How to optimise your business for agent experience (AX), not just AI visibility

Imagine asking AI to book a holiday for you based on your brief and your budget. Not just suggest destinations. Actually check availability, compare prices, and complete the booking for you.

That's not far off. 74% of consumers say they'd let an AI agent handle routine tasks on their behalf, things like renewals, complaints and reorders, according to Accenture's 2026 Consumer Pulse Research. We're moving from asking AI to research something towards asking it to actually do something.

The gap in AI visibility

Most of the AI visibility conversation right now is about being found. Does ChatGPT cite you. Does Google's AI Overview mention you. Do you show up when someone asks an AI a question about your industry. We've written before about why that measurement matters, and it does. But being found is only the first half of the problem. The second half is what happens next. If an AI system finds your business, can it actually execute a task with what it finds.

Computer use agents are already here

There's a name for this pattern too: computer use agents, AI systems that operate software the way a person would, by clicking, typing and navigating a real browser, rather than working through an API. Coding agents like OpenAI's Codex are already being set loose to click through an app, test it, find where something breaks, and fix it themselves.

Both of the two biggest AI labs now ship a version of this for your own browser. ChatGPT for Chrome, launched by OpenAI in May 2026, and Claude in Chrome, Anthropic's own extension, both let an AI agent act inside your real, signed-in browser session. Not read a cached version of your site. Click through it, fill in a form, navigate between tabs, and complete a task, live.

Whether it's Anthropic's Claude or OpenAI's ChatGPT doing it, the direction is the same. Agents are moving from reading your site to operating it.

The industry is starting to call this agent experience

The industry is starting to call the business side of this agent experience, or AX. It sits next to AI visibility, which is about whether AI can find and cite you. AX asks a different question: can it execute a task for you.

It sits alongside user experience too, which is about how a person experiences your site. AX asks how an AI agent experiences it, and whether it can act on your behalf once it's there.

What agent experience adds

Agent experience adds to the four layers we already examine at Xyra: discoverability, entity clarity, content extractability and decision journey readiness. We check whether you can be found and cited accurately and whether someone who lands on your website can act on it.

Agent experience tests that last step more deeply. Can an AI complete the task on your behalf rather than merely find where to start? The holiday booking example is the clearest version of this: can it find the hotel you want, check the price, and complete the booking?

The critical leap: can AI actually execute a task

This is genuinely difficult and sets a much higher bar than being found. It requires a site to be structured for agent usability as well as discoverability.

It's one thing for an AI system to read your homepage and summarise what you do. It's another for it to work out your pricing, find your contact details, or walk through your onboarding process accurately enough to explain it to someone considering working with you. Pricing is sometimes hidden behind a form. Services are described in marketing language rather than plain structure. Contact paths assume a human is scanning the page with their eyes, not a system trying to extract a fact and act on it.

There's no standard metric for this yet, and that's not an oversight. Task completion by AI agents is still new enough that nobody has agreed on what a reliable measure looks like. Anyone claiming a definitive score for it right now is ahead of where the field actually is.

Why task-based testing is the right approach

The only credible way to assess this today is through task-based testing, treating AI like a user and evaluating whether it can complete real-world scenarios. It's the same approach usability testers have always used with unfamiliar systems: give it a task and watch what happens.

Practical questions worth testing on your own site:

• Can it book a consultation slot, not just find your calendar link

• Can it complete an enquiry form on your behalf, not just locate it

• Can it add the right service to a quote and get you a price, not just find your pricing page

• Can it finish a checkout, not just find the product

• Can it compare two of your packages and pick the right one for a stated need, then act on it

Score each one simply. Did it succeed, did it partially succeed, or did it fail.

How we test for it

We're not introducing an AX score, and we'd be wary of anyone who claims to have cracked one this early. What we do is test whether AI systems can accurately understand, describe and execute common customer journeys on your site, the same way we'd test any other part of the decision journey. It's grounded in what actually happens when you ask an AI system to do something with your business, not a theoretical checklist.

That shift, from optimisation to usability, is going to define the next phase of AI-driven growth. Run a quick AI Visibility Scan to see where your site stands today, or get in touch if you want to know whether AI can actually execute a task on your site, not just find it.

---

Frequently asked questions

What is agent experience (AX)?

Agent experience is how well an AI system can understand, navigate and execute a task on a website on a person's behalf, not just read and summarise it. It's an emerging term used alongside AI visibility to describe the practical side of AI readiness.

How is agent experience different from AI visibility?

AI visibility measures whether AI systems can find and cite your business. Agent experience goes further and asks whether AI can actually complete a task once it gets there, like finding your pricing or working out how to book with you.

What are computer use agents?

Computer use agents are AI systems that operate software the way a person would, by clicking, typing and navigating, rather than working through an API. Codex for Chrome and Claude in Chrome are both examples, letting an AI agent act inside a real, signed-in browser session.

Is there a standard way to measure agent experience?

Not yet. Task completion by AI agents is too new for an agreed metric to exist. The more reliable approach right now is task-based testing, similar to a usability test, rather than relying on a single score.

What kind of tasks should I test on my own website?

Try asking an AI system to book a consultation, complete an enquiry form, get a quote, or finish a checkout. See how far it actually gets, not just whether it finds the right page.

Does improving AI visibility also improve agent experience?

It helps, but it isn't the same thing. A site can be visible to AI systems and still fail at task completion if pricing, contact details or processes aren't structured for an AI to act on.