# Agentic Sites: Building Hyper Personalized Websites — Carlos Sanchez, Adobe

## Executive summary

The presentation details 'Agentic Sites,' a system for building hyper-personalized websites that adapts content in real time based on user intent and browsing behavior. Instead of generating entire pages, the architecture focuses on customizing specific content blocks (e.g., hero sections, product lists) while grounding all generated content within the site's existing corpus to maintain strict brand guidelines. Key technical advancements include continuous model evaluation across multiple providers for both accuracy and speed, achieving impressive generation latencies as low as 1.1 seconds.

## Key takeaways

- Hyper-Personalization (Audience of One): The goal is to personalize pages in real time based on the user's intent or actions, moving beyond one-size-fits-all web experiences. This drives higher engagement and conversions.
- Block-Level Personalization: Personalization occurs at the block level (e.g., hero card, product list) rather than generating entire pages, which is crucial for adhering to strict brand guidelines and preventing hallucinations.
- Speed is Critical for Conversion: The system emphasizes speed, noting that a page generation time of 1.1 seconds was significantly faster than the runner-up's 4.6 seconds, demonstrating that low latency is critical to user experience and conversion rates.

## Technical details

- Architecture & Delivery: The stack uses AMH delivery for edge content serving, powered by a back end that utilizes LLM services (e.g., Cerebras) for personalization logic. The system is composed of dynamic front ends and customizable blocks.
- Content Grounding & RAG: The entire site acts as the corpus, and generation is grounded on this existing content to ensure accuracy and compliance with brand guidelines. This allows for automatic authoring by using AI for multiple layers of personalization.
- Model Evaluation Pipeline: A continuous evaluation process runs across a variety of models and providers, assessing both accuracy and speed (latency). The tool Promptfoo is used to evaluate prompts against multiple local and OpenAI-compatible models.
- Performance Metrics: The system demonstrated an average page generation latency of 1.1 seconds using the Gemma model running on Cerebras, compared to a competitor's 4.6 seconds. The LLM time was measured at approximately 1 second.

## Practical implications

- Marketers can define personalization strategies using natural language and use analytics to drive the continuous loop of adaptation.
- The system supports automatic authoring by generating multiple site variations without manual creation.
- Content can be pre-generated (e.g., 'For You' recommendation pages) based on gathered user signals, optimizing for speed when the page is requested.

## Topics

AI, Hyper-Personalization, Web Development, LLMs, Edge Computing, Build Engineering, Adobe Experience Manager (AEM), AMH delivery, Cerebras / Gemma 4 model, Promptfoo

Source: https://www.youtube.com/watch?v=jebp4V0vh30