Agentic Commerce Is Not the Next Voice Shopping, but It Is Still Unproven

We have a habit of falling for the same patterns in tech. A few years ago, the industry was convinced that voice shopping would fundamentally rewrite the retail playbook. We were told we would simply speak our needs into the air and the world would deliver. Then, the numbers stopped adding up, the hype faded, and we moved on to the next big noun.

Now, the conversation has shifted to agentic commerce. On the surface, it looks like the same story. But if you look at the plumbing, something is different this time. The risk is no longer just a failed trend, but a structural shift in how transactions happen, often without the merchant even knowing they have opted in.

The Shift From Input to Decision

To understand why this matters, we have to distinguish between an interface change and a structural change. Voice shopping was an interface change. When you used a voice assistant to buy something, you were still the primary actor. You made the choice, you confirmed the cart, and you were present for the transaction. You simply swapped a keyboard for your voice.

Agentic commerce is different because it targets the decision itself. The goal is for the purchase to happen without the human being in the room. The FIDO Alliance has actually created a specific category for this called Human Not Present. This isn't just a new way to shop; it is a new class of transaction. A useful companion note is from AI Discovery to Agentic Commerce, because it looks at a nearby part of the same system.

The distinction matters for anyone running a business. If you treat this as just another "channel" like voice or mobile, you miss the point. The tradeoff here is control. In voice shopping, the human remained the gatekeeper. In agentic commerce, the agent becomes the gatekeeper. The decision you should inspect is where the authority to spend money actually resides in your current checkout flow.

Four Factors That Separate Agents From Voice

Voice shopping failed to reach its trillion dollar potential partly because it required too much manual effort from the people selling the goods. Agentic commerce is attempting to bypass those hurdles entirely. The same pattern also shows up in Not the Platforms Selling It, where the practical question is how the signal becomes visible.

1. Zero Build Integration

In the past, if a merchant wanted to be part of a new wave, they had to build something. You needed an Alexa skill or a specific bot. Even recent attempts, like OpenAI's Instant Checkout, required integrations that eventually became obsolete when the surface disappeared.

The new approach is "invisible" integration. Shopify recently updated its UCP version to ensure these protocols work across storefronts without requiring merchants to change their existing setups. Mastercard's Agent Pay and Visa's Intelligent Commerce are designed to work with existing guest checkouts or web forms. Essentially, the agent just fills out the form a human would use.

This is a massive shift in power. The merchant no longer has to "join" the ecosystem; they are simply reachable by default. The tradeoff is that the friction which used to act as a form of consent is gone. You are now part of a system you may not have evaluated.

2. Payments Designed for Absence

Voice shopping relied on saved credit cards. Agentic commerce is building payment rails specifically for a buyer who isn't there. Stripe has introduced Shared Payment Tokens that are capped at a certain amount and expire. Google's AP2 uses signed mandates to prove what a user actually authorized, and x402 handles unpaid requests via specific HTTP headers.

This matters because it solves the security paradox of letting a machine spend your money. By limiting the scope and duration of the token, the risk is mitigated. However, the decision for the merchant is whether these tokens provide enough protection against fraudulent agent behavior.

3. The Role of Global Standards

Voice commerce was a walled garden. You had Amazon's API and Google's actions, and both were owned by the companies that profited from them. Agentic commerce is involving actual standards bodies. The W3C and GS1 have held workshops on this, and Google and Mastercard have donated protocols like AP2 and Verifiable Intent to the FIDO Alliance.

When standards bodies get involved, it usually means the technology is moving from a "feature" to "infrastructure." This increases the likelihood of long term viability, but it also means the shift is happening at a level deeper than any single company's marketing department.

4. Organized Merchant Resistance

Because the stakes are higher, merchants are fighting back in a way they didn't during the voice era. The Merchant Advisory Group has already begun setting conditions, arguing that merchants are being exposed to risks they didn't create and cannot manage.

They are proposing a "liability waterfall," which would place the financial responsibility on whoever controlled the function that failed. They are also insisting on the right to route their own payments. This is a direct response to the "invisible" nature of these integrations. If you are a merchant, the key question is: who is liable when an agent makes a mistake on your storefront?

The Gap Between Theory and Reality

Despite the infrastructure, there is a gap between the "Human Not Present" vision and what is actually happening. If you look at Google's current agentic checkout, it is still very much "Human Present."

Google's system can watch a price and buy an item when it drops, but it explicitly asks for permission first. The user must confirm the purchase and the shipping details. While this is the correct design for trust, it means it is still just an interface change, not a structural one. The true "Human Not Present" class is a target for 2026, not a current reality.

The trillion dollar projections we see in the news are based on the structural change, not the current permission based tools. We are seeing a mismatch between the marketing of the "agent" and the actual technical implementation.

The Danger of "Same Class" Forecasts

We are seeing some staggering numbers. McKinsey suggests $1 trillion in U.S. retail revenue orchestrated by agents by 2030. Gartner predicts that 90% of B2B purchases will be handled by agents by 2028, totaling $15 trillion.

The problem is that these numbers belong to the same class of prediction as the 2018 forecasts for voice shopping. Back then, consultants predicted a $40 billion market that never materialized. While the underlying technology of agents is more strong than voice, the habit of over projecting behavior change remains the same.

The lesson here is to be skeptical of the timeline. Machine led buying is a massive leap in human behavior. Just because the plumbing is being laid doesn't mean the water will flow at the predicted volume.

The Burden of Proof

The most telling evidence is often found in the documentation. When a technology is truly working and scaled, the providers brag about their partners. They name names.

If you look at the documentation for Visa's Trusted Agent Protocol, Visa Intelligent Commerce, Mastercard Agent Pay, or PayPal, you will find a striking absence of live merchants. They don't name a single one. Google is the outlier, naming a few like Wayfair, Chewy, and Quince.

the transparency of the rollout is inconsistent. Stripe is the only one providing a clear preview version with a date and a list of supported countries. Others describe their products as being "in the process of development" or as "representations of potential features."

This suggests that while the platforms are pushing the protocols onto merchants by default, the actual ecosystem of active, successful agentic commerce is still very small. The burden of proof is on the payment companies to show this works in the wild, rather than on the merchants to figure out how to integrate it.

For now, agentic commerce is a powerful set of tools looking for a proven use case. It is more than voice shopping, but it is far from the trillion dollar reality the forecasts suggest.