Anna Spysz uses a personal search for new studio headphones to explain the gap between an AI agent that recommends products and one that can complete a purchaseAgentic commerce lets AI agents discover, recommend, and purchase goods or services within delegated financial controls.. A merchant must expose structured product attributes, prices and policiesA commerce product feed is a structured, regularly updated dataset that supplies product details such as identity, price and availability to search or shopping systems. so the agent can compare options without guessing from an unstructured webpage.
The demonstration also shows why autonomy needs accountability. When the shopping agent starts pushing an expensive product too aggressively, Spysz traces the behavior back to its instructions and argues that merchants should preserve the attributes and reasoning inputs used for a recommendation so customers can understand and challenge the result.
For checkout, the agent uses a shared payment token rather than receiving the customer's raw card numberPayment tokenization replaces raw payment credentials with a limited token that can be used under defined conditions.. The payment provider enforces expiry, amount and currency restrictionsAgent payment authorization defines and verifies which payments an AI agent may make, including limits on value, recipients, timing, purpose, and transaction type., while the agent must ask for confirmation before placing the orderHuman-in-the-loop command approval requires a person to authorize a consequential AI-proposed action before the connected system executes it.. This separates an agent's purchasing workflow from the credentials and guardrails that keep the transaction bounded. The closing Stripe resource promotion is omitted.
Watch on YouTube
