use cases and opportunities for AI agents.
So the first question is, what are the most exciting AI agent use case you are seeing right now as a founder or key people in projects building AI?
So there's a lot happening right now, but a few stand out.
So there's a lot happening right now, but a few stand out.
I think DeFi is a big one, especially trading and automation where agents can execute strategies continuously.
Then you've got assistance to help users navigate Web3, which I think is really important for onboarding.
And something we're really interested in is agents tied to user-owned models where they can provide services and even monetize.
So it's not just about automation.
It's about turning AI into something that can actually generate value for users, if that makes sense.
Yeah, exactly.
It makes sense, Yeah, exactly. It makes sense.
Andrei?
Well, moving beyond the digital world,
the agents are now performing real-world businesses
services.
And for example, in our case at SmartSentinals,
we are using the proof of useful work where we basically bridge Web2 businesses also like medical offices from Romania.
And we are bridging a Web2 business with Web3 with the protocol of the proof of useful work where these agents are working off-chain and they settle these rewards towards
our user on-chain and we can scale the project basically everyone that's got an AI agent doing
whatever the use case it can be added to the network so in my opinion having AI agents on-chain
So in my opinion, having AI agents on chain,
it's an infinite possibilities that you can build
and do with them.
Great, Florence.
So I'm looking through some security operation agents.
Security is particularly a strong fit because
workflow is already too tool heavy, data rich, and time sensitive. Data agents can
try alerts, investigate anomalies, correlate signals, draft response actions, and help analysts analysts move faster. So yeah, I'm kind of into security agents that's
heavily on observation and governance
and least privilege controls for agent operating
at enterprise scales.
Interesting info guys.
Interesting.
Actually, right now, after listening to your answers,
now I'm going to skip the next questions
because I think you guys already answered this one.
So now let's go to the next question,
which is basically you're going to give a alphardrop
basically to ideas or maybe suggestion to the audience
or developers listening to this session.
So what's an idea that no one is building according to you and
the people who are listening to this AMA, this would build this?
So yeah, I think one idea I don't think is fully explored yet is agents that actually manage their own revenue streams.
So not just doing tasks, but operating more like a digital business.
For example, an agent tied to a specific AI model that can create content, distribute it and monetize automatically.
So instead of constantly interacting with tools, you're owning something that's generating value in the background.
That's very aligned with what we're building here at Coscription, turning AI models into assets that can actually earn.
Well, I think what the new developers or developers can build in a scam or stuff like that, having
Andrew?
chain analysis or stuff like that that require manual entry takes a lot of time.
And I think an AI agent that can investigate the source of the hack, the wallets that were
involved in this hack, perhaps you could get the results faster.
So yeah, this would be like an idea
for the developers to try and build.
I'll just start.
Sorry, it's a little bit laggy over there.
So I think agent reputation liability layer is very important because right now everyone is building agents that can do more but um very
few are building infrastructure that answer um that answer is like uh should this agent
be trusted will like hack my data like information and what is it
allowed to do or who is accountable when it fails how do I compare one agent to another in production
not just in a demo so I think like I think like agent security reputation and liability layer is
very important yeah I think in the future, hopefully,
more agent will focus on security and liability side.
Yeah.
Yeah, true, makes sense.
So now our last segment, where we discuss future
you guys gonna give some builder advice.
What will AI agents on BNN look like in the next 12 months, according to you?
So I think we're going to see AI agents move from experiments
to something much more integrated into everyday web through
usage. Right now, there's a lot of agents. It's sort of still very early, but the growth is already
crazy. We've seen a huge spike in on-chain agents just this year, especially on BNB chain.
So the next step is agents becoming the interface layer. So instead of users clicking through dApps,
they'll just tell an agent what they want and it handles everything in the background. We'll also start seeing more
multi-agent systems where different agents coordinate together to handle more complex
workflows. And then probably the biggest shift is agents becoming actual on-chain assets where
they can hold funds, execute strategies, and even generate revenue over time.
So overall, it moves from tools you use occasionally to systems that are
constantly running and acting on your behalf. At least that's my opinion anyway.
In my opinion, these agents in the following period of time will start to look more and
more like employees to do useful work for a business and you can monetize this business.
I think we should focus into going towards having utility rather than having a task that can be copied, like a trader agent that lots
of people can copy these agents and they can do the same trade.
But I think we need to focus to give these agents, these AI agents and the power that
these AI agents have to give them more utility and to try and combine real world businesses.
So to bridge Web2 with Web3, in my opinion.
We should focus on this and hopefully
the next period of time we will go towards this.
Yeah, makes sense, makes sense.
Makes sense.
Florence?
So I think AI agent on BNB chain will probably look less like standalone chatbots and more like production-grade economic actors.
The direction BNB chain itself is signaling is very clear that standardized
agent payment, agent identifier, reputation, verify vocabilities, and hybrid off-chain,
on-chain execution are becoming core infrastructure rather than experimental features.
experimental features and I think we will see a shift from personality-led agent to utility-led
agents in the future. So early Web3 agent got attention through memes, social presence,
and speculation. I think the next wave on BNB chain is more likely to be like task oriented,
trading copilots, research agents, customer facing service agent, and on-chain execution
agent, etc. So I think, yeah, so it's like an evolution
moving from personality to utility.
Okay, okay.
So now my last question is,
let's for example, I'm a developer watching this session.
And also I am so motivated developer
who really want to build something,
but I have no idea what to start with.
So if I'm a developer watching this,
what should I build in the next 30 days?
And one piece of advice that you guys want to give
to all the builders watching us
entering AI and Web3 right now.
Yes, I'd say don't overthink it
and just build something simple but useful.
Pick one clear use case,
maybe a trading assistant, content generator,
or automate something,
something you're interested in.
Build a loop where the agent can take an input,
make a decision, and actually execute something
so yeah even something basic connected to bnb chain for payments or access control is enough
to start if you're exploring ideas around user-owned models or monetization that's the
sort of direction we're pushing at co-scription as. But I think in 30 days, the goal would be, you know,
really just proving something works
and then iterating from there.
The most important thing, I think,
is just to give it a try and start small.
In terms of advice for builders entering AI right now
and Web3, I'd say just focus on real utility over hype.
It's so easy to get caught up in trends,
but what actually matters is building something people will actually use.
Start simple, solve a clear problem,
and just make sure your agent actually delivers value day to day.
And don't try too hard to decentralize everything from day one.
Use Web3 where it makes sense, you know, for ownership, monetization.
That's something we're really focused on at CoScription as well,
making it easy to go from using AI to actually owning and earning from it.
So, yeah, I think if you get usefulness right, everything else should follow.
So yeah, I think if you get usefulness right, everything else should follow.
Well, Clark said this very well. So Clark had a point here.
Utility. This is it. If it has utility, people will use it. If people will use it,
you can basically monetize it into a project. So this is the best answer that I can also give.
Utility and as an advice would be never to give up.
Keep trying, keep trying, keep trying until you find the right market share,
you find the right people, you find the right partnerships to get you to
to a successful launch never give give back and don't give back uh don't give up don't give up
yeah completely well thanks
well said yeah thanks for motivating developer Florence.
Yeah, I agree with both founders' recommendation.
So yeah, for me, I think also to start small first
and like to not start with a really broad idea.
And then I think it's just like, keep on trying
and then eventually you'll find the best like on the past or like
a specific past that you want to um achieve um so yeah and i think um please like also
while you're like being really creative creating an agent please make sure it transacts safely
and there's like um and it's like,
like I think right now it has to be like transact safely
and it be trusted inside real applications.
And yeah, I think just have fun and yeah.
Yeah, so the summary I can make,
the main thing should be the utility.
Utility is the king.
If you don't, it will take you anywhere.
So yeah, thanks for the advice and for the idea
and as well as I can say the key things you mentioned
about the utility.
So now the segment four is completed.
We are about to end our session, but before that we can,
let's give few seconds to our audience.
If they have any questions, they can ask us and comment.
But for now I can see,
we have already covered lots of important questions
and you guys have already answered very well
to all of my questions.
But let's give few seconds if they they have any questions then after that we can
conclude I think nobody have any questions, but still guys, if you are interested to ask any questions,
you can reach out directly.
Okay, I got a question.
So what kind of data do you store on SE for AI agents?
I think it's for Andrew. Can you repeat the question, please? What kind of data?
What kind of data do you store in SE for AI agent? In smart contract. Okay, it's smart contract. Okay. So in SC, do you mean SC as in short
for smart contract, the data? Yes, it's a smart contract. Yeah. Well, at SmartSettinels, we also
work with intelligent NFTs, where basically the metadata of this nft it's basically on chain
so um basically everything the device the ai agent it is running on the model the settings
of the ai agent basically everything is on chain is verifiable and this this is also something that
This is also something that we can use in the smart contract.
And we are also using gateways.
Each agent has its own smart contract that's basically a gateway towards the task that
he is performing.
So the agent is performing a task, we have a smart contract that acts as a gateway that outputs the interaction
of this agent on chain. I hope that this answers the question.
Claire? Claire, do you want to add something?
No, I think he summed it out perfectly.
Florence? Yeah, I think Nreal did a really good job answering the question.
Okay.
Okay.
So, I think no more questions.
We can conclude the session now.
And before that, I would like to thank you again for joining this session to make it
more successful.
And all audience, if you still have any questions,
please directly reach out to all of our guests and speakers
by their social media or using their community Telegram
group.
And you can have your questions to them directly.
And thanks for everyone participating in the AMA
and making it successful.
See you soon.
Have a good day, good evening, and good night.
Thank you for having us.
Bye.
Thanks for having us.
Great to chat.
Welcome. Thank you.
