Rapid Prototyping Process

Promptmanship: process, concepts, activities, and patterns

With no need for complex data engineering (although a small number of prompt examples is often required) and model training, foundation models coupled with prompt-based AI chains can support rapid development of AI services. This makes rapid application development (RAD) style methodologies highly applicable to AI chain development, as RAD emphasizes rapid prototyping and iterative delivery.

Considering that AI chain's requirements analysis, design, implementation, and testing are parallel and have no clear boundaries, we propose an AI chain process model similar to Rational Unified Process (RUP). RUP emphasizes identifying and addressing risks early in the project life cycle, which is beneficial for embedding RAI principles and risks in the AI chain process model in the future.

AI chain process consists of four phases: exploration, design, construction, and deployment.

AI chain engineering activities involve task modeling, system design (requirement analysis, task decomposition, separating AI and non-AI concerns, task workflow walk-through), AI chain implementation (prompt design), AI chain testing, and the co-pilot activity magic enhancing magic.

In the above figure, the bar next to an activity indicates the start and end of the activity, and the bar height indicates the intensity of the activity. We can see that each phase has focused (more intense) activities, but they all involve other activities to some extent (less intense). Therefore, the AI chain development process has both iterative and parallel characteristics.