GitLab engineer details workflow optimizations for AI-assisted development

A GitLab engineer has documented an optimized workflow for AI-assisted development after iterative improvements to the company's AI tools. The system integrates GitLab's existing features with custom agents to prioritize tasks, maintain context across sessions, and automate repetitive work. Key implementations include automatic context loading from active merge requests and todos, parallel work using git worktrees with isolated databases, encoded procedures for recurring tasks, and a Model Context Protocol server to reduce token overhead. This approach shifts development from prolonged focus sessions to distributed work in 15-30 minute blocks while maintaining code quality through AI-executed testing and human review.