Allaire publishes personal treatise on the "agentic economy"
Circle CEO Jeremy Allaire announced a personal paper, The Agentic Economy, on X on the evening of July 12 Eastern Time. He described it as a work that had been "years in the making" and built it around what he called the "convergence of intelligence and the economy." His central claim is direct: the agentic economy and the onchain economy are the same economy seen from two sides.
In Allaire's framing, once AI agents take on work historically done by employees inside companies, value transfer between those agents has to happen over open, programmable networks at machine speed. He argues that such a network is, in practice, blockchain and stablecoin-based financial infrastructure. The treatise has its own website, agenticeconomytreatise.com, with a 60-second summary, a short read, the full paper, an audiobook, and visual maps.
Allaire labeled it "a personal work" rather than a formal Circle product release and invited readers to enter at whatever depth they preferred.
AI and blockchain as a new convergence
The first section of the paper treats the meeting of AI and blockchain as a recurring pattern rather than a coincidence. Allaire points to three historical examples: steam power and factories, electricity and telegraphy or telephony, and computing and the internet. In each case, he argues, two independently maturing technologies arrived at the same time and amplified one another.
He places AI and blockchain into that same pattern. On one side sits an operating system for intelligence, made up of models, agents, and prompts. On the other sits an operating system for the economy, covering money, settlement, and ownership. He also points to three compressing exponential curves: falling marginal intelligence costs, rising model capability, and a declining labor share of output. That combination, he argues, weakens the traditional structure of the firm.
The firm becomes a coordination layer
Allaire builds this part of the paper around Ronald Coase's transaction cost theory. Firms exist, he argues, because coordinating each task in the open market has historically been more expensive than organizing work internally. If the marginal cost of cognition approaches zero, that boundary starts to recede. Work can be broken into tasks and assigned to agents, while the company shifts from a thick operating structure to a thin coordination layer.
He also identifies four capabilities he says agents cannot take over: Judgment, Care, Accountability, and Taste. Human involvement, in his description, falls into two modes. One is "in the loop," where a person approves each step. The other is "on the loop," where a person sets goals and supervises the system while it runs. As agents become capable of longer autonomous execution, people move from the first role toward the second.
Money for agents: par, redemption, and deterministic finality
Section 3 is the most technical part of the treatise. Allaire says money used in an agent-driven economy must satisfy three conditions at once: stable par value, redemption on demand, and deterministic finality. If any one of those fails, agents would need to reassess whether the money is good on every transaction. He argues that is economically unworkable at machine speed.
That leads him to a classical-sounding conclusion. The monetary base for this system, he says, needs to be full-reserve, bankruptcy-remote, and backed by central bank-grade assets such as short-term Treasuries, overnight repos, and central bank cash. He ties that framework to the recently passed U.S. stablecoin regulatory structure and says it matches the product structure of USDC.
Allaire summarizes the shift as "Velocity replaces Leverage." In the traditional banking model, the same dollar is lent multiple times to create a money multiplier. In an onchain model, he writes, the same dollar can be locked, lent, redeemed, and locked again within seconds, producing similar economic effect through speed rather than through synthetic dollars carrying systemic risk.
He also argues for what he calls a "yield firewall." Money itself should carry zero yield, in his view. The moment a holder seeks yield, that position enters a credit market and takes on credit risk. If that wall is broken, for example through direct interest payments on stablecoins, the monetary instrument turns back into a credit instrument and weakens the safety case.
From per-seat subscriptions to pricing by work
In Section 6, Allaire turns to business models. Over the past 30 years, software-as-a-service has largely been sold on a per-seat subscription basis, charging a monthly fee for each human user. He argues that this pricing axis breaks down when the user is an agent rather than a person. Agents do not have work-hour limits or attention limits and can run thousands of tasks at once.
His view is that pricing will move toward units of work, with usage-based pricing, outcome-based pricing, and bundled subscription models existing side by side. He extends that logic to micropayments. In his telling, the internet did not fail to build micropayments because settlement was too expensive. It failed because humans face mental transaction costs: repeatedly deciding whether to approve tiny payments is more exhausting than the payment itself. If the buyer is a machine, that friction disappears. Micropayments then become practical, not mainly for content access, but for pricing labor between agents.
From there, he argues that models become a cost layer while agents become the business layer. Base models, he says, will tend toward commoditization, while value accrues to agents that own customer relationships, proprietary data, workflows, and accountability for results. He also says frontier model owners will not simply accept commoditization, producing what he describes as a "barbell" structure: many mid-layer tasks handled by agents, with the hardest frontier work still delivered directly by model companies.
Reserve income and concentration risk
Section 8 focuses on concentration. Allaire's test is whether a layer combines increasing returns to scale with a non-forkable bottleneck. He says application and agent layers can be forked, and model layers are moving toward commoditization. The areas more likely to concentrate, in his view, are identity and the "override key." A verified identity can be reused across marketplaces, creating winner-take-most dynamics. An override key that can reverse a deterministic core represents, at the limit, control power over the system.
He then points to a form of concentration inside his own industry: major stablecoin issuers capturing reserve income. Allaire calls that income a "policy artifact" because current rules block issuers from passing that yield to holders as interest. Since regulation created that rent, he argues, regulation can also reallocate it through competition, supervision, or forced ecosystem sharing. He adds that "reasonable policy may require more yield to flow to holders."
He ultimately frames concentration as a geopolitical issue. Bottlenecks that gather rent can also be weaponized. Fast economic interconnection may raise the cost of conflict, he writes, but it does not remove the leverage itself. He uses the outbreak of World War I in 1914 as the historical comparison.
Published after Circle Agent Stack
The paper was released about two months after Circle formally launched Circle Agent Stack on May 11. That product set includes Circle CLI, Agent Wallets, Agent Marketplace, and Nanopayments built on Circle Gateway.
At the time, Allaire said in a company announcement: "The next phase of the global economy will be driven by AI and agents... developers and autonomous agents need financial infrastructure built for machine speed, extreme cost efficiency, and global availability and interoperability, and this is only possible with stablecoins and onchain infrastructure."
Several reading formats and early engagement data
The treatise website offers multiple entry points, including a sixty-second thesis, a short read, the full treatise, an audiobook, and visual maps. According to the source material, the X post reached 33,380 views, 494 likes, 82 reposts, and 285 bookmarks within six hours.
