When an AI agent writes production Drupal code, your architecture must answer questions it never had to before:
- What is the canonical source of truth for entity relationships?
- Where does business logic live if an agent can't infer it from file proximity?
- Which contracts are stable enough to be AI-navigable?
Modern AI tools often fail on complex Drupal architectures due to context loss and copy-paste friction. This live demonstration showcases how terminal-native AI agents operate directly inside local development containers to plan, build, test, and refactor Drupal 11 features seamlessly.
Drupal has always been the CMS of choice for complex, community-driven websites. But can Drupal serve the developers themselves — as a platform built by developers, for developers?
Attendees will witness real-time agentic execution using ddev-claude-code inside a live DDEV environment. The session covers setting up governance files (CLAUDE.md), delegating Drush commands to agents, leveraging Model Context Protocol (MCP) servers, and enforcing security and PSR-4 coding standards automatically in agent loops.
This session walks through building the Daily Developer Toolkit, a real open-source site that catalogs tools, tutorials, code snippets, and events for the global developer community. Every piece of this stack is open source.
Run-of-Show (45 Minutes):
00:00–00:10 | Governance & Setup
Boot Claude Code and walk through an enterprise CLAUDE.md: scope boundaries, forbidden paths, DI enforcement, coding standards, and why governance must precede every agent invocation.
00:10–00:25 | Agentic Module Construction
Use claude -p to scaffold a custom Drupal 11 module with PSR-4, a Plugin Manager, and wired DI — live, inside DDEV. Watch the agent run ddev drush commands and export configuration without leaving the terminal.
00:25–00:35 | TDD and Self-Correction Loops
Direct Claude Code to write PHPUnit tests, execute them via ddev exec phpunit, and observe the self-healing loop: agent parses failures, refactors the implementation, and re-runs the suite until green.
00:35–00:45 | Daily Developer Toolkit and Q&A
Walk through the three slash commands every Drupal developer should run before pushing code — /code-review for anti-pattern detection, /security-review for Drupal-specific attack surfaces, and /simplify for quality refactoring. Includes reusable prompt templates for scaffolding, service wiring, and debugging, plus claude -p in non-interactive mode for CI pipelines.
Architectural patterns:
- Spec-first contracts — CLAUDE.md as architectural documentation, not just tooling config
- Service boundary discipline — why DI-only code survives agentic rewrites; \Drupal::* statics don't
- Test-as-specification — PHPUnit tests as the architectural handshake between human intent and agent execution
- Config-as-truth — why config/sync/ must be complete before an agent can safely scaffold
Key Takeaways:
- How to write a CLAUDE.md that keeps agents safe, scoped, and standards-compliant in a Drupal 11 project.
- A repeatable daily workflow: Start → Develop → Review (three passes) → Ship — with the commands at each stage.
- Six prompt templates developers can copy and adapt for modules, services, tests, debug traces, and migrations.
- How to run Claude Code non-interactively (claude -p) in CI for automated DI and security checks on every commit.
Attendees will leave with a framework for auditing their own module architecture for AI-navigability — and a clear set of patterns that survive the agent loop. A live build-out is used not to showcase the tool, but to stress-test three architectural assumptions in real time — and to show where they break.
Claude Code is the pressure test that reveals whether your Drupal architecture is actually well-structured — and this session teaches you to build for that standard.