Prompts for Code Generation with AI

You type "write a function to manage users" and press Enter. The AI responds with 847 lines of code. The entity includes a repository, a service, a controller, a DTO, a mapper, a validation decorator, and a README detailing the architecture. When you request code without precise specifications, the AI doesn't fill gaps — it fills them with its own knowledge.

This is impressive but often irrelevant to your actual needs. The problem isn't the AI; it's that "manage users" isn't a clear specification. It's an open invitation, and without constraints, AI tends to output everything it knows, much like a coworker explaining atomic clocks when asked the time. In a previous tutorial, AI was used as a learning tool to understand programming concepts.

Now, the goal is different: you want functional, usable code. To achieve this, you need to specify your requirements clearly before asking for code. A well-defined specification is crucial. Vague prompts like "Write a function to process orders" can lead to solutions that work in one context but not yours. A concrete specification, on the other hand, ensures the generated code aligns with your needs.

For example, a function to process orders should accept an Order object with specific fields, perform calculations, return the processed Order, and raise appropriate exceptions. Providing such details helps the AI generate code that fits your context. Elements of a solid specification include the function name and signature, input and output types, error handling, technical constraints, and what not to do.

These constraints guide the AI to avoid unnecessary complexities. The scaffold approach is recommended: start with the file structure, then implement each component step by step. This method ensures you have a solid foundation before adding specifics. For instance, first request the file structure of a Flask REST API for a todo app, then implement each component sequentially.

This approach allows you to verify each step before proceeding, preventing errors from accumulating in a long chain of code. Breaking work into short, focused prompts enhances the AI's ability to reason and provides more accurate results than attempting to solve everything at once. Adding a context block at the start of each conversation can also improve the quality of generated code.

This block, containing project details like Django version, database, libraries, and testing framework, guides the AI in generating code that fits your specific environment. Once you set the context, you don't need to repeat it for every prompt. Starting with test-driven development (TDD) and writing failing tests first is another effective strategy.

By specifying the expected behavior in tests before coding, you ensure the AI generates code that meets your exact requirements. This method reveals edge cases you might have missed, helping you catch potential bugs early.