Insight
November 17, 2026
What Is Agentic Commerce and Why Does It Matter?
Executive Summary
- Agentic commerce, in which artificial intelligence agents independently make purchasing decisions and execute transactions on behalf of consumers and businesses, could fundamentally change the relationship between consumers, financial institutions, and the payment infrastructure underpinning the economy.
- The emergence of autonomous transactions presents significant implications for banks and payment networks, potentially reshaping deposit competition, payment processing, and the role of stablecoins and other digital assets in facilitating machine-to-machine payments.
- Although agentic commerce remains in its early stages, its implications for mortgage lending, financial intermediation, and the competitive relationship between banks and nonbank technology providers warrant attention from policymakers seeking to preserve competition without unnecessarily extending the regulatory perimeter around emerging technologies.
Introduction
Artificial intelligence (AI) is becoming capable of something more consequential than answering questions or generating a reasonably convincing email: It can act. Agentic commerce represents a potentially significant change in how commercial activity is conducted. An agent (authorized by a consumer) might open an account, move money, purchase a service, or arrange financing without requiring human intervention at every step.
Much of the early discussion has focused on online shopping – asking ChatGPT to buy more milk. The more consequential effects are likely to appear in financial services, where consumers and businesses routinely face repetitive decisions involving complex pricing, competing providers, and high administrative costs. An agent capable of comparing deposit rates, refinancing a mortgage, or executing routine payments could materially change how customers interact with banks and other providers. At the same time, payment infrastructure designed for autonomous transactions raises the possibility that stablecoins acquire a larger role in commercial activity – high speed transactions benefitting from the instantaneous nature of transmission.
The more significant development is not that consumers may one day outsource grocery shopping to artificial intelligence, but that the companies controlling the interfaces through which financial decisions are made could gain substantial influence over the distribution of deposits, credit, and payment services. Agentic commerce may weaken the switching costs and customer inertia that have protected established institutions while inserting a new layer of technological intermediation between consumers and providers. These are familiar questions of competition, consumer protection, and the regulatory treatment of nonbank activity, set in a less familiar technological context.
What Is Agentic Commerce?
Agentic commerce refers to the use of AI agents to conduct commercial activities on behalf of users, including research, comparison, selection, and execution of transactions. Its defining feature is that the agent moves beyond simply a recommendation and is authorized to take action toward a commercial objective subject to the user’s instructions and permissions. Here the agent may execute some or all of the transaction independently – for example, to identify the highest-yielding insured savings account and initiate a transfer, or to compare refinance offers and submit an application once projected savings exceed a set threshold.
Autonomous transactions require more than language models. The agent must interact with institutions and merchants, access account information, authenticate itself, demonstrate authority, and settle through an appropriate payment rail. Current implementations vary: Some retain a final consumer confirmation, while others operate within predefined permissions.
Several firms are building the necessary pieces. OpenAI’s Agentic Commerce Protocol and Google’s Universal Commerce Protocol aim to facilitate agent-merchant interactions. Visa’s Intelligent Commerce and Mastercard’s Agent Pay for Machines focus on secure initiation of transactions using authorized credentials. OpenAI’s current approach leaves merchants with their existing processors; Visa’s work centers on agent-specific credentials and controls that map transactions to consumer instructions. Mastercard’s initiative is designed for continuous, high-frequency machine-to-machine payments.
These are not theoretical. In October 2026, Visa and two business-to-business payment companies, BILL and Billtrust, announced a live pilot allowing authorized AI agents to initiate and complete approved business invoice payments through supplier portals, using agent authentication and tokenized credentials to automate steps that previously required manual log-in and submission. The technology is early, but it shows agentic commerce moving beyond consumer shopping into routine commercial activity.
Banking and the Future of Financial Intermediation
The most immediate implications concern the bank-customer relationship. Banks have long benefited from relatively durable relationships; consumers often keep deposits, cards, and other products at the same institution for years even when better terms are available. The friction of comparison and switching has provided a measure of protection. AI agents could reduce that friction substantially.
An agent monitoring deposit rates, fees, minimum balances, and insurance coverage could identify opportunities to move funds. The same logic applies to credit cards, borrowing costs, and refinancing. What now requires research and several account-opening processes could become routine. Demand for financial products may become more price- and quality-sensitive, forcing institutions to compete more aggressively on rates, fees, and service rather than relying on inertia. Funding conditions could become more sensitive to rate changes and competing offers, raising the importance of deposit pricing and liquidity management.
A deeper question follows. If consumers increasingly discover and select financial products through a general-purpose AI platform, the platform – not the bank – may become the primary interface. Technology companies could gain considerable influence over funds distribution, particularly if agents favor certain providers on the basis of referral arrangements, proprietary data, or platform incentives. A platform compensated for directing customers to a particular savings account has incentives that may not align with the consumer’s interest in the best available terms.
Agentic commerce could therefore reduce dependence on traditional institutions while creating a new dependence on technology platforms. Banks would still accept deposits, extend credit, and manage liquidity, but they might increasingly rely on third-party platforms for customer acquisition and retention. Those platforms could become powerful intermediaries without performing traditional banking activities or facing bank-like prudential requirements. Banks will remain central to intermediation and settlement; the more immediate question is whether the customer-facing layer migrates toward technology firms.
Housing Finance and Mortgage Lending
Housing finance is another area in which informational and administrative burdens are high. Purchasing a home requires coordinating agents, lenders, insurers, title companies, and other parties while comparing products with different pricing and eligibility rules. An agent capable of coordinating elements of this process could reduce friction.
An agent could compare mortgage products across lenders, accounting for rates, APRs, fees, points, closing costs, and eligibility rather than simply the lowest advertised rate; organize documentation; track requirements; and communicate with lenders. Once originated, similar agents could monitor refinance opportunities or more favorable insurance and servicing arrangements. Lenders are already using AI for document processing, communications, valuation, and aspects of underwriting; agentic applications could extend coordination across stages of the process.
Lower administrative costs could improve market efficiency and make it easier for smaller lenders to compete. The same logic applies to ancillary services – title insurance, property insurance, escrow, and closing payments – where consumers often have limited visibility into pricing. There are limits, however. Housing transactions involve large sums, individualized underwriting, and legal obligations that cannot be fully delegated without safeguards. An agent that misreads a borrower’s circumstances or relies on discriminatory information raises fair-lending concerns, as noted by the Government Accountability Office’s 2025 report on property technology for homebuying.
The banking discussion centers on ongoing customer relationships and the potential for continuous, low-friction switching of deposits and other products, which raises questions of funding stability, platform control of distribution, and the migration of the customer-facing layer away from banks. Housing finance, by contrast, involves episodic, high-stakes, multi-party transactions with substantial documentation, legal obligations, and individualized underwriting, so the primary implications concern efficiency gains in product comparison and process coordination rather than continuous deposit competition. Housing also carries more specific fair-lending and consumer-protection risks – particularly around automated recommendations that could affect access to credit or housing.
Crypto and the Future of Payments
Perhaps the most interesting implications concern the role of stablecoins in autonomous transactions. Traditional payment systems were designed around human-initiated purchases. AI agents allow for different requirements: high volumes of small transactions, continuous operation, and interaction with other agents without human involvement at every step.
Stablecoins are well suited to some of these use cases – or even potentially necessary. Dollar-denominated stablecoins can move value between digital wallets and across networks, potentially allowing agents to pay without conventional card checkout. Programmability and the possibility of low marginal costs may make them attractive for machine-to-machine payments, especially where individual transactions are too small to justify traditional arrangements. An agent purchasing data, compute, or software services could settle small amounts automatically as services are consumed; a business agent might purchase capacity or pay other agents for discrete tasks.
The commercial case is not simply that stablecoins are digital. Certain blockchain systems can settle around the clock, across borders, and without routing every transaction through conventional card or account-based arrangements. These advantages depend on network fees, congestion, custody, conversion costs, and reliable on- and off-ramps; stablecoins are not automatically cheaper. Visa is working on agent-specific credentials and stablecoin settlement; Mastercard’s Agent Pay for Machines is designed for rapid programmatic payments with settlement across cards, accounts, and stablecoins. Crypto-native efforts are attempting to establish similar rails.
Stablecoins could serve as an alternative settlement mechanism for certain flows, reducing reliance on traditional intermediaries. Established networks are unlikely to cede the ground; the outcome is more likely coexistence and competition than wholesale replacement. For banking, stablecoins allow consumers and businesses to hold and transfer dollar value outside traditional deposit accounts, potentially weakening the link between payment services and bank funding. The Federal Reserve’s December 2025 analysis and the February 2026 New York Fed staff report examine how stablecoin adoption could displace deposits, raise funding costs, and affect credit provision. Agentic commerce could make these questions more immediate.
Stablecoins do not remove banks from the payment chain. Issuers still rely on institutions for reserve management, custody, and conversion. The potential shift is in the nature of banks’ participation: from maintaining the direct customer relationship to supplying underlying infrastructure for transactions conducted on other networks. Agentic commerce may supply a concrete commercial application for stablecoins, particularly for high-frequency, low-value flows.
What Are the Risks?
The most immediate concern is the effect on bank funding and deposit stability. Deposits have historically been relatively sticky; consumers do not continuously monitor rates or move funds in response to small differences. Agentic systems could change that, allowing continuous comparison and relatively frictionless movement. Greater competition for deposits is desirable; less predictable funding and greater sensitivity to rate changes or perceptions of risk are the corresponding costs.
Banks losing deposits to better rates is not itself a problem. The concern arises when movements become unusually sensitive to common information or market events, especially in stress. If agents independently identify the same warning signals and recommend or execute withdrawals, outflows could accelerate in ways that are harder to anticipate or manage. This is not a prediction of crisis; it is a recognition that the technology could make it easier for customers to leave at the moment funding is most vulnerable.
A related channel is the migration of transaction balances toward stablecoins. Agents optimizing for speed and cost may favor digital assets for cross-border or continuous machine-to-machine payments. If stablecoins become a widely used alternative to transaction deposits, banks could face additional funding competition and effects on credit capacity. Federal Reserve research already flags these issues; autonomous agents could accelerate adoption.
Moving money from one insured bank to another intensifies competition among banks. Moving money from a deposit account into a stablecoin can change the composition of bank funding and the relationship between payments and credit intermediation. There is also the risk of correlated decision-making. Even independent agents may rely on similar models, data, and optimization criteria, producing similar responses to the same signals. In ordinary conditions this may improve efficiency; in less ordinary conditions it could encourage herding or create operational bottlenecks. Concentration of agentic services among a small number of technology providers raises additional questions about operational resilience and third-party dependence.
These risks should not be overstated. Common models do not produce identical decisions, and the possibility of correlation is not evidence of inevitable instability. Still, the speed and scale at which autonomous systems can act make operational resilience, third-party dependencies, and the allocation of responsibility for agent-initiated transactions more relevant for supervisors. Technical capacity to initiate a payment is not the same as authority to do so; existing rules on authentication, fraud prevention, and consumer protection will need to operate clearly when software rather than a human initiates the transaction, including clarity on who bears the loss when an agent acts outside its instructions or a disputed payment arises. The policy task is therefore to determine whether existing safeguards around bank liquidity, payment-system resilience, third-party operational risk, and authorization and liability remain adequate. The answer is unlikely to be a new regulatory regime for AI agents themselves; it is more likely to involve ensuring that existing requirements continue to function as the technology changes how institutions interact with customers and markets.
None of these risks necessarily require a wholesale reconsideration of the existing framework. Banking, payments, consumer-protection, and fair-lending laws already govern many of the activities agents may perform. That a transaction is initiated by software does not, by itself, change the economic function or justify extending bank-style prudential requirements to every technology provider involved. Where genuine gaps appear, targeted clarification may be warranted. Premature regulation risks imposing costs on emerging providers, entrenching incumbents, and limiting the competition agentic commerce could introduce. Regulation should follow the activity and the risks it creates, not the novelty of the technology.
Conclusion
Agentic commerce could insert a new layer of intermediation into banking, housing, and payments, weakening the customer relationships and switching costs on which many established models depend. The potential benefits – more competitive products, lower transaction costs, and new payment options – are real. The same technology could make deposits more volatile, create new dependencies on technology platforms, and accelerate the migration of certain payments away from traditional banking infrastructure.
Policymakers should, however, resist the urge to invent a new regulatory regime simply because transactions are increasingly initiated by software. Existing rules should continue to apply according to the activities performed and the risks created, with targeted adjustments where gaps emerge. An agent that helps a consumer find a better deal on milk is a welcome development. Large numbers of agents deciding simultaneously to take their money and leave the U.S. banking system is a different matter.





