Stablecoins Seen at Core of AI-to-AI Commerce at EastPoint: Seoul 2026

Micropayments for APIs and data set to surge

Card and blockchain payment rails could converge

Stablecoins could emerge as a core payment tool as “agentic commerce” spreads, with AI agents increasingly finding and paying for goods and services on behalf of people, panelists said. Handling repeated micropayments in short bursts will require payment rails that are programmable and can settle around the clock.

A panel at EastPoint: Seoul 2026 on Sept. 28 at the Westin Josun Seoul Parnas in Seoul’s Samseong-dong district discussed payment infrastructure for the AI agent economy. The session was moderated by Professor Lee Jong-sub and featured Joe Hwang, regional lead at Google Cloud; Cindy Shi, chief marketing officer at Kite AI; Jensen Tung, co-founder of Virtuals Protocol; and Dennis Won, chief technology officer at Newton Labs.

Hwang said AI agents could eventually develop to the point of fully replacing human judgment in transactions. He added that the shift would likely be gradual rather than immediate, moving from a structure in which humans make decisions to one in which agents transact autonomously.

“One of the main new areas to emerge as AI agents spread will be micropayments,” Hwang said. “A new economy could form in which countless small transactions take place in a short period, such as API calls or database searches.”

Panelists also said different payment methods could be used for consumer purchases and transactions between AI agents. Shi said end users in e-commerce will still feel more comfortable paying with cards and conventional currencies. By contrast, stablecoins are better suited to areas where machines make frequent micropayments, such as API fees and data purchases.

She added that card networks and stablecoin rails are likely to converge over time. Consumers may continue using familiar payment methods at the front end, while settlement between agents in the background takes place through blockchain networks and stablecoins. Use cases remain separate for now, but stablecoin adoption could widen as agent-to-agent transactions increase and demand for faster settlement grows.

Tung said AI agents will need more than basic payment rails to carry out complex commerce. Systems must also be able to automatically enforce transaction terms. “Payments and commerce should be viewed separately,” he said. “Commerce includes complex issues such as contract terms, whether services were delivered and how disputes are handled.”

He said machine-to-machine commerce could be handled more efficiently if two AI agents agree on transaction terms through smart contracts and use them to escrow and settle funds. In that setup, smart contracts could automate verification of terms and payment while reducing the risk of disputes between agents.

The panel also said agents will need autonomy, but with tightly defined limits on their authority. Shi compared assigning tasks to an AI agent to handing a child a credit card. Rather than simply passing over a card, agents should receive different permissions depending on the task, country, transaction type and spending cap.

“If agents are going to process transactions autonomously in milliseconds, there needs to be proper delegation of authority,” Shi said. “Kite AI applies a structure that sets permissions separately for humans, agents and individual task sessions.”

Won said blockchain has strengths in fast settlement, but still lacks sufficient tools to approve and control transactions made by agents. “Stablecoins are money that operates through software, but to use them safely, you need an approval system that sets transaction limits and conditions,” he said. “You also need to be able to verify an agent’s identity and whether it complies with regulations.”

Responsibility was also identified as an unresolved issue. It remains unclear whether liability should fall on the user, the AI model developer or the platform when an AI agent makes a transaction outside the user’s instructions. Won said an agent’s identity is the starting point for determining responsibility when problems arise. Standards are needed not only for know-your-customer checks on humans, but also for verifying the agents themselves.

Hwang said existing financial accountability frameworks could also be applied to agentic commerce. If there is a problem with the delivery of goods or services, responsibility could lie with the seller. If a problem occurs within the scope set by the user, the user could bear responsibility.

“Google’s role is to provide the infrastructure and tools to develop and deploy AI agents,” Hwang said. “Because AI models can produce incorrect results, deterministic controls such as identity checks, security safeguards and spending limits need to be built alongside them.” He added that if an agent acts beyond the authority it was given, responsibility on the part of the platform or service provider should be examined.

Kim Soo-hyun, Hankyung reporter / Hwang Doo-hyun, Bloomingbit reporter ksoohyun@hankyung.com