Summary
Carlos Sanchez, Principal Scientist at Adobe working on Adobe Experience Manager (AEM), demonstrates ‘Agentic Sites’ — a system that personalizes website content in real time based on inferred user intent. Rather than generating entire pages, the system selects and customizes modular content blocks (hero cards, product listings, navigation, calls to action) while keeping all generated content grounded in the existing site corpus via RAG, a design choice driven by brand-guideline compliance requirements.
The technical stack pairs AEM’s edge delivery service for content with Cerebras for fast inference, a requirement Sanchez emphasizes on screen: personalization must complete in under one to two seconds or the UX benefit evaporates. Model selection is treated as a continuous evaluation problem using Promptfoo, which benchmarks multiple providers (including Amazon Bedrock) against 15 accuracy prompts and latency targets — and the winning model varies by site type. Marketers define personalization strategies in natural language, which feeds an analytics-driven improvement loop.
Sanchez also shows a ‘For You’ recommendation page that pre-generates suggestions as a user browses, grouping query signals into persona clusters so that marketing teams can control how many intent segments to support. The talk includes a live demo and candid discussion of cost tradeoffs from multiple LLM calls per page view. It is a practical look at how a mature enterprise content platform is integrating AI personalization into real brand deployments without sacrificing editorial control.
📺 Source: AI Engineer · Published August 29, 2026
🏷️ Format: Hands On Build
