Building Drupal AI as Vardot's Full-Time Contributor

I'm an engineer by education, a Drupal developer by profession, and a gamer by passion. My childhood passion for gaming is what drove me into AI, as gaming was one of the major early drivers of AI.

When Vardot joined the Drupal AI Initiative as a Gold Sponsor, the company committed one full-time contributor to the work. That contributor is me. For the last several months, every hour of my workweek has gone into one thing: building AI into Drupal alongside developers from more than 30 partner organizations around the world.

This is what that experience has actually looked like.

What it's Like To Be Part of Building Drupal AI Initiative

On any given month, more than 50 developers are actively contributing to the initiative across companies, including 1xINTERNET, Acquia, FreelyGive, Salsa Digital, Pantheon, QED42, Kalamuna, and Tag1 Consulting.

We don't all sit in the same room. We coordinate through online channels and a shared resource-planning sheet that tracks who is doing what each month.

What surprised me most was how different the working styles are across companies. We discuss, we negotiate, and the code that ends up shipping is usually better for it.

What I Built Inside the Drupal AI Initiative

My contribution record on drupal.org now shows more than 86 credits across the Drupal AI ecosystem work spread across the AI core module, AI Agents, AI File to Text, AI Logging, Document Loader, and the Context Control Center. The biggest single piece is the Document Loader module.

Document Loader gives Drupal a consistent, plugin-based way to ingest documents from any source: PDFs, CSV files, Word documents, spreadsheets, HTML pages, APIs, and Parquet files.

Before this module existed, every developer who needed to load documents into Drupal for AI had to write their own ingestion logic. That meant duplicated effort across the ecosystem and inconsistent results when AI agents tried to read the data.

Document Loader 2.0.0 shipped on March 19, 2026. I worked on the release alongside Nick Opris and Rob Loach (Kalamuna), building on architecture work that started inside the Context Control Center effort led by Marcus Johansson at FreelyGive. I'm now a co-maintainer of Document Loader on drupal.org, along with AI File to Text, AI Agents Debugger, and Unstructured.

On the AI core module itself, my record shows 27 contribution credits in the last year. A lot of that work falls into three buckets: guardrails (input-length limits to defend against denial-of-wallet attacks, multi-guardrail-set support on InputInterface, schema fixes), agent skills and Drush generators (scaffolding for AI Provider plugins, Field Widget Action plugins, API Explorer plugins, Automator Type plugins, and Guardrail plugins), and editor integration (MDX editor event support and the tooltip component for the AI Dashboard). AI 1.4.0 pulled most of this together: streaming-aware guardrails, structured-content validators, and improvements to the entity reference automator.

Numbers aside, the part I'm proudest of is that Document Loader unblocked work on several other modules that were waiting on it. When you're building infrastructure, the success metric isn't your own feature shipping. It's how many other people can finally ship because of what you built.

Image

How My Vardot Background Helped

Four years at Vardot is what made this work possible.

Vardot's engineering culture is built around high standards and heavy automation. Every change goes through a structured review process. It's the kind of discipline that feels slow on a Monday and pays off on a Thursday when something in production doesn't break.

When I joined the Drupal AI Initiative, I started working alongside senior developers from around the world. The thing that kept me steady was the muscle memory from those four years at Vardot and the fact that Vardot is a Drupal Diamond Certified Partner with that level of practice behind every line of code we ship.

What I gained from the initiative, on top of what I already had, is hands-on experience designing and building AI agents, end-to-end guardrails, plugin scaffolding, document ingestion, and automator integration. It's a skill set I cherish and continue to develop.

Why Drupal Works Well for AI

Drupal is object-oriented and structured. When I write an AI agent that needs to read or modify content, I'm not writing prompts that try to wrestle blob text into a usable shape. I'm working with entities, fields, and a permission system that already exists. The agent operates inside the guardrails the platform provides, not guardrails I have to invent.

The other thing I notice every day: AI in Drupal is implemented as a first-class entity, not a feature bolted on top of the CMS. The AI module ecosystem is designed the way Drupal designs everything else. That's why it works.

This is what people mean when they say Drupal AI is production-ready. It's not that it's perfect. It's that the foundation is solid enough that you can put it in front of real editors at real organizations without being afraid of what happens next. It's also why those same modules can ship inside Varbase, our Drupal distribution, and behave predictably from day one.

AI From a Developer's Standpoint

AI is an incredible tool, and it has changed how I work. I write code faster. I debug faster. I ship more in a day than I used to ship in a week.

But the work of building real software is still detailed, structured, and meticulous in a way AI can't fully match yet. Every pull request still goes through human review. AI accelerates the work, but it doesn't replace the engineer doing it.

What I'm Working Toward

I set goals the same way I do as a gamer. Pick the next milestone, make it specific, work toward it, and don't stop until it's hit.

Right now, my goal is to make Vardot's contribution to the Drupal AI Initiative a real, visible part of our track record in the Drupal community, and to see Drupal AI itself reach its full potential as the leading open-source AI-powered CMS.

I'm already past the first checkpoint. The modules I co-maintain are downloaded and used. The 86 credits on my profile keep going up. The next checkpoint is the next release. And the one after that.