When:
Wednesday, September 30, 2026 - 15:00 to 15:45 CEST
Room:
Van Oldenbarnevelt Room
Tags:
development & coding, ai
Track:
SVG
m&b icon_new brand
development, ai & agentic architecture

SDCs, Canvas, and the Agent That Builds With Them

SDCs, Canvas, and the Agent That Builds With Them

Matt Glaman (mglaman)

You already know how to build an SDC. Now what if your agent knew how to build one too — then dropped it into a page, verified it rendered correctly in the browser, and asked you to review the diff? This session walks through a page-building workflow driven by an AI agent that speaks your component system.

Prerequisite

Familiarity with Drupal theme development
Working knowledge of SDCs (structure, YAML schema, Twig templates)
Basic comfort with CLI tools; no prior AI agent experience required

Target Audience

Drupal theme and front-end developers

Outline

SDCs are a natural primitive for AI-assisted generation: self-contained, schema-driven, and with a predictable file layout. Drupal Canvas uses them as the building blocks for page composition. MCP tooling gives an agent direct read and write access for content creation and site building. Browser automation closes the loop by verifying rendered output without manual testing.

Why the SDC file layout maps cleanly to what an agent can generate reliably, including a skill that produces a complete SDC bundle (Twig, YAML, CSS) from a natural language description.

How an MCP server lets an agent read, create, and update content directly and where it needs guardrails.

Using a browser agent to catch layout regressions without writing a single test.

A live demo threading SDC generation, MCP page assembly, and browser verification, including where the workflow breaks down and what you still own as a developer.

Learning Objectives

Generate production-ready SDC file bundles using an AI agent skill tuned for Drupal's component format.
Understand how MCP tooling gives an agent direct access to Drupal's page management layer.
Apply browser automation to verify rendered output during the component development loop.
Identify where human judgment remains essential in an agent-assisted front-end workflow.

Experience level
Anyone