Blockchain for AI Agents: How Agentic Commerce Works on Sui
Sui outlines how its blockchain infrastructure supports AI agents performing autonomous economic tasks, emphasizing atomic transactions, speed, and programmable spending limits for safe, scalable agentic commerce.
Automated traffic now exceeds human traffic online, and AI agents are poised to similarly dominate economic activity by executing payments, trades, and settlements on behalf of users. Traditional payment systems, designed for sequential human approvals, struggle to keep pace with agents that make decisions in under a second. This mismatch creates delays, fragility, and operational risks when agents attempt multi-step workflows like moving cash into yield-bearing stablecoins.
Sui addresses these challenges with Payment Intents, which bundle multiple actions—such as querying rates, swapping assets, and depositing funds—into a single atomic transaction. This ensures that either all steps complete successfully or none do, eliminating partial executions and the need for manual intervention. The approach contrasts with Web2 systems and most blockchains, which cannot group unrelated actions into a single, verifiable settlement.
Sui’s design supports high-throughput agentic workflows through programmable transaction blocks capable of executing up to 1,024 operations in parallel, while its object model prevents collisions between independent actions. During a public experiment in July 2026, AI agents processed over 6 million transactions per second using off-chain channels that settled to mainnet, demonstrating scalability without overwhelming the base layer.
For machine-driven payments, Sui offers gasless stablecoin transfers and is testing confidential transfers that hide transaction amounts while preserving sender-recipient visibility. Typical transactions finalize in about 390 milliseconds at under $0.02, making sub-second, sub-cent payments feasible for agents executing thousands of microtransactions on behalf of users.