El Agentic Commerce ya llegó… Pero no le queda bien a todos...

The AI agent is not replacing Google in Mexico; it is replacing the family WhatsApp group.

That single finding changes everything about how retail media should be built in this market. But before talking about implementation, we need to ask ourselves a more uncomfortable question: should you implement it right now, in your category, for your consumer?

Agentic commerce is already here. It’s real. It’s accelerating. And like a Paul Smith suit, it’s beautifully made but only works if it fits you well.

The study that reframed the problemIn December 2025, a multinational consumer goods company conducted a discreet experiment in three cities. Same product category. Same question: how do you prefer to discover and evaluate what to buy?

The results told three completely different stories.Dallas: 62% preferred an AI agent, 18% traditional search, and the remaining 20%, human recommendation.São Paulo: 17% an AI agent, 12% searching, and the remaining 71%, human recommendation.Monterrey: 48% an AI agent, followed by 20% who preferred searching and 32% human recommendation. That 32% in Monterrey is the number most vendors ignore. It's the number that breaks the imported manuals. In Mexico, the AI agent does not compete with a search bar. It competes with someone’s mom’s opinion.340 million messages that no algorithm is capturing.Mexico has the highest rate of product recommendation messages shared between users across Latin America. According to data from Meta's WhatsApp Business API from Q3 2025, that figure amounts to 340 million monthly messages just in the consumer goods category. Before a consumer in Monterrey decides what to buy, there is often a conversation happening in a family chat that no retailer, no AI system, and no attribution model is currently measuring.The AI agent is not competing against a search bar. It is competing against someone’s mom’s opinion.Trust is the real metric and the hierarchy is specific.A Kantar Mexico 2025 study mapped how much urban Mexican consumers trust different sources when making purchase decisions:

87% Family or close friend
62% AI agent trained with retailer data
54% Store employee
49% Organic search results
31% Paid advertising

An AI agent trained with firsthand retailer data ranks second, above a human employee, above Google, above any ad you might run. That’s extraordinary. But it only holds if the agent earns its position through relevance and precision, not just availability. The moment the AI agent feels generic, the consumer returns to the family chat.The 41% that changes the product roadmap.A retailer from northern Mexico instinctively understood this. In Q4 2025, they added a single functionality: a button that allowed users to share an AI-generated recommendation directly on WhatsApp.The result: a 41% increase in the conversion rate for shared recommendations, compared to those that were not.That button did not replace social conversation. It joined it. The AI became a participant in the family chat rather than a substitute. This is the design principle that U.S. manuals completely overlook: in Mexico, the goal of an AI agent is not to close the sale.

It is to deserve to be shared.

Not all categories are wearing the same suit

Agentic commerce is not a switch you activate across your entire business. It is a series of decisions category by category, audience by audience, about where behavioral willingness already exists — and where you first need to build it.

Consider three categories operating in the same Mexican market, facing completely different realities:Tech E-commerce: A platform selling electronics already serves a consumer comfortable with AI-assisted comparison and data-driven recommendations. The agentic layer is a natural extension of existing behavior. The challenge is technical and commercial (catalog structuring, bidding strategy migration), not behavioral adoption. This category is ready for the suit. It fits well.Supermarkets: Purchase cycles are high frequency, low consideration, and deeply habitual. The consumer is not searching; they are restocking. Here, agentic commerce does not need to persuade; it needs to anticipate. Subscribe & Save is the closest existing proxy, but the fact that 61% of Mexican subscribers use it only for free shipping indicates that the behavioral base is not fully consolidated. The suit exists in the right size. But the consumer needs one more test before they are ready to wear it.Beauty Brand: Purchase decisions are intensely personal, socially influenced, and emotionally charged. Kantar's trust data is probably sharper here. An AI agent recommending a foundation shade or a skincare routine is navigating territory that feels intimate. The consumer needs to believe the agent knows them, not just their purchase history. That requires a depth of personalization that most retailers are far from being able to offer. Using this suit before it is tailored not only looks bad: it damages the relationship.Not all consumers arrive at the same timeThe category dimension is only half the equation. Consumer willingness is equally fragmented and equally misunderstood.A 28-year-old professional in Monterrey with smartphone habits comparable to Dallas and a Netflix subscription since 2019 is not the same target as a 45-year-old cash-paying shopper in a suburban supermarket in Guadalajara. Both are "Mexico." Both will interact with agentic commerce on completely different timelines, through completely different trust thresholds, and for completely different reasons.Treating them as a single audience is not just strategically lazy. It is the reason that the $8 million tech investment we described in Week 1 generated 60% below projected revenues.The counterargument a CMO might defendThe adoption of agentic commerce in Mexico is projected at 8-12% for 2026. The sponsored products model still captures the overwhelming majority of high-intent purchase moments. Reallocating significant resources towards agentic preparation before behavioral demand exists is a strategic luxury that most organizations cannot afford.That is a legitimate and defensible position as long as it is a conscious choice, not an omission.The CMO who says "we have evaluated our category, our consumer willingness, and our infrastructure, and we are monitoring agentic commerce while optimizing our current model" is making a solid strategic decision. The CMO who says "we haven't thought about it" is building revenue streams with an expiration date they haven't read.The structural threat persists: by 2028, most product searches in Mexican urban areas will be driven by agentic commerce.When an AI agent presents a recommendation instead of ten ordered results, the bidding model for position collapses. Position one captures 35% of clicks today. An agent does not have a position one.

The question is not whether the suit will be worn. It is whether yours will fit well when the time comes.