The Deterministic Gap: Why Agentic Commerce is Stalling at Checkout

The tech and financial sectors are currently captivated by the prospect of "agentic commerce," a vision where automated software handles the discovery, negotiation, and purchase of goods on behalf of consumers. As The Register first reported, this trend has surged since early 2025, fueled by staggering projections. Mastercard recently published a report noting that ICSC and McKinsey project US consumer retail agentic commerce could reach $1 trillion in revenue by 2030, while Gartner expects AI agents to intermediate over $15 trillion in B2B spending by 2028.

**Opinion:** While these figures are designed to thrill investors, they ignore the operational reality of the checkout page. We are attempting to plug probabilistic models—systems based on likelihoods and patterns—into a retail environment that requires absolute, deterministic precision. In commerce, "close enough" is a failed transaction.

As Lindsay Walker, a product manager at Hedera AI Studio, explained during The AI Conference in San Francisco, the current infrastructure of e-commerce is built for humans, not AI. The core friction lies in the delegation of authority. While AI is viable for assisting with product selection, the process breaks down when users delegate the actual decision-making to an agent. Walker argues that because money and value are too critical, transactions cannot be run by probabilistic models; instead, they require "deterministic gates" that cannot be bypassed or left open to interpretation.

This friction is already manifesting as a wall of resistance from the platforms themselves. The Register reports that Amazon has blocked Meta's Muse bot, citing terms of use restrictions, and previously blocked Perplexity's AI bot, leading to a legal battle involving a preliminary injunction.

Beyond the corporate gatekeeping, there are basic functional failures. Walker noted that despite the emergence of so-called "agentic wallets" or "agentic cards," these tools are often non-functional. In one example shared by Walker, an AI agent was unable to complete a checkout because it could not log into an account to access a shipping address, nor could it interact with the specific DOM element on the webpage required to enter a credit card number.

There is also the issue of identity and trust. In traditional commerce, the "card present" standard helps merchants manage liability. With AI agents, cards are never present, necessitating a new mechanism to tether a human's identity to their agent. Furthermore, there is a lack of transparency regarding how these agents make decisions or whether vendors influence those choices—a concern compounded by the fact that US lawmakers are already scrutinizing AI-driven personalized pricing.

Ultimately, the industry is overestimating the demand for this automation. Low-cost items are already easy to purchase with a few clicks, and high-cost items require a level of consideration that most consumers are unwilling to delegate to a bot. As Walker suggests, while agentic commerce may become a reality within five years, it requires a total overhaul of how intent and identity are signaled in the digital marketplace.