March 29, 2026

Your Next Customer Isn't Human

Think about the last time you bought something online.

You probably browsed. Scrolled past hero banners. Opened tabs. Compared reviews. Got distracted. Came back. Checked a Reddit thread. Looked at the return policy. Added to cart. Abandoned it. Got a retargeting ad. Came back again. Bought it.

That whole journey, from intent to purchase, was designed for you. The colors, the urgency cues, the social proof badges, the checkout flow. Twenty years of ecommerce optimization, all built on one assumption: a human is looking at this page.

That assumption is breaking.

Machines at the counter

In January 2026, Google CEO Sundar Pichai announced the Universal Commerce Protocol at the National Retail Federation conference. It's an open standard that gives AI agents a common language to discover products, compare prices, and complete purchases across merchants and payment providers.

Google isn't alone. OpenAI and Stripe co-developed the Agentic Commerce Protocol. Shopify launched Agentic Storefronts. Amazon's Rufus is already driving $12 billion in incremental annualized sales, with users 60% more likely to complete a purchase when the AI assistant is involved. Perplexity's "Buy with Pro" saw a 5x increase in shopping queries after launch.

These aren't concept videos. Target customers can browse and buy through Google's Gemini assistant today. Instacart offers checkout directly inside ChatGPT. Stripe's new Shared Payment Tokens let AI agents initiate payments using your saved card without ever seeing the number.

The numbers tell the story: 45% of consumers already use AI somewhere in their buying journey. McKinsey estimates agentic commerce could redirect $3-5 trillion in global retail spend by 2030. Gartner says 90% of B2B buying will be AI agent-intermediated by 2028.

What shopping is for humans

Here's what's easy to forget in the rush to optimize for agents: shopping, for people, was never purely transactional.

You walk through a store and notice a color you hadn't considered. You try on a jacket that looked terrible online but fits perfectly. You ask the person at the counter what they'd recommend for a dinner party. You buy the expensive olive oil because the label made you feel like you were in Puglia for a second.

Even online, the best shopping experiences work because they understand something about desire. You don't always know what you want. Sometimes you're not buying a product, you're buying a feeling, an identity, a small act of self-expression. The browse is the point.

David Lauren, speaking about Ralph Lauren's new AI agent "Ask Ralph," said it plainly: "Nobody needs a flannel shirt. Showing a customer how to put it together is what makes us unique."

AI agents don't browse. They don't wander. They don't get seduced by packaging or persuaded by a well-timed popup. They parse structured data, evaluate attributes against a spec, and execute. An agent buying dish soap will never notice that your brand redesigned the bottle.

Not a replacement, a new surface

This isn't the death of human shopping. It's the birth of a second channel that works differently.

Think of it as a split in the funnel. Some purchases are transactional: replenishing supplies, reordering what you already know works, finding the cheapest option for a commodity. These are ripe for agents. You shouldn't have to spend twenty minutes buying paper towels.

Other purchases are experiential: exploring a new brand, discovering a style, making a considered choice where taste and context matter. These remain deeply human. Nobody wants an AI to pick their wedding gift.

The unlock is that agents handle the first category better than any human interface ever could. An agent can compare every dish soap on the market in seconds, weigh price against reviews against your past preferences, and buy. No dark patterns needed. No conversion optimization. Just a machine making a rational decision on your behalf.

This frees the human surface to be what it always should have been: less transactional, more meaningful. When the baseline purchase is handled by an agent, the moments when a person actually visits your site become higher-stakes and higher-intent. Those visitors aren't comparison shopping. They're choosing to be there.

What this means for brands

Your product data is now your storefront. When an AI agent evaluates your product, it doesn't see your homepage. It reads your structured data: titles, descriptions, attributes, pricing, inventory, reviews. If that data is messy, inconsistent, or thin, you're invisible. Shopify's new Catalog system auto-infers categories and extracts attributes specifically because most merchants' data isn't agent-ready. Answer Engine Optimization is the new SEO, and it's harder because machines are less forgiving than humans.

You need two design strategies. One for people, one for machines. The human surface should lean harder into what machines can't do: storytelling, editorial, sensory experience, community, taste. The machine surface needs clean schemas, real-time inventory APIs, reliable checkout endpoints, and response times measured in milliseconds. If your API can't answer a query within the latency threshold, the agent skips you.

Brand matters more, not less. When an agent narrows a category to three options, the deciding factor is often the brand signal embedded in the data: ratings, review sentiment, return rates, consistency. The brands that invested in genuine quality and reputation have a structural advantage in an agent-mediated world. The ones that relied on clever landing pages and retargeting have a problem.

Trust becomes an architecture problem. The UK's Information Commissioner's Office published a framework for agentic AI data protection in January 2026, identifying eight risk categories from automated decision-making to hallucination-driven data accuracy issues. When an agent buys something on your behalf and it's wrong, who's responsible? The protocols are still being written. Stripe reports fraud rates near zero in early agentic transactions, but identity resolution remains unsolved.

First-party AI is the new loyalty program. Home Depot built Magic Apron. Nike built NikeAI. Ralph Lauren built Ask Ralph. These aren't chatbots. They're brand agents that understand the full catalog and the customer's history. The strategic question isn't whether to participate in Google's or OpenAI's marketplace. It's whether you also build your own agent that knows your brand better than any third-party ever will.

The protocols

The infrastructure is moving fast. Here's what's live or launching:

Universal Commerce Protocol (UCP) by Google, with Walmart, Target, Wayfair, Lowe's, and payment networks Mastercard, Visa, American Express, and PayPal
Agentic Commerce Protocol (ACP) by OpenAI and Stripe, with Target, Sephora, Nordstrom, Best Buy, Home Depot, and 25+ ecosystem partners
Shared Payment Tokens by Stripe, enabling credential-free agent-initiated payments bounded by time and amount
Agentic Storefronts by Shopify, connecting merchants to ChatGPT, Perplexity, and Microsoft Copilot with one setup

OpenAI initially tried "Instant Checkout" inside ChatGPT, then pulled back, rerouting users to retailer sites. Their explanation was revealing: the initial version "did not offer the level of flexibility that we aspire to provide." Translation: merchants didn't want to lose the customer relationship. The tension between platform and brand is already shaping the architecture.

What to do now

If you're a product or brand team, three things matter in the next twelve months:

Audit your product data. Not your website, your data. Is every product described with structured attributes an agent can parse? Are prices and inventory available via API in real time? Are your reviews machine-readable? This is table stakes and most brands aren't there yet.

Design for two audiences. Your human surface should get more editorial, more experiential, more worth the trip. Your machine surface should get more structured, more reliable, more fast. These are different problems with different metrics.

Pick your protocol position. You can be a merchant on Google's and OpenAI's platforms (most will). You can build a first-party agent that represents your brand (some should). You can do both. But doing neither means you're waiting for the market to decide for you.

75% of attendees at NRF 2026 are either implementing or actively planning agentic commerce. The question isn't whether this is happening. It's whether your product surface is ready for a customer that doesn't have eyes.