DrupalCon Rotterdam 2026 wrap-up

How the conference responded. The Drupal CMS product lead's case for Drupal CMS (34:52–40:35) pitched governance against vibe-coding tools, and the code components lead advised starting without headless (38:28–41:32). A Canvas Tools session showed outside agents building through Canvas autosave (5:51–9:45) as the human gate. On reputation, one developer showed Drupal's desire gap in developer surveys (6:08–8:53), the marketing panel explained how advocacy credits will work (22:49–25:49), and Angie Byron (webchick) challenged everyone to go to AI meetups and not pitch Drupal (34:02–39:08), but listen first.

Mars and World Cancer Day: Canvas for non-developers, AI as the first moderator

Anthony (Mars, global web delivery) and Charles Andrew Revkin (UICC, World Cancer Day), opened by a Pantheon co-founder · Watch the keynote (1:06:33)

The end-user keynote opened with a short UN video on standardising hundreds of UN sites on Drupal (4:03–6:51). Mars then described consolidating more than ten CMSs into one Drupal platform after GDPR, and reported 400+ properties, 54 countries and a 40% lower total cost of ownership (14:02–16:27). These are the company's own figures, with no baseline shown. Mars explained why it chose Canvas while it was barely released (19:17–21:55): no licence fee, core-native, and the direction of Drupal AI. A recorded demo showed a novice editor building a page (24:41–28:38) in minutes, replacing a multi-team ticket process.

World Cancer Day's story was more nuanced. A chatbot for collecting cancer stories was dropped because it felt impersonal. When the stories arrived in volume, Drupal AI became a first-layer moderator (51:28–55:56): personal data is stripped, stories are marked approved, flagged or rejected, and humans see a reason for each decision. The lessons were decision support, not decision making (57:01–59:54), and measuring trust as well as efficiency. In a separate interview, the team said editors now review maybe 10% of stories at most (10:25–12:33). No accuracy figures were given.

What was announced or shown

• Mars is standardising new builds on Canvas; its first four sites went live in April 2026.
• World Cancer Day 2026 used Drupal AI moderation, with approved stories publishing immediately.
• Pantheon's sponsor slot pitched agentic site creation and mentioned support for the MCP module and the Drupal AI initiative.

How the conference responded. Mars and its agency went deeper in a session on how Mars chose Canvas over Layout Builder and Site Studio (10:25–15:07) and the Canvas KPIs it reports (34:22–37:27). The figures there do not fully match the keynote's, so treat them as indicative. Other case studies backed the "human decides" line, including EPSO's eight-week AI assistant rollout (5:04–7:32) and VisitEurope's tone-of-voice guard rails (18:15–20:53).

Opening: "free as in speech, not free as in beer"

DrupalCon Rotterdam hosts, Dutch Drupal Association speakers, and Tiffany, interim CEO of the Drupal Association · Watch the opening ceremony (54:15)

Most of the opening was logistics and fun, including a Dutch take on the Netherlands as a Drupal content model (22:35–25:53). The substance came from Tiffany, the Drupal Association's interim CEO. She marked the 25th anniversary, then described shrinking agency margins and deal sizes (30:35–33:18) next to a record 126 certified partners. Her central point was that somebody else has bought the round (33:18–34:21): security, release engineering and infrastructure are not free, and the same agencies keep paying. She said the DA is not regenerative (34:21–35:55), framed contribution as makers, patrons and advocates, and said the certified partner programme will become an ecosystem-wide recognition badge (35:55–39:57). What the badge will require was not stated.

What was announced or shown

• DrupalCon North America 2027 in Orlando (44:57–47:04), at Easter.
• DrupalCon Europe 2027 in Edinburgh (49:20–53:45), with a Highland cow mascot replacing Miffy.

The DA board tightens its finances and prices a European federation

Drupal Association board, with the interim CEO (ex officio) · Watch the public board meeting (48:24)

The board passed several financial resolutions in public: monthly reporting by line of business (4:06–7:52) from 1 January, unrestricted money to reserves and contingency before new hires, and written terms for every grant. The 2027 budget will show the AI initiative as a separate restricted fund; the DA holds that money and pays contractors only when the initiative approves (14:15–16:35). The interim CEO said these were policies the DA once had and had lost, and later referred to "a huge deficit we're digging out from".

The Q&A turned to the proposed European Drupal federation. The board said it is not the current priority, and set a condition for reviving it: about 25 organisations committing around 250,000 euros (32:30–36:35) of new money over two to three years. Members from the floor argued the sum is small. One proposed that local associations raise 10,000 to fund a lobbyist (45:58–47:21) first. The search for a permanent executive director is planned from January.

How the conference responded. The Drupal.org engineering panel (1:40–4:49) described a budget that has been flat for ten years, plus roughly two million of hidden costs. It wants an opt-out telemetry so it can ask the largest users of drupal.org infrastructure to pay their share.

Wrap-up: Drupal 12 beta 1 lands during the closing

DrupalCon Rotterdam hosts · Watch the wrap-up (27:10)

The hosts reported 876 participants from 47 countries, 236 of them newcomers (9:31–11:35), and announced that Drupal 12 beta 1 had been released (15:19–16:20) about an hour earlier, ready for contribution day. They repeated the advocacy ask and listed coming events, including DrupalSouth on the Gold Coast in May (25:30–26:45).

The tone the keynotes set. Confident about the product, sober about everything else. Dries and the end users said Drupal now matches its rivals on editing, front-end choice and AI. The Drupal Association then spent its stage time on money: who pays, who governs, and what a federation would cost. AI was presented throughout as something Drupal governs, with revisions, permissions and humans approving, rather than something that replaces editors.

Canvas is now the page-building story, with its gaps stated openly

Canvas was everywhere, and most speakers were frank about where it stops. Mars chose it while it was barely released (19:17–21:55). Mike Herchel (Acquia, Dripyard) showed that good Canvas sites depend on components: container queries and a canvas edit mode (7:26–12:47) so components survive any slot, and semantic colour tokens that guarantee contrast (21:30–27:01). He also listed what is missing, such as restricted slots and fieldable pages (32:11–34:30). Ted Bowman (Acquia) built a full site from components, Views and templates, covering how template props bind to fields (13:13–18:33) and Canvas taking over page templates (35:58–40:17). He noted that translation is symmetric only (42:49–45:08).

Not everyone agreed on the architecture. In a "deathmatch" with Display Builder, one speaker said Canvas abuses config as a component registry (16:05–19:54), and the pair concluded that both are viable for different users (27:19–30:39). Pierre (UI Suite) argued that theming is broken and that SDCs can now implement form inputs (10:57–16:37) after a small core change in 11.4. The code components team showed Canvas Workbench as a Storybook alternative (17:56–22:03), and server-side rendering (24:49–26:18) is available to Acquia customers now, with an open-source release targeted for October 2026. The media library's shared alt text and crop also carry into Canvas, which is why one team proposed per-use image overrides (4:23–8:13). Layout Builder stays in Drupal 12 (26:55–28:27), and Canvas is not formally named as its replacement.

So what: Canvas is ready for new builds where editors own free-form pages, provided you invest in a component library and design tokens first. For multilingual, Views-heavy or heavily governed sites, check the current limits (content templates for nodes only, no per-component permissions, symmetric translation) before you commit.

Drupal AI moves from roadmap to governed, shipping features

The Drupal AI initiative showed products rather than promises. AI search (10:14–14:07) answers from site content and refuses off-topic questions. Content review turns editorial rules into scored checks, and a long workshop showed how those rules and thresholds are built (1:19:08–1:24:41), along with input and output guardrails (25:02–28:43) and testing retrieval before building a front end (53:06–58:14). The Context Control Center keeps brand rules and terminology as governed, scoped content (15:43–21:05), and the Drupal AI UX lead showed why tone needs defined positions, not adjectives (7:44–11:33). The engine is a partner model of funded full-time contributors (29:22–32:06). Partners get the tested demo recipe, which a questioner gently challenged: the demo itself is not open source (40:15–42:29).

The case studies were measured and candid. EPSO's assistant produced a content-gaps report (12:24–15:02) that turned unanswered questions into an editorial backlog. VisitEurope tested its tone scorer (20:53–24:31) on a small set and admitted it had tuned and measured with the same system. Helsinki's Uni Help fixes content gaps through an MCP loop (21:10–25:23). Scolta skips the search server altogether, with the index shipped as static files (11:22–15:12) and optional AI on top. Several gaps remain open, including sharing context across many sites (38:25–44:05).

So what: AI search, content review and shared context are now real options for Drupal sites, including government ones, and the public demo is a usable sales and evaluation tool. Few talks gave accuracy or cost data, so plan your own evaluation and keep a human approving every change.

Agents work from the outside, through tools rather than prompts

A clear technical consensus formed: let outside agents act on Drupal through typed tools, and keep Drupal's permissions and revisions as the guard rails. The Canvas Tools work runs the same Tool API plugins over Drush locally or MCP remotely (1:36–5:51), with every change landing in autosave for review. It is beta, and the speaker flagged no public Canvas API (26:12–30:01). A site-building team found that live discovery tools beat prompts (6:20–10:40) and that validation errors that list valid options (14:46–18:19) let the model correct itself. The Drupal AI workshop made the same point: make everything available as a tool (38:05–41:43).

Others showed the engineering detail. FlowDrop's creator argued that most "hallucination" is missing facts, and that letting the model fail (7:15–11:02) lowers it. He found that simple commands beat generated JSON (25:57–31:05). ECA now has a generated guide and an MCP server (28:49–31:55) so agents can build and validate models. A Lullabot contributor said MCP matters most for editors, not developers (16:43–19:05). An EPAM prototype lets an agent drive Xdebug through a written plan (25:50–30:40) instead of editing code.

So what: If you maintain custom or contrib modules, the job ahead is exposing their capabilities as typed tools (Tool API, and the attribute syntax Dries prototyped once it settles). Much of this stack is alpha or beta, so pilot it on non-production content first.

AI in delivery: faster work, but humans must own it

Talks about using AI to build and maintain Drupal were notably sober. The Drupal AI contribution workshop named the core problem: saving yourself ten minutes can cost a reviewer an hour (4:06–7:33). A 19-year-old favicon issue showed why the AI's obvious fix was wrong (31:59–36:54). The AI module maintainers run a deterministic-first triage pipeline (15:21–20:05) with tightly sandboxed agents (20:05–24:32). A Views maintainer argued that clean code is now context engineering: thin hooks and injected services (19:26–23:32) plus static analysis and tests as guard rails (23:32–29:19).

Migration was the most common AI use case. Drunomics needed one ticket per component (22:05–27:41) after its first attempt failed. A Milan developer turned analysis into small single-question skills (15:26–20:11), and the AI Content Migrate module calls the AI once per page template (11:03–13:23). The failure stories were just as useful. One Dropsolid developer ended up with green tests and no idea what the code did (10:26–13:04). A Rotterdam agency recalled the Replit agent that ignored a freeze instruction (9:17–13:23), and offered four habits for owning AI-written changes (13:23–17:06). Two talks warned about people: AI is breaking the junior-to-senior pipeline (12:00–16:17), and Drupal in a Day organisers found few junior jobs for the students they attract (31:41–35:58).

So what: The working pattern is small tasks, deterministic steps where possible, sandboxed agents and a human who can explain every merge. Agencies should also decide now how they will keep training juniors, and how they will price work when AI makes velocity easy to inflate, a point made in a talk on gamed delivery metrics (12:19–15:28).

Digital sovereignty gives government Drupal a new argument

Sovereignty ran from the keynote to the hallway. Dries said Europe's dependence on US and Chinese technology is a challenge Drupal is ready for (10:42–12:02). Speakers were careful about what sovereignty means. An Athens agency warned of the API trap (9:00–12:28), where open-source Drupal on proprietary cloud identity and AI is still locked in. It proposed a sovereign service model for agencies (15:22–19:17) with tested exit plans. The audience then debated whether data location equals control (37:04–41:13). Edinburgh's law school argued that sovereignty is the ability to change course, and showed a university chat front end over many models (15:26–21:40). The European Commission explained how EU open-source policy shapes its 300+ Drupal sites (2:51–5:34), though a challenge over US AI models (26:59–29:33) exposed the tension.

The strongest government talks were about shared platforms. Dutch central government unpacks small recipes into each project (9:06–12:57), and reserves 5% of capacity for maintenance (29:27–35:41). It was frank that later platform changes must be reapplied by hand. LocalGov Drupal shared the hard lessons of its co-op (14:40–16:59). Catalonia's health service uses Drupal and Commerce as an app marketplace with hard boundaries (8:57–13:41). A Berlin agency showed Webform and ECA as a low-code platform that still needs people (23:54–27:40).

So what: For GovCMS and other government buyers, the useful framing is not "open source equals sovereign" but replaceable components, client-owned repositories and pipelines, and exit tests written into contracts. Shared platforms work when the owner can refuse one-off features and funds maintenance.

Drupal CMS and site templates become the front door, as core slims down

The Drupal CMS team kept repeating one line: Drupal CMS is just a Drupal site (7:43–10:44), built from recipes with no lock-in. Drupal CMS 2.2 brings a sensible multilingual setup and Canvas Translate (28:27–32:13). In Drupal 12, core gets leaner, with the toolbar, Olivero and default content types going (21:08–26:55). UX research showed a revised strategy that adds agencies and the mid-market (3:51–5:16), and survey findings on why some people still do not try it (7:34–11:25).

Site templates were pitched as the easy way to share a vertical. The Drupal CMS architect defined one as a recipe for a 90% finished site (2:36–5:20) and walked through exporting and publishing one (28:35–33:58). The marketplace for selling them is on the back burner (33:58–36:39) at the DA. The speaker listed Dripyard's Meridian as the one paid template in the installer, and mentioned Morpht's Convivial template, saying he thought it was premium but had not used it. Martin Anderson-Clutz told how Event Platform became the Summit template (9:10–14:12), and non-Drupal testers asked for full-page previews and accessibility ratings (12:30–15:15) in the marketplace.

So what: Evaluators can treat Drupal CMS as plain Drupal with a better start. Agencies with a repeatable vertical can package it as a site template now, but should not count on a DA-run marketplace for income, and should plan for the unsolved problem of updates after a template is applied.

Core engineering: faster pages, better tests, modern APIs

Below the AI headlines, core work was strong. Nathaniel Catchpole (catch) showed placeholder and fibers tricks (19:24–23:25) behind large cold-cache gains, and gave practical advice for site owners (36:06–39:36): update, measure and write performance tests. Another speaker showed how to maintain performance budgets in CI (15:24–18:21). Testing was a theme of its own. Tag1 argued that test levels are not test types (17:10–20:11), and Open Social cut a suite from three hours to 6.5 minutes (35:07–39:05) by moving to Playwright.

APIs are modernising too. Procedural hooks are deprecated from 11.5, and Rector handles most of the conversion (31:05–35:25), though form callbacks still need hand fixes. Core now has its own CLI, and Moshe Weitzman explained why it is in core (2:25–6:21) while Drush stays. HTMX suits simple interactions, not multi-step forms (25:56–29:07). On drupal.org, issues are moving to GitLab (7:59–11:00) and the security team handled a record security window (28:03–31:03), which it expects to become normal as AI finds more vulnerabilities. One AI talk found and diagnosed a Drupal 11.4 query regression (30:07–35:48).

So what: Staying current on Drupal 11.x pays off in speed. Plan hook conversion and Drupal 12 testing now (beta 1 shipped during the conference), and treat drupal.org's GitLab move as the new normal for contribution.

Accessibility: automation and AI help, but do not certify

Four accessibility talks agreed on the basics. Automated tools catch most issues by volume but only a minority of WCAG criteria (1:15–3:31), and AI helps with triage, drafting and explanation (10:01–12:42) but cannot be a gate. A Canadian advocate urged teams to use AI to build tools, not answers (7:04–10:12), such as Axe in Playwright and scripted screen readers, and to treat AI alt text as a draft for human review (19:00–22:08). Another speaker covered overlay lawsuits, the FTC accessiBe order and the European Accessibility Act (6:22–9:07), and showed that prompting for WCAG reduced errors but did not remove them (11:21–14:17).

For editors, Editoria11y's maintainer stressed picking your battles (10:18–15:36) so editors see few, relevant alerts. The European Commission keeps detection rule-based and AI for remediation (7:39–11:09). The contextual media talk argued that context-appropriate alt text is best practice, citing editor research (13:26–16:38), and set out a path to core by Drupal 13 (36:42–40:41).

So what: An all-green report is not compliance. Government clients need automated checks in CI, trained people, testing with disabled users, and AI only as an assistant whose output a person signs off.

Money, maintainers and the reputation gap

The project's sustainability came up far beyond the DA's own sessions. The Drupal.org team said there is no more blood to squeeze from the stone (1:40–4:49). Editoria11y's maintainer is trying a paid add-on subscription (31:44–37:06). An Indian community organiser said that more owners, not more helpers (6:33–8:51) reduce dependency on a few people. A Finnish team noted the 'someone else will do it' problem (11:48–14:42) in issue queues.

Reputation was the other half. One developer argued Drupal is invisible where developers learn (10:02–12:19) and fails the 15-minute test (14:48–18:18) against Laravel and Statamic. The marketing panel called the reputation gap the number one challenge (11:28–15:26) and heard an audience push to get off drupal.org and onto social video (28:50–32:35). Two search talks agreed that AI search is mostly good SEO: manage bots rather than block them (2:21–7:29), and skip llms.txt and AI rank tracking (23:42–29:01). Agency case studies, such as Sharp's land-and-expand relationship (19:41–23:06), are exactly what the advocacy programme wants published.

So what: Contribution now explicitly includes funding and public advocacy, and the DA plans to recognise both. Writing up client work, clipping talks and speaking at non-Drupal events now earn credit. Budget for it as part of being a Drupal business.

Technologies and tools

What was said about each technology, the talks that discussed it, and a signal for a typical team: Adopt (ready to use), Watch (promising, not yet proven), Hype (claims ahead of evidence).

Where people disagreed

• Ship AI fast, or review it properly? Dries asked maintainers to tool-enable modules at scale, while the contribution workshop warned that AI output shifts the cost to reviewers. The AI initiative accepts vibe-coded prototypes as ideas but never merges them.
• Canvas or Display Builder? Canvas speakers called it ready for real sites; a Display Builder team argued Canvas reinvents entity and config handling. Layout Builder stays in Drupal 12 and core will not name Canvas as its replacement.
• Open source, but the demo is for partners. Drupal AI's modules are open source, but its tested demo configuration and recipes are a partner benefit, which a questioner raised.
• Sovereign means where, or who controls? Several talks said hosting location is not sovereignty, and an audience example from Ukraine showed moving data to a US hyperscaler saved it. The European Commission was challenged over using a US AI model.
• Everyone benefits, few pay. The DA called itself not regenerative and plans a badge and telemetry to bring in big users. Whether non-agency users will join, and how government users will accept domain reporting, is open.
• A European federation: wish or plan? Community members pressed for one; the board says it cannot fund it and set a 250,000 euro bar. Some argued the sum is small.
• Juniors versus AI-boosted seniors. Agencies say AI plus seniors replaces juniors, yet Drupal in a Day and an Agiledrop manager warned the talent pipeline is breaking.
• Agents inside Drupal or outside? Drupal AI builds agents inside the CMS, while Canvas Tools and the ChatGPT demo favour outside harnesses reaching in through tools. Most speakers expect both.
• Self-reported numbers. Mars's TCO and Canvas KPIs, the AI search install counts and most AI accuracy claims came without baselines or independent checks. Mars's session figures do not fully match its keynote.

Talk index

Takeaways by audience

Evaluators of Drupal

• Judge the current product: Drupal CMS 2.2, Canvas and Drupal AI are a different proposition from the Drupal many people remember. Try the public AI demo at drupal.org/ai/demo.
• Governance is the pitch: revisions, permissions, audit and human approval are what Drupal offers against vibe-coding and AI-first platforms.
• Check the gaps: Canvas still lacks per-component permissions and asymmetric translation, and much of the agent tooling is alpha or beta.
• Read case figures carefully: Mars, World Cancer Day and others gave strong numbers that are self-reported.

Editors

• AI as reviewer, not author: content review scores pages against your own rules and leaves each suggestion for you to accept or skip.
• Assistants can do content operations: the ChatGPT demo restored a set of translated pages with a revision trail under the editor's name.
• Images in context: today, changing alt text or crop in the media library changes it everywhere; a contextual override is proposed for core.
• Accessibility checks: a well-configured Editoria11y shows a few useful alerts rather than many noisy ones.

Marketers

• AI search is mostly good SEO: manage bots, write explicitly and consistently, put summaries near the top, and skip llms.txt.
• Tone needs defining: write down what good looks like (tone positions, conventions, structure rules) before asking AI to enforce it.
• Join the advocacy programme: clips, posts and case studies submitted at drupal.org/advocacy earn credit.
• Measure what AI answers say: track citations, referrals and mentions rather than a rank that does not exist.

Site owners

• Update and measure: moving to current Drupal 11.x removed hundreds of cold-cache operations per page in core tests.
• Drupal 12 beta 1 is out: core drops some modules and defaults, so check what your site relies on.
• Own your exit: keep your repositories, pipelines and keys, and keep AI and identity behind replaceable layers.
• You are accountable for AI on your site: speakers cited the Air Canada chatbot ruling; keep guardrails and logs.

Developers

• Convert hooks: procedural hooks are deprecated from 11.5; Rector handles most of the work.
• Expose capabilities as tools: typed Tool API plugins and MCP make modules usable by agents and workflows.
• Test smarter: push tests down the pyramid, add performance budgets in CI, and consider Playwright for end-to-end tests.
• Own what you merge: small prompts, read the diff, failing test first, and never commit what you cannot explain.

Site builders

• Learn Canvas templates: content templates, page templates, patterns and Views together can build most of a site without code.
• ECA is getting easier: the new modeler, token picker and AI assistant lower the barrier for config-based workflows.
• Package what you repeat: export a finished site as a site template with Drush site export and share it.
• Know the alternatives: Display Builder suits builders who want deep use of Drupal's entity and context model.

Themers

• Components first: Canvas is only as good as its components; design for the common editor case and label props in plain language.
• Use container queries and tokens: components must survive any slot, and semantic colour tokens keep contrast correct.
• SDC now covers form inputs: a small core change lets SDCs implement form elements, usable in UI Patterns and Display Builder.
• JavaScript is first class: code components bring React, TypeScript and Tailwind into Canvas, and Drupal 12 has no default front-end theme.

The Drupal community

• Talk louder: the reputation gap was named the top challenge; clip talks, write posts and speak outside Drupal events.
• Fund the commons: the DA says the current model is not sustainable and is broadening partner recognition to all who benefit.
• Grow owners and juniors: hand over whole areas to new owners, and keep hiring and mentoring entry-level people.
• Respect reviewers: short, human-written messages, attached logs, stated testing, and discuss the problem before posting AI code.

AI generated. Sources: the DrupalCon Rotterdam 2026 session playlist and the Driesnote on YouTube.