In the past, discussions about AI agents focused on whether they could "plan" or "reason." However, in the past year, the tone has clearly changed. The question now is: how can we enable agents to truly get things done on their own, rather than getting stuck waiting for confirmation at every step.

The banking system has already piloted AI agents in actual business, and regulatory authorities have begun to face the risks that "speed" and "autonomy" may bring. At the same time, payment institutions have long passed the experimental stage of treating stablecoins as serious cross-border settlement tools. When these two forces collide, the problems become evident—if agents are to really do their jobs, there must be a payment method that is neither awkward nor hinders efficiency, and can also be controlled.

Most projects follow the approach of "adding a layer of proxy capability" on top of existing payment systems. However, KITE's logic is the opposite. Its reasoning is straightforward: traditional payment stacks were never designed for proxy purposes. Rather than patching it up, it's better to build "proxy payments" from the ground up as infrastructure.

KITE has highlighted three core contradictions: speed must keep up with the machine's decision-making rhythm, safety should not make the client appear too reckless, and autonomy cannot be stifled by the approval process. These three points cannot be perfectly compatible; balance can only be sought amidst the conflicts—this is the real engineering challenge.

When talking about "faster payments," many people still think of the scenario where card transactions are a few seconds faster. But the agent has completely rewritten the arithmetic. They do not just occasionally transfer a sum of money, but rather interact continuously, in batches, and at very small units. This logic only holds if settlements are cheap enough and the timing is timely enough. KITE is directed towards this.