What should I include in a project plan or spec before asking an AI to write code?

To get the best results from an AI, treat it like a highly skilled junior developer: provide clear instructions, context, and constraints. Before starting, ensure you have these elements in your plan:

1. The Core Objective

• The "What": Define exactly what the code should do.
• The "Why": Briefly explain the purpose. This helps the AI choose the best approach (e.g., optimization for speed vs. readability).

2. Technical Stack and Constraints

• Language & Frameworks: Specify the exact version if it matters (e.g., "Python 3.12," "React 18").
• Dependencies: List the libraries you prefer (e.g., "Use Pandas for data processing," or "No external libraries, only standard library").
• Environment: Is this for a browser, a server, an AWS Lambda function, or a local script?

3. Functional Requirements

• Input/Output: Define what data goes in and what should come out. Include examples of expected formats (e.g., "Input: CSV file; Output: JSON object").
• Edge Cases: What should happen if the input is empty, malformed, or unexpectedly large?
• State/Logic: Describe the "happy path" (the ideal scenario) and any logical branching required.

4. Style and Quality Standards

• Formatting: Do you have specific style preferences? (e.g., "Follow PEP 8," "Use functional programming style," or "Strictly typed with TypeScript").
• Documentation: State clearly if you need code comments, type hints, or a separate README file.
• Testing: If you want unit tests, ask for them specifically (e.g., "Include PyTest unit tests covering the core logic").

5. Scope Boundaries

• "In Scope" vs. "Out of Scope": Be clear about what you don’t want the AI to handle. This prevents "code bloat" and keeps the output focused.

Example Template

If you are preparing a prompt, you can use this structure:

Role: You are an expert [Language] developer. Context: I am building [Project/Feature]. Task: Write a script that [Action]. Input: [Format]. Output: [Format]. Constraints:

• Use [Library/Version].
• Include error handling for [Specific Scenario].
• Keep the code modular and well-commented.
• Do not include [Feature X].

Pro-tip: If the project is large, break it into smaller modules. Asking for a complete, complex application in one go often leads to lower-quality code. Ask for the core logic first, then build the supporting features step-by-step.