Building a Web That Works for Everyone: Why AI Improvements Can't Leave Humans Behind

markdown formatted blog content

The web is at an interesting crossroads. Every week brings new proposals, APIs, and standards designed to help AI agents navigate, understand, and interact with websites. It's a land grab of sorts—everyone wants to shape what the AI-powered web looks like.

But here's the thing: the web wasn't built for AI. It was built for humans.

Last week, the Safari team dropped some wisdom that should be required reading for anyone proposing changes to how websites work for AI agents. Their core argument? If something is hard for an AI to use, that's not an AI problem—it's a web standards problem.

"When a site's actions are hard for an agent to use, that is a gap in the page's own semantics, and the fix should be to close it in the platform's shared layers (HTML and ARIA), where the user, assistive technology, and agents all benefit."

This is the kind of thinking that built a resilient, accessible web in the first place. The same principles that make sites work for screen readers and keyboard navigation can make them work for AI agents. We're not talking about building separate tracks here—we're talking about building better foundations.

The Accessibility Parallel

Here's what really strikes me about the Safari team's argument: they frame AI agents as a form of assistive technology. Think about that for a second.

An AI that books your flights, fills out forms, or completes transactions on your behalf is essentially an extension of you—the user. It should interact with sites exactly as you would. And if sites need special accommodations to work with these agents, those accommodations should benefit human users too.

This isn't hypothetical. There's currently discussion in web standards circles about creating machine-readable ways for websites to declare their identity and authority. The goal? Help AI understand which sources are trustworthy.

But here's the irony: humans struggle with this exact problem. We all know the feeling of landing on a site and wondering, "Is this legit?" If we're solving this for AI, shouldn't we solve it for humans too?

The W3C's Priority of Constituencies

The W3C has a guiding principle that should inform every standards discussion:

"If a trade-off needs to be made, always put user needs above all."

It continues: "User needs come before the needs of web page authors, which come before the needs of user agent implementors, which come before the needs of specification writers."

Notice AI agents aren't mentioned. But if they're going to stick around—and let's be real, they are—we need to figure out where they fit in this hierarchy.

My take? User needs come first, and AI agents exist to serve those user needs. They're a means to an end, not the end itself.

What This Means for Your Website

If you're building or hosting a website today, this debate might seem distant. But here's the practical takeaway: good web development practices are more important than ever.

Semantic HTML isn't just "nice to have" anymore. It's the foundation that lets everyone—humans, assistive technology, and yes, AI agents—understand and use your site. Proper ARIA implementation, clear navigation, machine-readable metadata—these aren't just accessibility wins. They're future-proofing your web presence.

When you register a domain, set up hosting, or configure DNS records, you're not just building technical infrastructure. You're building a digital home that should welcome all visitors, human and artificial alike.

The Bottom Line

We should be skeptical of any proposal that makes the web better for AI but worse for humans. The web is a public resource, and it should serve the public—developers, businesses, and end users alike.

Before we rush to add "AI training wheels" to the web, let's ask: Does this help everyone? Can we solve this problem at the standards layer? Are we building something that makes the web more accessible, or are we creating another fork in the road?

The Safari team's response to WebMCP wasn't just about one API. It was a reminder of what makes the web work: shared standards that serve everyone.

Let's build a web worth living in—for all its inhabitants.