The Future of Agentic Commerce

AI agents are beginning to autonomously browse, evaluate, and purchase products. What does this mean for how we design ecommerce platforms?

Last year I watched a colleague ask an AI assistant to find a replacement HEPA filter for a specific air purifier model. It took about ninety seconds. It read three retailer pages, noticed that two listings had the wrong model compatibility buried in the Q&A section, and picked the third. No banner ad influenced it. No lifestyle photography influenced it. No urgency badge influenced it. It read specs, cross-checked a compatibility claim, and made a decision.

I have spent most of a decade building systems that optimize for the opposite buyer. At AliExpress US I helped build an affiliate growth engine with more than 10,000 partners across 40-plus countries that lifted revenue about 30 percent, and nearly every lever in it assumed a human on the other end who could be persuaded. Creator content, social proof, urgency, hero imagery, a well-timed coupon. At Next2Market we drove 350 percent GMV growth across consumer electronics brands including Samsung, TCL, Levoit and Anker, and an 18 percent conversion rate improvement, largely by getting better at the psychology of the product page.

Agentic commerce does not attack those levers gradually. It routes around them.

The buyer we optimized for is not the buyer who is arriving

Almost every conversion technique in modern ecommerce is a workaround for human cognitive limits. People cannot hold twelve variants in working memory, so we simplify. People are loss averse, so we show scarcity. People discount future value, so we bundle. People are visually driven, so we invest enormously in photography and video.

An agent has none of those constraints and none of those weaknesses. It can read every review, every spec sheet, every return policy, and every price across forty merchants in the time a person spends scrolling one page. It does not get tired at variant nine. It does not respond to a countdown timer, because it can compute that the timer resets.

The uncomfortable implication is that a meaningful share of what ecommerce teams call conversion optimization is really friction management for humans, and it produces zero value for an agent. Worse, some of it produces negative value. Interstitials, email gates, lazy-loaded specs, compatibility information stored in an image rather than text, size charts rendered as a PNG. These are all invisible to a system that reads structure rather than pixels.

What agents actually reward

When I look at the early data on how assistants pick products, the pattern is not mysterious. Agents reward merchants who are legible. Specifically:

Notice that none of this is aesthetic. The parts of a storefront that consume the largest share of most teams' budget are the parts an agent ignores entirely.

The two-audience problem, and why most teams will get it wrong

The obvious reaction is to build an agent-facing layer. A clean feed, a product API, structured data markup, maybe a dedicated endpoint. That is correct as far as it goes, and most merchants should do it. But it is the easy half.

The hard half is that you now have two audiences with genuinely conflicting incentives, and for several years you will serve both simultaneously. Humans still need the photography, the story, the brand. Agents need the spec table. If you let those two representations of the same product drift apart, you have created a trust failure waiting to happen, because the agent will quote a spec the human-facing page contradicts, and the customer will blame you.

I have seen a small version of this problem already. Managing a catalog across affiliate networks means the same SKU appears in a dozen feeds with slightly different attributes, and reconciling that is genuinely painful work that nobody gets promoted for. Agentic commerce makes that reconciliation a first-order business risk rather than an ops annoyance.

The merchants who win in agentic commerce will be the ones who treat their product data as a product, with an owner, a quality bar, and a roadmap. Everyone else will treat it as a feed, and feeds rot.

Where the margin goes

Here is the part I disagree with most of the commentary on. The common prediction is that agents commoditize everything, price becomes the only variable, and margins compress toward zero. I think that is half right and badly incomplete.

Agents do compress the margin available from confusion. If your pricing power came from a customer not knowing the comparable product was 12 dollars cheaper two tabs over, that pricing power is gone. Fine. That was never a durable advantage anyway.

But agents are optimizing for a specified outcome, not for price alone. When someone asks an assistant to find a good air purifier for a nursery, the agent weighs noise level, filter cost over three years, certification, and the reliability of the seller. That is a multi-variable optimization, and multi-variable optimizations reward products that are genuinely better on a dimension the buyer cares about. It is a more honest market, not a cheaper one.

What actually gets commoditized is the middle. Undifferentiated products sold on the strength of marketing spend have nowhere to hide. Products with a real, articulable, verifiable advantage get found by agents more efficiently than they ever got found by humans, because the agent will actually read the spec sheet that no human read.

What discovery looks like on the other side

Affiliate and creator channels are where I expect the most interesting disruption, partly because that is where I have spent the most time. At AliExpress I ran a creator commerce program that onboarded around 400 creators a month and grew to more than 2,000 active creators, driving about 40 percent GMV growth. The mechanic there was attention: a creator has trust with an audience, and that trust transfers to a product recommendation.

Agents change the shape of that but do not eliminate it. An agent making a recommendation is going to draw on the same corpus that humans draw on, which includes reviews, teardowns, comparison content, and expert opinion. Content that is genuinely evaluative and specific becomes more valuable, because it is the raw material an agent reasons over. Content that is thin affiliate filler becomes worthless faster, because an agent can tell the difference between a review that tested the product and a review that paraphrased the spec sheet.

My working assumption is that the creator economy bifurcates. Personality-driven content keeps working for humans who want to be entertained into a purchase. Rigorous, testing-based content becomes infrastructure for agents. The middle, which is most of the current affiliate content on the internet, gets squeezed out from both directions.

What I would do right now

If I were running growth at a consumer brand today, I would not rebuild my storefront for agents. That is premature. I would do four cheaper things.

The broader point is that agentic commerce rewards substance and punishes theater, and most ecommerce organizations are staffed and budgeted for theater. That is not a moral failing, it is a rational response to the buyer we had. But the buyer is changing, and the organizations that adjust early will spend the next few years quietly winning queries that their competitors never knew they lost.

I do not think the human-facing storefront goes away. I think it stops being the only front door, and a lot of teams are going to be surprised by how quickly the second door starts carrying traffic.