Drupal AI Playground Experiments Highlight Recipe-Based Development Workflow

Experiments with a Drupal AI “playground” show how developers are integrating AI-assisted tools into local workflows while replacing scripted installation practices with recipe-based configuration. The setup, documented by Jacob Rockowitz, combines Drupal, DDEV, and Claude Code to create reproducible environments for testing AI-driven development.

A central shift described in the post is the move from custom bash scripts and Drush-based installation commands to Drupal Recipes. Recipes define configuration and assemble features during installation, allowing developers to replace procedural setup steps with modular and repeatable configuration.

The post describes Recipes as composable building blocks that can incorporate upstream configurations from Drupal CMS while supporting project-specific customisation. One example includes automatically configuring API keys from a private directory, extending beyond what standard distributions typically provide.

The experiment also highlights limitations in AI-assisted development. While using Claude Code to generate configurations, difficulties arose when applying Composer patches using newer workflows based on patches.lock.json. The issue required manual intervention and updated commands, reflecting gaps in tool awareness and documentation coverage.

The post concludes that AI-assisted development remains an iterative process rather than a replacement for existing workflows. While the experiment results in a fully recipe-based installation approach, effective use of AI tools depends on context, guidance, and continued developer oversight. More details are available in the original blog post.