I found a bug on my own site last week that I'd have missed any year before this one: the blog listing page rendered beautifully in every browser, and contained precisely zero links to my blog posts in its HTML. Everything was fetched client-side. A human saw a working page. Google's cached view saw a dead end. And an AI agent, the newest visitor to arrive on every website, saw nothing at all. Here's the thing that's changed: agents don't browse like search crawlers did. The crawlers of the last decade indexed your pages into a giant database. The agents of 2026 are operators: they arrive with a task, read the page like a visitor would, click the links that exist, and leave. No rendering pipeline, no JavaScript execution, no benefit of the doubt. If the link isn't in the HTML, the page doesn't exist.

The new accessibility checklist

The good news is this overlaps almost perfectly with old-fashioned good engineering:

• Server-render your critical content. If a page's purpose is to link to things, those links must be in the response body. Client-side fetches are invisible to the eyesless visitor. This is the exact bug I found: the fix was rebuilding so the listing server-renders, which it should have done all along.
• Semantic HTML is agent UX. <article>, <nav>, <time>, real heading hierarchy. An agent navigating by structure does with your page what a screen reader does, and both reward the same discipline. Accessibility for humans and for machines turns out to be the same work.
• A complete, current sitemap. Your sitemap.xml is the closest thing an agent gets to a map of your intentions. Mine was the only reason anyone could find those posts at all. If yours is stale or partial, you're publishing a book with missing chapters.
• Machine-readable summaries. Structured data, clean meta descriptions, and emerging conventions like llms.txt give agents (and their humans) an honest abstract of what you offer. Think of it as metadata for a visitor that will never scroll.

The mindset shift

For twenty years, "works in the browser" was the whole job description. Now there's a second audience that consumes your site through its HTML like a text file, and it's growing faster than your human traffic. The sites that win its attention are the ones that are honest about what they contain, which is, conveniently, the oldest SEO advice there is.

A quieter take

There's a pleasing irony here. I've spent the last few years putting AI agents into businesses, and now I'm optimising my own site to be readable by them. The lesson generalises: when you build for agents, you build for the machines your future visitors delegate to. The web used to be documents for people, then apps for people. It's becoming both of those plus a landscape that machines traverse on people's behalf. Your newest visitor doesn't have eyes. Write your HTML like it matters, because it's how they'll all see you.