Robert Menetray details 20-60-20 AI workflow for Drupal development

Drupal developer Robert Menetray says artificial intelligence now performs roughly 60% of his development workload under a structured process that preserves human control over planning and review. In a detailed account of his workflow, Robert describes a “20-60-20” model dividing work into 20% human planning, 60% AI execution and 20% human validation.

After more than a year integrating AI into client projects, he reports that the approach has reduced development time by approximately half while maintaining, and in some cases improving, code quality. He attributes these gains to disciplined specifications at the outset and strict review policies at the end of each task.

The first 20% of the process is entirely human-led. During this phase, he defines module architecture, identifies required hooks, services and plugins, documents dependencies and outlines performance and security constraints. According to Menetray, insufficient specification results in weaker AI output and increased rework during review.

The central 60% is handled by AI in one of two modes. For complex or highly customised work, he uses a supervised approach, reviewing outputs in stages and redirecting the system as needed. For repetitive or well-defined tasks such as migrations, module scaffolding, refactoring, test generation or static analysis fixes, he employs a loop-based system in which the AI iteratively generates code, runs tests and corrects errors autonomously.

A key rule in both modes is that the AI never performs version control actions independently. Menetray says all commits and pushes are executed manually after review. He argues that this final 20% is critical for detecting architectural flaws, performance risks, security concerns or violations of Drupal best practices.

Automated testing plays a central role in enabling this workflow. Unit, kernel and functional tests act as validation mechanisms that allow the AI to iterate without supervision. Menetray notes that test coverage has increased since adopting AI assistance and describes testing not only as a quality safeguard but as the infrastructure that makes extended autonomous execution possible.

He cautions against what he describes as a “5-95” model, in which developers provide minimal guidance and deploy AI-generated code without thorough review. While he expects AI capabilities to shift the balance further toward 10-80-10 in the future, he maintains that experienced developers remain essential for architectural decisions, quality assurance and long-term maintainability.

Menetray concludes that AI can significantly accelerate Drupal development, but only within a framework combining detailed planning, automated validation and disciplined human oversight.