What is agent perception?

What is agent perception?

When a human opens a web page, they see a rendered layout: headings, spacing, a nav bar, maybe an illustration. An agent almost never sees that. It sees whatever its tools return: the raw HTML a fetch tool downloaded, the Markdown a browser-to-text converter produced, or a JSON blob from an API. That gap between what you designed and what the agent actually receives is agent perception, and it’s the first thing to get right in Agent Experience Design.

A page can look great to a human and be nearly unreadable to an agent — content buried behind a client-side render with no server-rendered fallback, a critical instruction that only exists as alt text on an icon, a table whose meaning depends on visual alignment rather than markup. None of that survives the trip through most agent tooling.

Designing for agent perception means asking, for every piece of information: if this arrives as plain text with no styling and no visual context, does it still mean what I intended?