Stephen: If you're wondering why I'm smiling ear to ear, 'cause the Around The Coin podcast has two of the coolest guests that we could ever imagine, Finance, Itai from Dynamic. They're both under the Fireblocks label, and we're talking all about agentic AI. If you listen at the end of the episode, they give you really good advice if you're an employee or a founder, how to leverage some of these AI agents.

But we talk about the actual use cases of agentic commerce agentic AI where does compliance and cybersecurity fit into it.

We talk all about the use of stablecoins and potentially other tokens in this invisible background agentic economy that we're gonna see that's flourishing. This is not about buzzwords. This is deep understanding of where agentic AI is today, who's using it, how compliance teams are leveraging it, and how founders are building these amazing companies.

This is gonna be one of our favorite podcasts, anytime you have two cool founders like Itai and Tal. So listen in, and then reach out to them on Twitter or LinkedIn, because we're a fan favorite of some of their family members apparently.

Listen in to this full podcast. It's amazing.

Welcome to the Around The Coin podcast. This might be the coolest podcast on Around The Coin 'cause we have two of the coolest founders that I know. We have Itai and Tal. Maybe give a brief discussion, Itai. This is not your first time on the pod. I think people know you're a fan favorite. Plus me, you, Tal, have all been on the Public Key podcast, me listening in the background.

So maybe tell us a little bit about yourself, Itai, what you've built, what you're working on at Fireblocks, and then we'll go to Tal

Itai: Sounds great. Yeah. So to your point, f-first, thanks for having me. Uh, second, yeah, longtime listener, second time caller. Uh, but, uh, you know, a little bit about, uh, myself. Um, you know, I started-- about four years ago, we started a company called Dynamic, which was backed by a16z and Founders Fund. Uh, the premise was pretty straightforward, which is if you kinda fast-forward a couple years and look at your phone, every app on your phone is gonna have a wallet component to it.

Uh, it's gonna move money globally through, uh, crypto rails. Uh, you might not know about it, you might not care about it, but that's pretty much gonna be, uh, what happens in the future when you open every app on your phone. That's actually happening today, which is pretty exciting, so our prediction is actually becoming true.

Every, every year since starting the company, we said five years, and now all of a sudden it moved from five years to now. Uh, and then, uh, about six months ago, uh, we were fortunate enough to partner with Fireblocks and, and, uh, join the Fireblocks team to essentially expand, uh, the mission of Dynamic. So we, we get to do the same thing just on, on bigger scale.

So we now get to com-- uh, to partner with companies like Papaya Global and Western Union and, um, uh, giant exchanges and, and, and many more, uh, the Circles of the world, et cetera, to essentially bring that vision of every app on your phone, uh, should have a wallet imple-- uh, integrated into it, uh, to life. So that's pretty much, uh, what we do for a living, uh, now as part of the, uh, larger Fireblocks organization.

Stephen: And Tal, you took reconciliation, backend operations, accounting, and, actually made it cool, uh, in Web3. So maybe give us a little bit of background on what you've

built

Tal: Sure. First of all, thanks for having me. I will say that

Itai is one of the best founders out there and super hands-on, and I will never forget, I think we were on the same flight, uh, can't remember from, from where to where, and then I was like: "Yo, Itai, what you gonna do during the flight?" I think I'm gonna take a nap.

And you were like: "What do you mean, man? I write blog posts every time I fly. I can like... I write like three or four blog posts every time." And I sort of, I felt so bad

Itai: I was just making you look bad. I slept the entire flight. I was, uh,

Tal: Okay. Okay. I'm re- so I'm happy that we closed that, that story now because I

Itai: that's, that's right

Tal: was like, there we go. Then like every time we go on the flight I'm like, "What will he tie do? I'm gonna write a blog post." Anyway, um, so I'm Tal, I'm the CEO and co-founder at Trace Finance. Uh, we started the company early 2022, um, making accounting back office reconciliation for digital assets cool, as you said.

Um, we scaled really quickly over 250 customers, uh, working with some of the largest names in TradFi, exchanges, asset managers, and we were acquired at the beginning of the year by Fireblocks for $130 million. Um, and Fireblocks was just like such a, an amazing integration because, you know, from day one the goal was to target all Fireblocks, uh, customers.

And it takes me back to like the, um, September 2022, a few months after we started the company, um, uh, there was an event called Spark, which is a, uh, Fireblocks, uh, event, uh, for their customers. The first time they did it, uh, ever, I think since then there like three or four Sparks. And I had to kind of, uh, sneak into the conference.

Of course, like there was lots of security at the conference. I was caught at, at the beginning, they kicked me out, but eventually I was able to buy a ticket and get in. So it's like such a crazy ride from like trying to sneak into the conference to be acquired by them and, and be part of the team. So, uh, it's an amazing, uh, uh, privilege to work with, uh, Fireblocks and tap into this huge market and get kind of the enterprise buy-in that we are seeing today.

And we can talk about it later about like what kind of customers we, we serve and like how it's, it's evolving now under Fireblocks umbrella. Um, but yeah, accounting, reconciliation, audit for digital assets, that's what we do at Trace Finance.

Stephen: curious, you know, other than your, you know, illegal disguises to sneak into the conference, what do you think gave Trez Finance the edge in a space where there's a lot of accounting, a lot of, you know, c- traditional companies transitioning to Web3? What do you think gave you the advantage? Was it your crypto native aspect that approached the industry?

What are your thoughts on that?

Tal: I, I think that actually the fact that other teams that were building similar products, uh, were usually, uh, CPAs in their past and had like this finance background, and Eylon, our co-founder, and I, we come from more of this, uh, cybersecurity background. So the way we looked at the problem was very similar to like cybersecurity network defense.

And we said, okay, every wallet is an endpoint. If we sit on the endpoint and we read all the movements in and out, okay, packets moving in and out, we're able to go into the metadata and analyze it, then essentially we'll have all the data. And when it comes to accounting, essentially accountants are data scientists, right?

They just never got the glory of being a data scientist. And if you're able to get all the data normalized and reconciled, the ability to build accounting and back office applications on top of it becomes really easy. So when we, when we started TRES, that was the beginning of 2022, there was this influx of, uh, different blockchains and exchanges, and our competitors built an accounting application that was more of an application based on three or four networks that were live at the time.

So when they started seeing the scale of networks, things started to break, and they couldn't provide these customers with the coverage. While for us, adding new blockchains, new exchanges and more, more data sources was native to how, uh, uh, we built the infrastructure from the get-go. So our ability to scale and serve any customer saying to them, "No matter where you work, what venue, what asset, what DeFi protocol, what exchange, what bank account, we can give you visibility to everything and do the back office for everything," that resonated, and we were able to kind of take off and say kind of bye-bye to our competitors and, and, yeah, win the race.

Stephen: Yeah, I don't think traditional accountants are gonna think that way when approaching a new industry, and it definitely makes sense. Uh, Itai, your team authored this report, you know, Agentic Finance and Stablecoins. You know, agentic commerce has been, I think, overused. Um, why don't you give us the lay of the land?

There's been a lot of conversations around AI agents doing everything from booking our flights to transacting on our behalf. You know, every day there's a, there's a new Instagram or TikTok post about, "AI agents are gonna take my job and your job, and Megan's-- everyone's job is gonna be taken by AI agents."

Why don't you level set the landscape of what you're seeing right now and, you know, maybe the press for this for creating this amazing report?

Itai: Yeah. So first, I think your job is safe. So in case you're wondering, I think you're good. But, uh, yeah, to, to your point, uh, I, I think the, the, you know- I, I don't know today if there's any more debate around whether, uh, AI agents will b-be a part of our life, right? But I, I think we all agree on this podcast and anyone listening is that, uh, AI agents will be a part of how we kind of interact with the world, right?

They're gonna be helpful for us in our day-to-day life. They already are, right? So we wake up in the morning, and we interact with, uh, with, you know, your Claude or Claude Code or, or, uh, uh, Codex or ChatGPT or anything of that sort, right? And, and there's no debate around that. The, the second order is will A...

Will, will folks that help us on a day-to-day basis also be able to take action for us, right, on our behalf? And that's, that's where payments come in, right? To take action, you need some sort of way to pay the other side. And so I think a couple months ago and a couple years ago, there was a debate of whether that would be possible or how we will do it.

I don't think there's a debate around that anymore, right? I, I think we can all agree that in a couple years, um, not-- you will be able to delegate some sort of authority to someone to, um, buy your ticket to whatever or transact on your behalf. Uh, we can also, I think, all agree that, uh, in a couple years, agents will just say, "Hey, this, um, specific thing that I wanna do is paid versus free.

Uh, let me just pay for it, and I have some sort of budget, and, uh, I will complete my task that way." Right? And so all these l- kind of, uh, hypotheses are kind of coming true, and I, I don't think there's a debate today anymore about whether that will happen. But the question is how, right? The question is what type of rails, uh, will enable these agents to organize your trip for you or to, uh, compare the different, uh, shoes that you, uh, potentially...

I was just buying shoes this week, therefore the shoes analogy. Um, uh, and by the way, I bought them through, uh, or I did most of the work through ChatGPT, uh, and I will report back on whether that worked well. But essentially, uh, whether you're buying shoes or whether you're compiling a report like this and you wanna get a bunch of data sources, the key question is not whether agents will take action on your behalf but how they take a-action on your behalf, right?

This is where the report comes in. The report pretty much, uh, says that, uh, today's payment system is very much a human in the loop payment system. That inherently means that cards, um, that, uh, ACH, et cetera, is built, uh, around a human, meaning I initiate a payment or I accept a payment. There's some sort of human-to-machine interaction or alternatively a machine-to-human interaction of some sort, right?

Uh, I go to a store and I buy, um, and therefore the type of mechanism for payment, uh, can be a certain way. In our, um, you know, in our world it's, it's usually a credit card, right? It takes the shape of a token that you provide to someone in an exchange of money. But the system is built for, you know- You pay 50 times a day, 10 times a day, you know.

Uh, if you're going crazy, you will pay fifty. I, I don't know who pays fifty times a day with their credit card, but, you know, maybe. Um, you know, uh, but what if an agent has to pay a thousand times a day or ten thousand t-times a day, right? And what if the payment isn't a hundred dollars, but rather a hundred or a thousand payments of a dollar, right?

At that point, the credit card system kind of breaks, right? Because you have base fees and you have, uh, transaction costs, and it, it doesn't necessarily scale, and it doesn't necessarily scale to, um, kind of agent-to-agent k-kind of, uh, interaction. That's where crypto, uh, rails come in, right? And the report that we talked about, and this was the longest intro of all time for a report, but the report, uh, that we're talking about is essentially saying that the right rails for agents to do this and act on your behalf are rails that move money instantly at under a subset-- uh, kind of sent cost and do this without you knowing.

And the best rails for do that-- doing that globally are crypto rails. And on top of it, you have a bunch of innovations like X, X four two and MPP, uh, and others to kind of try to align that. But at the very basic level, all we're really saying is actually something pretty straightforward, which is agents need a better payment system that can handle a thousand actions per day or ten thousand actions a day versus fifty actions a day or ten actions a day, and the best way to do that is crypto rails.

I will cover, and we'll get into this. I'll end it with one thing, which is most users listening, most, uh, folks, my, uh, kids or my wife or my dad should not care that these are crypto rails, should not care that these are stablecoin. Don't need to know about any of this stuff, right? It's fun for us, um, uh, crypto geeks to talk about, but it's just better, faster money movement that happens at scale.

Stephen: Tal, this sounds like a back office nightmare. You know, AI agents making payments to each other, transacting using different blockchains and different stablecoin providers. How

are you gonna be able to manage this from a backend perspective? And then are you gonna use AI agents to manage some of these interactions and transactions that are being made by other AI agents?

Tal: Yeah. So, uh, what you do is you answer AI with AI. That's the, that's the way to go. Um, we, we launched, uh, last month our, uh, MCP and also inside Tres we have, uh, Vara, which is our new AI assistant. So essentially the idea is that back office in general is super repetitive, right? Labeling transactions, going through their accounting rules, making sure the cost basis is aligned, looking for the gaps.

Now all these things, they are very much the same every day. The problem is the scale of transactions which we just talked about, like if there's an agent that's in these transactions, even if it's micro transactions, just becomes absolutely impossible to track. And the only way to actually keep track of everything and make sure that you are compliant and, and your books are closed and you're able to align, uh, your numbers for audit is using software and the next level is us-using agents.

And we've built not only the autopilot that will take care of all these repetitive, uh, issues, but also we built a bank or let's say a marketplace of skills that our team has developed in-house, so our customers won't need to do that. So essentially, once you connect your, uh, Tres finance to Claude, and by the way, we already talked to Entropic and we are now listed as an official partner on their, uh, marketplace, then you also have a bunch of skills you can just offset into, uh, your Claude UI and start closing your books or going through any kind of repetitive task that you already have.

Um, so it can happen on Claude, it can happen, uh, OpenAI, ChatGPT, it can also happen natively inside the Tres, uh, UI. Uh, but essentially it's kind of the perfect storm for us to, to leverage and offer our customers so much more with less

Itai: by the way, just to add to that, um, one thing to add to that is that it, it comes back to the exactly this power of like the way the trust system was built on scale, right? And the ability to kind of scale this type of stuff, right? Because coming back to my point of people should not know about this, and that's either individuals or businesses.

It should just kind of work, right? And it works across kind of acceptance of payments on if you, you run, uh, your business on a Fireblocks treasury can just accept it, and reconciliation of that with trust kind of just works, right? It, it kind of magically works whether it's ten transactions or a million transactions, right?

That is the, the beauty of this, is that we wanna build out these agent tools across kind of an agent having a wallet and payment to an agent receiving a wallet and reconciling that, right? All in a way where we really care, but no one else should, right? And so that's kind of the end-to-end experience here, um, that is only na- uh, available with like s- tools that work at scale

Stephen: And I'm curious, both of you are working on Fireblocks. I just spoke with Yatsu at Animoca Brands. They've built their,

Animoca Minds, which is their, you know, AI agent for everyone, uh, solution. But they also have the Moca Network and talking about digital ownership. When there's gonna be millions of agents being deployed, what are your thoughts about digital ownership and identification when it comes to the agents?

Maybe this is a great segue to X for, you know, 02 that you're a part of. Maybe kind of talk to us about verification, identification, uh, and all those things that matter as I'm assuming to the majority of Fireblocks customers.

Itai: Yeah. A-absolutely. I can start and, and Tal, please, uh, feel free to kind of jump in. But to your point, it, it's a really interesting cha-- It, it's, um... When you start getting, uh-- And all of these are kind of moving from experiments to live kind of, uh, being in the wild. And what you see is that when you start moving from, you know, uh, agents trying out transactions to actually doing this at scale, you run into these second-order problems, right?

And one of the big second-order problems is identity for agents, right? Which is, okay, I'm an agent. Do I get my own wallet? How do I identify, right? Uh, I, you know, a human has an email address or a phone number or some sort of, uh, KYC information, right? And it's-- the, the web is very much built for humans, right?

How do I identify, um, yeah, who I am as an agent? How do you prevent, uh-- The web has been building for, for many, many years to prevent bots, right? Uh, like the entire CAPTCHA industry is pretty much like, uh, let, let me ensure only humans interact with my site. And now we're pretty much switching that to only agents interacting with my site, right?

So the entire architecture has to completely flip, which means, okay, how do I identify that I'm not a rogue agent? How do I, uh, ensure that, you know, I'm paying for what I have? But how do I maintain persistency? So let's say, um, as an example, I ask for a wallet, right? And I get a wallet. How do I make sure that when I lose my context on my end, right, Claude loses context once in a while, how do I maintain that wallet, right?

How do I ensure that I don't lose money, right? So all of these are unsolved questions, some of which are addressed already today with X for two, which is that handoff. It's actually kind of a, um, oh, it's built in partnership with Base, but it's actually kind of chain-agnostic, right? It works on Solana, et cetera.

Uh, is starting to define, um, those handshakes for agents of how do I move money, right? I have a wallet. Let's say, uh, me as an agent have spun up a dynamic wallet Uh, m- the merchant has a Fireblocks account, right? How does that communication happens, right? How, how do we actually move money from one to two?

What is the start of the conversation? What is the reply or what's called a facilitator, which is, uh, the Fireblock side or, like, the, the recipient side, right? And what's that conversation back and forth, right? And that's what X42 is. Um, there's a different competing one, uh, called MPP. Google has their own, right?

So the first line of o- uh, order of business is let's figure out the communication between two parties, between, um, a recip- a sender and a recipient. Then the second order is essentially saying, "Okay, wait, how do you identify them over time," right? So all of these are unsolved problems that are being discussed today.

For-- It is, it is an extremely fun part of the life cycle of, uh, the moving to agentic payments because you're starting to see those second-order problems and third-order problems, uh, that you could not see at the beginning. We're gonna run into-- I'll end with this, which is y-you, okay, you move money around.

Now how do you do subscriptions, right? How do you do recurring payments, right? So all these types of questions are being, um, solved with these standards. And one cool thing is the industry is coming together around those standards rather than trying to invent, you know, uh, their own thing and trying to build kind of proprietary solutions.

Stephen: Good luck for these AI

agents to try and solve the CAPTCHAs. It

takes me three or four times to figure out which is the wheel and,

you know,

Itai: Oh, I've gained it up. Yeah

Stephen: If I can, if I could get an agent to solve those for me, I'd be a lot

quicker on the internet.

You know, one thing from the report, it shared some interesting, like, agentic trends that you're watching in 2026.

Tal, maybe I'll start with you. Compliance agent teams was one of them.

You've worked cl- closely to embed in, you know, many compliance and probably legal departments and accounting departments. What does a new look compliance team leveraging AI agents look like going into 2027?

Tal: I think that with everything that's happening with AI, the ability to stay a SaaS platform only has kind of lost its purpose, and there is a lot of pressure coming from the-- these compliance teams to get much more done in the way that they want it to be done, right? Because they have the option of going and building things in-house that are very custom to what they need.

Now, the option exists doesn't mean that they're gonna go and execute on that because of course it also requires a lot of resources. The way Tress is envisioning, I would say back office, accounting, compliance, is that we want to give the compliance team or the finance team an army of agents that doesn't only give them the option to interact with the data, but also takes care of all the tasks that they need to do. Okay? I'll give you an example of something we've implemented in Tress. Inside the Tress platform, soon you will see kind of a Kanban, uh, um, worksheet. And if there is a gap in some, in some task or some process or some workflow that you're building, let's say closing the books, there's a reconciliation issue, you can actually open a ticket and see the ticket inside Tress.

An agent that we build will go and read these tickets and start solving it for you. It will move it into in progress. So essentially you get this army of agents that are like an extension of your team. You create the tickets for them. Sometimes they'll even create the tickets already for you, but you'll have visibility into what they're working on, and they'll start moving it into done and closing the workflows for you and the gaps for you, giving you the output, showing you what the issues were and the root cause for those.

So you also have some kind of, um, audit trail, right, if you need to go through audits, et cetera. So instead of having to kind of manipulate some kind of SaaS platform that maybe answers seventy-five percent or eighty percent of your workflow, now you have the, the SaaS itself, the agents working for you on top of it that know the system better than you, right? Taking care of the whole workflow from A to Z

Stephen: And I'm curious, you know, maybe Itai, you can help me with this one. When it comes to blockchain analytics, obviously as a wallet service provider, you understand that wallets have to identify things like sanction screening. Because each of these agents should have like a, a base wallet or something similar, will we need to do-- will we need to reinvent the wheel when it comes to blockchain analytics?

Or can we just, you know, take that same address, pop it into Chainalysis or TRM or Elliptic or Score Chain and be able to identify risk without having to r- you know, revolutionize the way we do blockchain intelligence and analytics?

Itai: Yeah, it's a good question. I, I think t-to your point, there, there are, uh, really phenomenal tools today, right? So s- you know, today, if, if, uh, um, y-you, uh, want to essentially, uh, check whether transactions or there's, uh, sanction screening on, on an address, you have Chainalysis and you have TRM, uh, and you have kind of, uh, transaction simulation tools like Blockade and Hypernative, which are extremely powerful tools, right?

Uh, that is gonna accelerate, right? Not decelerate. Uh, because, you know, crypto doesn't live in its own la la land. It lives in the same, uh, world of, of, of kind of compliance and AML and KYC r-regulations, uh, that, that exist today, right? And so short answer, it doesn't go away, right? It doesn't go away. Um, it actually gets accelerated.

It gets more baked in, right? When-- Let's come back to, uh, uh, you know, within Agentic for a second, right? When I go to a store, uh, or when I buy something online, uh, through an Amazon, right? And I put in my credit card, there's a lot of checks that happen in the background, right? There's checks on whether I'm high risk for fraud, right?

Whether my-- there's matching between what I provided on my credit card and what p-I provided on my address. There is a history of the card, right? Is this a brand-new card? What's the limits of... Has the system seen, you know, uh, returns or others in different parts of the organization? So all of these are kind of fully baked in, and me as a user, I really don't care.

I just swipe my card, right? In the same way, you're gonna see the same with agents, and you're gonna see the same with crypto transactions in general. Which is you're gonna have all these super powerful tools. Um, you know, there's gonna be an exchange in standards around and, and some of them, you know, we've, we've recently announced like the, uh, an OTL, uh, the OTL Network.

But there's gonna be, um, some standards around communication and how you pass information, how you pass KYC information. What is really, you know, agent KYC information, how you do AML checks, how you do travel rule checks. Uh, but they're gonna move more and more to the

background so that folks just assume it's a part of the

network versus something that you explicitly need to

think about.

Um, so in short, doesn't go away. Actually becomes more and more important. Uh, it just becomes more and more,

um, kind of invisible to the average user

Stephen: super interesting. You've mentioned a couple use cases. Tal mentioned the use case of AI agents within the TRES Finance platform. You know, but I think the average person maybe listening to this is like, "Okay, AI agents sound great. I just listened to Gavin Baker on the podcast. He's talking about DRAN.

Like, things are going way over my head, and I think I'm pretty invested in this space." What are some use cases that you're seeing, especially leveraging the report, where you're seeing agentic AI being used today, not within a SaaS maybe perspective, but like the average user is really taking advantage of this technology?

Itai: So, f-first, uh, to whomever is listening and thinks they're crazy for this going over their head, know that you are not, right? We are still very much in experimentation phase, and there's still not, like, giant use cases of agentic money movement on crypto rails that you're missing out. You are not behind the curve, okay?

Everyone, ev-- Like, these are still very much at the tail end of experimentation to the beginning of, of, of, uh, kind of production use. And we are very much, uh, you know, to paraphrase, uh, Jeff Bezos on, like, day one, right? Like, uh, uh, I think I'm-- I, I think that's him. I think he claims credit for the day one statement.

Uh, but essentially, uh, we are very much at the beginning. So the, the, the use cases are very experimental, ex-uh, experiential, exper-- I can't pronounce that word, but just for every listener, just assume that I've said it. Uh, the, the use cases we're seeing are the following. First, you're seeing the extreme use cases.

You're seeing the open call use cases, right? You're seeing the, uh, or Hermes use cases, which is I have a long-running agent that works on my behalf. It moves from a conversational agent to an assistant, and at that point it needs to buy things, right? And so that's a place where you're starting to see some, uh, kind of money movement, which is let me give that agent a dynamic wallet, right?

Let me provision one for them or delegate access to my wallet, uh, to them with some sort of rules, uh, and policies around what the agent can do on my behalf. So I have a wallet that have money in it, and I give the a-agent access for two days for up to a thousand dollars, right? So I delegate access, and then the agent takes action on my behalf.

So that is one use case we're, we're actually starting to see. A second one is more on the data kind of, uh, purchasing. So there's, um, a company called Parallel, uh, um, out of, uh, used to be the CEO of Twitter, uh, uh, and now created this company. They've done very well. And they are-- uh, and their hypothesis is to say, "Look, you know, you need a search engine not just for humans, but you need the equivalent for agents," right?

And the way to interact with that is X four two and MPP, et cetera, right? And so that's a place where you're seeing use cases exist, which is agent saying, "Look, I need to compile a report. I need to search these ten places. I need to scrape these ten websites. I need this data that's proprietary. Let me interact with kind of an API endpoint, and let me exchange money."

So that is the second place we're seeing it. So those are kind of two extremes, uh, where we're seeing this interaction. Besides that, still a lot

of experimentation. No one still, to my knowledge, is buying full, uh, vacations with agents. Uh, if you are, please reach out. I wanna talk to you. Um, and please buy one vacation for me as well.

But, uh, essentially, that is, uh, it's very much still in experimentation. Uh, Tal, I don't know if you want

Tal: No,

you you should talk to Goldie because he told me that he's actually, uh, doing something like that for vacations. Anyway, I f- I feel,

Itai: Okay, fair enough

Tal: I, I feel, I feel very much behind the curve. Like, I don't know who's using a wallet to buy things with AI. Like, I had to fight with Claude the other day to kind of reverse something he did.

It messed up my whole thing. So yeah, I'm still at the, uh, the basics of, uh, of using AI, to be honest

Stephen: I'm curious, Tal, you, you know, you're in the backend operation business. I'm assuming chargebacks is, you know, one of those interesting things to your point, Itai, you were saying, you know, people are using their credit cards, they don't see anything, but when they do see something that's out of the ordinary, they have no problem picking up the phone and calling and, you know, disputing charges.

Tal, what are your thoughts on chargebacks, how agents are gonna make payments? You know, this is a very gray area, especially when it comes to even regulations. What do you think is gonna happen over the next two or three years? 'Cause we're already seeing friendly fraud and chargebacks for things that people did actually purchase.

I can only imagine when agents are, even if it's $1,000, that's huge if you're in the US or Canada, even, you know, other countries where that's a huge amount of money to, you know, take a loss on that because your agent overspent or spent on things that you can't actually now get. You know, they spend it on vacation that doesn't really exist because I'm assuming criminals are now gonna start selling things that don't actually exist, hoping that agents will purchase them

Tal: First of all, I have to say, um, I'm an angel investor in a chargeback kind of startup called Just Shout Out. Um, and I think it's just a crazy line of business at the moment with everything that's got to do with, I would say fraud and chargebacks when it comes to AI, that is gonna absolutely explode. Okay?

Uh, like we already seeing, by the way, the rise of, uh, AI cybersecurity companies. Only in Israel there were like four or five different acquisitions, one after the other. And every time I talk to my friends who work at venture capital firms, you can just see the money that is being raised for these kind of companies is absolutely crazy.

And if you think about it, okay, everyone... I would say organizations that don't want to get hit by this crazy train called AI are going full on, on AI and building these agents and using Claude and, and, uh, OpenAI and, and ChatGPT and whatever. The idea that you have so many agents running on so many data sources and are so, like, much integrated into the company, this creates such a crazy attack surface.

Like at, at Fireblocks, what we've been doing with AI is absolutely amazing, but it took time, okay? Because of Oded and the whole security team, they took the time to understand what is the right way to actually implement AI, what is the right way to give the team at Fireblocks infrastructure to build agents on top of the data without creating this new attack surface that we all know how sensitive, uh, the data and, and, uh, at, at Fireblocks is, right?

So everything has to be done so carefully, and I'm really worried for some organizations that are just going all in, flat out, just like opening up everything. Um, and then if you take that into consideration that now some of these agents are gonna get, uh, wallets or credit cards and start running, uh, free, um, yeah, I, I...

If you're thinking about building a startup, I would definitely say those are the spaces that I would, uh, I would focus on because I think there's, uh, a lot to be done.

Stephen: Because both of you work in a very, I think, you know, you know, both have Israeli company. Like, there's this unison when it comes to Israel, Israel and cybersecurity companies, especially coming out of Unit 8200. Are you worried that Mythos will affect those companies, or are these companies excited to leverage Mythos to s- further support their work?

Itai: I don't know if, Tal, if you wanna jump in first or...

Tal: You, you go for it

Itai: Yeah. May-maybe two, two things. First, just on the previous point, by the way, just, just so folks know, like, uh, a lot of the agentic stuff, it's no different than, um, you know, um, I've, I've, I have a kid that's at, at the age where he has his own iPad, and one of the risks is him clicking and buying things without my approval, right?

And how, how does that work with chargebacks, et cetera? And so the answer is controls, right? Like Amaz- uh, like, uh, Apple gives you a bunch of controls around parental controls and what can happen, what needs a push notification, et cetera. You're gonna see the same with agents. Not to compare my kid to an agent.

He's, uh, uh, much cooler. But, uh, essentially, um, you know, you're gonna get, uh, controls, and we provide policy controls around what an agent can't do, what an agent needs to ask permission for, um, you know, how do you cancel a process midway through. So all these would exist for agents as well. Now, to your point about mythos, uh, et cetera, this is like the golden age of cybersecurity, right?

You're able to find so many, um, potential things that you couldn't find before, right? Like the, the, the-- in cybersecurity, you had to focus on specific things where you had a hypothesis for attack vectors, and now you can do like a boil the oce-- literally the reverse of what y- folks have been telling you since you grew up, which is don't boil the ocean.

Uh, you can now boil the ocean. Right? You can literally, you can literally run an agent daily for y- or, or for days on end to try to constantly test your system. And so it creates this acceleration of security, not deceleration of security. Yes, the counter is that someone will try to find it, but by definition, it actually b-becomes, systems become more robust, not less robust, right?

Which is, it, it, it becomes this really fascinating, um, kind of time of I think we're gonna get to a bunch of humps along the way and a bunch of like, uh, exploits, et cetera, and then we're gonna literally get to a golden age of everything is extremely, uh, highly secure because you're just gonna have these agents checking and rechecking and trying to think about all the creative and combining and connecting systems, um, to try to find whether they find a flaw.

And we think about this in Fireblocks every day. Every day. Like this is a, a-- I would argue that fifty percent of my conversations are around that, right? We think about security. This, this is, um, it's, this is true for all companies, and it's doubly true for companies in crypto, right? And so we have conversations around this every day.

We think about how state actors use this. It is, it is by far one of the most critical things we do, uh, and we're very excited about it because all, all of all the cool stuff you can do with this and all the extra protections you can, you can build.

Stephen: Tal, I wanna ask you, it

seems like Trust Finance kinda solved the interoperability, you know, issue, right? That that's how you beat out a lot of your

competitors by being able to plug into so many

different blockchains. W- do you think we're walking into another interoperability issue going into agentic AI?

They're using different chains, possibly different stablecoins. Are we walking ourselves right into, you know, what, where the travel world is right now today, where Coinbase has their own solution and, you know, some are open loop and some are closed loop. Talk to me about the interoperability issues we're gonna have in the agentic commerce world

Tal: uh, uh, to be honest, I'm not that worried about that because I think that the, the data issue, at least on the Trace side, is kind of solved and we are like, we are working with these organizations in kind of understanding how they are building out their networks and their protocols, so we are kind of ahead of the curve.

And when they do launch and they become, uh, uh, move into production, we are already there to catch the data. I think actually what is more interesting is the ability for more players to start playing in a digital asset, uh, economy with stablecoins. And we are seeing today, I'll say Fortune 500 companies looking into how they wanna manage their treasury in digital assets 'cause they understand what they can do with stablecoins and how this affects the whole treasury matter.

I'll ju- just give you an example, okay? Let's say I'm a company that has 30, 40 subsidiaries around the world. What I need to do at the end... at the beginning of every month, I need to top up all the bank accounts, right? So they can pay rent or pay their, their employees or pay vendors, et cetera. Now, the fact that they need to move it at the beginning of the month, even though maybe their subsidiaries don't need to pay those payments at the beginning of the month, they can pay it like two or three weeks later.

We think- we're talking about billions of dollars that are being moved, um, not generating yield, right? 'Cause they've already moved to, to those, uh, subsidiaries. The fact is that now you can actually do everything with digital assets. We can use something called just in time liquidity. What does that mean?

Now, as a treasurer, I can say, "Okay, I don't have to move the assets at the beginning of the month. I can actually have them in some kind of yield-bearing, uh, uh, vehicle." Okay? It could be a tokenized money market fund, it could be some kind of borrowing or lending protocol, any of those. And when my subsidiary needs to make the payment 24 hours before, let's say, then I can liquidate it very easily, right, everything's running on the blockchain, liquidate the position and using stablecoins just move them.

So I also avoid the fees, I'll also avoid the, the, the foreign exchange, uh, and the currency, uh, um, change that just kills the margins, okay? And also, I'm still generating yields for another 7 to 10 days, which when it comes to billions of dollars can be millions that you save every month, okay? So I think that the amount of huge companies, conglomerates are gonna be understanding they can save tens of millions of dollars a month by just moving part of the treasury, if not all of the treasury, to digital assets with stablecoins and having all these new tokenized, very, uh, low risk, uh, vehicles on-chain is gonna change how we are interacting today with, uh, with the market and what kind of players are even touching digital assets.

Stephen: I'm curious, Tal, you mentioned stablecoins there. Um, this is a question to either one of you. Stablecoins run at a premium, right? But when you're talking about hundreds of millions of dollars and funds locked up, you know, due to Forex or the weekend, it makes sense for a lot of companies. Itai, does it make sense to only use stablecoins when it comes to agentic commerce because you are paying a premium for the stablecoins?

And, like, that makes a lot of sense to me when you're talking about hundreds of millions of dollars. But when you're talking about micropayments, although the speed is there, is this the most efficient, you know, token to be used for agentic AI?

Itai: it's a really great question. The answer is it depends. It-- There is no winner take all in the financial market, right? And the easiest way to figure this out is to go to Europe and see that you have credit cards and you have like ten other successful, uh, country-by-country payment systems, right? Or if you go to Brazil, you have Pix, which is phenomenally successful, right?

Uh, there is no one system and there is no one protocol. Uh, you, you, you're gonna have multiple forever. And that's true for agentic payments as well. When an agent pays a human or an agent goes to, uh, a merchant on Shopify or whatever and clicks to pay, the best interface for that today is credit cards, right?

Because the merchant already accept those, right? So the distribution is there. Uh, so I don't see that changing for a while, right? Not, not because it's less efficient to pay with stablecoins. It's more, um, but you have these built-in systems of, uh, and we mentioned this, fraud, et cetera, um, that, that are built, built on, on the stablecoin side but don't exist at that scale yet, right?

So, um, that interface I don't think changes. I think agents get credit cards and interact with those. But the interfaces where I do think we're gonna see an acceleration of kind of stablecoin-based payments is the agent-to-agent interfaces, like the parallel web-type interfaces, where an agent needs to complete a thousand transactions, where an agent needs to start compiling multiple sources or multiple resources, right?

When an agent wants to pay for some sort of MCP integration that, um, uh, they couldn't do before, right? When an agent-to-agent is the interface, then by definition, y-y-you know, cards don't really work there. And stablecoins are phenom-- a phenomenal fit for those. They're, they're literally programmable money, right?

They're money meant to be moved at giant scales, uh, automatically. It's like this magical thing literally built, uh, for, um, for this type of, of money move. So there I do think we're gonna see, um, this huge rise of, of kind of stablecoin-based, um, uh, payments and agents carrying these embedded wallets, uh, and just being able to move money around.

Stephen: I have two questions for you. You know, with the cost of using these tokens at a pretty high

rate going through Anthropic and Claude and all these things,

then these micropayments might, you know, if, as Tal said, maybe Goldie's looking for a

vacation, but if Goldie spent, you know, $1,000

that week on tokens, that savings that he's getting from using an agent to find him a vacation is kind of canceling itself out.

Where- are we close to the point where, you know, agentic AI makes financial sense to us, or is it still kind of early stages where it's still a high cost, but eventually once we get those costs down, you know, the, the ocean is ours?

Itai: Yeah, the, I, I don't think it's around saving efficiency necessarily. I don't think you get, um, you know, the, the reason to give an agent a wallet is not because it's gonna be more ch-cheaper for you to buy your vacation. It's just the way in which, uh, an agent can execute on the task at hand, right? It's just a more efficient way for the agent to execute.

So it means that you can, uh, uh, kind of... the agent has less blocks, right? Uh, if you, for anyone, you know, uh, building with Claude Code, uh, one of the annoying things is that unless you dangerously, dangerously skip permissions or use auto mode, uh, every couple minutes it asks you a question and kind of gets blocked, right?

And so the name of the game is how do I unblock my agent, uh, more often with guardrails, with these policies, and with these security-type, uh, controls, and that's, um, you know, that's where there's value in these agents. You're not gonna get-- I, I hope no one t- uh, leaves this, uh, podcast and kind of h- uh, feels like they're gonna get cheaper vacations.

You're not. I hate to break it to you, you're not. Uh, but

you are gonna just have an assistant that can potentially do things on your behalf, not just give you information, but actually act on your behalf and be your personal assistant. And the way it works is just gonna be the most efficient way it knows how to work, which is to move money on-chain

Tal: A-agents are

gonna drive the prices of vacation up because you're gonna see so many agents visiting the website searching for the, uh, for the same flights. You're just gonna see, oh, demand is high, and so the algorithm is gonna start spiking up. Um, so,

and, and I just wanna say,

um, I don't know if you saw, but I think it was like a couple of days ago, the CRO of Uber said that within four months they used their whole budget for AI at Uber.

Within four months for the year, they finished all the budget for AI, and he could not say, uh, clearly that it generated any value, which is crazy

Stephen: And I think that's across the board, right? At a enterprise level, people are using the agents. They're seeing that you can do things quicker, but is that mean that there's an ROI there? Tal, I wanted to ask you, you know, with this hu- the word hu- or this term human in the loop seems to be used by a lot of traditional compliance folks that don't want to give up their job, and they're like, "Hey, yeah, agentic AI is great, but you need humans in the loop."

But if we look back in just April, you know, the human in the loop cost, you know, DeFi almost $600 million with just two protocols. How-- Are we working towards no human in the loop, or because humans make mistakes? Or are, you know, is this more of a like, we need to keep it there because we need to comply with regulators that aren't gonna be comfortable that there's no human in the loop at all?

Tal: I think if you would ask me this question, I would say like eight months ago, I would pitch you like human in the loop is how we need to do things. And now I'm like, you know what? Maybe not. Um, we've, we've seen a few cases where, uh, humans interfere. At the end of the day, for regulatory purposes, you'll need a human to sign off, right?

But the amount of humans you'll actually need is gonna go down by like ninety percent, and you'll probably have one human who doesn't even have to be the most professional human in, in that profession. He can just sit there and have a good enough common sense to be like, "This is good. I've reviewed this.

Makes sense. I'm gonna sign off on this." But the progress that we are seeing today with the agents that we build and how they are analyzing tasks, es- uh, you know, it also took us time to get there because you have to train them, you have to give them context, you have to explain to them what they actually need to do.

If you don't do that, then yeah, you'll need a human in the loop. But if you know, like kind of what you wanna get out of that agent, you can get to like really crazy results that are better than an army of humans at a fraction of the price and the time. And at the end of the day, you will need some kind of human to, uh, to sign off, but it's not gonna be humans in the loop.

It's gonna be a human in the loop for all the agents. That, that's the way I see it at least.

Itai: M-maybe, maybe just to add to that, I, I think it's a problem that extends beyond just, uh, you know, crypto and stablecoin payments, right? It's a problem that you're gonna see with everything that, uh, AI-related, right? You see this-- I'm, I'm in the Bay Area. We-we're fortunate to have Waymos here, which are self-driving cars, right?

And then, uh, the, the question is not necessarily just whether the car can drive by itself. There's a lot of regulatory and compliance questions as to who's liable, right, when the car makes a mistake, right? How does that work if the car doesn't stop or if the car runs a red light or if it stops where it shouldn't or, or God forbid, like, it accidentally, uh, doesn't stop in time, right?

And so all these questions are questions that need to be figured out in parallel to kind of the, the, the actual technology side, right? Which is hard on its own. You now have the full regulatory side of, you know, to Tal's point, can the agent file a report, right? Can, can, uh, can an agent move money on be-my behalf?

And if the agent made a mistake and moved a million dollars, uh, s-uh, to, to the wrong place, who's liable

for that, right? Like, how does that work? And so these are all open questions, um, that, that

are, are gonna be, uh, conversation topics, uh, not just within crypto, but in general within, uh, the world

of, of fancy automation and, and AI, um, in the next couple of years.

Stephen: I'm curious. The report shared some insights. I think it was Nano Rodriguez from Bitso who stated that, you know, agentic AI could be, to your point, uh, Itai, the invisible rail behind global payments. Uh, but he saw the biggest opportunity in LATAM. Is there a chance that it's the emerging, you know, markets and their institutions?

We've seen, you know, countries like Bermuda that are gonna get the... Are gonna be far in advance than, let's just say, the US or the EU

Itai: Absolutely. Right? You, you see this, uh, if, if you think back 10, 20 years, right? Uh, the, the, the US and, and, you know, parts of Europe had, uh, you know, a bunch of desktop computers and, and everyone got a laptop, et cetera, and everyone has a bank account, et cetera, right? And there was a lot of, uh, conversation around innovation with mobile first in, you know, uh, neobanks, et cetera.

But where it got most prevalent is actually in, in different parts of, uh, of Africa, right? In, in different, i-i-in either Nigeria or South Africa or, or Ethiopia, et cetera, where they just skipped a generation, right? They just went straight to mobile first to kind of neobanks first to, to... A-and essentially moved, hopped o-o-over a full kind of technological step and moved two steps ahead of time, right?

And if, uh, i-i-into much more innovation, right? Uh, and you realize that kind of the, the European market and, and the US market has actually been slower as compared to some of those, right? Like Pix in Brazil is this magical thing, right? That, that does not exist in a bunch of other parts of the world.

They've just skipped this entire, uh, kind of technology thing. Uh, we, we think in a lot of ways you're gonna see the same thing with stablecoins, which is established markets tend to move slightly slower to adapt some of these technologies to establish compliance rules around those because the need, the pain is not, uh, as big, right?

But in, in places like i-in LATAM, et cetera, and, and where Bitso has this, uh, you know, phenomenal kind of vantage point in to seeing, uh, to seeing, um, what's being built, you're just gonna have that skip, and you're just gonna go straight to a lot of this innovation that, that, that we can't see from, from here.

Stephen: And I think that happened in

Canada. You know, to your point about picks, we've been talking about open banking in Canada for years, and

it just ha- we haven't gotten any closer. We've now, you know, labeled stablecoins. I think they're changing the

regulations, but stablecoins are securities. And you can see how other countries are just kind of like

leapfrogging over Canada. And because there's a lot of large institutions, legacy institutions that don't wanna see this change because of, you know, the market share that they have. Uh, but we have seen regulators clamp down in places like the US when it comes to prediction markets, maybe at the state level. I'm curious, Kyle, do you think that the agentic AI industry should slow down a little, take a more uniformed approach to regulation and in the adoption of risk of the average person?

Um, or do you think, like, we just kinda let it fly and let the agents kind of do their thing?

Tal: Le- let it fly. At the end of the day, the market will dictate where the regulators should be. You need to break everything, like Uber did, by the way. They just, like, caused chaos, and then the regulator had to kind of, uh, um, pick up, uh, e- everything that was, like, scattered around and, uh, help them, uh, because it was just too big for them to now start regulating.

So same thing with, uh, AI in general and, uh, AI in finance. Uh, like, whatever we can do, we just need to push forward and, and change the world, and the regulators will just need to keep up. Because if we wait for them, it will never happen

Stephen: Itai, I'm curious, is there, if there could be one page you could add to this report, because, you know, as soon as you press post, something news comes up in the headlines. Uh, is there anything, any topic that you think you would have loved to cover more of or at least highlight, like something to look forward future about?

Itai: Yeah. I, I think actually there's a topic that's extremely relevant today. We launched, uh, kind of the, the open protocol at Fireblocks, which is this stack O-- we call it OTL, right? The open protocol layer. Uh, and it talks about this, you know, what we talked about throughout this conversation, which is you have the rail and then you have the compliance system, right?

And how do you do compliance in a distributed system at global scale? How do you coordinate between different players, between the Revoluts of the world and the Bitos of the world, in how money gets moved on stablecoin rails and in an agentic way and-- or in general, right? And how do you do it with travel rules?

How do you do it with the different compliance controls, et cetera, right? And so one thing I would add is a full section, I think there is a section of compliance, so maybe I'll buff it up, uh, of, of kind of an open protocol for doing this at global scale and how the industry has to come together as a whole, uh, to actually kind of answer some of these questions that can't be answered at the local maximum.

So not every player can kind of choose their own solution, but you gotta work as an industry in order to kind of a-align on, on these standards around compliance and standards around kind of money movement here, um, um, you know, both at regular scale on crypto rails and agentic s-scale doubly so. So that, that would be, you know, the one pl-- Uh, obviously I'm a little bit biased considering this is the thing we announced today, but, uh, you know, so check it out and shamelessly go to Fireblocks and look at the blog, uh, we published around that.

But, uh, e-essentially, uh, it is, it is becoming n-not like a second-order question, but like one of the primary questions, and I, I, I wish we would've spent more time on it.

Stephen: If, if you're getting into agentic, you know, AI right

now today, what, where does Fireblocks fit

into this ecosystem? Or is Fireblocks the ecosystem that you go

to and they have all the

integrations, all the partners to get you from an end-to-end solution? I'm curious what your

thoughts are, Atal.

Tal: The, the Fireblocks is gonna be the infrastructure for everything that's got to do with digital assets. Okay? Be- and, and the reason is, and we go back to what we talked about, uh, when we mentioned security, okay? When it comes to digital assets, there's one thing you can't compromi- compromise on, right? And that's security.

And then on top of that, you can build different layers like, uh, payment that is, uh, leading the way or working with your banks, uh, et cetera. On top of that, you can start building the, uh, um, AI agents for payments and protocols, but everything has to sit on the infrastructure of Fireblocks. If not, the most basic layer, which is trust, just doesn't exist

Stephen: And I'm curious, if you were to build today, Itai, how do you create a moat with the, you know, the use of these AI agents? It seems like anyone can spin up a cybersecurity company i- if they wanna spend enough on AI. How do you build a moat if you're building in digital assets today? Or maybe how do you build an agentic AI company that can compete?

Itai: Yeah. So it, it's a, it's a very good question. And I, I spend, I would argue probably seven of the last seven days I've touched cloud code and tried to kind of vibe code things, right? So I've, I've, uh, I, you know, I, I, I play, uh, with, with this stuff and I try to build stuff pretty much on a day-to-day basis. Uh, when you go to-- I was, uh, you know, fortunate enough to take computer science in college and, uh, one of, uh, the first things they tell you when, when you sit there in computer science, uh, school is like, "Don't reinvent cryptography."

Like, there's one place where you should not innovate. It's sort of like accounting. Don't innovate in accounting, right? Like, the, the, the-- Or innovate in systems around it, but, like, they're pros-- Like, do not innovate in cryptog- Do not roll your own cryptography. It is a terrible idea, right? It is not something you can vibe code because the cost of, of, uh, getting it wrong, it's math, right?

You-- It is very, very high, right? And it-- That is our entire premise at Fireblocks and, you know, at Trust and Dynamic, is we focus on that. We have cryptographers, we have kind of world-class researchers, we have teams trying to poke holes in all of this. We work with something called MPC, so multi-party computation, uh, in order to provide security for a thing that literally holds funds at global scale, right?

I think we, we've moved, uh, at Fireblocks, six trillion dollars, uh, uh, last year. Or like, uh, these insane numbers, right? And so there are many things you should be able to build with a two-person team. Um, you know, and many companies that are gonna be extremely innovative with very, very, very, very small teams.

Uh, the core cryptography and policy controls and compliance controls around those are where you start getting into scary territory of the, the, you know, the thing where you should let folks that do this for a living and let folks that think about this every morning and every night kind of take care of it.

In the same way that you, you know, um, it, it, it just, uh, it's a, it's a really important math type of the equation, really important security type of the equation that you kinda can't get wrong.

Stephen: You know, you're both founders.

I guess technically now you might be employees of Fireblocks.

Itai: That's

Stephen: I'm listening to this podcast and I'm like, "AI's taken my job." The Block just let go

of 40% of their staff. You

know, they want a flat organization that all reports into Jack Dorsey. you know, it seems like we're m- you know, moonwalking away from AI is gonna steal everything.

It seems like both Anthropic, Sam, and, you know, both Anthropic are walking back their statements of like, "AI is gonna kill your job tomorrow." What are your thoughts? If you're an employee, you're working at a digital asset firm, you're seeing layoffs across the board, but you're also seeing hiring in very niche areas.

How do you approach this industry now, especially both of you have been in it for so long? How would you employ- approach it as an employee versus a founder?

Tal: I'll say that, um, I'm gonna take-- Like the, the first view I'm gonna, I'm gonna give you is as a founder, I can say that the ability to scale products, features, operations, customer success, support, dev is just amazing with AI. So I would say that for Tress, if it wouldn't have been acquired by Fireblocks, we would probably reduce our hiring by, I would say, ninety percent and make sure that everyone in the organization is leveraging AI to get more out of what they're doing.

So not firing, but saying with, with the exact team we have today, how can we, um, reach and, and achieve more? As a, as an employee now, as a team member at Fireblocks, I can say that, um, every person should look at what kind of value they bring to the business, what kind of value can bring using AI, and also understand what AI can actually do to replace them at what they do.

And they have to look at that equation and understand what they need to change and how they operate on a day-to-day basis, uh, at their current job to make themselves, um, very difficult to replace and bring much more value. So the manager says, like, "This team member is not only contributing more than what he contributed before, but he's also enabling the rest of his team to get more."

Um, so that's what I would do. It, it's difficult 'cause you have like your day-to-day and you are most of the time just like putting out fires. But I do recommend to any person listening to this podcast, to this episode specifically, to think about if they don't have time to really become proficient in AI and tools, they have to stop now and really take the weekend or take a few, uh, I don't know, a leave from their work for a few days, say they're sick or I don't know what, and just sit and understand and analyze what they are doing today and what they need to change to become, uh, someone who's irreplaceable at their current job.

Itai: yeah, maybe just to add to that, I, I think if we kind of reflect back, uh, twenty years ago, right? And the rise of Google, or at this point I'm d- aging, I'm dating myself, it's like twenty-six or twenty-eight years. Um, you know, uh, n-now I'm sad as I continue this rest of the answer. But, uh, you know, there was a competitive advantage for anyone who knew how to use Google, right?

There were classes that folks remember of, you know, learn how to use Google, and those that are able-- were able to search well, uh, had a built-in competitive advantage. They, they-- because th-they had access to the world's knowledge, right? I-in their fingertips. And then the, the difference was whether you used it and you cared about it or you didn't.

You literally now have, uh, as an employee, a, like, ac- or a founder, access to these magical tool sets, just pure magical tool, tool sets, right? And if you, you know, the word being used around is agency, and I, I kind of like it, which is like, if you have the agency to build with those, you literally get superpowers.

You literally get superpowers. And the, the-- it's not about number of employees or t- like, it is about folks that wanna leverage superpowers or folks that sit there and, like, are like, "Oh, great, superpowers." You know, they're there. Uh, that is the differ-- you can literally build incredible things today, uh, and, and it's just whether you want to, and whether you wanna learn, and whether you wanna push yourself, uh, to doing that.

And so, you know, um, to Tal's example, if Dynamic wouldn't be independent, it wouldn't be around less hiring, it would be around folk- hiring folks that I think can take advantage of these superpowers that exist today. And that would be the cool thing.

Stephen: are you guys, either of you growing your teams internally, or are you same, same

philosophy? You're taking what you have and seeing how you can exponentialize that using AI

Tal: I, I will say that like when we, when we were acquired, I presented, uh, to the CFO our, uh, headcount need for 2026.

Um, we have used 0% from the headcount that was allocated to us, and we won't be, uh, using it

Stephen: Really interesting. Last question of the day. You mentioned learning. Where do you both learn from? Like, where do you either get insights about AI and agentic? 'Cause we joked around before the podcast started recording about, you know, Tal sent you a message, Itai, you didn't even see it until late. Like, where-- And obviously, with the time difference, you just woke

up.

But where do you, where

do you go to when you need inspiration and to learn from? Is it, you know, maybe so- through your, your VC network and, you know, crowdsourcing ideas and, you know, knowledge through the, you know, professionals that you've already built relationships with?

Tal: Ty, I just wanna say I did, I did expect you to put me on the VIP list. So when I did send you a message, it went to your-

Itai: I think that's right. I, I, I think that's right. If, if, if there's no other takeaways from this podcast, everyone should know that I've, I've now changed my Slack notifications for Tal, uh, and they turn into a phone call, whether it's three AM or four AM, that, uh, that's, that's, that's the way we roll. Um, yeah.

Yeah, yeah, yeah. Uh, but to, to your question, maybe from my lens, uh, I, I, you know, I'm, uh, I, I'm a big fan of Twitter. I think it has, uh, some really, uh, um-- I refuse to call it X, but, like, a big fan of Twitter. Uh, and, you know, ha- uh, that's a very good place to, to consume things. But honestly, just by doing. I, I think you're, you're, you learn so much more just by trying out, you know, try out Eleven Labs and see what, uh, voice agents are like.

And try out, uh, you know, Cloud Code versus Codex or Cognition AI or any one of those tools, or Factory AI, or just try them out. There's so many of these magical tools out there. Uh, y-you're not gonna, you know, be able to just read your way through it. Um, you know, it's sort of, it's sort of like me saying, "Hey, read your way through, like, learning how to snowboard."

Uh, like, it's not-- It doesn't work, right? Just try it. You know, you're gonna fall, and it's gonna be fine, and you're gonna get better, right? But that is the way to, to im- you know, embrace yourself. And the same sim- same with stablecoins. Try them out. It's extremely-- You can implement Dynamic with, uh, Cloud Code in, like, four minutes.

Uh, and you can build your own thousand agents, all with wallets, all moving stablecoins within, like, thirty minutes. Try it out. It's not that difficult.

Stephen: I love it. Tal,

last word

Tal: Last words.

I also put Itai on the VIP list on my Slack, if you want to know that. Um, I, uh, I use Eylon, my co-founder, who's a technical genius. Uh, too much noise for me, Twitter, newsletters,

uh,

taxi drivers are telling me how cool AI is. I'm like, "This is not for me." Eylon, what actually works? Use this, use that, check this out.

I'm like, "Okay, this is gold." So, uh, yeah, f-find a trusted advisor that can clean up the noise for you

Stephen: I love it. And I know we can both find both of you on LinkedIn. You're fairly active there, obviously Twitter as well. Um, and then go to the Fireblocks website. We're gonna have the report in the show notes. Really interesting insights from the report. And we appreciate the two coolest CEOs I know under the Fireblocks label joining us today.

Uh, and I can't wait for this episode to air. Thank you both

Tal: Cheers.

Itai: Thank you for having us.

‍