Agentic Payments
Abstract
The required financial infrastructure for an agentic future
Description
It’s increasingly consensus that the era of AI-powered agents is quickly approaching. If that’s true, are the systems we have today fit for purpose in that future state? Menial tasks like purchasing groceries or buying plane tickets are commonly cited use cases, and because these are automations of existing behaviors, it seems likely that our existing infrastructure can be bent to enable these new capabilities with little resistance.
But as purchasing behavior matures beyond the mundane, new controls will be required to provide the assurance and stability that the market will demand. Custodians and escrow systems are the obvious examples, particularly those that have been natively built for the blockchain. Multi-party computation and mulit-sig approval will likely extend into agent-first systems to ensure that human managers maintain control over significant funds. These systems are rigid and slow by design, which may prove fatal to the immense velocity of micro / atomic payments. At a minimum, KYC/AML architectures will need to be revamped because they can no longer rely on the traditional idea of an identity.
he most curious question is whether agents will cohabitate our internet or build their own entirely. The web is a human invention, built for our understanding of biologically based sentience. We’re a far cry from achieving eight billion internet users, but it’s very possible that we’ll see 80 billion agents. The internet, as it’s built today, will be crushed by the weight and volume of what’s coming. Does this mean we reinforce what we have or build an entirely separate agentnet? Is this agent-first web fully headless? Will we soon need human on-ramps?
Working Thoughts
Most of my conversations with financial operators tend to gravitate toward payment revocability rather than robust approval flows. For many, the immutability of blockchain solutions is a no-go. This suggests that either more centralized solutions will take hold, or stopgap products like agent-focused cyber insurance will be required.
The combination of near-frictionless atomic transactions and 24/7 autonomous execution would unlock entirely new markets that seem preposterous today. My personal white whale is autonomous vehicle traffic orchestration—self-driving cars that actively bid for passing and yielding. Time = money.
There’s an obvious challenge to deploying against this theme in its obviousness. Agent systems are peak tech zeitgeist right now—to the point where one must question if all of the alpha, at least the kind attainable by a fund of Fly’s size, has already dissipated. I’m sensitive to this critique, though I find it unconvincing. Despite the surge in venture interest, we’re still very early in deployment, and if agents represent a new platform, then we really can’t recognize the exponential curve from our current position in time.
It is true, however, that competition will be fierce in this arena. The requirements for founder domain pedigree, execution velocity, and learning speed are a level or two higher than normal. You also probably need to have a uniquely strong reality distortion field for resource magnetism, which tends to be unambiguous in my experience.