The new challenge for digital product teams: building experiences that work for people and the agents acting on their behalf.

The dual-interface problem

Every website today faces an emerging reality: your visitors increasingly aren’t human. AI agents are browsing your pages, parsing your content, and making decisions about whether to recommend you, transact with you, or move on. This creates a design challenge without precedent—serving two fundamentally different types of users simultaneously.

Humans need visual hierarchy, intuitive navigation, and emotional resonance. They scan pages in f-patterns, respond to color and imagery, and make decisions based on feeling as much as information. Agents need none of this. They parse structure, extract meaning from semantic markup, and evaluate content based on relevance and reliability.

The question isn’t which audience to prioritize. It’s how to serve both without compromising either.

Semantic structure over visual design

For decades, web design has been primarily visual. We’ve optimized for how things look, trusting that structure would follow. For the agentic web, this inverts. Structure becomes primary; visual presentation becomes a layer on top.

This means taking semantic HTML seriously. Proper heading hierarchies (H1, H2, H3) aren’t just accessibility nice-to-haves—they’re how agents understand your content’s organization. Schema markup isn’t optional metadata—it’s how machines know what your business does, where you’re located, what you sell, and why you’re credible.

Structured data becomes your agent-facing interface. JSON-LD schemas tell AI systems Not just what your content says, but what it means. A product page with proper schema markup communicates price, availability, reviews, and specifications in a format agents can instantly parse and compare against competitors.

Api-first thinking

The most agent-ready organizations are those already thinking API-first. If your website is just a visual layer over well-structured data and services, you’re positioned for the agentic web. If your website is a collection of pages with content locked in HTML, you have work to do.

This doesn’t mean building public apis for everything. It means architecting systems where data and functionality are cleanly separated from presentation. When an agent needs to check inventory, it shouldn’t have to scrape a webpage—it should be able to query a service directly.

The Model context protocol (MCP) accelerates this shift. Organizations exposing their services via MCP servers make themselves immediately accessible to any mcp-compatible agent. It’s the difference between requiring every visitor to walk through your front door versus offering a service window for those who know exactly what they need.

Preserving the human experience

Here’s the tension: Optimizing for machines can make experiences worse for humans. Pages stuffed with schema markup can feel sterile. Content written for semantic clarity can lack personality. Interfaces designed for data extraction can feel transactional rather than engaging.

The solution isn’t choosing one audience over the other—it’s layered design. The semantic structure serves agents. The visual and emotional layer serves humans. When done well, they’re complementary: clear structure improves human usability too; good content works for both audiences.

Think of it like architecture. A building needs solid engineering (structure for machines) and thoughtful interior design (experience for humans). Neither is optional. The best buildings excel at both.

Practical implications

For product teams, this means several shifts:

Content strategy must consider machine consumption. How will an agent summarize this page? What will it extract as key facts? Is the most important information structured or buried in prose?

Technical SEO becomes foundational infrastructure, not an afterthought. Schema markup, clean URL structures, proper meta tags, and semantic HTML are table stakes.

Analytics need expansion. You’re already tracking human behavior. Now you need to understand agent interactions—which AI systems are accessing your content, what they’re extracting, and how they’re representing you to users.

Design systems should separate structure from style. Build component libraries where the semantic meaning is independent of visual presentation. This enables consistent experiences for humans while maintaining machine readability.

The websites that win in the agentic era won’t look radically different to human visitors. But under the hood, they’ll be built for a world where half your audience doesn’t have eyes.