The Four Pillars
To generate consistent code, four elements must come together as shown in this diagram:
Image: Venn diagram showing the intersection of elements for consistent AI code generation
The Four Elements
1. Clear Requirements
Purpose: Defines (with detail and clarity) the functional and pseudo-technical requirements for the product or service idea you want to implement.
Where created: Use an advanced “thinking” chat model like Claude or ChatGPT.
Ref: Product Requirements Workflow
2. Good Prompts
Purpose: The clear, detailed ask for a given task. It can refer to the IDE rules and the requirements. However, you don’t need to repeat what’s in the rules or requirements in the prompt.
Where created: Create prompts manually at first but then refine them using a chat model like Claude or ChatGPT.
Ref: Prompt engineering and meta prompting is explained further in the Prompting Guidance
3. Rules for AI
Purpose: AI Rules define consistent and repeatable standards, patterns and conventions across your codebase.
Where created: Rule file formats are usually defined by the AI IDE tool. To help generate the rules themselves, use a chat model like Claude or ChatGPT.
Ref: Rules for AI
4. Capable Code Generation Model
Purpose: Use the most capable LLM model for the task you are running to get good quality results. Not all tasks need advanced models, so select the most cost-effective model that can achieve the outcome you want.
Where created: The AI IDE tools typically let you select which model to use when prompting the LLM.
Next steps
We recommend reading and understanding the detailed Prompting Guidance before you start.
