Last night I took the train into Manhattan for Pints, Prompts & Page Builders, an open source AI meetup at Automattic’s NoHo office. Evolving Web hosted the evening as a lead-up to EvolveDigital NYC 2026, coming November 19–20 to Civic Hall in Union Square. The premise was simple and a little chaotic: three presenters, three AI page builders, ten minutes each, one landing page.
WordPress folks, Drupal folks, and at least one attendee from Australia packed the room. Pizza disappeared fast. Then the demos began.
The AI landing page showdown
Each presenter built a landing page for the EvolveDigital summit itself. Same brief, wildly different approaches.
Acquia Source + Drupal Canvas
Dimitriy Kaplan, a solutions engineer at Acquia, showed a page built entirely through conversation with Grace, Acquia’s site-building AI. He presented alongside Alberto Andrade, Acquia’s senior director of solutions engineering. Dimitriy fed Grace content types like sessions and speakers, suggested a few fields, approved her plan, and got a full landing page in about 35 minutes. This wasn’t a static one-off, either. Every section lived in the CMS backend as structured content.
He then jumped into Drupal Canvas, a point-and-click page builder where humans and AI share the same component library. He prompted a new stat-band component, previewed it, approved it, and dragged it onto the page. His closing point landed with me: Drupal spent 25 years gated behind technical expertise, and that gate just came down.
Elementor AI
Anthony Bradshaw of Evolving Web ran a true live build with Elementor’s brand-new AI builder — his first time using it on stage, which takes nerve. The tool can’t browse live sites yet, so he fed it screenshots of the real EvolveDigital site to replicate. Some sections generated live; others he wisely pre-baked, because demo gods.
Two things stood out. First, everything the AI produces stays editable as regular Elementor components. Second, the inline AI offered three design variations per request without being asked. Responsiveness needed work, but for a product this new, the bones looked promising.
WordPress + Gutenberg + MCP
James Giroux, a technical account manager at WordPress VIP, opened with my favorite line of the night. The actual EvolveDigital site runs on Webflow — not open source — so he “liberated it to WordPress” before the event.
His demo connected Claude to WordPress through MCP, built on the new Abilities API. The Abilities API gives WordPress a standard format for defining tools that MCP can then expose. If you can imagine a capability, you can build it as an ability — no waiting on a platform vendor to ship it.
From one prompt, Claude pulled content from a Google Doc, mapped it to existing block patterns, and assembled a design-compliant draft page in roughly two minutes. Because the AI reused patterns already on the site, the output stayed on brand automatically. That matters enormously for enterprise governance. He capped it off with a front-end feature that answers visitor questions by generating a custom page on the fly.
When the hands went up, James’s WordPress build took the room’s vote, with Acquia’s demo a strong second — though every builder earned its applause.
The panel: AI in real content operations
John Eckman (formerly of Fueled and 10up) moderated a panel with Abhimanyu Gupta, director of engineering at the IEEE Communications Society; Richard Chiriboga, AVP of digital innovation at Smile Train; and Anne Cantera, founder of Elemental Intelligence. Their takeaways deserve their own conference track.
Your leadership owns your upskilling. Anne didn’t mince words: teams get in trouble for burning through tokens by Wednesday, yet nobody taught them spec-driven work. Handing someone a Ferrari without a driver’s license is a leadership failure, not a user failure.
Cite your sources, keep humans in the loop. Richard’s team at Smile Train works in over 75 countries, and translation became their hardest problem. Cleft-related terminology translates offensively in some languages, so native speakers review everything. Computers, it turns out, are not great at cultural sensitivity.
Don’t throw away 20 years of governance. Abhi reminded us that a CMS gives organizations accessibility, permissions, and review workflows. Jump straight to raw AI generation and you lose all of it. IEEE runs 2,400 conferences a year, so they’re rebuilding those guardrails deliberately with skill files and structured harnesses.
Most of the work happens before the first prompt. Anne’s token-saving advice: specs, ADRs, and UI kits leave the model nothing to guess at. She once let Claude interview her for eight hours to build a PRD, then compared it to a one-shot generated version. The difference was night and day.
Pilot small, review monthly. Smile Train started with a chosen cohort, legal review, and monthly check-ins before expanding. Now they’re moving to AWS Bedrock with Claude so more staff can use AI officially instead of in the shadows.
What I’m chewing on
Three demos proved the same thing from three directions: conversational site building has arrived across open source CMSs. The differentiator is no longer “can AI build a page.” It’s whether the output respects your content model, your brand system, and your governance.
John closed the panel with a challenge I keep replaying. If AI only helps us build the same mediocre websites faster and cheaper, who cares? The opportunity is building more accessible, more usable, genuinely better experiences.
I also left with a talk idea for the summit. Proposals close Friday, August 28, so if you have a case study or a project worth sharing, submit yours too.
See you at Civic Hall in November?
Leave a Reply