Rebuilding MPACER with Drupal 11: AI-Assisted Delivery, Human-Reviewed Work.

The rebuild required a fresh look at information architecture rather than a reproduction of the old website. MPACER needed a structure that could present its services clearly and support the content being prepared. That meant resolving how pages should be organised before committing their presentation to templates.

Approved designs also needed to become consistent, responsive pages. Desktop and mobile layouts had to share a coherent visual language while keeping content editable in Drupal. Hardcoding page copy into templates would have undermined that requirement. Authors needed straightforward forms and rich-text tools within approved layouts, supported by editorial governance.

A further challenge was evaluating AI assistance without surrendering control over architecture or publishing. Generating code and content was only part of the task. The implementation needed clear structures, review points and safeguards so that human expertise could shape the work and external assistants could contribute unpublished drafts for review.

Image

Conceptual illustration of the project flow, from requirements to editorial preparation.

Start with requirements and design

Requirements and a revised information architecture informed low-fidelity and high-fidelity Penpot designs for desktop and mobile. This established a basis for implementation that connected page structure with the business purpose of the website. The development task was to translate those decisions into editable Drupal pages with consistent presentation.

A custom theme uses Single Directory Components, shared design tokens, typography and native CSS. Reusable presentation components and shared visual decisions support consistency across pages. The significance is practical: changes to presentation can be managed through a coherent theme structure while content remains under Drupal’s editorial control.

Refine the content architecture

An initial shared Landing Page model used layout-dependent fields and conditional code. Dedicated page content types replaced it as the architecture developed. Existing content was preserved during the transition, migrated, and the obsolete content and fields were then removed. This refinement made the content structures more closely reflect the pages they serve, rather than continuing to accommodate different layouts through a shared model.

Articles and Case studies use structured fields, taxonomy-based authors and topics, lead images and captions. These choices give editors defined places for information and allow presentation to draw from predictable content structures. They also provide external assistants with explicit fields to populate instead of treating every entry as an undifferentiated body of text.

Make authoring and review work together

Markdown-based editing was evaluated before CKEditor 5 was selected for human rich-text authoring. Drupal Media Library selection and native inline Media embeds better matched the editorial requirements. This decision centred the author’s actual workflow: preparing text and selecting media through Drupal’s editing interface.

Author and Publisher roles, moderation and workflow buttons support human review and publishing. AI-assisted content creation fits into that process as draft preparation. It does not remove the editorial decision about whether material is ready to publish.

Connect assistants through controlled capabilities

Authenticated MCP/OAuth and custom Tool API capabilities enable external assistants to create and edit unpublished Article and Case study drafts. The implementation includes revision checks, retry protection and taxonomy handling, alongside Media import, lead-image attachment, inline Media embeds and structured closing CTA fields. These capabilities let assistants work with Drupal’s content model and media integration while retaining a human publishing step.

Drupal-specific agent skills and project instructions provide development context. The AI coding agent helped interpret requirements, prepare plans, implement substantial code and configuration, investigate problems and perform focused verification. Playwright CLI is available for focused browser checks. Its availability should not be confused with completed launch acceptance testing.

Hanmant’s involvement extended throughout this work. He reviewed implementation, anticipated downstream issues and directed architectural and editorial refinements. The shift in page modelling and the choice of authoring tools illustrate why experienced judgment matters: a working implementation must also fit future editing and maintenance needs. Expert review actively shaped the solution as it developed.

Prepare operational foundations

Webform handles enquiries and stores submissions for review; email delivery integration is deferred. Metadata, canonical URLs, sharing metadata, XML sitemap, robots policy and structured data were configured as part of the completed implementation. Production-host validation remains part of launch preparation. Configuration establishes capabilities to validate, rather than evidence of search visibility or a completed production release.

Delivered capabilities

The completed rebuild delivered responsive public-page templates and reusable presentation components. Page-specific content structures and editorial forms support content preparation. External AI-assisted draft creation, editing and media integration were exercised, with human review and publishing control retained in the workflow.

These implementation outcomes provide a practical foundation for organising content, maintaining consistent presentation and bringing external assistance into drafting. Authors can work within approved structures, while publishing decisions remain with people. The results described here are delivered capabilities; commercial returns and performance improvements have not been measured.

Completion and launch scope

The rebuild implementation is completed. Launch content preparation, production hosting/DevOps and production-host validation remain separate launch activities. Email delivery integration remains deferred. Completion of the implementation does not imply a production launch or completed launch acceptance testing.

The project demonstrates an approach in which AI assistance can accelerate implementation and repeatable work, while expert review supports precision, quality and maintainability. For MPACER, that approach took shape through actual code, configuration and editorial decisions. The delivered foundation supports the goal of presenting services effectively through a maintainable website. Its business impact remains to be assessed after launch.