Safely manage your Zendesk from the AI assistant you already use, via the Deltastring MCP. Beacon configuration platform
← Back to news
ai

Inside AMEX’s agentic commerce stack: How intent contracts and single-use tokens enforce AI transactions

Amex is constructing an agentic commerce infrastructure that permits AI agents to execute transactions on behalf of customers through intent contracts and single-use tokens—cryptographic mechanisms designed to enforce authorisation boundaries and prevent token reuse. The system currently operates within Amex's closed payment network, meaning agents can only transact within predetermined parameters rather than across open marketplaces. This architectural choice reflects a deliberate trade-off: security and control are maximised, but the black-box nature of how agents interpret user intent creates an auditability problem that could undermine customer trust when disputes arise. For CX teams already managing payment-related inquiries, this raises an immediate question: if an agent executes a transaction that a customer disputes, how will your support team access the decision logic that led to that transaction, and what documentation will you need to defend Amex's position?

The broader implication is that agentic commerce is moving from theoretical to operational, but in siloed, enterprise-controlled environments rather than open ecosystems. Amex's approach mirrors what we're seeing elsewhere—agentic AI-powered self-service can reduce customer support load—yet the payment layer introduces compliance and liability dimensions that self-service chatbots don't face. For support leaders, this means preparing for a new category of escalation: agent-initiated transactions that require human review not because the agent failed, but because the customer needs to understand why the agent acted as it did. The real tension isn't whether these systems work—they do—but whether the governance frameworks CX teams currently use in Zendesk, Salesforce, or Freshdesk can accommodate audit trails for AI-driven financial decisions, or whether you'll need entirely new tooling to manage liability and customer confidence.