AI Dev Summit

AI is rapidly changing how developers build applications, manage content, automate workflows, and create new digital experiences. As Drupal, PHP, and the broader open web ecosystem evolve, developers need practical ways to understand and apply these technologies.

Drupal developers, architects, and technical leaders share many of the same questions: how to integrate AI responsibly, where it can improve existing workflows, and how to move from experimentation to reliable production solutions.

This dedicated AI Dev Summit creates a focused space to gain practical insight, explore real-world use cases, and connect with open-source innovators building AI-enabled solutions across Drupal and modern web development.

• Registration

  8:00 - 8:45 | Nick Veenhof, Director, Developer Relations, Gitlab

  Introduction and Welcome

  8:45 - 9:00 | Thomas Schröpfer, Go-to-Market Program Manager AI Products at amazee.ai

  Humans and AI Agents, One Contribution Loop

  9:00 - 9:30 | Nick Veenhof, Director, Developer Relations, Gitlab

  My team at GitLab uses AI to accelerate open source contributions. The real barrier to contributing is not code quality. It is documentation & real enablement - which we call Co-Create. When documentation is public and machine-readable and enablement is really tangible, both humans and AI agents can onboard faster. What used to take a new contributor weeks now takes hours. I'll share the pattern behind this acceleration, the trade-offs we hit, and the cases where it breaks, with real numbers.

• Pluto: Infrastructure for AI-Native Development

  9:30 - 10:00
  Sam Zwaaij, Digital Strategy, Synetic
  Jordy Vleugel, Solution Architect, Synetic

  An inside look at Pluto, our AI-powered development platform. We'll explore how we combine hosted LLMs, background agents, n8n workflows, and modern AI coding tools to enable customers to build and extend their own solutions.

• FlowDrop in Production: Observable, Async, Cost-Controlled Agentic Workflows for Drupal

  10:00 - 10:30  | Shibin Das, Developer, Factorial

  A production-grade introduction to FlowDrop, the visual workflow editor for agentic Drupal — covering observability, async scale, and cost control.

  Agentic demos look magical. Agentic systems in production look different: unpredictable latency, runaway token bills, opaque failures, and flows that can't be replayed or audited. Enterprises need a harness.

  FlowDrop is a visual workflow editor that runs on your own Drupal runtime — no hosted control plane, no hidden backend. In this 25-minute session I'll introduce FlowDrop and walk through the three production concerns that matter once agents leave the demo stage:

  • Observability — every agent decision inspectable, replayable, and version-controlled
  • Async scale — execution that escapes the request cycle and handles long-running, parallel, and event-driven flows
  • Predictable cost — composing many small AI skills instead of one large prompt, with measurable token and latency budgets

  Demo that shows a real flow handing off between deterministic steps and an agentic decision point — including what observability looks like when something goes wrong. We'll close with where FlowDrop fits alongside ECA and how to start small.

• From Developer to Code Director: Strategic AI Acceleration for Drupal

  10:30 - 11:00  | Martin Anderson-Clutz, Senior Product Marketing Manager, Acquia

  As AI tools evolve from simple autocomplete extensions into sophisticated reasoning partners, the role of the developer is shifting. For technical leaders, this evolution offers a massive opportunity: transforming engineering teams from line-by-line coders into high-level "code directors." But how do we strategically harness this shift to accelerate both internal velocity and open-source contribution?

  In this fast-paced, 20-minute session designed for technical leaders, we will explore the practical realities of "vibe coding" within the Drupal ecosystem. Drawing on real-world engineering workflows, we will look at how leading models and assistants—including Gemini, Claude, Codex, and Copilot—act as powerful force multipliers.

  We will cut through the AI hype to examine how these tools can be used as strategic reasoning partners to solve complex architectural challenges, accelerate custom development, and supercharge community contributions.

  You will leave this session inspired by the future of AI-driven development and equipped with practical, high-impact tips you can bring back to your engineering teams immediately to maximize their impact and velocity

• Your AI Decisions Are Creating Risk: What Agency and Business Leaders Need to Own Now

  11:30 - 12:00  | Thomas Schröpfer, Go-to-Market Program Manager at amazee.ai

  AI is already part of how work gets delivered.
  
  It’s in the tools your teams are using, the platforms you’re recommending, and the solutions you’re putting in front of clients. It’s being introduced quietly often without much scrutiny.
  
  That raises real concerns around data privacy, sovereignty, and who actually controls the systems your work depends on. A lot of AI-enabled workflows depend on sending data to systems you don’t control, running on models you can’t inspect, in environments you didn’t choose.
  
  That creates a gap between what organisations believe they are responsible for, and what is actually happening.
  
  We can close that gap.
  
  We’ll look at how AI is really showing up in client work today in delivery pipelines, vendor decisions, and everyday tooling. Let's sort out where risk is introduced, how sovereignty is affected by common integration patterns, and what it takes to stay in control.
  
  We'll review:
  • Where AI is entering projects without clear oversight
  • How data moves through typical AI-enabled workflows
  • The implications of relying on closed vs open ecosystems
  • How vendor choices shape long-term control and flexibility
  • What transparency looks like when you actually need to explain a system
  • Approaches to governance that can be applied in active projects
  
  We don't need to slow teams down, but let's make sure the decisions being made are intentional, understood, and aligned with the responsibilities you carry.

• Context as a Platform: From Configuration to Custom Integrations

  13:30 - 14:00  | Marcus Johanssen, Tech Lead AI Initiative/Senior Developer, FreelyGive

  Drupal AI Context, also known as Context Control Center (CCC), helps Drupal sites provide structured, governed context for AI agents and workflows. But beyond the UI features, CCC is designed as an extensible platform developers can configure, customize, and build on.

  This session takes a technical look at how CCC works and how to extend it for real-world Drupal AI implementations.

  We’ll explore configuration patterns, including:

  * configuring scope plugins 
  * enabling advanced capabilities 
  * tuning agent-to-context selection behavior
  * controlling how context is filtered and assembled

  We’ll also walk through extension points available today, including: 
  * creating custom scope plugins 
  * extending or modifying existing behavior 
  * prefiltering context before selection 
  * adding new context source types beyond files and URLs

  Finally, we’ll look ahead to extensibility patterns under development, including remote and dynamic context connectors, reusable context templates, and integrations with standards such as MCP (Model Context Protocol).

  Whether you’re building AI-powered editorial workflows, custom agents, semantic search systems, or organization-specific AI tooling, this session will show how Drupal AI Context can become the governed context layer behind your implementation.

• Defending Drupal AI: Hands-on with the guardrails system

  14:00 - 14:30  | Luca Lusso, Lead developer, SparkFabrik 

  As AI features become central to web experiences, controlling their inputs and outputs is no longer optional. This session dives deep into the Guardrails subsystem within the Drupal AI module, exploring exactly what guardrails are and why they are critical for brand protection, compliance, and mitigating prompt injection. We will unpack the vital difference between deterministic guardrails (like rigid regex checks and structural validation) and LLM-based evaluators (such as topic classification, toxicity detection, and tone analysis). We will discover how guardrails are integrated into Automators and Agents, and how they operate globally. Finally, we will put theory into practice with a hands-on developer walkthrough, showing you which guardrails already exist, how to code a custom guardrail, how to configure pre- and post-generate sets, and how to test your defense layers live using Drupal's AI Explorer.

• Giving Authors Before-the-Save Power with AI-Powered Field Widget Actions

  14:30 - 15:00  | Mike Anello, Developer/Trainer, DrupalEasy

  The Field widget actions (https://drupal.org/project/field_widget_actions) module allows site-builders to put the power of AI directly at authors' fingertips on Drupal content entry forms in a controlled and structured manner. Most importantly, Field widget actions can do all of this before the author clicks the "Save" button, ensuring they have the opportunity to review and modify any AI generated content. 
  
  In this 30-minute session, we'll see a series of examples that will provide an overview of the Field widget actions module and how it integrates with the AI automators and AI content suggestions modules.

• Operationalizing AI Safely in Enterprise Workflows - Without Custom Code Everywhere

  15:30 - 16:00  | Jurgen Haas, Drupal Architect, LakeDrops

  I've spent the last year combining ECA's deterministic workflow engine with AI agents for the non-deterministic parts. The pattern works surprisingly well.

  Three real examples: 
  1. Context-aware content suggestions - Editors click a button, ECA checks the content type and user role, then fires different AI prompts based on context. Same workflow, different AI instructions per situation. 
  2. Automated comment moderation - ECA watches for new comments, passes them to an AI agent for toxicity/spam analysis, then routes to different approval queues based on the agent's response. Fully auditable. 
  3. AI-generated workflows - I built an ECA MCP server that lets AI agents read ECA documentation and generate complete workflow models from natural language prompts. The meta-loop: AI creates the automation that will later invoke other AI.

  What you'll see: 
  - Live workflow debugging showing AI input/output at each step (you see what the agent decided, not how) - Permission boundaries that keep AI from accessing what it shouldn't 
  - Config-based deployment (no custom code sprawl) 
  - The limitations I hit (AI hallucinations still need validation gates) 
  - The core insight: deterministic workflow control (ECA) wrapping non-deterministic AI decisions gives you the safety enterprises need with the flexibility AI provides.

  I'll show working code, real debugging sessions, and the mistakes I made figuring this out.

• Group Breakout Sessions

  16:00 - 17:00  | All Attendees

  • One for Guardrails/Security

  • One table for ECA

  • One table for Flowdrop

  • One table for Context Control Centre

• Breakout Presentations

  17:00 - 17:30  | Table Presenters

  Each table nominates one person to give a short presentation (5-10 minutes) on what they learned.