Most AI keynotes go like a magic show: lots of sparkle, a vague promise, and nothing you can take home. At DrupalCon Rotterdam, Dries Buytaert did something unusual. He showed AI features you can try today, explained how they work, and left the "trust us, it's magic" part out entirely.
A Fake University With Real Lessons
In his DriesNote, Buytaert set out Drupal's direction around three pillars: multilingual support, JavaScript and headless capabilities, and AI. The Drupal AI Initiative now has more than 100 contributors. Its showpiece is the Drupal AI Demo: Northmoor University, a complete university website with AI running on realistic content.
The demo scans content and flags weak items for accuracy, tone, and inclusive language. It lets editors fix problems themselves without calling in a specialist. It answers visitor questions with citations drawn only from the site's own content, drafts updates that a human must review before they go live, and logs every interaction. The team also showed AI-assisted component building for Drupal Canvas, where the assistant checks its own work against Canvas Workbench, ESLint, and the Canvas CLI. One more prototype stood out: it cuts a Model Context Protocol (MCP) server integration from about 1,000 lines of code to roughly 20 PHP attributes, using the Simple OAuth, MCP Server, and Tool API modules.
Guardrails as a Feature
Notice what Drupal chose to show off: human review before publishing, answers grounded in your own content, and audit logs. That isn't the flashy "AI writes your whole website" pitch. It's the pitch for organizations whose lawyers read the terms of service, such as universities, governments, and healthcare providers. That happens to be Drupal's home turf.
The detail most people will skip is the 20-attribute MCP prototype, and it may matter most. If exposing a Drupal capability to an AI agent becomes as easy as adding an annotation, every Drupal site becomes something agents can work with. The planned Rosetta Sprint fits that idea. It's a four-day session where key maintainers will standardize how Drupal capabilities are described to AI agents, orchestration tools, and people. In other words, Drupal is building a shared vocabulary so agents don't have to guess what your site can do.
Drupal didn't promise robot overlords. It promised AI that drafts, cites its sources, and waits for a human to approve. Frankly, that's the version most of us actually want.
If the Northmoor demo has you thinking about AI-assisted editing or MCP-ready workflows on your own Drupal site, let's talk about which of these pieces fit your setup today.