Jeremy Allaire’s The Agentic Economy makes a persuasive case that AI agents need economic infrastructure. Its more ambitious claim is that this infrastructure must be onchain. That conclusion is plausible, but it has to be earned function by function.
Source note: Jeremy Allaire. The Agentic Economy: The Convergence of Intelligence and the Economy. July 2026. Read the full treatise.
What This Is
Allaire’s argument begins with a useful picture of the firm. A company is not only a legal entity, a brand, or a collection of employees. It is an information system that coordinates many kinds of cognition: product decisions, engineering, sales, marketing, finance, legal work, support, and operations.
AI makes more of that cognition available as software. Work can be broken into agentic skills, routed through an orchestrator, checked against acceptance criteria, and recombined into an outcome. A small team can operate with capabilities that once required several departments. Some skills remain internal. Others can be discovered and purchased from outside the firm.
This is the strongest part of the treatise. As the cost of cognition and coordination falls, the boundary of the firm becomes less rigid. A capability that is clean enough to call inside a workflow may eventually be clean enough to sell into a market.
Allaire then takes the next step. If agents can work across companies, they need identity, money, contracts, settlement, and ownership. Those functions also need to operate continuously, programmatically, and across borders. His conclusion is that the agentic economy and the onchain economy are not adjacent trends. They are the same economy viewed from two sides.
That is the bet worth examining.
The Core Thesis
The treatise’s core thesis is that AI supplies a new labor layer while blockchains supply its economic form: programmable money, contracts, settlement, governance, and ownership. The first half of that claim is already visible in agentic workflows. The second is a proposed architecture whose value depends on the economic function being redesigned.
The Argument Map
The argument moves through firm decomposition, agent orchestration, money, credit, global coordination, software pricing, corporate form, concentration, and finally shared ownership. Its structure is cumulative: once work becomes modular and machine-executed, Allaire argues that each surrounding institution must become equally programmable.
Agents Clearly Need Economic Rails
An agent that only drafts text does not need much of an economy. An agent that purchases data, hires another agent, commits funds, signs a contract, or holds an asset does.
As soon as agents cross that line, several requirements appear:
- They need an identity tied to a responsible person or institution.
- They need scoped permissions rather than unlimited authority.
- They need a way to quote, pay, and receive payment.
- They need records that establish what was authorized and what occurred.
- They need rules for disputes, reversals, fraud, and failure.
- They may need to hold assets or working capital.
The treatise is right to move the conversation beyond models and prompts. A functioning agent economy is an institutional-design problem. The intelligence layer may decide what to do, but the economic layer determines what it is allowed to do, how counterparties trust it, and who answers when it fails.
This is also why the shift from software seats to software work feels credible. A human no longer has to sit inside every application. An agent calls a capability when it needs it. Pricing naturally moves toward usage, tasks, or outcomes, often wrapped in a commitment that gives buyers predictable costs. Foundation models become an input cost inside a specialist agent business, much like cloud infrastructure sits inside a software company’s cost base.
None of that requires a speculative future. The design pressure already follows from agents consuming services directly.
What Skeptics Would Challenge
Software-Native Does Not Automatically Mean Onchain
The treatise’s weakest move is also its most important. It often treats four properties as if they were interchangeable: programmable, internet-native, open, and onchain.
They are not.
A bank payment API is programmable. A platform ledger is internet-native. A conventional database can produce an auditable event history. A legal entity can own assets and delegate authority to software. A marketplace can coordinate agents using accounts, contracts, escrow, and internal settlement without placing every action on a public blockchain.
These systems have limits. They create intermediaries and platform boundaries. They may not compose cleanly across providers. Their records can be altered by whoever controls the database. Cross-border settlement can be slow, fragmented, and expensive. But those are tradeoffs to compare, not proof that the alternative must be onchain.
The useful question is narrower: where does an agent workflow need a neutral economic rail more than it needs the simplicity, privacy, reversibility, and legal clarity of a conventional one?
That framing turns a sweeping thesis into an architectural decision.
The Strongest Ideas
Where Onchain Rails Earn Their Place
Onchain infrastructure has a strong case when several conditions appear together.
First, the workflow crosses platform or company boundaries. If every participant already trusts one operator, that operator’s database may be enough. If no participant should control the shared ledger, neutral settlement becomes more valuable.
Second, ownership must be portable. A platform balance is usually a claim inside one system. An onchain asset can, subject to its design and regulation, move between applications without waiting for the original platform to grant an export path.
Third, the market needs open participation or composability. Agents may need to discover services, combine them, and settle with counterparties that were not integrated in advance. Open protocols can reduce the cost of negotiating each connection.
Fourth, the system needs verifiable scarcity or provenance. If an agent owns a license, credential, security, or unique right, a shared ownership record may matter more than it does for an ordinary API call.
Fifth, credible exit matters. Allaire’s civic argument depends less on transaction speed than on the ability to resist capture. Forkability, open standards, and portable ownership can discipline chokepoints only if participants can actually leave.
These are real advantages. They are also conditional. Many agent actions will remain inside trusted firms, regulated platforms, and existing payment systems because those environments offer privacy, customer support, reversibility, and known accountability. The likely future is not one universal ledger. It is a mixed architecture in which some functions move onchain and others do not.
Load-Bearing Assumptions
The Money Argument Is Ambitious
Allaire’s preferred monetary substrate is full-reserve stablecoin money with fast, final settlement on open networks. His logic is that machine-speed transactions cannot pause to assess the creditworthiness of every issuer. Money used by agents must remain fungible and redeemable without forcing each micro-transaction to price embedded bank risk.
That is a coherent design goal. The treatise goes further by arguing that velocity can perform work historically done by leverage. A dollar that settles and becomes reusable almost immediately can support more activity without being multiplied into layers of risky claims. Credit then sits above the safe money layer rather than being embedded inside the money itself.
This is one of the document’s most consequential claims, and it deserves more evidence than the treatise provides. Faster circulation does not by itself answer questions about liquidity under stress, maturity transformation, credit creation, monetary policy, or who absorbs losses. Machine underwriting may reduce the cost of evaluating small borrowers, but it does not eliminate adverse selection, correlated risk, manipulation, or the political choices inside a credit model.
The right takeaway is not that the monetary case fails. It is that the proposed stack is a hypothesis for financial architecture, not an inevitable consequence of agent adoption.
What This Means for Buyers and Operators
The Deeper Argument Is About Ownership
The treatise becomes more compelling again when it turns from jobs to income.
Allaire argues that the central risk is not necessarily mass unemployment. People may continue to work while their share of economic output shrinks. If agents occupy new tasks quickly, if their marginal cost tracks falling inference prices, and if the surplus they generate finances more software labor, then capital’s share can become self-reinforcing.
That shifts the policy question. Defending every existing job is unlikely to preserve bargaining power in an economy where cognition becomes cheaper. Broader ownership of the agents, models, infrastructure, and firms that capture the gains addresses the distribution of value more directly.
Tokens can lower the administrative cost of distributing ownership and governance. But the treatise correctly acknowledges that coordination technology does not defeat concentrated power by itself. Cooperatives and mutual institutions achieved durable shared ownership without blockchains. Other ownership movements failed because of law, capital, employers, and the state, not because they lacked a better ledger.
The technology can make ownership easier to express. It cannot decide who receives it.
That distinction may be the most important lesson in the entire document. Infrastructure creates options. Institutions and politics choose among them.
What This Means for Builders
A Better Test for the Agentic Economy
The agentic economy should not be evaluated as a referendum on crypto. It should be evaluated as a sequence of design decisions.
For each economic function, ask:
- Does the agent need to act across organizational boundaries?
- Is there a trusted operator, or does the system need a neutral shared record?
- Must ownership or identity travel between platforms?
- Is finality more important than reversibility?
- What privacy, compliance, and dispute-resolution obligations remain offchain?
- Who is accountable when the agent or the infrastructure fails?
- Does moving the function onchain broaden participation, or merely change the technical wrapper around the same concentration of power?
Allaire is probably right that AI agents will force the economic stack to become more programmable. He may also be right that stablecoins and open networks will become important parts of that stack.
But agents do not need blockchains because they are agents. They need them when neutral settlement, portable ownership, open composition, or credible exit solve a problem that conventional infrastructure cannot solve as well.
That is a smaller claim than “the agentic economy is the onchain economy.” It is also a more useful one for builders. It tells them what to inspect before choosing the rail.
What to Read in the Original
The highest-value sections are Section 1 on the decomposition of the firm, Section 2 on orchestration, Section 6 on the shift from seats to work, and Sections 8 and 9 on labor share and ownership. Sections 3 through 5 contain the most ambitious monetary and onchain claims and deserve the most skeptical reading.
Bottom Line
The treatise succeeds in showing that agentic work creates economic-design problems, not just model-design problems. Its onchain conclusion is strongest when neutral settlement, portability, openness, and exit are real requirements. Where those requirements are absent, conventional infrastructure may remain the better rail.
Source
Jeremy Allaire. (July 2026). The Agentic Economy: The Convergence of Intelligence and the Economy. Available at: https://agenticeconomytreatise.com/exports/The-Agentic-Economy-Treatise.pdf