In this Monkeys Sharing,
Cass explains how he used Product Requirements Prompts (PRPs) to build our AI stack kAIoh
• 100% AI Code Generation: The kAIoh project’s AI stack has been entirely generated by AI, marking a shift from Python to NodeJS due to proficiency challenges and problems with Python servers. The speaker emphasizes that they “didn’t touch a line of code, almost none”.
• Product Requirements Prompts (PRPs) as the New Development Standard: A new concept called Product Requirements Prompts (PRP) is introduced as the current approach to coding with AI. PRPs are markdown documents that detail the project’s early specifications, including functional, acceptance, and technical requirements, essentially serving as a blueprint for the AI.
• Iterative Process of PRP Creation and Code Implementation: The development process involves the user defining initial project goals, and then the Large Language Model (LLM) edits the PRP markdown file to prepare functional and technical analyses. This is an iterative process where the user reviews and refines the PRP with the LLM until it’s satisfactory before instructing the LLM to start the code implementation.
• Leveraging Context and External Knowledge for Prompts: It is crucial to provide the LLM with ample context, which can involve deep scanning of the project’s codebase, referencing existing code, and explicitly asking the LLM to search online for best practices or relevant libraries. Using “thinking models” like ChatGPT or JCLI to draft the initial PRP before refinement is also suggested.
• Shift from Coding to Reasoning and Validation: The core message is to embrace AI agents in all development tasks, as this paradigm shift moves the focus from direct coding to more abstract reasoning, precise specification, and thorough review of the AI-generated output. A key recommendation is to have the LLM write the prompt itself based on high-level goals, as this leads to more detailed and comprehensive prompts than a human might initially create.