The author outlines a systematic approach for product and system analysts to utilize Large Language Models (LLMs) in software development. The methodology involves a step-by-step process, starting from defining user stories to designing database structures and solution architecture.
The process begins with creating a project Readme.md to provide context and prevent model hallucinations. Each subsequent step uses the output of the previous one, with a focus on maintaining a 'Questions' block to clarify requirements with stakeholders rather than letting the AI make assumptions.
For the 'Repeat Order' feature example, the author demonstrates how to generate user stories with Given/When/Then criteria, followed by Use Cases that account for system failures and edge cases. The workflow ensures that technical decisions are deferred until the appropriate stage.
The final stages involve conceptual data modeling and physical database design using DBML. The author emphasizes using existing system entities and adhering to project-specific type conventions, such as using UUIDs for identifiers and bigint for currency in kopecks.
The solution architecture step identifies which existing services require modifications. By restricting the AI to use only known services, the author avoids the creation of unnecessary components and helps determine if the feature can be implemented within a single sprint.
