DrupalCon Rotterdam Speakers Put AI, Workflows, and Accessibility Into Practice

Drupal teams are asking software to do more work that once depended almost entirely on people: prepare coordinated changes, inspect content, validate accessibility, recommend revisions, and answer questions from structured information. Several DrupalCon Rotterdam speakers are concentrating on the point after those capabilities become possible: where automation should stop and a person should still have to decide.

That boundary appears across very different kinds of Drupal work. Ajit Shinde of Tag1 Consulting is looking at Workspaces as a controlled layer for content, configuration, and AI-assisted changes. Adam Nagy and David Galeano are presenting an agentic editorial workflow in which human approval is enforced by the system. Daniel Angelov, John Jameson, Christopher Torgalson, and Junaid Masoodi approach the boundary through accessibility, while Antonella Picarella applies a similar control problem to AI-assisted answers built from Drupal content.

The common thread is not resistance to automation. These speakers are using it, extending it, and building workflows around it. What changes from session to session is where human judgement re-enters the system: before publication, during accessibility review, when deciding whether generated code makes sense, or when determining which information an AI system should be allowed to use.

Ajit will present Beyond content staging: where Drupal Workspaces is heading on 29 September. In his response to The DropTimes, he described Workspaces as increasingly useful beyond its established content-staging role, pointing to Content Moderation, AI governance, configuration changes, previews, and safer publishing workflows.

The session examines how Workspaces can support coordinated changes that should not immediately reach production. Workspaces Extra extends that model through capabilities including wse_config and wse_preview, while tool_workspace is being explored for human-in-the-loop AI workflows. The practical attraction is that groups of machine-assisted changes can be isolated, inspected, and approved rather than being treated as trustworthy simply because an automated process produced them.

Adam and David reach a similar principle through the architecture of AI Assists, Humans Decide: Agentic Workflows in Drupal - A Drupal AI Hackathon Story. Their session is based on the solution their team developed at the European Commission's Drupal AI Hackathon in January 2026, which Adam said won first place.

The starting problem is editorial scale. European Commission editors manage hundreds of websites in 24 official languages while trying to keep facts aligned with policy, metadata complete, accessibility and consistency maintained, and existing pages current as policies change. Adam said the team used AI to assist that work while keeping the final decision with editors.

Drupal orchestrates the workflow rather than handing the entire process to an autonomous agent. Rule-based checks handle work suited to deterministic validation, while AI is used where contextual evaluation is useful. Proposed changes remain separate from published content until an editor approves them, making the checkpoint part of the system architecture rather than an instruction that a language model is expected to remember.

You May Like

Blog

A "Humanizer" Should Not Become a House Style Manual

Blog

A "Humanizer" Should Not Become a House Style Manual

Adam and David will also cover asynchronous execution, model integration, cost, speed, reliability, and data sovereignty. Adam describes the work as a functioning solution and a set of architectural decisions rather than a forecast of what autonomous agents might eventually do.

Accessibility exposes another limit: software may establish that a rule has been satisfied without establishing that the result makes sense to the person using the page. That difference is central to Daniel Angelov's The Reality of Accessibility Automation.

Daniel separates what remains for people into meaning, flow, recovery, and experience. An automated check can establish the presence of a technical feature without knowing whether it communicates the intended meaning or allows somebody to complete a task. His session will also examine why apparently conflicting estimates of automated accessibility coverage can refer to different measurements, and where AI can extend testing without replacing human evaluation.

John Jameson approaches automated checking from inside the editorial workflow. His Getting the most out of Editoria11y v3 session examines what teams can do once a live checker becomes part of everyday content work rather than a final scan.

John described users writing custom tests, matching alerts to user roles, mapping false positives, changing themes in response to checker behaviour, experimenting with automatic fixes, and beginning to integrate editorial style guidance. The result is not a universal pass-fail standard but a checker that can be adapted to the recurring problems of a particular editorial environment.

Christopher Torgalson of Annertech shifts the discussion from tools to responsibility across the wider team. His Accessibility Worst Practices session draws on remediation experience and the 2026 WebAIM Million study to examine recurring accessibility failures.

Christopher's argument widens responsibility beyond an accessibility specialist. Developers, designers, editors, and site builders all make decisions that can introduce or remove barriers, and he notes that many frequently occurring problems are also comparatively straightforward to address once teams recognise them.

Junaid Masoodi brings that problem directly into AI-assisted front-end development in Can AI Fix the Web? The Promises and Perils of "Automated" Accessibility. His concern is that generated ARIA attributes, alternative text, and focus-management code can appear technically plausible, pass automated checks, and still make the screen-reader experience worse.

Junaid is not asking developers to stop using Claude, ChatGPT, Copilot, or similar tools. He wants teams to change the role assigned to them. His practical recommendation is to use AI as an auditor that identifies risks and challenges assumptions rather than automatically replacing components, with human sign-off required before generated accessibility changes are merged.

Antonella Picarella applies the control question to AI-assisted answers. Her From Drupal Content to AI Answers: Learnings from EPSY session examines an assistance experience built on Drupal for public-sector content.

Antonella said the implementation controls which Drupal content enters the knowledge base, what retrieved information is supplied to the AI, and how the system is instructed to answer. That gives the team a defined information boundary instead of treating the language model as an unrestricted source.

Backend and usage statistics can show what people are asking, where they struggle to find information, and where existing material may be incomplete or unclear. The assistance layer therefore becomes a source of editorial feedback as well as a way to retrieve information.

These sessions put different technologies around the same operational question. Workspaces can hold changes away from production, agentic workflows can enforce approval outside the model, accessibility tools can expose problems without deciding whether an experience works for a person, and constrained AI systems can limit the information available to an answer.

The human role therefore extends beyond approving whatever a machine produces. It includes deciding what the system may change, what evidence is sufficient, what a user actually experiences, and what should happen when automation produces a technically valid result that is still wrong.