与 Circle 联合创始人兼 CEO Jeremy Allaire 讨论 AI Agent 将如何改造金融系统

Today on a priors we have Jeremy Allaire, the co-founder and CEO of Circle. We'll be talking about cryptocurrency, AI, agentic payments, AI evolving on the blockchain, and a variety of other topics.

Thank you so much for joining us today. It's a pleasure to have you.

Jeremy Allaire

It's great to be here. Thank you.

Maybe we can start with you giving a quick review of Circle—what you do and how you approach the world. I think we're going to be talking a lot about stablecoins, crypto, AI, and how all these things tie into the agentic future. I'd love to start with the origins of the company, what you're all up to, and go from there.

Jeremy Allaire

Circle's been around for a while. I co-founded the company over 13 years ago, in 2013. At inception, I was really excited about this idea that we could create a protocol for dollars on the internet.

I had been really excited about what was happening with technologies like Bitcoin and had been working on internet infrastructure for a long time. I got really excited about the idea that, if we had a protocol for dollars on the internet, potentially we could have a way to store and move value instantly, globally, frictionlessly, and ultimately at no cost.

The other idea we were really excited about back then was programmable money—the idea that eventually these networks, blockchains, would become like operating systems, and you could actually have machines that intermediate economic and financial activity on the internet, including autonomous software machines.

Back then, we didn't have generative AI or anything like that, but there was this idea of commoditizing the payment utility layer with very safe digital-dollar currencies and then having programmability with machines that are tamper-resistant and can run on the internet. That's what drove the founding of the company.

The view was that if we could do that, we could actually improve the financial system, make it safer, more accessible, and more efficient, and derive new utility for money that we hadn't had before. That was where we were.

Why is the dollar aspect of that important? If you look at a lot of the things that happened in cryptocurrency in the early days, it was really about creating things that were divorced from the traditional financial system, if possible, or were not dollar-centric.

For example, Bitcoin was in part a response to the Global Financial Crisis—the view that all sorts of weird bailouts happened there and, therefore, we needed some alternative financial infrastructure for the world.

Jeremy Allaire

Yeah. I believe in Austrian economic thought. I was studying Austrian economic thought in the early 1990s and had been interested in sound-money theory for a very long time. It was studying the impact of the Global Financial Crisis that drew me in, because my view was that there had to be a way to build a safer financial system.

The key issue was that I was interested in this idea of full-reserve money. In some ways, Bitcoin is full-reserve money, because there is no way to fractionally lend Bitcoin, per se.

Sound money means currency that's backed by something hard behind it—some asset?

Jeremy Allaire

It doesn't necessarily mean that it's hard-backed. Full-reserve banking, I should say, is different from fractional-reserve banking.

There's another major economic collapse, which was the Great Depression—the run on all the banks and all that. In the 1930s, there was a really big debate about the right construct for the banking system and the financial system. There was a proposal from a group of economists called the Chicago Plan.

The kind of ringleader was a Chicago economist. Actually, it might have been a Yale or Princeton economist at the time: Irving Fisher, who wrote a book called 100% Money.

The idea was that full-reserve money was essentially government-obligation money. It's still the obligation of the government, like the U.S. government in that instance, but you can have that and hold it without taking it and fractionally lending against it. You have a full reserve, but you can only lend full-reserve money.

That was a big proposal for how to structure the way the financial system worked, and it was lobbied very hard against by the banks. The banks really liked fractional reserve. They liked being able to have the inherent leverage and risk-taking. Instead, they convinced the government—or collectively with the government sanctioned and established—an insurance company called the Federal Deposit Insurance Corporation.

That was a corporate insurance model, but the risk-taking still existed. We've continued to face those issues. The Global Financial Crisis was an example of 30× leverage, 12× leverage, and 14× leverage against these base-layer assets.

My philosophy has been that, right now, in terms of general utility, our existing economic system depends on major reserve currencies like the dollar. My view is that that's going to continue for a while—maybe 30, 40, or 50 years. It'll continue for a while.

But what we want to do is construct a system that is, in fact, safer: a full-reserve form of money. That's what stablecoins are. That's what dollar stablecoins are. In fact, with the GENIUS Act that passed last year, it's been codified in law. You can't do anything with this; it's a very narrowly bound, narrow-money model.

In some ways, that original vision has now been established in laws around the world. Now we have to do more with it. We have to make it extraordinarily useful. You can lend that form of money as well; it's just that you can't do fractional reserve.

What are stablecoins currently backed by? My understanding is that stablecoin companies are big buyers of Treasuries, U.S. Treasuries, and other instruments like that. Could you explain a bit more about what tends to back these things?

Jeremy Allaire

Up until really the last couple of years, stablecoins like USDC had to be always one-for-one redeemable against very safe, liquid assets. We couldn't take risks outside of what was permitted under the payment-system laws that regulated us.

There are other people who didn't take that approach, but fiat stablecoins in this way were backed that way. Laws have now come into play in major jurisdictions, whether it's Europe, Japan, or the U.S., and we've been following whatever laws apply to us whenever they apply to us, obviously.

What that's really led to is an architecture that's now federal law: holding only short-duration U.S. government Treasuries or Treasury collateral that's overnight with global banks. That's very safe, overnight Treasury collateral for cash. Then there's some amount in cash for immediate liquidity, held in big custodial institutions like Bank of New York, which holds hundreds of trillions of dollars in assets.

That is essentially the architecture of USDC, and we're very transparent. We have daily transparency on most of it through a system we set up.

So USDC is a crypto token that anybody can effectively purchase, and in exchange for $1, you get 1 USDC. That USDC continues to be backed by a government Treasury, like a short-term T-bill, some cash, or some exchange.

Jeremy Allaire

Right. It's backed by Treasuries, repos, and short-duration T-bills. The average duration of the T-bills and that portfolio tends to be around 13 days, so it's super, super liquid. It sort of allows it to be treated as a cash instrument.

What do people do with it? What are the main use cases of USDC?

Jeremy Allaire

The conception of this is obviously a general protocol for dollars on the internet. In fact, the whole design is as general-purpose, general-architecture money.

We see it used at the very smallest end—for example, someone paying 25 cents for a digital object in a digital game built on a blockchain. We're also starting to see, and we'll come back to this topic, I'm sure, AI agents paying for the output of essentially the AI tokens of another AI agent. They're spending a dollar, 50 cents, 20 cents, and so on.

So there are super-tiny transactions at one end, all the way to the largest electronic trading firms in the world, which do huge amounts of capital-markets activity and settle multi-hundred-million-dollar transactions. The powerful thing is that it's all the same.

If I send you an email saying, “Hey, this is what I had for breakfast,” the payload is the same as if I sent you an email that had a CIA dossier attached to it.

USDC doesn't care. As a general architecture, it can be used across a huge range of things. We have everything from merchants on Stripe and Shopify that are using it, to Visa actually using it themselves to move money on their own internal network instead of using the legacy banking system, to lots and lots of neobanks and remittance companies that are using it as a way to move value. A great B2B fintech, Ramp, just yesterday launched USDC as the core of its treasury system. You can use it to pay invoices.

My sense is that some of the reasons people do this are, number 1, you can do it at any time. For example, if I send a wire, I know a lot of crypto companies that, when they raise money, ask you to send USDC because instead of hitting a wiring deadline in the afternoon, you can wire the money on the weekend. You can send money anytime.

Jeremy Allaire

It just works the way the internet works, right? Our expectation is that I can pick up my WhatsApp or my WeChat and communicate by video with anyone anywhere, and it just works. My expectation is, hey, if I make a piece of software and put it on the internet, billions of people can access it. I don't need to do something special. This is just internet-native, and it runs on internet protocols, so it behaves the way that any piece of data or content behaves on the internet, which is what most people's expectations are.

I was just trying to enumerate a little bit of what makes it a superior instrument for all sorts of purposes. One is 24/7 accessibility, two maybe some form of transaction fees relative to the volume, and then three, my sense is, it's also a way for people to participate in US dollars who often would not have access directly, so they use crypto almost as a proxy to—

Jeremy Allaire

Yeah, for sure. I think a store of value is a really big thing, and we see that. In fact, the law that was passed last year, the GENIUS Act, had as a big motivation for the administration—and this is something that we've been proposing and pushing for a long time—that this is a way to continue to export the dollar. We're now exporting digital dollars, and we're doing that all around the world. That's strategically important to the United States from a geopolitical and geo-economic perspective.

There are other things, too, which come from my own background. This is programmable money. There's never been programmable money. Essentially, our stablecoin network is just a public API on the public internet that anyone can plug into and use. If I'm a developer and I want global dollar settlement, and I want to provide that as a capability to my users, I don't have to ask permission. I can just connect to that smart contract, connect to that public API, and boom, I now have an application with global digital-dollar utility. That's really different.

Mhm. Yeah, and smart contracts are basically a way to write code that's wrapped around this money, which allows you to effectively have a virtual contract online. You can say, under XYZ conditions, pay this out. We're going to generate a financial instrument off of this, and it's based on this other layer that you can plug into.

Jeremy Allaire

Yeah, that is definitely the case. The idea of programmable money was, again, this early idea that we had, and smart contracts were sort of the original expression. When I looked at that 13 years ago, my view was that blockchain networks are operating systems, or they're going to be operating systems.

When we think about operating systems, we have lots of paradigms for that. We have mobile operating systems. The web was itself kind of an operating system, with a runtime, a programming model, and an object model. Clouds became big virtual operating system environments. AI foundation models are now essentially operating systems that execute tasks and other things. Blockchains are operating systems. They have compute engines and virtual machines, and you can write Turing-complete code. You can write software that runs on these.

But there are some really key attributes that make them different. The first is that the code is tamper-resistant. Once it's published, it's out there as a tamper-resistant machine. The second is that it's perfectly auditable. You can audit every single input and output of that machine in real time.

On a public blockchain, anybody in the world can look it up.

Jeremy Allaire

So it's like all the compute is publicly accessible, it's open source by nature, and that's really powerful as well. It also has transaction and compute integrity assurances. This is really key, and it ties back to AI as well. You want assurances that the machine is doing what it said it was going to do. You want the inputs and outputs to be provable, too, and the state of the machine to be provable.

These are things that were not easy to do in the past. These network computers, these operating systems, now provide for that. As we're moving into the AI-driven economic system, having those mechanisms becomes even more important. It happened to be important for financial transactions, where integrity, proof, auditability, and verifiability are intrinsic to a fiduciary apparatus. That was really key. But now, when we're dealing with autonomous actions in the economy, that also becomes extremely important.

It'd be great to talk about that because, probably 7 or 8 years ago, my friends and I used to speculate that the most likely place where AGI would emerge—which, again, I don't think is going to be the case in the future—would be on the blockchain. You had these, effectively, agents—very simple agents—running back then, in some sense, in terms of doing transactions on the blockchain.

You had these economic games that were multi-turn games, to some extent, that these actors could play. We said, "Isn't that a great place to basically evolve intelligence?" You have these multi-turn games, economic incentives, game theory, and you learn all sorts of lessons from that. Obviously, there's a very different world now with generative AI and foundation models. I'd love to hear your view of where agentic payments are going.

Jeremy Allaire

Yeah.

Is it going to be crypto? Is it going to be more traditional banking systems? Is it a hybrid? What do you think are the drivers of that?

Jeremy Allaire

I mean, there's a lot in there. There's a lot we could talk about. Maybe first, my own view is that we're going through a pretty steep curve right now. We're about 3 months into a pretty dramatic shift in the fundamental capabilities of technology, probably the most dramatic that I've ever seen in my own time in technology.

I think that shift is effectively going to mean a couple of things. The first is that more and more of the actual work that's done in the real economy—especially in what we call the white-collar economy, but in many areas of service and delivery and so on—is going to be conducted by AI agents. AI agents will be conducting the work, collaborating with each other, consuming services from each other, and purchasing, effectively, specialized intelligence or output. We're on a really interesting curve there. The agentic economy is being born as we speak.

Mhm.

Jeremy Allaire

In that world, we need a different infrastructure for financial intermediation.

Why?

Jeremy Allaire

We don't have an infrastructure that can support that. We don't have an infrastructure that can work globally, interoperably, instantly, and that can be programmed through software layers by arbitrary pieces of software. That doesn't exist.

We need an infrastructure where the agents themselves can dynamically create and spin up different kinds of financial endpoints. We need transactions that can scale potentially into the billions or trillions of transactions. We don't have that. We also need the ability to handle transactions at micro scale. Consuming a certain amount of intelligence might be 5 cents or 10 cents, as it is with these. We need that to work in real time, again, between any piece of hardware or software anywhere in the world.

Isn't that really all the stuff that people have been talking about for a long time in terms of just crypto?

Jeremy Allaire

The benefits of crypto, or whatever, haven't really become possible until just the last couple of years. It really took third-generation blockchains to deliver on this. Today, you can actually look at transaction volumes of USDC, which is by far the most transacted digital currency in the world—way more than anything else. Transaction volumes have grown incredibly off of a monetary base that's also growing, but the transaction volumes are growing way faster.

And that's because money velocity has picked up. The cost to transact is now sub-cent, reliably. And so, when you take out the cost, you can do more transactions. And so, with Arc—which we can come back to—we now have an infrastructure where we can conduct transactions for a millionth of a penny, which just was never feasible before.

Yeah, tell us more about Arc, because I know that you folks are rolling this out as your own blockchain, et cetera. I would just like to learn more about what it is, what the use cases are, and what the advantages are.

Jeremy Allaire

I would love to talk about that. I want to finish one other thought on this identity piece, which is, in addition to this financial infrastructure that's needed in this world and the role that will play—and it ties back to your actual question and stuff that you were thinking about before—my own view is that agents, and seeing what happened with OpenClaw and Moltbook and all this stuff, is really interesting because it showed that you could actually see emergent forms of cooperation, interaction, and engagement amongst AIs. And that's pretty powerful, and clearly we're at the front edge of that. There's going to be a lot more of that.

If you have AI agents that are from around the world, they could be generated from lots of different models and LLMs and the like, and they need to coordinate. They need a trustworthy medium where they can do that, where they can instantiate an entity, where they can store value in that entity, where they can execute and arrange contracts that intermediate the work on the tasks, and where all of it is real-time, mathematically and computationally provable.

And so blockchain infrastructure now actually gives us the building blocks for this. When I say “agentic economic activity,” most people think, “Oh, that's e-commerce or payment.” It's not. Agentic economic activity is actually about how the organization of what we used to think of as labor and capital—essentially, how this organization of compute work happens—and what kinds of corporate forms might emerge in that world to do that.

I'm actually quite interested in that. That does tie to Arc, because that's a design space that we care about. We describe Arc as an economic operating system. This goes back to a comment I made earlier, which is that these networks are operating systems. We're moving now from the early-adopter era, which you're very familiar with, which was mostly around speculation on different things. There were some interesting things like NFTs, but we're now moving very squarely, because of stablecoins, into the real economic activity side of this.

And I think my view is that, as we go forward, the substance of what we think of as contracts, the substance of what we think of as corporations, is going to be software machines themselves. And so we're going to see this progression. Arc, as an economic operating system, is conceived of as a compute environment for laying down all of the building blocks of economic activity, whether that's storing value, moving money, instantiating a corporate form, or manifesting and intermediating complex contracts.

A lot of this stuff, which was conceptual a long time ago, is now real. We have a legal basis for it. We have regulatory clarity for it increasingly. And what's interesting is that the drivers of this machine economy are actually machines. Our view is that Arc is designed for this moment—a moment when machines are going to play a larger and larger role in all of the output of the economic system.

So if I look at a lot of the blockchains that people have found exciting over the last few years, obviously there's Bitcoin, which was almost purposefully designed in a certain way to make it a little bit less adaptable to all these new things that are happening now.

Yeah.

Solana, Ethereum, et cetera, have in the past been the traditional places where people have thought about ways to build smart contracts and a lot of the types of things that you just described. What do you think is the difference between some of these more traditional L1s or blockchains and what you're doing at Arc?

Jeremy Allaire

Yeah. So, a few big things. I think the first is that, as I think you were sharing or we were talking about before we started recording, a lot of the designs on blockchains from, let's call it, the early-adopter phase, were sort of like, “Hey, we're going to build something that is completely censorship-resistant or outside of the reach of governments.” It's sort of like, “We're building an alternative universe,” and that's the goal.

I think decentralization is itself a good goal. But I think as we move from early adopter to mainstream scaling, whether it's a major company like Walmart or a household that's thinking about how they store their wealth, the intermediaries—and there will continue to be intermediaries; we're not all going to be our own bank—have obligations in terms of the robustness of the infrastructure that they have to run.

Arc is actually set up with a number of features. One is that it's actually a known validator set. The infrastructure operators of Arc are major financial infrastructure companies, and those are companies that are held to these very high standards for infosec, compliance, reliability, and availability.

So, who are some of the validators on your network?

Jeremy Allaire

We haven't announced the validators yet, but that will come in due course. The model, at a high level, is that you have financial companies, financial infrastructure companies, including possibly large technology companies, that are responsible for running the infrastructure. It's a distributed infrastructure, but because of that, we're able to provide assurances. We're able to provide assurances that the bad guys aren't running your transactions, and we can also provide assurances that transactions actually have settlement finality. They can't be hard-forked; they can't be reorged. You can get what's called deterministic settlement finality, and that's really important, whether it's a security or a piece of cash or whatever it is, in essentially hundreds of milliseconds.

The other piece is that it's designed with real money as the foundation. There's not a volatile gas token. USDC is actually the default native token, which is now, under the law, essentially a legal form of electronic money.

Uh-huh.

Jeremy Allaire

So you have real dollars as the way that people understand it. To a company that's doing this, it's like, “I pay AWS credits. I understand how to budget for that—my treasury, my operations, my compliance, et cetera.” So this allows actual usage to make sense both to the user, to the developer, to the corporations, and to the FIs that deal with this.

So that's really important. And then I think the other thing is that we've been building in a lot of primitives that are important to the way that payment systems work, the way that capital markets work, and the privacy requirements that are needed in some of these cases, while still allowing for compliance to happen.

We've purpose-built this for a different set of participants who need to run on it, and we've had the advantage, as Circle, of working with many of the leading financial institutions that are getting into this space over the last couple of years, whether it's Visa, BlackRock, Bank of New York Mellon, or any of these types of companies. We've been able to work with them, and we've been able to work with governments around the world to hear from central bankers all over the world about what's important as they think about allowing this internet infrastructure to run the financial system.

We're trying to incorporate a lot of the requirements that they have. And so that's very different. I think it's a different design space, and I think these new distributed network operating systems will need to support the real economy's activity, not a kind of shadow economy. That's just substantively different. There are a lot of other technical things I could talk about that are part of it, but those are helpful in terms of framing this.

What else do you think is interesting that's happening in the crypto world today outside of stablecoins and related infrastructure? I know that there was a whole wave of things that people were doing. On the infrastructure side, there were ZK rollups and a variety of approaches in the last few years. Besides stablecoins, is there anything that you think was especially interesting or that will be impactful? Or are a lot of these things infrastructure looking for solutions? I'm curious how you think about crypto at large right now.

Jeremy Allaire

Yeah, I think there's a lot of attention that has gone into what I'll broadly call scaling models. If you take as your design center that these are network computers, and these network computers are really good at establishing record-keeping that is public and available to all and performing computing on those records that's public and available to all.

That’s a general utility space that’s super interesting.

Uh-huh.

Jeremy Allaire

A lot of this has been, “Okay, how do we make sure that can scale?” As an example, in a world of billions of AI agents that are swarming and doing other things, scaling this is extremely important. ZK rollups, as an example, or zero-knowledge proofs more broadly, are a way of proving compute, which allows you to do compute off-chain and then prove it on-chain. That’s really important. These off-chain or trusted execution environments and other things that provide cryptographic proofs of compute or other kinds of assertions become really key.

A lot of that research is now becoming extremely valuable. The same thing goes for a lot of that research and development, which is critical to enabling privacy. We want all the benefits of open, interoperable, permissionless infrastructure, but we also want to be able to have privacy. Corporations don’t want everyone to see what they do, and we don’t want to be doxxed and all this kind of stuff.

That’s now coming into real production. That was research-y for a long time. Arc is shipping with built-in privacy primitives from day 1, which is again the result of a lot of work over a long time. Those are important pieces.

As more large-scale financial infrastructure comes over to this, as we move toward a world where the New York Stock Exchange or the biggest derivatives clearinghouses say, “Yeah, we’re going to move to an on-chain world,” the scaling stuff becomes really critical. I actually feel like now, more than ever, those big work streams are coming online.

If I use as a reference point the early internet and the early web, I spent a long time building on that in the early ’90s and all the way up until 2001. For me, it was 10 years, and it was still awful. You kept grinding, and it was, “How do we make this useful? How do we make this useful?”

Then a whole bunch of things happened in the background: Wi-Fi, broadband, and usable internet-connected devices. You could actually start to do stuff. You could deliver software over the internet. You could deliver media over the internet. You could do communications—real-time communications—over the internet. But it was 10 years in the desert, or longer, before you could even get there.

I feel that way about the blockchain space. It’s been a dozen years or so, and now we’re having the broadband moment. The demands of society and the financial system, the agentic economy—all of that is coming together at a really interesting time.

I guess the one thing that people have been talking about for a long time in the crypto world is the securitization of other assets on the blockchain.

Yeah.

That would be stocks. Should you be able to buy fractions of Berkshire Hathaway—

Yep.

—using crypto?

Yeah.

And should that be globally available? Given the success of USDC and other stablecoins, that’s made it even more interesting in terms of the approval model. When do you think that stuff will happen, and what approach do you think will be taken? How does that tie into the agentic world?

Jeremy Allaire

It’s totally happening. There’s a great site, if people are interested, called RWA.xyz. “Real-world assets” is sort of what that refers to: RWA.xyz. I’m very proud because there are tokenized stocks out there, and the most active tokenized stock today is not Tesla, it’s not the S&P index—it’s actually Circle. That was cool to see.

That’s cool.

Jeremy Allaire

We’ve also seen this growth in tokenized money markets—basically, on-chain Treasury bills. We actually operate the largest tokenized Treasury product, called USYC, or US Yield Coin. That’s grown quite fast as well. We run the largest tokenized euro as well, EURC. We’re definitely looking at this broadening out.

I think there’s a huge effort right now at every layer of the whole financial-system stack to go into tokenization. All the way down at the layer of the people who keep the records of the stock—if you’re familiar, like Computershare and companies like it—up to the layers that are the depository and clearing systems, like the DTCC, which most people don’t know but is actually the backplane of how all securities work, they’re moving to tokenize.

The actual brokers and exchanges want to take those tokens and support trading and distribution of those tokens. Nasdaq, the New York Stock Exchange—all of them are doing this as we speak. The SEC has been providing clear guidelines on how to do that. They issued guidance just about a month ago that basically said, “Here’s what you do in all these layers. Here are your obligations.”

We’re at a point where technology and the market’s desire are creating that. Right now, the interesting thing about tokenized stocks is that it’s mostly interesting for enabling people outside the U.S. to access them. That’s where a lot of the growth has happened, because not everyone has access.

Mhm.

Jeremy Allaire

If you’re used to going the other way, right? There used to be Chinese stocks that were basically held in third-party instruments that you could purchase—

Yeah.

—on stock exchanges, so you could participate in some of the Chinese-listed entities and all.

No, there are definitely some packagers, like ETFs and funds and stuff, that are mirroring this, for sure. But I think this is similar to when the web became available or broadband really hit scale. A lot of times, people just think, “I have this existing product. I can now put it over here. Here’s the TV show; I’m going to put the TV show over here.” Or, “Here’s the game. It used to be on a CD, and now you can download it.”

I think the really interesting thing is: What can you do that you couldn’t do before? What kind of utility gets unlocked? Whether it’s fractionalization, how you can borrow and lend on these things, or how you could package them together in different ways, AI could play a pretty significant role in that as well.

It seems like it could really tie into some of the prediction-market stuff that’s been happening, because to some extent, the world’s biggest prediction markets are actually stock markets—or financial markets, I should say. Prediction markets themselves are becoming a kind of parallel infrastructure for people who participate in stock markets, right?

In fact, the biggest adopters, it seems, of the market makers of prediction markets are actually the people who are trying to figure out what reality is and what that means for companies and equity shares and stuff, and the interplay or whatnot.

Jeremy Allaire

We’re seeing that. USDC powers Polymarket, for example. The same people who are trading derivatives over here on oil or Bitcoin over here are also moving their money quickly using USDC over here to figure out what’s going to happen in some event.

The one other thing that I think has been happening recently is that there have been a couple of papers focused on basically tying proof of work to generic inference work.

Yes.

In other words, can you tie those two things so you’re being very GPU-efficient in terms of what you’re doing, but also you can effectively generate—

Yes.

—income or revenue through GPU usage while you’re using it for inference for other purposes?

Jeremy Allaire

I’m really interested in that. Again, the little conversation we were having before we recorded: proof of work was itself an innovation, and essentially, the exhaust of the proof of work of Bitcoin is just the exhaust of energy consumption. In some ways, it’s waste. The energy is wasted.

I think the idea of essentially inference compute—as GPU inference compute—as proof of work, where the work itself is the inference, and using that as the underlying basis for a proof-of-work cryptocurrency, is pretty interesting. It would potentially be something that could align with the monetary principles of something like Bitcoin, but actually be productive proof of work. That’s really interesting.

My own view—and this goes back a long time—is that people have axiomatically assumed, “Well, Bitcoin is the thing. It got the network effects. It has all of this.” I’ve always said, “I don’t know what we’re going to be using in 10 years.”

Jeremy Allaire

Like, we don't know. Now, Bitcoin has lasted a really long time. But I think the paradigm shift that we're seeing in energy infrastructure, in the performance of the conversion of energy into intelligence, and in the computer layers in that, certainly opens up a new avenue to think about this that wasn't readily available 15 years ago.

So, if you were to give one piece of advice to agents, it would not be “buy Bitcoin.” I'm just checking.

I have a better question. So, in terms of, say, if we were to think out 10 years, what does that world look like in your mind? And obviously, we're going through a period of intense change.

Yeah.

I'm finding it incredibly hard to predict the future right now in terms of just what's going to happen in AI, much less AI plus crypto plus the global economic system plus everything else. So, given all that, and putting that aside, what is your vision of the future?

Jeremy Allaire

Yeah. Well, I mean, a couple of things I would say. The first is sort of the thing that everyone is debating right now is the pace of AI diffusion, right? So, what is the pace of AI diffusion, and what does that then imply in terms of the amount of change that we're going to have to deal with? And so, that's all debatable, right? You hear Dario debating that versus others and so on.

But it definitely feels like the diffusion limiters are, in some cases, bureaucratic, in some cases, legal, in some cases, human risk or other things, right? We have these limiters that are there. But it does seem like the pace of diffusion is accelerating and will continue to accelerate, and that's pretty dramatic.

And so, I guess my own view—it’s very rooted in my own political and economic philosophy—is that we have a real opportunity to create essentially new social, political, and economic organizational structures. In many ways, we have to. There's a kind of, in these periods, whether it's the Enlightenment and the Industrial Revolution and other things, there's a new definition of the social contract.

And that new definition of the social contract is then, in turn, reflected in social, political, and economic ordering and the mechanisms that we use for those things. It feels to me like we're going to be forced through that. I think that's simultaneously terrifying and exciting, et cetera. I am of the view that we are going to have a kind of lag effect between the disruption and the establishment of those new institutional forms.

But at the same time, I actually believe new institutional forms are going to be emergent out of this. And so, as I talked about earlier, the formation of these kind of on-chain organizations that have different forms of governance and contracting, and a mixture of human and agentic actors—that seems like we're going to have a lot more of that. We're probably going to have huge proliferation of that.

And it may be that those corporate forms are the most productive corporate forms that we've ever seen in economic history. And then they'll need to be kind of like an overlay into the governance systems of political organizations and systems as well.

So, my view is we're going to have, simultaneously, all around the world, a renegotiation of the social contract, and it's going to require new systems of participation in economics and governance that we haven't had. That's very high-level mumbo-jumbo-y, but it's also sort of how I think about it at a high level.

That's super interesting. Have you ever heard of a book—“Lady of Mazes”?

No.

It's like a sci-fi book from, I don't know, 15 years ago about the post-AI world. Part of it is that, as a big enough block or demographic emerges in human society, an overlay AI agent observing everything spawns a specific agent that represents that viewpoint, which then is part of this sort of virtual senate.

I see.

Negotiating policies. So it's kind of this interesting view of how you can spontaneously spawn these sorts of systems from a governance perspective, which is kind of cool.

I would love to read that.

What is your prediction? And so, if you look at a lot of prior waves of technology, their actual impact on GDP has been difficult to tease out.

Jeremy Allaire

Mhm.

Right? So, the productivity gains of the internet versus actual GDP growth, or things like that, have been notorious, and there are all sorts of reasons for that. It could be measurement. It could be deflationary aspects of some of these things. It could be a variety of things. How do you think about the GDP impact of AI? So, if you think ahead 5 years—

Yeah.

And, you know, what do you think the global economy is?

Jeremy Allaire

Yeah.

10% bigger? 50% bigger? 3 times the size? Does that even matter as a metric anymore?

Jeremy Allaire

Yeah, I mean, I see this debated all the time, and Cathie Wood's talking about, you know, we're going to have 10% GDP growth for the 2030s, et cetera. I don't quite know what to think.

I mean, I think it's quite plausible that the giant leaps that we see in productive output in a huge range of industrial to other commercial services really drive a very significant discontinuous jump in GDP at an absolute level. It'll have, in many ways, probably less meaning than we've historically had with GDP and the kind of economic well-being indexes that we think about.

The risk here is that GDP growth effectively is a sort of capital capturing more capital at the expense of humans. That's the real risk. And so, GDP growth generally has been a really great thing, and the question is: Will it remain a great thing? Do we have the new social contract to deal with that yet?

I guess my general view, just as a technologist, talking about seeing what we see with diffusion and other changes, is that it does feel like we have the potential for double-digit GDP numbers in the 2030s. That's not unrealistic to me.

Not that that's going to be uniform all around the world, but certainly in large, large parts of the world, that seems very achievable based on what I see.

Mhm. Amazing. Thank you so much for joining us today at No Priors. Super interesting conversation. Thank you.

Find us on Twitter at No Priors Pod. Subscribe to our YouTube channel if you want to see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.