Your homepage loads in 1.2 seconds. Your hero image A/B tested to a 14% lift. Your scarcity nudge (“Only 3 left!”) converts beautifully. None of it matters to the AI agent that just became your customer.
The agent doesn’t see the image. It doesn’t feel urgency. It parsed your product schema, checked your delivery SLA against the buyer’s threshold, cross-referenced your return policy with three competitors, and either added to cart or moved on—in the time it took you to read this sentence.
We’ve spent two decades optimizing commerce experiences for human psychology. That work is becoming obsolete. A January 2026 IBM Institute for Business Value study found that 45% of shoppers already use AI to help with the buying journey.
Consumer behavior has shifted. The infrastructure hasn’t.
Every funnel metric, every conversion rate, every heatmap, all of it was designed to model human attention. Machine buyers don’t linger on a page. They don’t scroll. They make structured requests, receive structured responses, and execute decisions algorithmically. If your analytics can’t distinguish agent traffic from human traffic, you’re flying blind on a growing share of your transactions.
What agents actually evaluate
Here’s what an AI shopping agent cares about:
- Structured data availability. Can the agent reliably extract price, SKU, variant specs, availability, and shipping estimate without scraping HTML? If your product data lives in rendered JavaScript that requires a full browser context to hydrate, many agents either can’t read it or won’t bother. Schema.org markup, clean JSON-LD, and machine-readable APIs are now table stakes for agent-accessible commerce.
- Policy clarity. Return windows, warranty terms, compliance certifications, geographic restrictions. Agents representing buyers with specific requirements—a procurement system enforcing Net-30 payment terms, an assistant honoring a user’s “only ships within 5 days” rule—will filter you out instantly if this data is buried in a PDF or requires parsing ambiguous prose.
- Decisional confidence signals. Human buyers respond to social proof like reviews, ratings, and “bestseller” badges. Agents respond to something different: data completeness and consistency. An agent querying your product API and finding mismatched dimensions between your PDP and your structured data doesn’t think, “I’ll ask for clarification.” It downgrades your reliability score and moves on.
- Response determinism. If an agent queries your inventory API and gets different availability data on two sequential calls because your cache is stale or your CDN is serving inconsistent responses, you’ve introduced ambiguity into a system that runs on certainty. Human buyers tolerate a little inconsistency, but agents flag it as a trust failure.
Where the human customer still shows up
This doesn’t mean human-centered design is dead. It means its scope is changing.
The human moment has shifted upstream and downstream of the transaction. A person still decides what they want and sets the constraints their agent operates within. They’re not browsing your site, but they configured the agent that is, and that configuration reflects human preferences, priorities, and values. Getting that upstream preference layer right—being the brand a person trusts enough to authorize their agent to buy from—is still a human design problem, which makes it a brand and trust problem.
The human moment also returns post-purchase. Delivery exceptions, product questions, and returns generate human escalation. Customers who bought via agent but had a poor support experience still churn, still leave reviews, and still tell people. The service layer remains deeply human even when the purchase layer doesn’t.
What disappears, or at least shrinks dramatically, is the middle: the persuasion layer. The merchandising, the personalization engine trained on click patterns and hover behavior, and the carefully sequenced email nurture flow. If the agent is the buyer, that whole apparatus is operating on the wrong audience.
What machine-readable commerce actually looks like
An agent-ready commerce experience has a few characteristics that most platforms don’t currently prioritize.
First is a headless, API-first product catalog where every attribute that influences a purchase decision, including price, availability, lead time, compliance status, bundle eligibility, and substitution rules, is queryable programmatically and returns consistent, versioned responses.
Alongside that, you need structured policy endpoints. Not a /shipping-policy page, but an endpoint that returns your shipping rules as structured data, including carriers, regions, cutoff times, and SLA tiers. Same for returns, warranties, and any compliance certifications relevant to your category.
Promotions need the same treatment. A discount that lives only in a banner graphic is invisible to an agent. If your promotional logic isn’t exposed through your pricing API, or at minimum through machine-readable markup, agents will consistently see your full price while your competitors’ discounts are surfaced automatically. And inventory needs to be honest and machine-checkable in real time, not batch-synced. Agents don’t have patience for “usually ships in 3–5 days” when a competitor returns a deterministic delivery timestamp via API.
None of this is technically exotic. Most of it is a priority problem. These are capabilities that engineering teams have been able to build for years, but that commerce organizations haven’t demanded because their primary user was human.
The design discipline that doesn’t exist yet
Here’s what bothers me most about where the industry is: Nobody owns this.
Ask a commerce team who is responsible for the human shopping experience, and you’ll get a clear org chart: UX, CRO, merchandising, personalization. Ask who is responsible for the agent shopping experience, and you’ll get a long pause followed by “probably engineering?”
That’s not good enough. Machine-readable commerce isn’t purely a technical problem. It requires the same kind of deliberate, tested, iterated design work that human-centered UX does. Someone needs to be asking, what does an agent encounter when it queries us? Where does it get confused or give up? What data are we failing to expose? What inconsistencies are costing us agent-driven conversions we can’t even see in our analytics?
This is a new design discipline. Call it agent experience design, machine UX, whatever you want—the name matters less than the fact that it needs a dedicated owner with the authority to change things. Right now, at most companies, it has neither.
The web professions have absorbed new disciplines before. Accessibility practice, performance engineering, and API design each started as someone’s side concern before becoming a professional function with its own standards, tooling, and community. Agent-experience design is at that same early stage, except the adoption curve of AI buyers is significantly steeper than any of those predecessors.
In fact, McKinsey projects that by 2030, the U.S. B2C retail market could see up to $1 trillion in orchestrated revenue from agentic commerce, with global projections reaching $3 trillion to $5 trillion. The organizations that haven’t figured out who owns this problem soon will find out the hard way.
A practical starting point
If you want to audit where you actually stand, start by querying your own product data the way an agent would.
Use your public API or structured data markup, not your CMS or other internal tools, to retrieve price, availability, delivery estimate, and return policy for your 10 best-selling products. Check whether the data is complete, consistent, and deterministic. Check whether your promotional pricing is reflected and whether your inventory is real-time or lagged.
Most teams that run this exercise find gaps they didn’t know existed, because nobody had ever looked at their commerce experience from the outside in as a machine. That’s where the work starts. And right now, almost no one has started it.
