Adobe Principal Scientist Carlos Sanchez demonstrated an "agentic site" at the AI Engineer World’s Fair that assembles personalized web pages in real time based on a visitor's intent. The system interprets browsing signals and search queries to retrieve relevant content from a company's existing corpus, composing a unique page for each user rather than selecting from predefined segments.
- The technology generates a full page in one to two seconds at an estimated inference cost of one to two cents per page.
- It uses large language models to reorganize copy and product selection around specific intents, such as outdoor camping or conference research.
- Adobe is currently presenting the concept to customers for experimentation rather than deploying it broadly.
The approach aims to create an "audience of one" by dynamically tailoring experiences to individual users and their agents, potentially increasing conversion rates in commerce applications.