Designing Products for the New User: AI Agents

AI agent doesn’t experience a moment of “Oh, I love this design!” Instead, it systematically filters against parameters.

From our partner:

Reforge Build is the AI prototyping tool for product teams

Vibe coding is fun, but AI prototyping is changing product development. Reforge Build is designed around your product team’s workflow. Start from your product → plan with AI that knows your customers and features → explore multiple variants → collaborate internally → validate with customers, all without leaving the app.

---

For the last decade, success in online shopping has been synonymous with operational design excellence. A pristine product page, seamless checkout flow.

That advantage is becoming more complicated.

As shopping moves through AI agents, a meaningful share of commerce will happen through interfaces that don’t rely on visual design at all. Your product page won’t be seen by human eyes for an increasing percentage of transactions.

AI Agents don’t feel delight, they reason.

This doesn’t mean beautiful interfaces are going away. We now have to deal with an uncomfortable reality: While some share of shopping continues through traditional UX, an increasingly important share will be decided by something entirely different: the training and reasoning of agents evaluating your product.

---

Understanding the future of ecommerce

Picture two shopping journeys:

Journey 1 (Traditional UI): A customer browses your website, discovers a product through beautiful product photography and compelling copy, adds it to cart, and checks out. Your design, your brand storytelling, your visual hierarchy all of it shapes this decision.

Journey 2 (Agent-Driven): The same customer says to an AI assistant, “Find me sustainable running shoes under €100.” An agent queries your product data systems, evaluates your offering against specific parameters, and recommends it (or doesn’t) based on structured data and attribute-level reasoning. Your product page is never seen.

Both journeys will exist simultaneously. But their relative importance is shifting.

For product teams, this creates a complex challenge: you need to excel at both design-driven commerce (for humans) and intent-driven commerce (for agents) at the same time.

From Visual Design to Designing for Intent

This is a fundamental mindset shift, but not a replacement an addition to your product thinking.

PMs and designers have spent years asking: “How do we make this experience feel effortless and delightful?” The answer was usually found in thoughtful UX patterns, smooth interactions, and beautiful aesthetics. This question doesn’t disappear. Humans will continue shopping on websites and apps, and design will remain important for that segment.

But now, a second question emerges alongside it:

“How do we make our product logic, data, and value proposition transparent to an AI system making a split-second decision?”

The agent doesn’t experience delight. It performs reasoning. When a user delegates shopping to an agent with the instruction “Find me sustainable running shoes under €100,” the agent evaluates your product against a mathematical framework. It checks:

Whether your product attributes clearly signal sustainability (materials, certifications, third-party verified data)
Whether your price is truly transparent or hidden behind obscure discounts
Whether inventory is accurate in real-time
Whether your return policy is crystal clear and machine-readable
Whether your brand reputation signals trustworthiness through structured trust indicators

The agent doesn’t browse your product page hoping something catches its eye. It doesn’t experience a moment of “Oh, I love this design!” Instead, it systematically filters against parameters: price ceiling, specific attributes, availability, shipping time, and seller reliability.

What this means for your team:

Your design and UX investment still matters for customers who shop through traditional interfaces. But you now need to allocate significant resources to understanding how agents reason about your products something your design team has no experience with yet.

Designers need to shift additional focus to interface architecture for agents. This means:

Structured product data that’s comprehensible to language models. Not just product descriptions written for humans, but tagged, categorized, attribute-rich data that an AI can reason through.

Conversational content that helps agents understand the nuances of your product. If your product has trade-offs (a running shoe optimized for speed but less cushioning), your content should explicitly articulate this so agents don’t misinterpret.

Trust signals encoded in data, not just visual design. Where once you’d use a badge or icon to build trust, now you need verified, machine-readable certifications and guarantees that agents can verify programmatically.

Product managers need to expand their definition of great product design. The old question ”Is this delightful?” remains important for human shoppers. But you must add: “Can an AI agent understand the true value and positioning of this product without ambiguity?”

Source: 2026, Decoding AI consumers Whitepaper from Statista

The Discovery Inversion: From SEO to Data Architecture

For years, the ecommerce discovery funnel looked like this: customers land on Google or a marketplace, browse search results, click through to your site or listing, and make a purchase.

SEO was the religion. Backlinks, keyword rankings, content strategy entire teams were dedicated to gaming visibility in search results.

Agentic commerce is creating a parallel discovery funnel. The agent doesn’t start on Google. It starts with a user intent, then scans across dozens of marketplaces and seller sites simultaneously, evaluating what’s available, comparing options, and delivering a single curated recommendation.

For the portion of commerce flowing through agents, your visibility is determined not by search rankings, but by whether an AI system can find you, understand you, and verify you.

The New Moat: Machine-Readable Product Data

Reuters projects that AI agents will account for 25% of ecommerce by 2030, totaling nearly $500 billion in sales annually. That’s not a niche that’s a massive parallel channel alongside traditional ecommerce.

And the brands winning in that channel aren’t necessarily the ones with the best websites or the most Instagram followers. They’re the ones that mastered structured, machine-readable data.

When an agent asks “What running shoes are available under €100?”, it doesn’t visit 20 websites and read product descriptions like a human would. Instead, it queries product data systems. It’s looking for:

SKU and real-time inventory
Price in structured format
Physical attributes (shoe size range, weight, materials) encoded in standard taxonomies
Seller information and fulfillment details
Delivery timelines

If your product data is incomplete, inconsistent across channels, or buried in unstructured text on a webpage, the agent can’t reliably reason about your product. It might skip you entirely in agent-mediated transactions, even if human shoppers would love your product.

One study showed that implementing semantic schema markup (structured product data) led to a 192% increase in add-to-cart actions and a 278% boost in transactions. That’s the power of being discoverable to machines.

Recommendations Reimagined

For the past five years, the recommendation engine has been king in ecommerce. Every platform invested heavily in algorithms that learned from user behavior, then surfaced products that similar users liked.

Social proof amplified this. Customer reviews, ratings, and “Other customers also bought” sections weren’t just trust signals they were the primary discovery mechanism for many shoppers. If a product had 4.8 stars and 10,000 reviews, it won. If it had 3 stars and 20 reviews, it lost.

This created a winner-take-all dynamic where established products with critical mass of reviews dominated recommendations, while emerging products struggled to get traction.

Agentic commerce is changing this calculus but only for the portion of commerce flowing through agents.

The Agent solves for an algorithm

When an AI agent makes a recommendation, it’s not optimizing for “what similar users liked.” It’s optimizing for what best solves the user’s stated intent within their constraints.

The agent has context that traditional recommendation engines lack:

The user’s explicit goal (”I need running shoes for cold weather under €100”)
User history and preferences (shoe size, gait type, previously purchased brands)
Real-time availability and pricing across all channels
Detailed product attributes (temperature rating, material composition, weight distribution)
Seller reliability and fulfillment speed

The agent then performs reasoning that looks more like human judgment than algorithmic automation. It doesn’t just count reviews it analyzes review sentiment for the specific use case. It doesn’t just rank by price it evaluates value relative to the user’s constraints.

Example: Two running shoes. Shoe A has 15,000 reviews and a 4.7-star rating. Shoe B has 300 reviews and a 4.8-star rating. Shoe B is newer, lighter, and specifically designed for cold weather. Shoe A is a bestseller but more general-purpose.

Traditional recommendation engines might recommend Shoe A (safety of the crowd). But an agent trained on the user’s context might recommend Shoe B because it better matches the specific use case, even though it has fewer reviews.

What Product Builders Should Do Now

Agentic commerce is already happening, and adoption is accelerating rapidly.

This isn’t a future scenario it’s a parallel channel emerging now. Product teams need to prepare for a bifurcated ecommerce landscape where some shoppers use traditional interfaces and others use agents.

In the human-facing channel (traditional UI):
Continue optimizing for discovery, conversion, and delight through design, content, and brand storytelling. This remains hugely important.

In the agent-mediated channel:
Stop thinking about website UX. Start thinking about data architecture, structured attributes, and agent reasoning. Build dashboards that show you: How often are you appearing in agent recommendations? What’s your win rate against competitors in agent-driven searches? Which product attributes drive agent recommendations?

Across both channels:
Audit your product data architecture. Is your data machine-readable? Complete? Consistent across channels? This is now your primary product surface for agent-mediated commerce.

Explore whether a branded agent makes sense for your company. Some companies will build proprietary agents; others will optimize for discoverability in third-party agents. Both strategies have merit.

Redefine your success metrics to account for both channels. Website traffic and conversion rate matter. But now also measure brand visibility across AI platforms. How often are you mentioned in LLM responses? What’s your share of voice in agent recommendations?

What does it mean for Designers?

For human-facing interfaces:
Beautiful, thoughtful design remains central. Continue investing in UX excellence, visual design, and brand expression through your website and app.

For agent interfaces:
Learn to think in structured data. Your new design canvas (for this channel) isn’t Figma it’s your product information architecture. Understand schema markup, data taxonomies, and how to represent product information in ways machines can reason about.

Transition to conversation design. If agents interact with your products conversationally, understand how to make that experience clear and contextual. What questions should an agent be able to answer about your product?

Design for trust and verification. Visual design builds trust with humans. Explicit trust signals, transparent policies, and verifiable information build trust with agents. Both matter now.

Audit your current design against agent reasoning. Put on your “agent lens” and ask: Could an AI system reliably understand what this product is from our current product page and data?

---

Resources & Further Reading

• Mirakl. “How AI shopping agents are rewarding brands that master data-driven discovery.” November 2025.

• BCG. “Agentic Commerce is Redefining Retail - How to Respond.” October 2025.

• Constructor. “How AI agents for ecommerce are changing the shopping experience.” November 2025.

• Coveo. “Agentic AI in Commerce: What It Is and Why It Matters.” July 2025.

• Hosanagar, K. “Hey AI, buy my product: How LLMs are rewriting the customer journey.” July 2025.

• Perplexity. “Brand Visibility is the North Star for AI Search.” July 2025.

• Prefixbox AI. “Making Products Discoverable in the Age of AI Agents.” August 2025.

• PWC. “The rise of agentic commerce on buying behavior.” August 2026.

• Euroshop. “How agentic AI might redefine the customer journey.” June 2025.

• Alokai. “How Agentic AI Changes the Buyer Journey.” November 2025.

• McKinsey. “Agentic commerce: How agents are ushering in a new era.” October 2025.

• Jakob Nielsen. “Hello AI Agents: Goodbye UI Design, RIP Accessibility.” February 2025.

• Nielsen, J., & others. “From Ranking to Relevance: How AI is Redefining Brand Visibility.” June 2025.

• ChannelEngine. “Agentic commerce: The next leap in marketplace automation.” August 2025.

• Syde. “Agentic Commerce: When AI Shops for Your Customers.” September 2025.

• Commerce Tools. “How Agentic Commerce Is Transforming Retail and Beyond.” November 2025.

• LogicBroker. “The Rise of Agentic Commerce: How AI Shopping Agents Are Changing Ecommerce.” July 2025.

• BigCommerce. “Ecommerce AI Agents in 2025 (Shopping’s Next Big Shift).” November 2025.

• LinkedIn. “How LLM’s are quietly transforming shopping and product discovery.” April 2025.

• Adobe. Statistics on AI adoption and shopping agent usage trends 2024-2025.

• Insider. “5 Best AI Shopping Assistants for Ecommerce 2025.” September 2025.

• BrightLocal. “2025 Local Consumer Review Survey.” 2025.