Field note2026-09-15T15:35:35.873Z

"AI agents will replace your website." No, and here's what actually happens to it.

An AI agent shopping on someone's behalf doesn't need a website the way a person does. That's not the same as the website disappearing. Its job changes, from the primary place a sale happens to one surface among several reading from the same underlying facts.

5 min read

The Selfe team
Agentic commerce infrastructure
The website was never the target. It was always the venue an agent had no reason to visit in the first place.

The myth assumes the website is still the unit that matters

A homeware retailer hears "AI agents will replace your website" and pictures the site going dark, replaced by some chat interface nobody's built yet. That's the wrong image. A more accurate one: the site keeps running exactly as it does now, for the people who still want to browse it, while an agent buying on someone's behalf reaches the same product data through a different door entirely, one that was never a page to begin with.

"Replace the website" only makes sense if a website is the thing an agent would otherwise use. It isn't. An agent doesn't scroll, doesn't click, doesn't read a hero banner. It checks specific facts, variant, stock, price, delivery, and none of those facts require a page to exist. The website was built to persuade a browsing human. An agent was never the audience it was persuading.

What actually happens to the site

Nothing dramatic, and that's the point. People still land on product pages from search results, from links shared in messages, from browsing a category. The site keeps doing that job. What changes is that it stops being the only door. A retailer's real facts, the same ones the site displays, become directly reachable by an agent too, without the agent needing to load, parse or interpret the page at all.

The retrofit version versus the real one

The retrofit instinct is to build an AI-friendly version of the website, a lighter page, a chat overlay, something an agent can supposedly parse more easily. That's still solving the problem as a page problem. The real fix skips the page step entirely for the agent's route in, while leaving the human-facing site untouched for the people who still want it.

Where this shows up for a business

An agent buying on someone's behalf reaches a retailer's real product data directly through Selfe, the same facts the website shows, without ever needing to load the page, while the site itself keeps doing its job for every browsing human who still lands on it.

Should we stop investing in our website?

No. It's still doing real work for browsing humans. This is about adding a second door, not replacing the first one.

Do we need a special "AI version" of our site?

No, that's still a page-shaped answer to a problem that isn't about pages. Connect the underlying data directly instead.