Stablecoins as settlement infrastructure for the agent economy

An AI agent buying access to a data feed needs a way to pay the supplier. A stablecoin can make that exchange programmable across a compatible blockchain and service interface. Whether it improves the business process depends on the entire journey: how funds reach the wallet, what the supplier accepts, when the obligation is discharged and how both parties obtain money they can use.

Institutions should assess stablecoins as a candidate settlement asset for a defined transaction route. The strongest case is likely to involve a genuine requirement for on-chain delivery or an accepted, programmable payment interface. Agent capability alone does not establish that requirement. Existing cards and bank services can also support automated instructions, subject to provider capabilities and controls.

This distinction matters to the investment decision. Authorising software to spend and choosing the money it spends are separable choices. The companion agent-authority analysis addresses the first. This piece examines the second from a treasury and operating perspective.

What the new payment interfaces establish

In May 2025, Coinbase launched x402, an open protocol using HTTP's “402 Payment Required” response to request a stablecoin payment for a web resource. Its launch description shows a client requesting a resource, receiving payment details, submitting a signed payment payload and obtaining the resource after a facilitator verifies and settles the payment.

That mechanism is relevant to a proposed agent buying individual API calls without a separate subscription arrangement. It gives the application a structured way to discover a price and request payment. The launch article's examples of agents purchasing compute, data and services are proposed uses. They are not evidence of widespread institutional adoption, audited transaction economics or demand for each example.

Other designs are explicitly open to multiple rails. Google's September 2025 AP2 announcement describes cards, stablecoins and real-time bank transfers within a payment-agnostic approach to agent authority. DBS and Visa's February 2026 pilot reported real-world agent-initiated food and beverage purchases using cards. A stablecoin-only strategy would therefore need a reason beyond the assertion that agents cannot use conventional payments.

Singapore's BLOOM initiative, announced in October 2025, also considers tokenised bank liabilities and well-regulated stablecoins. Its initial areas include distribution and clearing, programmable compliance and exploration of agentic payments within predefined limits. The announcement sets out collaboration and trial objectives. It does not establish that every contemplated route is commercially available.

Receiving a token changes the exposure

This analysis concerns issuer-backed, fiat-referenced stablecoins. Algorithmic and crypto-collateralised arrangements require a different risk assessment. Even within this narrower category, tokens differ in reserve arrangements, holders' rights, redemption access, network support and issuer controls.

When a supplier receives a stablecoin, it receives the token and associated rights under that arrangement. It does not necessarily receive a bank deposit in its operating account. If it sells the token through an intermediary, the conversion price and availability matter. If it redeems with the issuer, eligibility, processing times and payment arrangements matter. A fast token transfer can coexist with a slower route to usable bank funds.

The BIS Annual Economic Report 2025 provides a consequential counterargument to treating stablecoins as universal money. It assesses them against singleness, elasticity and integrity: acceptance at par within the monetary system, flexible provision of settlement liquidity and safeguards against illicit activity. The BIS argues that stablecoins fall short as the monetary system's mainstay, including because fully backed issuance requires upfront funding.

That is a system-level policy assessment. It does not establish that every regulated stablecoin use case is uneconomic. It does require a treasury sponsor to address issuer exposure, prefunding and financial-crime controls rather than infer monetary equivalence from a one-dollar target price.

Compare a route rather than an asset label

The following comparison identifies questions to resolve with providers. It does not rank unspecified products; capability, access and contractual rights vary.

Existing regional infrastructure deserves a place in the comparison. MAS and the Bank of Thailand launched the PayNow–PromptPay linkage in April 2021. That announcement concerned participating banks and a bounded retail transfer service. It demonstrates a conventional route for faster regional transfers; it does not prove that the service offers corporate agent APIs or handles every wholesale requirement.

For a Singapore business paying a Thai supplier, compare the provider's actual route, eligible parties, currency conversion, limits and recipient experience. Repeat the exercise for another country. “ASEAN cross-border payments” describes a region, not one interchangeable financial service or regulatory permission.

Count prefunding and conversion in the economics

Suppose, illustratively, an agent makes many small purchases from approved data suppliers. The blockchain fee per purchase may be low. The organisation may still need to fund a wallet, obtain the token, maintain an operating balance, pay a facilitator and reconcile thousands of items. The supplier may need to convert the proceeds to its local operating currency.

Use a full-cost model: acquisition and FX cost, payment and service fees, custody and control costs, the opportunity cost of prefunding, reconciliation and exception labour, and the cost of obtaining usable funds at the destination. Avoid comparing a token transfer fee with an all-inclusive bank quotation. Both alternatives need the same starting point and end condition.

Small transactions may benefit from a standing prepaid account or aggregated invoice instead. Paying each call separately adds transaction evidence and cost that another model might avoid. Conversely, a pay-per-use interface may reduce unused subscriptions. Measure the real workload and compare these commercial arrangements before choosing the rail.

Likewise, a service that moves value outside bank operating hours may still rely on bank windows for initial funding or redemption. Establish the weekend and stress scenario, including a larger-than-expected withdrawal, before valuing continuous availability. Define how much liquidity must remain accessible through another route.

Specify what can fail after settlement

A confirmed token transfer does not prove that the purchased data arrived, was licensed correctly or met the agreed quality. Connect the invoice or service request to payment and fulfilment records. Determine whether the commercial contract treats token receipt, a number of network confirmations or later conversion as discharge of the obligation.

The recovery plan should cover a wrong recipient, an unavailable facilitator, suspended withdrawals, a token trading below its reference value, a network interruption and compromised signing authority. A cross-network transfer may introduce additional bridge or conversion dependencies. Record them if the actual design uses them; do not assume a single-chain transaction inherits every cross-chain risk.

Custody changes the control task. An institution managing signing keys must protect and recover them. A third-party custodian adds its own access, service and insolvency questions. In either arrangement, define how approvals reach the signer, which recipients are permitted and how new instructions can be stopped. Do not assume the transaction carries a card-like dispute process. Establish the actual issuer, intermediary and commercial remedies, including whether recovery is available after a completed transfer.

Read Singapore's policy in the right tense

MAS finalised its stablecoin policy framework in August 2023, covering single-currency stablecoins issued in Singapore and pegged to the Singapore dollar or a G10 currency. The announcement set out requirements for reserves, capital, disclosures and redemption at par within five business days. Those features describe a regulatory design, including eligibility for a specific “MAS-regulated stablecoin” label.

On 1 September 2026, MAS opened consultation on legislative amendments to implement that framework. The consultation also proposes approaches to multi-jurisdictional issuance, recognition of some foreign-issued stablecoins and additional safeguards. It closes on 16 October 2026. At this article's research cutoff, these amendments and additional positions are proposals under consultation. The 2023 announcement should not be used to describe every Singapore stablecoin business as licensed under a fully implemented framework.

Before approving a route, verify the specific issuer, service provider, permissions and token status. A payment-service licence, a commercial claim of regulatory alignment and eligibility for a stablecoin label are separate statements. For a regional route, confirm the counterpart jurisdiction's treatment as well.

Make the settlement choice an operating decision

Ask treasury and operations to produce one route record: payer and recipient entities; currencies; token and network if applicable; providers; funding source; maximum exposed balance; completion condition; conversion/redemption path; costs; failure owners; and legal or regulatory questions requiring resolution.

Then trace a normal transaction and an adverse one from bank funds to supplier fulfilment and the books. Use the digital-asset exception tabletop to structure the discussion, adapting its fictional lifecycle to the actual route. Choose the stablecoin where this comparison demonstrates a useful capability and an acceptable exposure. Keep the alternative payment path available where the proposal's economics or recovery depends on conditions the institution cannot yet demonstrate.