On September 8, Meta introduced Muse, a complimentary personal AI assistant capable of navigating browsers, completing digital paperwork, negotiating prices, and executing transactions on behalf of its user. Twelve days later, individuals deploying it to shop on Amazon encountered a blocking notification stating that continued access by an unauthorized AI agent violates Amazon's Conditions of Use. Third-party shopping agents, an Amazon spokesperson asserted, should operate transparently and respect service provider decisions regarding participation. Meanwhile, Amazon's proprietary shopping assistant continues to purchase goods from alternative retail sites on behalf of its customers.

Prior analyses have suggested that AI agents threaten to divide numerous markets strictly down the middle between the lowest-priced option and the ideal fit. The pressing issue revolves around who ultimately secures ownership of the customer, and the Amazon dispute paints this as a winner-take-all battleground. However, investigative reporting reveals a more complex reality: the customer relationship is fracturing into distinct pieces claimed by different entities. This leaves ample strategic ground for enterprises that will never develop their own proprietary agent, including the vast majority of B2B suppliers.

Discovery Separates From the Point of Sale

The gravest strategic danger does not stem from surrendering the checkout screen. According to Leonid Myroniuk, a former senior product manager at Visa, the paramount risk lies in losing the critical moment when a buyer decides whom to place their trust in.

OpenAI signaled alignment with this view in March by pulling back from native checkout capabilities inside ChatGPT. The company admitted that its initial Instant Checkout iteration lacked the desired flexibility, prompting them to let merchants manage their own checkout processes while focusing squarely on product discovery. By August, this strategic pivot helped propel their emerging advertising business to an annualized revenue run rate of $1 billion.

For years, travelers have naturally decoupled these two phases, notes Ankur Malhotra, a former agentic AI product leader at Expedia. The digital environment trusted for exploration and research is not necessarily the one trusted for the final booking. honeymoon planners often compile options in a collaborative spreadsheet before booking through a specialized travel platform. Corporate procurement frequently follows a parallel trajectory, with engineers compiling shortlists while the purchasing division executes the transaction.

The Four Distinct Pieces of the Customer Relationship

Modern business relationships comprise discrete operational components, and agentic AI is empowering diverse contenders to compete for each one:

  • Discovery: The phase where consumer needs are articulated and options are filtered. AI platforms hold a dominant position here, monetizing through paid placements, transaction fees, or direct user subscriptions.
  • Authority to Spend: Encompassing the credentials, thresholds, and permissions that enable an agent to disburse funds, whether utilizing a consumer credit card or adhering to corporate procurement guidelines. Financial institutions and card networks are actively constructing this layer, exemplified by Mastercard introducing its Verifiable Intent standard alongside Google to maintain a tamper-resistant audit trail of user authorizations.
  • Recourse: Determining accountability when an AI agent purchases the incorrect item. This landscape remains deeply ambiguous; during a September address regarding agentic commerce, Federal Reserve Governor Christopher Waller raised the fundamental question of who bears financial responsibility when an AI makes a flawed purchase.
  • Fulfillment and Service: Covering post-sale execution and record-keeping. Traditional merchants are fiercely guarding this territory. When Google integrated hotel reservations into its AI Mode last August, it preserved the hotel or booking platform as the official merchant of record responsible for managing customer service.

Traditional banking institutions hold a natural structural advantage regarding the second and third components, given their existing control over user credentials and payment dispute mechanisms. Javier Rodríguez Soler, global head of corporate and investment banking at Madrid-headquartered banking giant BBVA, noted that consumers do not choose a dining establishment based on their bank. People buy the meal and subsequently utilize the credit card, he observed.

Nonetheless, Rodríguez Soler views agentic commerce as a monumental commercial opportunity. Once consumers begin transacting through fully autonomous agents, financial institutions aim to position themselves squarely as the indispensable middleman. Demonstrating this intent in July, BBVA executed its inaugural AI-agent-initiated transaction in collaboration with Visa, leveraging biometric passkeys to authenticate the cardholder securely.

Context Alone May Not Secure Customer Loyalty

Initial assumptions often lean toward the premise that whichever company's AI agent possesses the deepest contextual understanding of a user will successfully lock that customer into its ecosystem. Technology analyst Ben Thompson champions this perspective, predicting that most individuals and businesses will rely on a single agent rather than a fragmented array.

Conversely, Rodríguez Soler urges caution against absolute certainty, drawing from his personal habit of running Claude, ChatGPT, and Rogo—a specialized tool tailored for investment banking—simultaneously. The specific operational context Claude retains about him remains entirely invisible to ChatGPT and vice versa, a phenomenon he describes as lacking true stickiness. Users may deliberately choose to compartmentalize the intelligence these systems accumulate about them.

Furthermore, technology platforms are actively streamlining context migration; Google Gemini now facilitates the import of memories and historical chat logs from competitive assistants. Conversely, spending authority, recourse, and fulfillment operations remain considerably more difficult to transfer, meaning customer trust will likely anchor to whichever entity controls those foundational pillars. A 16-country Accenture survey underscored this sentiment, revealing that consumers exhibited more than twice the level of trust toward agents deployed directly by established brands or retailers compared to those offered by pure-play AI firms.

Writing in the Harvard Business Review, Sangeet Paul Choudary contends that competitive advantage increasingly favors organizations capable of supplying actionable data to help agents optimize their decision-making. Entities that manage fulfillment and service possess this precise operational visibility, tracking post-transaction realities that remain opaque to a standalone AI agent.

For most enterprises, the operational imperative centers on identifying which specific components of the relationship to secure. While few B2B suppliers will ever control product discovery, they can optimize their digital architectures to ensure agents can easily locate and trust them. The majority can successfully anchor fulfillment and service operations, converting performance metrics—such as delivery punctuality or system uptime—into verifiable evidence readable by machine agents.

Some firms can strategically absorb recourse liabilities by guaranteeing operational outcomes, thereby minimizing the risk calculations a buyer's agent must process. Myroniuk, who predicts B2B sectors may spearhead agentic commerce adoption, emphasizes that machine-driven transactions must achieve absolute traceability and auditability from the initial prompt to the final outcome.

Amazon’s decision to block Muse may well represent a calculated defense of the assets it already commands, ranging from user credentials and order histories to physical fulfillment warehouses and return desks. While few enterprises command Amazon's market leverage, any supplier can rigorously evaluate which operational components should remain anchored to its own infrastructure if an AI agent were to execute every incoming order, and fortify those strategic beachheads immediately.