Can AI Fix Drupal Issues Without Writing Code?

Fixing a bug without typing a single line of code sounds unrealistic—until it isn’t. Gábor Hojtsy documents a real Drupal contribution where AI generated the solution, tests, and commits, while he focused on direction, review, and accountability.

The issue involved language names not being extracted correctly for translation in a Drupal module. Instead of debugging manually, Hojtsy used Claude with minimal setup, asking it to analyse the problem and propose a fix. The system identified the root cause, generated code changes, added test coverage, and prepared commit messages as part of the workflow.

However, the process required continuous oversight. Initial tests appeared correct but missed edge cases, requiring further prompts and corrections. The AI was also unable to run the test suite in the given environment without explicit guidance, leading to additional intervention before results could be verified.

At multiple stages, the role of the contributor shifted from writing code to steering the process. Hojtsy reviewed outputs, corrected assumptions, and ensured that fixes worked across versions, including backports. While the system accelerated development, responsibility for accuracy and quality remained with the human contributor.

The workflow also included AI-generated issue comments and summaries, with manual disclosure to align with Drupal’s contribution policies. This highlights an emerging need for transparency as AI-assisted development becomes more common in open-source ecosystems.

The experiment suggests that AI can compress the path from problem to solution, but does not replace developer judgment. Instead, it redefines contribution as a process of supervision, validation, and ownership rather than direct code authorship.