What happens when money no longer moves between people at all?
The next wave of demand for crypto-native payments and settlement infrastructure does not come from retail users or even institutions. It comes from AI agents that can initiate, route, and execute transactions on their own. That shift is pulling two of the sector’s most important crypto infrastructure players, Circle and Coinbase, toward the same conclusion. Both now believe that AI agents will become economic actors in their own right.
They are building different layers of the same emerging stack. One is building programmable digital money for AI agents. The other is building the wallets and transaction rails that those agents will use.
Circle is creating the underlying infrastructure for AI agents. It is doing this through its Agent Stack, a set of infrastructure tools designed for machine-native commerce. It includes programmable wallets, micropayment rails, and transaction capabilities that allow value transfers at extremely small increments, down to fractions of a cent.
The design reflects a simple constraint. AI agents will not behave like humans. They will not bundle payments, wait for approval cycles, or operate within predictable billing rhythms. Instead, they will generate continuous streams of low-value interactions across APIs, data services, and automated systems.
Traditional payment rails struggle with that model because they are built for human behavior.
Circle is building for a different cadence entirely. The firm is pushing value transfer closer to computation speed. In this model, USDC stops behaving like a digital dollar for people and starts functioning as a unit of value that software systems can execute against in real time, embedded directly into workflows rather than layered on top of them.
The aim is to make digitized money readable and spendable by machines operating at their own speed.
Coinbase approaches the same transition from a different layer of the stack. Instead of redefining money, it focuses on redefining access to it.
The company already operates a widely used crypto wallet product, but it is now extending that infrastructure for AI agents that need to hold and spend value under tightly controlled conditions. These wallets allow agents to transact in stablecoins such as USDC, but only within predefined boundaries set by users or institutions. Those boundaries include session-level caps, per-transaction limits, and programmable rules that define what an agent can and cannot do. The intent is to formalize autonomy within software-defined guardrails.
Coinbase connects this architecture to broader machine-to-machine payment rails, including protocols such as x402, which enable automated payments for APIs, digital services, and content access.
The design avoids forcing AI agents into banking systems built around identity verification and human accountability. Instead, it treats them as constrained participants in financial systems, with permissions defined in code rather than identity documents.
That shift repositions the wallet from a human-centric financial tool into a programmable execution layer for autonomous transactions.
But it also introduces a harder question that the industry has not fully resolved. Financial systems assume accountability sits with a person or institution. When agents act independently, that assumption becomes harder to enforce in real time, especially as transaction volume and complexity scale.
This is already starting to surface in adjacent parts of financial services. American Express, for instance, has begun exploring this risk surface through its newly launched Agentic Commerce Experiences (ACE) Developer Kit. It introduces what the firm calls an “industry-first” commitment to protecting card members from charges linked to AI agent errors, provided the agent is registered and the user authenticates intent. This comes as agentic systems begin to participate in financial decision-making flows.
Coinbase is building on the belief that the next phase of crypto will be driven by infrastructure that allows software itself to participate directly in transactional activities.
Circle and Coinbase are building adjacent layers of a shared system. Between them sits a broader reconfiguration of financial infrastructure, where software is starting to initiate and execute transactions instead of human actors.
This is the early shape of what some in the industry are beginning to describe as an agentic economy, a system in which AI agents become active participants in financial flows rather than passive analytical tools sitting on top of them.
Stablecoins sit at the center of this shift because they offer what traditional rails struggle to provide at scale: programmable, always-on settlement that can operate without human timing or institutional settlement windows.
But the implications extend beyond payments. These moves point to a financial system where interaction, execution, and settlement collapse into a single layer of software logic. The core design question then becomes how much of finance should be designed for non-human participants from the outset.