The era where humans served as the sole final arbiters and executors of commercial transactions is drawing to a close. At the heart of this tectonic shift lies American Express (Amex), which is developing a sophisticated infrastructure for what is known as "agentic commerce." This is an ecosystem where autonomous AI agents do not merely suggest products but are authorized to negotiate, decide, and ultimately pay on behalf of the cardholder.
Amex’s approach is built on two pillars: "intent contracts" and "single-use tokens." While the idea of automated purchasing is not new, Amex’s technical implementation attempts to solve the biggest problem of AI in financial services: trust and control. However, the current implementation within a "walled garden" raises questions about interoperability and the transparency of decisions made inside the AI’s "black box."
The Anatomy of Intent: Contracts and Constraints
In traditional e-commerce, a user selects a product and clicks the "buy" button. In agentic commerce, the user defines an intent, such as "find me the best flights to London under 500 euros departing this Friday." This intent is converted into a digital contract. The intent contract acts as a set of parameters within which the AI agent is allowed to act. If the agent finds an option that violates even one term (e.g., a cost of 501 euros), the transaction is automatically blocked at the network level.
This structure allows Amex to shift the burden of authorization from the moment of payment to the moment the intent is defined. Single-use tokens further enhance this security. Instead of the AI agent having access to the full credit card number, it receives an encrypted key valid only for that specific transaction and amount. In this way, even if the agent is compromised by malicious software, the financial damage is kept to a minimum.
The Black Box Problem and Auditability
Despite the technical prowess, Amex faces a significant challenge: justifying AI decisions. In the banking sector, auditability is non-negotiable. When a human makes a purchase, accountability is clear. However, when an AI agent selects one vendor over another, the decision-making process often remains opaque. This "black box" makes it difficult for regulators to ensure there is no bias or unfair competition.
Amex is attempting to address this issue by participating in international protocols and standards for agentic commerce. However, for now, the system operates primarily within its own payment network. This means an Amex AI agent can easily communicate with merchants already integrated into the company’s ecosystem but struggles to function in an open, global web where merchants use different standards.
Economic Implications and the New Consumer Psychology
The transition to this model will radically change marketing. Companies will no longer target human consumers with flashy advertisements but will instead target AI agent algorithms with optimized data and APIs. The "attention economy" will give way to the "efficiency economy." For Amex, this move is strategic: if it can become the preferred network for AI transactions, it will secure its role in a future where humans delegate their daily purchases to digital assistants.
In conclusion, American Express’s initiative represents a bold step toward the full automation of economic activity. Its success will depend on whether it can balance absolute security with the need for an open, transparent system that wins the trust of both users and regulators. The stakes are high, as agentic commerce is not just about payments; it is about redefining the very concept of commercial agency.
