AI agents cannot pay for much yet; Visa and SFA paper maps the fix
Source: Fintech News SG
A StraitsX, Visa and SFA whitepaper argues agents need one common layer reaching cards, banks, country networks and stablecoins — plus fresh thinking on authority and liability — before autonomous buying scales.
Smart shopping agents can hunt down products and compare prices, but they still struggle to pay for much of anything. Closing that gap is the thrust of a September whitepaper from StraitsX, Visa and the Singapore Fintech Association, which contends the money-movement layer beneath autonomous software has to be rebuilt before agents handle serious volume.
Standardising on a single channel is the wrong answer, the authors argue. What agents actually need is a shared layer that routes them to whichever method fits — cards, direct transfers between banks, country-level networks or stablecoin settlement. Stablecoins earn special praise: their programmability and near-instant finality suit the torrent of tiny, high-frequency payments machines will make to each other.
Machine-to-machine spending gets particular attention. An agent buying an API call, a chunk of compute or a dataset mid-workflow needs that bill settled instantly and cheaply, and the paper reckons stablecoin rails are purpose-built for exactly that.
Two thornier problems follow. Accountability comes first: who authorised an agent, whose behalf it acts on, and who absorbs the loss when it strays past its mandate. Then there is the rulebook itself — identity, delegated authority and transaction-level controls all need to interoperate, and today's security and compliance regimes were never drafted with self-directed software in mind.
The stakes are regional. Deloitte's 2026 outlook has Asia-Pacific taking roughly two-thirds of fresh retail spending worldwide over the coming five years, backed by a consumer base exceeding 4.3 billion. Only 29% of the region's consumer businesses have adopted agentic AI so far, a figure Deloitte expects to reach 76% within two years. Store formats are already shifting — China and Singapore outlets are layering in electronic shelf labels and digital operations — while JD.com runs about 14,000 AI agents, among them a Demand Agent and Courier Assistant, reportedly carrying more than 18% of tasks across its businesses. Some forecasts put agent-handled purchases at a quarter of global e-commerce by 2030.
Why it matters for Singapore: The SFA helped write this blueprint, and Deloitte name-checks local stores in the push toward intelligence-laden retail, so the city-state is not a bystander here. Should interoperable, well-governed agent payments arrive early, Singapore's standing as a regional fintech and payments hub hardens; stall on fragmentation and that edge dulls. The demand for rules that keep pace with autonomous activity lands squarely on Singapore's regulators and industry groups.
