Six Principles for Getting Agentic Commerce Right

Agentic commerce—the idea of AI agents acting on a customer’s behalf to discover, evaluate and purchase products and services—is still emerging, but the conditions that make it necessary are already here.

AI agents have moved beyond just answering questions to now taking action. Payment infrastructure is starting to catch up. In such an environment, learning to shape your agentic commerce experiences responsibly, with your customers in mind, is crucial.

The previous article of our agentic commerce series argued that conversational commerce is how most brands will take their first real step into agentic commerce. But consumers aren’t yet ready to hand over full transaction autonomy to an agent.

Closing this gap is a design problem. How will your brand choose agentic experiences that earn trust?

Here we will illustrate six principles for designing agentic commerce experiences. Using real-world examples, we will show you how these principles can help you navigate customer expectations to answer questions like: What elements of the shopping experience do people want to continue to control themselves? What shopping tasks are they willing to delegate to an agent?

Why principles, why now

Customers now encounter brands across an expanding range of digital touchpoints, from search engines and marketplaces to AI assistants and autonomous agents. As these journeys become more fragmented, brands can no longer shape every interaction directly. Increasingly, AI agents are researching, comparing and making decisions on customers’ behalf.

In such agentic commerce experiences, clear principles are essential. They help provide guardrails that ensure every product, service and interaction reflects the brand’s intent, even when the brand is not directly involved in the decision-making process. Internally, they give teams a shared yardstick: what to build, what to reject and what memorable means for an experience no one on your side will directly touch.

The six principles below are a starting point, meant to be adapted to your goals, your values and what you know about your customers, market and competitors. We’ll start with the one the other five depend on.

Principle 1: Trust as the universal currency
As commerce moves from screens we operate to systems that act for us, trust needs to intrinsically become part of the product. Customers are delegating judgment, and brands need to earn that delegation by making the experience feel safe through tangible measures such as permissions, limits, reversibility and proof. These measures deserve the same design care we give to flows and content. As our last Futurescape report puts it, trust is a universal currency. Confidence grows when a system behaves predictably, explains itself at the moments that matter and makes it easy to hand over more control or take it back. Consider Stripe’s Agentic Commerce Protocol. It issues Shared Payment Tokens as credentials that carry a spend limit, merchant restrictions and an expiry and can be revoked in real time. The permission is visible and reversible by design: a genuine control surface the customer can see and act on.

Principle 2: Readiness as the new relevance
Agentic commerce offers a way out of the personalization arms race. Instead of following people around the internet hoping to catch them in a buying mood, brands can prepare to be useful the moment they’re called upon—by a customer or by that customer’s agent. In conversational commerce, this readiness is the new relevance: being reliably useful when it counts, wherever the customer happens to meet you. Your product knowledge is structured, your proof is machine-readable and your value is legible to a system that will never see your carefully art-directed homepage. This is already concrete. In January 2026, Google’s Universal Commerce Protocol went live with Shopify, Target, Etsy, Wayfair and Walmart among its launch partners; OpenAI’s Agentic Commerce Protocol feeds structured catalogs into ChatGPT; Perplexity runs its own merchant program. The through-line is unforgiving: agents can only recommend what they can parse. When a brand’s product data is incomplete or unstructured, its products vanish from the set the agent considers. Readiness is the price of being considered at all when and where the customer needs it.

Principle 3: Agency is felt, not performed
Models are getting better at reading the person in front of them: shifting the options they surface, adjusting tone, knowing when to be direct and when to hold back. They are not there yet, and the brands that get there first will offer something that feels genuinely tailored. The test of a good conversational commerce experience is simple. Does it leave the person feeling more capable or does it leave them feeling managed? The point is for agency to be genuinely felt. In practice that means designing a spectrum of participation—assistive, delegated, fully autonomous—and letting customers move along moment to moment, deciding for themselves how hands-on to be. Some moments should stay manual on purpose, because the human part is the whole point. Google’s Agent Payments Protocol shows what this can look like under the hood. It splits a purchase into two separate grants: an Intent Mandate that lets an agent search and negotiate, and a Cart Mandate that authorizes the actual buy. A customer can hand over the first and hold on to the second—or, once trust is established, switch on a “human not present” mode and let the agent close on its own. Same system, different postures, chosen by the person rather than the conversational commerce platform.

Principle 4: Show, don’t tell—multimodal by default
Agentic commerce collapses the old line between physical and digital, which means brands have to work in more than one mode at once. A conversation on its own is a thin storefront. Many conversational commerce experiences on the market today feel like a chatbot bolted onto a storefront, and customers can tell. The strongest agentic commerce experiences weave language together with images, motion, sounds and other elements the customer can actually manipulate. Dialogue conducts the interface (or the interface-free experience), cueing the right comparison, the right photo, the right video, states that shift as intent sharpens. This is where generative, dynamic design systems earn their keep: they let an experience recompose itself in real time while keeping a brand’s expression intact. Reading builds some confidence. Seeing and hearing builds more.On Sephora.com, if a customer searches for an everyday foundation, the answer isn’t a paragraph to parse. A shade gets matched against the customer’s own skin tone, a short clip shows how it sets through the day, and it sits side by side with what they were already comparing. When asked “does it work under sunscreen,” the panel updates rather than the chat restarting.

Principle 5: Optimize for long-term human outcomes
An agentic commerce system negotiates, recommends and buys on someone’s behalf week after week. As such, conversion becomes a poor barometer of success because it gets measured only over the next five minutes after the performed action. By using conversion as a scoreboard, you end up inheriting the responsibility for whatever happens downstream: a customer’s budget, their mental load, their confidence, the environmental cost of what you help them buy. Handled badly, an agent becomes one more source of decision fatigue. Handled well, it works against a life goal, respects a stated preference for durability or lower carbon, and might even recommend the smaller purchase, or none at all. The bar is whether people buy better and feel more in control over time. There’s a harder question underneath, and it’s worth saying out loud: Who does the agent actually work for—the customer, the merchant, or the conversational commerce platform? Once agents hold the checkout button, small choices about defaults and rankings compound into large effects on what people spend and save. Optimizing for long-term human outcomes is how a brand answers that question in the customer’s favor before anyone has to ask.

Principle 6: Progress through principled experimentation
No brand arrives at agentic commerce in one move. The path is iterative: set a clear north star, agree on your principles, then run tight experiments against them. Prototype a single interaction. Test one degree of autonomy. Try a multimodal pattern and watch what actually lands. Then let the evidence decide what comes next. This keeps progress moving without a redesign of everything at once—the principles become the filter for what you test, what you shelve, and what you refuse outright. Amazon.com’s “Buy for Me” is a study in restraint. Rather than switching on full autonomy, Amazon shipped a deliberately narrow version: a clearly labeled button, a limited set of partner brands, the customer’s explicit approval on every order, and no visibility into their other purchases. It’s an experiment with the guardrails built in from the start—exactly the shape early agentic commerce bets should take.

These six principles won’t land the same way for any two brands, and they’re not meant to. Use them as a lens for your own decisions. What matters most is starting. Pick the principle your organization is furthest from living up to, run one focused experiment against it and let what you learn point to the next. How fast any of this arrives will vary—by how comfortable customers are, how mature the category is, how ready each brand is.