Something changed quietly over the last two years. AI assistants stopped only answering questions and started doing things: opening websites, reading pages the way a person would, clicking through menus, comparing options, filling in forms. Every major AI company now ships some version of this, an assistant that can operate a browser on the user’s behalf. The industry calls them agents. Your future customers will just call it “I asked it to find me someone”.
Play the scene forward, because it is not hypothetical. A busy founder tells her assistant: “Find me three web designers who work with service businesses, get their prices, and send an enquiry to the best fit.” The agent visits a dozen sites. On some, it finds the services, the prices and a contact form it can complete. On others, it meets a popup it cannot dismiss, a menu that only opens on hover, prices locked inside an image, a form with unlabeled boxes. Guess which businesses end up in the shortlist, and which one gets the enquiry.
Agents fail where humans grit their teeth
Here is the useful mental model: an agent experiences your website like a first-time visitor with no patience and no ability to guess. Everything that mildly annoys a human stops an agent cold.
A button that is not really a button, just a decorated rectangle that reacts to a script. A contact form whose fields have placeholder hints but no actual labels a machine can read. A phone number that exists only inside a graphic. A chat widget covering the very button it is supposed to help find. Content that only appears after scripts run, which some agents never do. And the front door problem: some sites block the AI systems’ fetchers outright without knowing it, so the agent never even gets inside.
None of these things stop a determined human. All of them can end an agent’s visit, and the agent does not complain or retry. It just moves to the next site on its list.
The agent-ready checklist
The good news is that agent-readiness is not a technology purchase. It is mostly the discipline of building plainly.
Real links and real buttons, written as what they are, so any software can recognize them. Form fields with visible, properly attached labels: name, email, message. Your phone number and email as actual text a machine can copy, not pixels. Prices stated in text on a real page, because “contact us for pricing” filters out agents even more efficiently than it filters out humans. One obvious path to contact from every page, which you should have anyway. And the machine-readable layer describing your business underneath it all, the same one that feeds AI search.
Read that list again and notice what it also is: an accessibility checklist, and a conversion checklist. A site a screen reader can navigate is a site an agent can navigate. A form a distracted human completes in thirty seconds is a form an agent completes in two. There is no trade-off anywhere in it, no budget spent on robots at the expense of people. The same plainness serves everyone, which is the strongest kind of investment.
Where this is going
Nobody knows how fast agent-driven enquiries will grow, and I am not going to invent a projection. What I know is the direction: the AI companies are all building toward assistants that act, and the sites that fail agents today are the same sites that already lose impatient humans. Making your site operable by software is not a bet on one future. It is the same work that pays this quarter in human enquiries.
Every site I build works this way by construction: real elements, labeled forms, text you can copy, no tricks between the visitor and the contact button. If you want to know whether an agent could actually get from your homepage to a submitted enquiry, that walk-through is part of the free audit, and the result tends to surprise owners more than any speed score does.