Adobe is building websites that assemble themselves around each visitor, in real time, from a company’s existing content. At the AI Engineer World’s Fair in San Francisco, Adobe Principal Scientist Carlos Sanchez demonstrated what the company calls an “agentic site,” as reported by Latent Space. The system reads a visitor’s intent, pulls relevant material from the company’s own content, and composes a personalized page on the fly. Adobe calls the approach an “audience of one.”
What stands out here is the timing. Sanchez insists this isn’t a someday concept. “Many people don’t even think it’s possible to generate a web page on the fly,” he told Latent Space. “People think it is future-looking. No, you can do this. It’s not the future, it’s the present now.”
What was demoed
Sanchez showed a site that treats browsing behavior and search queries as signals. The system sorts those signals into an intent category, then uses an LLM to build a page to match:
- Intent buckets: exploring, researching, or preparing to purchase.
- A camping example: a visitor interested in the outdoors got a coffee-machine site whose copy, product picks, and supporting content were reorganized around making coffee outdoors.
- Open-ended queries: type something like “Europe AI conferences” and get a page composed specifically around that request.
The key design choice: the site’s existing content is the grounding corpus. Adobe’s system retrieves from that material instead of asking a model to invent an experience from scratch. That keeps output tied to real products and copy, not hallucinated filler.
Why this matters
Personalization has been a website goal for decades, but it’s mostly meant picking from a predefined menu. A retailer recommends an item based on a past purchase, or drops you into one of a few audience segments. That’s been the ceiling. An agentic site raises it: the page itself becomes the variable, not just a recommendation slot inside a fixed template.
For AI engineers, two practical numbers matter:
- Latency. Adobe evaluates models for speed, not just accuracy. “We don’t want the site generation to take more than one or two seconds,” Sanchez said.
- Cost. He put current inference at “one to two cents per page,” and expects it to drop. “This is where we are today. In six months, who knows where we’re going to be.”
The honest caveats
Adobe hasn’t broadly deployed this on production customer sites. The company is pitching the concept and hunting for organizations willing to experiment. Commerce is the obvious first use case, since personalization ties directly to conversion, but Sanchez says it could fit “anything that needs more conversion and has a big matrix of user types or personas.”
He’s also candid about the uncertainty. “With AI, it’s very easy to build things, but it’s hard to know what to build,” he said. “We build things and then we find the customers.” Adobe isn’t alone in that fog. Site owners are weighing chat interfaces, structured content like WebMCP, generative UI, and personal agents all at once, plus figuring out how to pull users in from third-party AI platforms.
The bigger shift: two kinds of visitor
Here’s the part worth sitting with. Websites won’t only be personalized for humans. As personal agents get more capable, a user may hand off research or purchases entirely, and the agent could arrive carrying a far richer read of that user’s preferences than cookies ever could.
Sanchez expects sites to serve both. A personal agent might quietly reorder toilet paper, while a person buying a jacket still wants to inspect it and click buy through a visual interface. So sites will need to support different levels of delegation, not treat “agentic commerce” as one pattern. Technologies like WebMCP could expose structured tools to an agent, while MCP Apps could bring product experiences into the user’s chat. An A2A backend might let agents transact without ever loading the visual site.
“That’s still what everybody’s trying to figure out,” Sanchez said. “But there’s going to be agentic targeting, for sure.”
The traditional website probably won’t vanish. Its job is what changes. If you build or run web experiences, the move now is to get your content clean and retrievable, because both your human visitors and their agents will soon expect a page built for them. More detail is available at the original Latent Space report.