
Talking Drupal Talking Drupal #538 - Agentic Development Workflows
8 snips
Feb 2, 2026 Matt Glaman, principal software engineer at Acquia and Drupal maintainer; Andy Giles, veteran web developer and founder focused on modern Drupal themes. They explore agentic development workflows, how agents differ from autocomplete, orchestration of agents and tools, Drupal Canvas CLI and syncing front-end components, UUID pitfalls, validation tools, and security/privacy trade-offs with AI-driven workflows.
AI Snips
Chapters
Transcript
Episode notes
Have Agents Run Validation And Fix Errors
- Use self-validation tools (linters, tests, MCP/playwright) so an agent can run checks and correct its own output.
- Matt shows agents can run PHPStan/PHPCS, read errors, and iterate until validations pass before returning results.
Chunk Work And Chain Agents For Large Tasks
- Break large features into small agent tasks and run multiple agents in parallel for parts like content generation, image creation, and Canvas page building.
- Andy Giles describes chaining agents: write markdown, generate images, then use Playwright to place content in Canvas.
Drupal Config Makes Agents Practical
- Drupal's structured config management and YAML recipes make it a strong fit for agentic site building.
- Andy notes agents can generate field and content type config programmatically, but UUIDs must be produced by tools to avoid LLM hallucinations.
