A millionth of a penny: Jeremy Allaire on the payment rails of the agentic economy

Elad Gil opens No Priors with Jeremy Allaire, co-founder and CEO of Circle, on the pile-up of cryptocurrency, AI, and agentic payments. Allaire dates the company to 2013. At inception he wanted a protocol for dollars on the internet: a way to store and move value instantly, globally, with as little friction and cost as the rest of the network. He had been watching Bitcoin and had spent years on internet infrastructure. The other founding idea, he says, was programmable money — the bet that blockchains would become operating systems, and that machines, including autonomous software, would intermediate economic activity on those systems.

Generative AI did not exist yet. What they thought they could commoditize was a safe digital-dollar payments layer, with programmability on tamper-resistant machines that run on the internet. If that worked, the financial system could be safer, more accessible, more efficient, and capable of uses money had not had before. The episode's through-line is that the world those 2013 notes described is the one arriving now, and that correspondent banking is the wrong substrate for it.

Gil asks why the dollar mattered, given that early crypto, and Bitcoin in particular, was a response to the global financial crisis and a wish to leave traditional finance. Allaire reaches back further than 2008. He had been reading Austrian economics and sound-money theory since the early 1990s. The crisis pulled him in because he thought a safer system had to be possible, and the idea he kept returning to was full-reserve money. Bitcoin, in that frame, already is full reserve: you cannot fractionally lend the coin itself. Full reserve is not the same as backing by a scarce metal. It is the opposite of fractional-reserve banking.

The last time the United States had this argument in public was after the Depression, when banks failed in waves. A group of economists advanced the Chicago Plan. Irving Fisher, whom Allaire places in that debate and who wrote *100% Money*, wanted government-obligation money that you could hold and even lend, but not lend fractionally against. The banks lobbied hard. They liked the leverage. What the country got instead was the FDIC, a corporate insurance wrapper around the same risk-taking. The 2008 crisis, Allaire says, was 30x, 12x, 14x leverage against the base layer all over again.

His philosophy is that the real economy still runs on major reserve currencies, the dollar among them, and that this will remain true for decades. What you can do is make the instrument safer. Dollar stablecoins are that full-reserve form. The GENIUS Act, passed the previous year, writes the model into U.S. law as narrowly bound narrow money: you may lend the money, but you may not run fractional reserve against it. Versions of the same architecture now exist in Europe, Japan, and other jurisdictions. The original vision, he says, is in the statutes. The remaining job is to make it extraordinarily useful.

USDC, under those rules, is one-for-one redeemable against very safe liquid assets. Circle holds short-duration U.S. Treasuries, overnight Treasury collateral at global banks, and cash for immediate redemptions at large custodians such as Bank of New York Mellon. Average duration on the T-bill book is about thirteen days, liquid enough to treat as a cash instrument. One dollar buys one USDC; it stays backed by treasuries, repos, and short T-bills. Circle publishes daily transparency on most of the reserve through a system built with BlackRock. Other issuers, Allaire notes, did not all take this constrained path. Fiat stablecoins that did are now living inside federal law rather than around it.

What do people actually do with USDC? Allaire's picture is a general-purpose architecture, used at both extremes. At the small end, someone pays 25 cents for a digital object in a blockchain game. Agents are already paying other agents a dollar, 50 cents, 20 cents for model output. At the large end, the biggest electronic trading firms settle multi-hundred-million-dollar capital-markets transfers on the same instrument. He reaches for email: the payload of "this is what I had for breakfast" and a classified dossier ride the same protocol. USDC does not care. That indifference is the point of a general architecture.

The middle of the range is already commercial. Merchants on Stripe and Shopify take it. Visa moves money on its own internal network with it instead of the legacy banking rails. Neobanks and remittance firms use it as a value-transfer layer. A B2B fintech, Allaire says, had just put USDC at the core of its treasury system so invoices could be paid anywhere. Crypto companies raising money ask for USDC so the funds can arrive on a weekend instead of racing an afternoon wire cutoff. Gil ticks off three advantages: it is available 24/7, fees can be tiny relative to volume, and people without a U.S. bank account can still hold dollar exposure.

Allaire agrees, and he adds a geopolitical clause. Store of value is a large use. A motivation for the GENIUS Act, one Circle had been pushing, is that digital dollars are a way to keep exporting the dollar. That, he says, is strategically important to the United States. For developers there is a more local fact. There has never been programmable money in this sense. The stablecoin network is a public API on the public internet. If you want global dollar settlement in your product, you do not negotiate access. You connect to the contract, and the application has digital-dollar utility. Smart contracts are code wrapped around that money: pay under these conditions, or spin a financial instrument off the same layer.

Programmable money was the early idea; smart contracts were the first expression. Thirteen years ago Allaire's stronger claim was that blockchain networks are operating systems, in the same lineage as mobile OSes, the web, cloud runtimes, and now foundation models that execute tasks. They have virtual machines and Turing-complete code. They also have attributes the others do not. Published code is tamper-resistant, a machine you cannot quietly rewrite. Every input and output is auditable in real time, on a public chain, by anyone. The compute is open-source by nature. Transaction and compute integrity can be assured cryptographically. You want to know the machine is doing what it said, and you want inputs, outputs, and state to be provable. That used to be hard. It is native here.

Those properties already mattered for fiduciary finance. They matter more once autonomous software is acting in the economy. Gil recalls a speculation from seven or eight years ago: that AGI might emerge on a chain, because simple agents were already running transactions, playing multi-turn economic games, and living inside game-theoretic incentives — a plausible nursery for evolved intelligence. Foundation models took a different path. He still wants Allaire's view of where agentic payments go, and whether the rails are crypto, banks, or a hybrid.

Allaire says we are about three months into the steepest capability curve he has seen in a career in technology. More of the real economy's work, especially white-collar work, will be done by agents that collaborate, consume each other's services, and buy specialized intelligence. The agentic economy is being born as they speak. The financial plumbing cannot support it. There is no stack that is global, interoperable, instant, and programmable by arbitrary software; that lets agents spin up their own financial endpoints; that can scale toward billions or trillions of transactions; and that can clear a five- or ten-cent purchase of intelligence in real time between any hardware and software on earth. Those properties were talking points in crypto for years. Allaire says they only became real with third-generation blockchains in the last couple of years.

USDC is now, in his account, the most transacted digital currency in the world. Volumes are growing much faster than the monetary base because velocity is up and the cost to transact is reliably under a cent. Take the cost out and frequency explodes. On Arc, which he will describe next, a transaction can be done for a millionth of a penny. That, he says, was never feasible before. OpenClaw and Moltbook interest him less as products than as evidence that cooperation and interaction can emerge among AIs. The question he actually wants answered is what organizational forms appear when agents start to accumulate assets, take on liabilities, instantiate entities, store value, and execute contracts in a medium that is real-time mathematically and computationally provable. Agentic economic activity, in that frame, is not a checkout button. It is a new way of organizing labor and capital.

Circle describes Arc as an economic operating system: a compute environment for storing value, moving money, instantiating a corporate form, and intermediating complex contracts. Allaire's claim is that the substance of contracts and corporations will increasingly be software machines, and that the drivers of this machine economy are themselves machines. Legal and regulatory clarity, he says, is finally catching up to ideas that used to be conceptual. The design center is a moment when machines produce a larger and larger share of economic output.

Gil asks how that differs from Bitcoin, Solana, and Ethereum, the L1s people already use for smart contracts. Allaire splits the history into two design spaces. Early-adopter chains optimized for censorship resistance and a universe outside the reach of governments. Decentralization is a good goal. Mainstream scale is a different user: Walmart, a household storing wealth, and the intermediaries who will still exist. "We're not all going to be your own bank." Those intermediaries have obligations for robustness. Arc starts from that fact.

The chain uses a known validator set. Operators are major financial-infrastructure firms, possibly including large technology companies, held to high standards of information security, compliance, reliability, and availability. Circle has not named them yet. Because the set is known, Allaire says, you can promise that random bad actors are not running the ledger, and you can promise deterministic settlement finality: no hard forks, no reorgs, irreversibility in hundreds of milliseconds, which cash and securities require. There is no volatile gas token. USDC is the native asset, now a legal form of electronic money, so a company can budget chain costs the way it budgets AWS. Privacy primitives, payment and capital-markets components, and compliance hooks are built in from day one, shaped by years of work with Visa, BlackRock, BNY Mellon, and central bankers. The distributed OS, he says, has to support the real economy's activity, not a shadow economy.

Asked what else in crypto is real besides stablecoins, Allaire points at scaling. If these networks are public computers for public records, they have to survive billions of swarming agents. Zero-knowledge proofs and ZK rollups let you compute off-chain and prove the result on-chain; trusted execution environments do related work. The same research is what makes privacy possible on an otherwise open, permissionless base — corporations do not want a world-readable ledger, and people do not want to be doxed. That stack was "researchy" for a long time. Arc ships with privacy primitives on day one. When the New York Stock Exchange or the large derivatives clearinghouses move on-chain, scaling stops being a white paper and becomes a production constraint.

He analogizes from his own internet years, the early 1990s through 2001: a decade in the desert of making the thing useful, until Wi-Fi, broadband, and decent connected devices finally let you deliver software, media, and real-time communications over the network. Blockchain, after about a dozen years, feels like that broadband moment, and the agentic economy is arriving at the same time. Tokenization of other assets is no longer a speech. The site rwa.xyz tracks it. Allaire notes, not shyly, that the most active tokenized stock is not Tesla and not the S&P — it is Circle. Circle also runs USYC, which he calls the largest tokenized Treasury product, and EURC, the largest tokenized euro. Computershare-class registrars, DTCC, Nasdaq, and the NYSE are all moving. The SEC, about a month before the taping, issued layer-by-layer guidance. Right now the practical win of tokenized stocks is letting people outside the United States buy U.S. names. Gil notes that Chinese listings once used third-party vehicles for the reverse trip. Allaire's interest is the new utility: fractionalization, borrowing and lending, packaging, with AI in the loop.

Prediction markets sit in the same picture. Gil observes that the world's biggest prediction markets are the financial markets themselves. Allaire says USDC powers Polymarket, and that the most intense users often look like people already in equities and derivatives, moving dollars quickly to hedge a view of events. A newer research line ties proof of work to GPU inference: the work *is* the inference, so mining produces something besides heat. Bitcoin's proof of work, Allaire says, is energy exhaust. Productive proof of work could still rhyme with Bitcoin's monetary principles. He will not axiomatically assume Bitcoin is what we use in ten years. Network effects are real and the coin has lasted, but converting energy into intelligence is a paradigm that did not exist fifteen years ago. Gil jokes that if you could give agents one piece of advice, it would not be "buy Bitcoin."

Gil asks for a ten-year picture in a moment when even AI, let alone AI plus crypto plus the world economy, is hard to forecast. Allaire starts with the live debate over the pace of AI diffusion, the one Dario Amodei and others are having in public. Limiters are bureaucratic, legal, and human risk appetite. Diffusion is still accelerating, and that is both exciting and frightening. Periods like the Enlightenment and the Industrial Revolution, he says, forced a new definition of the social contract, which then showed up in political and economic order. This period will force the same rewrite. There will be a lag between disruption and new institutional forms. He also believes those forms will emerge: on-chain organizations with mixed human and agentic members, on-chain governance and contracts, possibly the most productive corporate forms in economic history, which will then need an interface into political government. Around the world, a renegotiation of the social contract is already on.

Gil recommends Karl Schroeder's novel *Lady of Mazes*, a post-AGI story in which, when a bloc of people gets large enough, an overlay AI spawns an agent to represent that view in a virtual senate. Allaire says he would love to read it. On GDP, Gil notes how badly the internet's productivity showed up in the official numbers. Cathie Wood has talked about 10% GDP growth in the 2030s. Allaire does not quite know what to think, and then he does: giant leaps in industrial and commercial output could produce a discontinuous jump in GDP, even as the metric means less than it used to. The real risk is that the growth is capital capturing more capital at the expense of humans. Whether GDP remains a good thing depends on whether a new social contract is in place. As a technologist watching diffusion, he thinks double-digit GDP numbers in the 2030s are not unrealistic — not uniform everywhere, but achievable in large parts of the world.