Your product has a second user now. As of May 2026, Cloudflare Radar reports bot traffic at 57.2% of HTML requests. Human traffic: 42.8%. The crossover arrived over a year earlier than forecasted.
This changes what "user experience" means. An agent navigating your interface doesn't see your careful layout. It reads the accessibility tree. The semantic markup underneath your visual layer is the only interface the agent has. Affora's design research found that agent performance depends directly on the interaction meaning available through its interface representation.
The accessibility tree. The same infrastructure you've been told to build for years, for different reasons.
The overlap nobody planned
The arXiv paper "From Human Interfaces to Agent Interfaces" defines agent interfaces as needing structured inputs, explicit semantics, and deterministic execution. That list is WCAG compliance stated in different vocabulary.
InfoWorld made the connection direct: agent operability is a strict superset of accessibility. Everything an agent needs includes everything accessibility needs, plus more. The performance gap is not marginal. Switching from screenshot-based to DOM-native execution cut per-action latency from 2-5 seconds to under 500ms and token cost by an order of magnitude.
The accessibility tree is the agent's interface. It always was. Nobody designed it that way.
The infrastructure is degrading
The timing makes this worse. WebAIM's Million 2026 study found 95.9% of home pages had WCAG failures, up from 94.8% in 2025. 56.1 detected errors per page, a 10.1% increase. DOM elements per page rose 22.5% in a single year, to 1,437.
WebAIM attributes the rising complexity to increased reliance on third-party frameworks and AI-assisted coding practices. The same AI tools generating more code, faster, are degrading the structure that agents need to read that code's output.
The constraint that clarifies
Nielsen Norman Group identified the shift: the assumption that "user" means "human" needs updating. And the incentive follows. Investing in accessibility has always been the right thing to do, but now there's a clear business case for it, because the accessibility tree is how the majority of your traffic interacts with your product.
This is constraint inversion. Adding "an AI agent must be able to operate this interface" as a design requirement doesn't restrict the product. It clarifies it. The same way adding "a screen reader must be able to navigate this page" forced developers to write semantic structure that benefited every user.
Google's Natively Adaptive Interfaces documentation invokes the curb-cut effect: features designed for people with disabilities benefit everyone, the same way wheelchair ramps serve parents with strollers. TestParty's research quantifies the pattern. 80% of people who use captions are not deaf or hard of hearing. The voice recognition market exceeds $30 billion annually, built on technology originally developed for users who couldn't operate keyboards.
Affora's research confirmed that substantial visual variation remains possible when semantic meaning is preserved. Designing for agents doesn't compromise visual design freedom. It constrains the semantic layer, and that constraint does the thinking that sloppy code skips.
The second user test
As Johnny Butler wrote: "A chatbot is just a new way to talk to the old interface. An agent-accessible app is something an agent can actually use."
Every product now has a second user. You can treat that user as a cost: a separate API layer bolted on after the fact. Or you can treat the second user as a constraint that the product was missing.
The products that will feel best to use in two years are the ones that added the second user as a design constraint this year. Not because agents matter more than humans. Because the constraint of serving both forced the clarity that serving one never demanded.
.webp)
.webp)
.webp)
