I was working on my personal blog and ran it through Google Lighthouse and I noticed something new:
Agentic Browsing.
That caught my attention.
Why would Lighthouse care whether an AI agent can browse my website? I started digging. Then I found myself fixing an accessibility issue in my own code that was causing both my Accessibility and Agentic Browsing audits to fail. That was the moment it clicked. The web isn’t about to become agentic. It already is.
And I think UI/UX is about to gain a new branch:
AIx — AI Experience
UX has a new user: AI
OpenAI describes Astra as its most capable model for computer use and browsing. Suddenly, “Can an AI agent use my website?” doesn’t sound like a future question. It sounds like a web-development requirement.
For decades, we’ve designed websites around one fundamental assumption: There is a human on the other side of the screen.
A human can look at a page and immediately understand that this is a button:
┌──────────────────────┐ │ Buy now │ └──────────────────────┘
An agent doesn’t necessarily “see” it the way we do. It needs the underlying page to communicate:
Role: button Name: Buy now Action: purchase product State: enabled
That is where I think AIx — AI Experience starts to make sense. UX optimizes the experience for humans. AIx optimizes the experience for AI agents. Not instead of UX, but alongside it.
The web is gaining a second interface
SEO exists largely to help search engines discover and understand that website. But nowadays, an increasingly common workflow looks different:
Human ↓ AI Agent ↓ Search / Browse / Compare ↓ Website ↓ Agent performs task ↓ Human gets result
The agent might be asked:
“Find me a good hotel in Tokyo next weekend.”
Or:
“Compare these three laptops.”
The website is no longer simply something the human reads. It’s something the agent operates. And that changes what “accessible” means.
Your accessibility tree is becoming an agent interface
This is one of the most interesting things about Google’s Agentic Browsing work. AI agents can use browser information, including the accessibility tree, to understand pages and interact with controls. That means many of the practices we’ve considered “accessibility work” suddenly have another consumer: The AI agent.
Consider this:
<div onclick="submitForm()"> Submit </div>
A human can probably figure out what it does. But compare it with:
<button type="submit"> Submit </button>
The second version gives the browser much more semantic information. That information can become part of the agent’s understanding of the page.
Google’s Agentic Browsing audits specifically look at things such as accessible names, roles, relationships, and whether interactive elements are properly represented in the accessibility tree.
This leads to a surprisingly simple AIx principle:
Don’t make an AI agent infer what your HTML already knows.
AIx isn’t just accessibility
Accessibility is the foundation, but I think AIx goes further. Imagine an ecommerce website. A human sees:
Nike Running Shoes $150 Size: [42 ▼] [ Add to Cart ]
The human understands the workflow almost instantly. An agent needs to understand:
Product name: Nike Running Shoes price: $150 Select name: Size options: 40, 41, 42, 43 selected: 42 Button name: Add to Cart action: add product to cart
The visual interface can stay exactly the same. But the machine interface underneath it needs to be meaningful. This is where AIx starts overlapping with:
- accessibility
- semantic HTML
- structured data
- information architecture
- API design
- browser automation
- agent tools
- WebMCP
It becomes less about how the website looks and more about whether its meaning and capabilities are exposed clearly.
I learned this from my own website
This wasn’t just a theoretical rabbit hole. While investigating the Agentic Browsing score on my personal blog, I found many accessibility problems.
My Lighthouse Agentic Browsing score went from 50 toward 100, while the Accessibility score was also addressed from 97 toward 100. The PR documents the exact changes and the tests used to verify them.
That’s the part I find fascinating. I didn’t make my website more “AI-looking”. I made it more semantically correct and that made it more usable by machines.
That’s AIx.
From clicking buttons to exposing capabilities
This is where WebMCP becomes particularly interesting. Suppose an agent wants to search an ecommerce site. Without an explicit interface, it might have to:
Find search box
↓
Click
↓
Type query
↓
Press Enter
↓
Wait
↓
Parse results
↓
Find filters
↓
Click filter
↓
Parse againThat’s browser automation. It can work, but it’s fragile. Now imagine the website exposes a capability:
search_products(query, category, maxPrice)
The agent doesn’t have to reverse-engineer the UI. The website is effectively saying:
“Here is one of the things I can do.”
That’s the idea behind WebMCP, an emerging web standard that allows websites to expose structured tools to AI agents. This is a significant conceptual change.
A website stops being just a collection of pages.
It becomes a collection of capabilities.
— Is the browser becoming the new API?
Stability matters too
Agents also benefit from predictable interfaces. A button that moves because an ad loads might be mildly annoying to a human but can cause an automated workflow to fail. This is why Google’s Agentic Browsing work also considers layout stability. Good UX and good AIx increasingly overlap.
So is Agentic Browsing the new SEO?
Here’s where I would be careful.
SEO isn’t dead.
Google’s own documentation distinguishes between agents searching the web and agents actually using websites. Traditional SEO still matters when agents need to discover your content. But once an agent reaches your website, a new question appears:
Can it actually use it?
So I think the evolution looks something like this:
SEO
│
└── Can they find me?
│
▼
AI / GEO
│
└── Can AI understand my content?
│
▼
AIx
│
└── Can an AI agent understand my interface?
│
▼
Agentic Web
│
└── Can an agent actually perform the task?> SEO optimizes discovery.
> AIx optimizes interaction.
> Agentic Web technologies optimize action.
That’s why I don’t think Agentic Browsing replaces SEO. I think it becomes the next layer.
And advertising gets very interesting
This is the part I’m most curious about. Traditional advertising is optimized for human attention.
Tomorrow, what if the decision-maker isn’t the human? Imagine asking:
“Find me a good pair of running shoes under $150.”
The agent doesn’t care about the flashy banner. It cares about price, reviews, shipping, availability, return policy, discounts — and potentially commercial incentives.
The ad isn’t competing for an eyeball anymore. It’s competing for an agent’s decision. That could create an entirely new advertising ecosystem.
The next Google Ads might not be designed to win clicks. They might be designed to win agent decisions.
There’s a dark side to AIx
If agents become users of the web, websites will eventually try to influence them. That creates a new security problem.
So AIx can’t just be about:
“How do I make agents use my website?”
It also needs to ask:
“How do I make that interaction trustworthy?”
That’s going to become a major engineering discipline of its own.
A practical AIx checklist
If you’re building a website today, you don’t need to rebuild everything.
Start with the fundamentals.
1. Make your HTML semantic
2. Make the accessibility tree meaningful
3. Keep the interface stable
4. Expose structured information
5. Identify your website’s capabilities with MCP
6. Test with Lighthouse
7. Test the actual user journey
The ultimate goal isn’t:
“Does Lighthouse give me 100?”
It’s:
“Can an agent successfully accomplish the task a human came here to accomplish?”
That’s the real AIx metric.
The web isn’t changing. It’s expanding.
I started this investigation because I wanted to improve a Lighthouse score on my personal portfolio/blog. I ended up changing how I think about web development. We’ve spent decades designing websites for humans. Then we built standards so search engines could understand them.
Now we’re building another layer so AI systems can understand and operate them. I think that’s the beginning of AIx — AI Experience.
Designing digital experiences for humans and machines at the same time.
SEO taught us to make websites discoverable. Accessibility taught us to make them usable by more people.
AIx may teach us to make them understandable and actionable for smart Agents.
And this user may not have eyes 👀
