How do you let an AI agent change enterprise content without giving it publishing authority?
Enterprise content teams want AI to find content, coordinate updates across sites, and prepare changes for review. Connecting an agent to a CMS is only the beginning: the system must decide what it can read, what it can change, and who can authorize publication.
This demonstration-led session follows a product-message update across three regional websites using synthetic enterprise content. With Adobe Experience Manager (AEM) and the Model Context Protocol (MCP) as the technical setting, we will walk through four distinct stages: propose, validate, approve, and publish. The agent will identify relevant content, preserve regional variations, and prepare a before-and-after comparison for a human reviewer. Publishing authority will remain outside the agent's permissions.
We will deliberately introduce a product claim that is not supported by an approved product reference and an attempted unauthorized action. Attendees will see how checks against that reference and server-side permission checks catch these failures, how a reviewer can reject or revise a proposal, and how an audit record connects the request, proposed changes, validation results, and approval decision. We will distinguish instructions in a prompt from controls enforced by the application.
Attendees will learn to:
- Separate content discovery and draft preparation from approval and publication.
- Design narrowly scoped MCP tools and enforce permissions at the application boundary.
- Check product claims against an approved product reference and validate regional constraints before a reviewer approves a change.
- Test failure paths and record enough evidence to investigate or recover from an incorrect change.
The session will provide a reference architecture and a practical evaluation checklist. The demonstration uses AEM, while the design principles apply to Drupal, WordPress, and other enterprise CMS platforms.
Intended audience: CMS developers, technical architects, and engineering leads evaluating AI-assisted content operations.
Prerequisites: Familiarity with CMS workflows and basic APIs. No prior AEM or MCP experience is required; the relevant concepts will be introduced.
Format: 40-minute presentation and demonstration, followed by 5 minutes of Q&A.
Presenter background and related work:
Website: https://satishkchoudhary.com/
Related AEM/MCP presentation: Connect AI to Adobe Experience Manager with MCP | Setup & Demo — In Simple Terms with Satish
https://www.youtube.com/watch?v=U0k0nQNuzYc
This related video introduces AEM/MCP integration; the proposed session extends those concepts to enterprise content approval and permissions.