Agentic Commerce: Not Voice Shopping, But Unproven

The Structural Shift Beyond Voice Interfaces

The tech industry’s history with automated purchasing is littered with overpromises. In March 2018, OC&C Strategy Consultants forecast that voice shopping would surge from $2 billion to over $40 billion by 2022 across the U. S. and UK. That market never materialized. Prior to that failure, Facebook launched M, an assistant designed to book restaurants and chase refunds, which remained in private beta before shutting down in January 2018 after just two and a half years. This pattern raises a critical question: Is agentic commerce simply voice shopping rebranded with new terminology? The answer is no. Unlike previous waves, this technology may already be operating on your website without explicit merchant consent, introducing four unprecedented elements that distinguish it from earlier attempts at automation.

Agents Make Decisions; Voices Just Change Input

Voice shopping was fundamentally an interface evolution. Users still made the final decision, confirmed the transaction, and remained present throughout the process. Switching from typing to speaking altered the input method but not the fundamental structure of commerce. Agentic commerce claims to operate differently by executing purchases without human presence. The FIDO Alliance categorizes this as Human Not Present, treating it as a distinct transaction class separate from human-present interactions. Equating agentic commerce with voice shopping is a category error that ignores this structural shift.

This distinction is vital because agentic commerce introduces capabilities that voice interfaces never possessed. It removes the requirement for merchants to build specific integrations, establishes payment systems designed for absent buyers, engages formal standards bodies, and has prompted organized merchant resistance. These factors create a landscape where the technology operates independently of direct merchant approval, setting it apart from the friction-heavy adoption models of the past.

No Build Required for Merchant Participation

Every prior iteration of automated commerce required merchants to actively build something. Whether it was an Alexa skill, a Messenger bot, or a voice-optimized catalog, participation demanded effort. OpenAI’s Instant Checkout, for example, engaged roughly thirty Shopify merchants in six months before being retired, all of whom had built custom integrations for a surface that no longer exists.

Agentic commerce flips this model. On September 4, Shopify enabled UCP version 2026-08-25 across its storefronts, stating the change “doesn’t require any changes to your existing integrations.” Similarly, Mastercard’s Agent Pay offers a compatibility path called DTVC, allowing merchants already accepting Mastercard to process agent transactions without new work. Visa’s Intelligent Commerce states that agents pay through “guest checkout, key entered, web form or through an available merchant API,” effectively filling out forms a human would use. Through platform defaults, network compatibility paths, and automated form completion, merchants can become reachable without agreeing to anything new.

Payment Infrastructure Built for Absent Buyers

Voice commerce relied on saved cards, lacking specialized infrastructure for autonomous transactions. Agentic commerce introduces payment mechanisms specifically scoped for scenarios where the buyer is not present. Stripe’s Shared Payment Tokens are limited to one merchant, capped at a specific amount, and expire automatically. Google’s AP2 uses signed mandates to prove exactly what the user authorized. Additionally, x402 responds to unpaid requests with HTTP 402 and a payment header. These tools provide a level of control and verification that voice shopping never achieved, creating a foundation for secure, automated transactions.

Standards Bodies Drive Adoption, Not Just Tech Giants

Unlike voice shopping, which operated solely within Amazon’s skills API and Google’s actions, agentic commerce involves rigorous standardization processes. The W3C and GS1 hosted a two-day workshop in Zurich on September 8 and 9 to discuss these protocols. Major tech companies have contributed significantly to this ecosystem. Google and Mastercard donated AP2 and Verifiable Intent to the FIDO Alliance, while Google donated A2A to the Linux Foundation. This collaborative approach to standard-setting contrasts sharply with the closed-loop ecosystems of the voice era, suggesting a more robust and interoperable future for the technology.

Merchants Organize Against Unchecked Risk

The merchant community is actively pushing back against the risks associated with agentic commerce. At its Zurich session, the Merchant Advisory Group highlighted that without proper safeguards, merchants are “exposed to risks they did not cause and have limited ability to identify or manage.” Their proposal includes a liability waterfall that assigns responsibility to whoever controlled the failing function, along with demands that merchants retain the right to route their own payments. No equivalent position paper existed for voice shopping because merchants were never asked to assume such significant liabilities.

Google’s Agentic Checkout Remains Human-Present

Google’s agentic checkout is currently live on Search and in AI Mode with named merchants including Wayfair, Chewy, Quince, and select Shopify partners. The system monitors prices and executes purchases when they drop using Google Pay. However, Google emphasizes that it does not act autonomously in these initial deployments. As stated by the company:

> We’ll always ask for your permission first, and only buy after you’ve confirmed the purchase and shipping details.

Because the user confirms both the purchase and shipping details, this implementation remains human-present. By the distinctions outlined above, this constitutes an interface change rather than a structural one. Asking for permission is the correct design approach, and it aligns with current realities. The FIDO Alliance’s Human Not Present class is targeted for harmonization by the end of 2026, not active deployment today. Therefore, the trillion-dollar projections surrounding agentic commerce are based on a structural change that is not yet available to consumers.

Forecasts Mirror Past Overpromises

Current predictions for agentic commerce echo the inflated estimates of the voice shopping era. McKinsey projects $1 trillion in U. S. retail revenue orchestrated by agents by 2030. Meanwhile, Gartner predicts that 90% of B2B purchases will be handled by agents by 2028, intermediating $15 trillion. These figures are comparable in nature to OC&C’s $40 billion forecast. While there is no claim that these predictions are incorrect, history shows that similar forecasts about machines buying on behalf of humans were off by an order of magnitude. Those who cited them moved on once the discrepancy became apparent.

The Burden of Proof Lies With Proponents

Removing the merchant’s obligation to build also removes their decision to join. Shopify states that these protocols run across millions of merchants, implying that a vast number of storefronts now speak protocols their owners never evaluated, under liability terms that remain unsettled. This represents a much larger trust requirement than the Alexa-skill era, where building something served as both friction and consent. You knew what you had joined. Now, inclusion is often default, relying on a platform’s judgment rather than individual choice. Consequently, the burden of proof belongs to those selling this technology, not to merchants absorbing it.

Transparency remains critically low. Visa’s Trusted Agent Protocol, Visa Intelligent Commerce, Mastercard Agent Pay, PayPal, and Stripe do not name any live merchants in their documentation as of September 14, 2026. Only Google names Wayfair, Chewy, and Quince. Furthermore, most major players fail to publish verifiable version numbers or statuses. Visa’s products carry no version number or date. Mastercard’s carries none. PayPal’s carries none. Stripe publishes 2026-04-22.preview and is the only payments company here that provides a timestamp. Status updates are equally sparse. Visa states its product “is in the process of development and deployment” and that depictions “are representations of potential features.” Stripe lists 34 countries by name but labels itself as preview twice. The rest offer no clarity.

Crucially, no protocol on the table specifies who pays when things go wrong. This omission is the central demand of the Merchant Advisory Group, yet no solution has been provided. Publishing a version number, status, or naming a live merchant costs nothing. Naming a merchant requires only one partner’s permission, which Google secured. The fourth requirement,defining liability,is genuinely difficult, yet it is the one merchants have demanded since Zurich. Mastercard’s Agent Pay page calls this “a paradigm shift” offering “unparalleled convenience and efficiency,” citing challenges like “trust, security and interoperability.” Yet it names no merchant, carries no version, and states no status.

An unfinished technology is not inherently scandalous. Refusing to acknowledge its unfinished state while merchants’ storefronts already respond to agents is a deliberate choice. Five companies have made this choice, leaving merchants to bear the risk. The core premise of agentic commerce is that a machine can verify claims instead of taking them on trust, utilizing signed mandates, scoped tokens, and verifiable payment headers. None of this is documented in a way that allows independent verification. Until proponents publish versions, statuses, merchant names, and liability frameworks, outsiders cannot distinguish between a functional payments rail and a press release.