Summary
AI agents are moving from answering questions to taking actions and, potentially, spending money on a user’s behalf. The shift could reshape payment authorization, wallet controls and financial-service distribution, while stablecoins and blockchains emerge as possible rails for machine-driven digital commerce.
AI agents are becoming economic actors
The discussion around artificial intelligence agents is moving beyond chat interfaces and content generation. Earlier generations of consumer AI were primarily evaluated on how well they answered questions, summarized information or created text. The newer agent model is more operational: an agent can interpret a user’s objective, break it into steps, call software tools and, within defined limits, act on the user’s behalf.
That shift becomes consequential when an agent is able to spend money. An AI system that recommends a service is still an information layer. An AI system that selects a service, initiates a transaction and manages a recurring digital workflow becomes part of the financial infrastructure itself. It may influence which providers receive payments, which wallets hold funds and which networks settle transactions.
CoinDesk’s reporting on the subject highlighted remarks from ARK Invest CEO Cathie Wood, who suggested that investors may increasingly need to “follow the agents” rather than focus only on developers when assessing where technology is heading. Her comment was presented as a forward-looking observation, not as evidence that a particular agent payment system has already become established. Its significance lies in the question it raises: where will agents spend, and who will control the rails through which those expenditures occur?
Authorization becomes more complicated
Traditional payments are generally organized around a person, a business, an account and a specific transaction. The account holder provides authorization, a financial institution or payment provider applies its controls, and the merchant receives a settlement instruction. Agent-mediated payments introduce another layer between the user’s intention and the transaction.
A payment system will need to determine what an agent is allowed to do, for how long and under which conditions. Can it purchase only a defined category of digital services? Is there a spending limit? Can it pay a new counterparty without additional confirmation? Is it allowed to create a recurring obligation? What happens when the agent misunderstands the request or acts on inaccurate information?
These questions point toward more granular permission systems. An agent might be allowed to pay for a limited set of services but prevented from changing account settings. It might operate within a daily or task-specific threshold, with a human review required for transactions outside that scope. It might also need to use a designated wallet rather than access an organization’s primary treasury account. Such controls are design requirements under discussion, not established industry standards.
For institutional users, the distinction between operating permission and settlement authority is particularly important. Organizations commonly separate the people who initiate transactions from those who approve or reconcile them. If an agent is added to that process, a wallet or custody system would need to preserve those separations rather than treat the agent as a fully trusted account operator. Policy engines, transaction screening, key protection, multi-party approval and complete audit records could become as important as the underlying payment network.
Why digital settlement assets and blockchains are being considered
CoinDesk’s report cited industry views that digital settlement assets and blockchains could become payment rails for agents purchasing data, computing capacity and other digital services. The argument is that machine-driven commerce may benefit from programmable settlement and a digital unit that can be transferred across software environments. In theory, an agent could use a defined balance or permissioned wallet to pay for services without relying on a person to approve every individual transaction.
That possibility does not remove the practical risks. Blockchain transactions may involve private-key management, network fees, address errors, limited reversibility and questions about who controls the receiving account. Digital settlement assets also depend on the arrangements of their issuers, their redemption mechanisms, the transparency of relevant reserves and the rules of the jurisdictions in which they operate. A system that gives an agent automated access to such assets could amplify an error by repeating it quickly or across multiple transactions.
The important question is therefore not simply whether an agent can send a token. It is whether the surrounding system can constrain where funds may go, identify the counterparty, detect unusual behavior, stop execution and resolve an erroneous payment. For institutional use, a policy layer that limits assets, destinations, frequency and transaction size may be more consequential than the choice of a particular chain.
Open networks versus concentrated platforms
The development of agentic finance also raises a competition question. The debate described in the reporting is not only about whether agents will use blockchain-based payments. It is also about whether agent commerce will develop on open networks or become concentrated among a small number of banks, payment providers and technology platforms.
Open networks could offer interoperability among different agents, wallets and service providers. They may allow users to change providers without abandoning an entire ecosystem. At the same time, open systems can make identity standards, responsibility and cross-network risk controls harder to coordinate. Different providers may apply different rules to the same transaction, and it may be unclear which party is accountable when an automated payment fails.
Platform-centered systems may provide a more unified user experience. A single technology company could combine the agent, identity layer, payment account and customer-support channel. That arrangement may simplify permissions and dispute handling, but it could also increase dependence on one provider and concentrate behavioral, payment and transaction data. It may also give the platform significant influence over which merchants and financial products an agent can access.
For consumers, the central issue is not only convenience. They need to understand what the agent can do, what it cannot do and how to intervene. Clear consent screens, comprehensible transaction histories, timely alerts and effective cancellation mechanisms will be important if users are expected to delegate spending decisions to software.
Regulation and liability remain unresolved
Agent-driven payments create difficult questions for existing financial and technology rules. If an agent purchases the wrong service, responsibility could be disputed among the user, the model provider, the wallet operator, the payment institution and the merchant. If the system cannot explain why a transaction was approved, audit and consumer-protection processes become more difficult. Cross-border activity could add further uncertainty around identity checks, anti-money-laundering obligations, data protection and jurisdiction.
These are not solved market facts. They are infrastructure and governance issues that will have to be addressed before autonomous spending can become broadly dependable. A capable model needs to be paired with verifiable identity, explicit permissions, transaction records and a reliable way for a person or organization to suspend its actions. Without those safeguards, greater automation could also magnify fraud, account compromise, mistaken payments and compliance failures.
The most important change may therefore be a new definition of what needs to be controlled. Existing systems focus heavily on accounts, credentials and individual transactions. Agentic systems will also need controls over behavior: which tools an agent can call, what conditions must be met before a payment is made, how frequently it can act and when it must stop and request review.
What to watch next
Wood’s suggestion to “follow the agents” reflects a changing framework for analyzing financial technology. Market observers may increasingly look beyond model capability and user counts to ask where agents obtain data, how they pay for computing resources, which wallets or accounts they control and which providers become embedded in their workflows.
That does not mean digital settlement assets or blockchains will quickly replace existing payment systems. Adoption will depend on whether authorization can be made precise, whether institutional controls are strong enough, whether liability is clearly assigned and whether regulators permit the relevant activities in particular jurisdictions. The more likely near-term development is a gradual layering of agents onto existing payment and custody systems, followed by experiments with programmable settlement where the use case is sufficiently narrow.
For wallet, custody and payment providers, the competitive focus may consequently shift from access to a transaction rail toward control and verifiability. The systems most likely to support responsible automation will need to show not only that an agent can act, but also that its authority is limited, its decisions are recorded and its actions can be stopped when circumstances change.
Source: link