Adobe demonstrates agentic websites that assemble pages in real time based on user intent

Adobe Principal Scientist Carlos Sanchez demonstrated a prototype system at the AI Engineer World’s Fair that interprets a visitor’s intent and composes a personalized web page in real time. The system groups signals such as browsing behavior and search queries into intent categories like exploring, researching, or preparing to purchase, then uses an LLM to assemble a page suited to that intent.

In one example, a visitor interested in camping received a version of a coffee-machine site whose copy, product selection, and supporting content were reorganized around making coffee outdoors. In another, a user querying “Europe AI conferences” received a page composed specifically for that request.

Adobe calls this approach an “audience of one,” grounding the generated page in the company’s existing content rather than inventing an experience from scratch. The company evaluates models for both accuracy and speed, targeting generation latency of one to two seconds and estimating current inference costs at one to two cents per page.

Sanchez said Adobe is presenting the concept to customers and seeking organizations willing to experiment, with commerce as an obvious initial use case. He acknowledged uncertainty about whether agentic sites will become widespread, noting that with AI it is easy to build but hard to know what to build.

Sanchez also discussed the need for websites to evolve for both human visitors and AI agents, as personal agents may soon carry richer expressions of user preferences than traditional signals like cookies. He suggested websites may need to support different levels of delegation, from autonomous reordering to interactive product inspection.

The prototype is part of a broader industry exploration of AI-driven web experiences, including chat interfaces, generative UI, and structured content standards like WebMCP, as businesses seek to connect with users across third-party AI platforms.