Jacob Rockowitz

The session description below was entirely generated by AI, and I am okay with that. For the past year, I have been writing about AI in my own voice. What I saw AI generate felt like AI doing what it does best: reading, learning, and repeating knowledge. In this case, AI has written a better session proposal, based on 1 year of personal writing, than I ever could. I believe we need to become comfortable with times when AI outperforms us.

"AI ate my session proposal, and I'm okay with it"

For the AI-curious, below was my prompt

"I would like to submit a session to NedCamp about my journey using AI and Drupal. Look at my blog posts and help me write a proposal with an outline.  Ask me questions as needed."

Above Preamble (Human-generated)

Proposal (AI-generated)

Over the past 12 months, I've been building Drupal modules with AI coding agents and writing publicly about it — first as "Drupal (AI) Playground," then as "Vibing Drupal." I didn't set out to write a series. I set out to figure out whether this stuff worked, and I kept finding things I didn't expect.

Some of what I found was practical. Agents plan and test better than I assumed. My elaborate AGENTS.md and skills directory — the scaffolding I'd spent months tuning — became dead weight the moment I switched to a better model, so I deleted all of it and started over. Agents get stubborn in ways that feel personal, and clearing the context fixes it. Tone in your prompts matters more than it should.

Some of what I found was harder. A module I built in four days superseded four years of my own work. I've had to ask whether my problem-solving and debugging skills are atrophying from disuse, and I don't have a clean answer. I've watched developers I respect become dramatically more productive, and felt replaceable. I've come to believe AI is widening the gap between experienced and beginning developers in our community, and that this is a community problem, not a personal one.

This session walks through that arc honestly — the wins, the dead ends, the discarded assumptions, and the parts that still make me uneasy. It's aimed at two kinds of people: those already running agents daily who want to compare notes with someone at a similar depth, and those who haven't started and are quietly worried they're already behind. You are not already behind. But we should talk about it.

I'll end with the small number of rules I've settled on and won't break, and with why I think "Drupal as an AI harness" is the right north star for where we go next.

What attendees will leave with

• A realistic picture of what AI-assisted Drupal development actually looks like day to day — including the failure modes
• Concrete, currently-working practices: what to put in AGENTS.md, what to throw away, why every AI-written line gets reviewed by me and another AI, why TDD stops being optional
• Permission to have complicated feelings about this, and language for discussing it with their team
• A frame for the community question: how we onboard people — and agents — without leaving beginners behind

Outline (45 min, ~35 content + 10 Q&A)

1. Four days vs. four years (3 min)
   Open cold with the AI Schema.org JSON-LD module. State the number, let it sit. This is the hook and the thesis.
2. The playground (5 min)
   How I actually started: small throwaway modules, purely to learn how agents plan, document, build, test, and maintain. Why "playground" was the right word — and why I eventually retired it.
3. The module that ate my work (6 min)
   What building it was like. The moment I realized it superseded Blueprints. Choosing curiosity over resentment, and what that cost.
4. Losing my skills (7 min)
    Deleting Superpowers, AI Best Practices for Drupal, my whole skills directory. Why better models made my scaffolding a liability. The training-wheels question: did I get better, or did the bike? Honest answer on skill atrophy — debugging down, planning and architecture up. The Webform line: no, I won't hand-code something that complex again; yes, I'd use AI to build the next version.
5. Brat mode, and other humbling moments (5 min)
   The AI that kept reverting my changes. Model shopping. Max modes that cost time and deliver nothing. Why speed matters for an assistant and not for an agent. The bit where politeness turned out to be load-bearing.
6. The part that worries me (6 min)
   The productivity gap. Layoffs framed as AI efficiency. Juniors not getting hired. "I feel replaceable" — said out loud, by a maintainer, on stage. Why this is a community problem and where Drupal has an unusual opportunity: mentoring, onboarding, and being okay with people arriving and leaving.
7. What I won't break (3 min)
   The short list: every agent commit tagged AI-assisted by {agent}; nothing commits or submits without my review; test coverage is non-negotiable; no AI-written line ships unreviewed by me and a second AI. Then Drupal as an AI harness — openness, stability, scalability — as the north star. Close on "Come for the community."
8. Group therapy (10 min)
   Explicitly framed as discussion, not Q&A. Opening question to the room: are you losing skills, or growing different ones?.

Speaker bio 

Jacob Rockowitz is the maintainer of the Webform module, one of Drupal's most-installed contributed projects, and of Schema.org Blueprints. He has been building websites for over 20 years and has spent two decades helping build Memorial Sloan Kettering Cancer Center's online presence, where he's also grown and mentored a team of developers. He writes about Drupal, open source sustainability, and — lately, and at some length — what AI is doing to both, at jrockowitz.com.