Summary
BlackRock’s Intelligent Economy white paper argues that as AI agents begin to purchase data, call software services and obtain computing power with limited human intervention, stablecoins and tokenized assets could become important machine-native payment and settlement tools. The report sees AI and digital assets moving from parallel technology trends toward a shared economic infrastructure.
Key takeaways
BlackRock: AI Agents and Digital Assets Could Support New Forms of Automated Commerce
- BlackRock defines agentic AI as systems that can use external tools and infrastructure, plan around a defined objective and execute multi-step tasks with limited human intervention.
- The report describes a potential complementarity between AI’s machine-native intelligence and digital assets’ machine-native money.
- BlackRock says digital assets designed to maintain relatively stable value and other on-chain assets could support always-on, low-value and highly automated machine-to-machine payments.
- Protocols such as x402 and ACP are cited as emerging approaches to agentic commerce payments and settlement, although the broader infrastructure remains under development.
- Standardized rights linked to computing capacity could become a new digital-asset use case for financing and programmable settlement.
- Agent identity, authorization limits, AML, KYC and KYA will be central to controlling autonomous financial activity.
AI and digital assets are moving toward the same economic layer
BlackRock’s Intelligent Economy white paper presents artificial intelligence and digital assets as two technology trends that are beginning to converge. Historically, AI has been discussed primarily as a tool for information processing, prediction and content generation, while digital assets have been framed as a new form of financial infrastructure. That separation is becoming less clear as AI systems gain the ability to call external services, access infrastructure and interact with economic networks.
The report’s focus is not simply on a human using an AI assistant to initiate a conventional payment. It is on agentic AI systems that can pursue a defined objective, make intermediate decisions and complete multiple steps with limited human involvement. An agent may need to purchase data, call a software service, obtain computing capacity or pay for resources as a task progresses. In that model, payment is no longer an activity that merely follows a human decision. It becomes part of the agent’s ability to complete its assigned work.
BlackRock describes the potential relationship as one between machine-native intelligence and machine-native money. AI can interpret information, formulate plans and select actions. Blockchains and digital-asset networks can represent assets in a machine-readable form and provide programmable rules for transferring and settling value.
Tokenization and language processing share a structural idea
One section of the white paper compares the structure of large language model tokenization with the tokenization of assets on a blockchain. Language models break human-readable text into smaller units, such as words, subwords or characters, and assign numerical identifiers to those units. Embedding layers and other computational processes then allow the model to analyze relationships among the resulting representations. The model’s numerical output can eventually be decoded back into language that people can read.
Asset tokenization uses a different technical process and serves a different purpose, but it also converts complex real-world information into a standardized representation that software can process. Cash, fund interests, securities and other ownership or economic rights can be represented as digital tokens recorded on a distributed ledger. Transaction data, permissions and other conditions can then be interpreted by a network, a smart contract or transaction scripts.
The two processes should not be treated as equivalent. Tokenization in a language model addresses information processing, while financial tokenization addresses the representation and transfer of value or rights. The similarity lies in the architectural function: both convert inputs that are difficult for machines to handle directly into discrete, standardized units. For an autonomous agent, this type of standardization could make it easier to discover, invoke and settle for external resources.
Agentic commerce needs machine-native payment rails
The white paper argues that the growth of agentic AI could increase demand for payment systems designed for machine-to-machine activity. Existing networks, including card systems and ACH, already support large volumes of automated payments. However, certain agentic workflows may involve transactions that are continuous, low value, high frequency and executed according to software rules. Account-opening requirements, interface constraints, settlement timing and fees could make some traditional rails less suitable for these use cases.
The white paper says digital assets designed to maintain relatively stable value, native crypto assets and other on-chain assets could serve as payment and settlement instruments on programmable networks.
BlackRock cites x402 and ACP as examples of emerging protocols being deployed or explored for agentic commerce and machine payments. Their mention does not mean that a single standard has already emerged, or that every automated transaction is suited to on-chain settlement. Practical deployment still involves questions about identity, authorization, error handling, disputes, asset conversion, network availability and compliance screening.
The most important distinction is between the ability to execute a transaction and the authority to execute any transaction. An agent should not receive unlimited economic permissions simply because it can interact with a payment network. Organizations connecting agents to financial or operational systems will need limits on transaction size, counterparties, service categories and time windows. They may also need auditable approval controls that distinguish actions an agent can take independently from those requiring human confirmation.
Computing capacity could become a larger digital-asset market
The report also identifies computing power as a potentially significant resource in the AI economy. Training and operating AI systems require processing capacity, and the importance of that capacity may grow as models become more capable, agents run for longer periods and applications become more complex.
The paper cites analysts who estimate that annual cloud revenue from hyperscale providers could exceed $1 trillion by 2030. That estimate illustrates the potential scale of the computing economy, but it does not establish that any particular asset or business model will succeed. For digital assets, the possible opportunity is to represent rights to computing capacity, service credits or related economic claims in a more standardized form and settle them through programmable mechanisms.
If a computing entitlement could be clearly defined, measured and verified, an agent might acquire resources according to a task’s requirements and settle payment based on actual use or pre-agreed conditions. Computing capacity, however, is not a fully interchangeable commodity. Chip type, location, network latency, availability, service quality and data-security requirements can all affect its value. Any tokenization model would therefore need to address how the underlying resource is described, delivered, measured and handled in the event of underperformance.
The same issue applies to financing. A digital representation of computing capacity would not eliminate the need to assess the quality of the underlying infrastructure, the enforceability of the claim or the risks associated with the provider. Programmability can automate a process, but it cannot by itself guarantee that the promised resource exists or will be available when needed.
Governance and compliance remain central constraints
Autonomous agents entering commercial and financial networks could create efficiency gains, but they could also magnify operational and security failures. Errors, excessive permissions, compromised keys, impersonated identities and automated fraud may spread more quickly when software is able to act continuously. Several basic questions therefore become unavoidable: Who does the agent represent? What exactly has it been authorized to do? Who is responsible for the resulting transaction? Can an erroneous or fraudulent payment be stopped, reversed or investigated?
The white paper points to AML, KYC and KYA, or know-your-agent processes, as elements that may become part of the infrastructure for autonomous economic activity. The fact that a transaction is initiated by software does not remove compliance obligations. It may require more detailed controls around identity, permissions, policy enforcement and records of agent activity.
On-chain transaction records can provide an auditable trail, but they do not automatically solve real-world identity, consumer protection or legal-liability questions. A wallet address is not necessarily a verified organization, and a smart contract can enforce encoded rules without determining whether those rules are appropriate for a particular user, jurisdiction or business relationship.
For institutions, the operational challenge will extend beyond choosing a payment network. It will involve controlling keys and signing authority, defining spending policies, monitoring agent behavior and separating autonomous execution from settlement finality. The more value an agent can access, the more important it becomes to use layered permissions, transaction limits and mechanisms for rapid suspension.
From conceptual convergence to practical infrastructure
BlackRock’s central thesis is that the spread of AI could create additional demand for digital assets, while digital assets could provide the machine-readable payment and settlement infrastructure required by an AI economy. AI interprets information and decides how to act; blockchains can represent assets and enforce programmable transaction rules. The two systems therefore have a potentially complementary role.
BlackRock says the practical outcome remains uncertain. Agentic systems must become reliable enough to perform complex tasks, payment protocols must achieve interoperability, digital assets designed to maintain relatively stable value must operate within applicable regulatory frameworks and tokenized assets must map clearly to enforceable rights in the real world. Each of these conditions is necessary for large-scale deployment.
The relevant question is consequently broader than whether one token or one protocol becomes popular. It is whether identity, authorization, payment, custody, settlement and regulatory controls can be integrated into a system that is both programmable and auditable. Machine-native money may reduce friction for some forms of automated commerce, but it also creates new requirements for governance and risk management.
BlackRock presents AI agents and digital assets designed to maintain relatively stable value as an infrastructure thesis rather than a finished commercial model. Their convergence could change how software acquires services and handles economic value, but scale will depend on technical reliability, legal clarity and robust controls developing together.
Source: link