Zum Hauptinhalt springen

Code Generation

AI is most effective for code generation when the task is well-defined and bounded. Use it to eliminate repetitive work, not to figure out what you're building.

What AI generates well​

  • Boilerplate (API routes, CRUD handlers, form schemas)
  • Data transformations and mapping functions
  • Regex patterns and parser logic
  • Type definitions from JSON or database schemas
  • Configuration files (Dockerfile, CI pipelines, linting rules)
  • Repetitive UI components with variations

What to avoid generating blindly​

  • Authentication and authorisation logic (verify every line)
  • Cryptography (never use AI-generated crypto without expert review)
  • SQL queries on untrusted input (always check for injection)
  • Business logic you don't fully understand yet

Generating from a schema​

Give the model a source of truth and ask it to derive code from it:

Here is my Zod schema:

const UserSchema = z.object({
  id: z.string().uuid(),
  email: z.string().email(),
  role: z.enum(['admin', 'member', 'viewer']),
  createdAt: z.date(),
});

Generate:
1. The TypeScript type (inferred from the schema)
2. A mock factory function for tests
3. An Express POST /users route that validates the body and returns 201

Generating from examples​

If you have existing code you want replicated in a different shape:

Here's how I handle a GET endpoint in this codebase:

[paste example]

Now generate a DELETE /users/:id endpoint that follows the same pattern.

Iterating on generated code​

Generated code rarely fits perfectly on the first pass. Use follow-up messages to refine:

  1. Fix types: "The return type should be Promise<User | null>, not Promise<User>"
  2. Adjust style: "Use named exports, not default exports"
  3. Add error handling: "Handle the case where the database returns null"
  4. Optimise: "This runs N+1 queries - rewrite it using a JOIN"

Reviewing generated code​

Before using any generated code, check:

  • Does it compile without errors?
  • Does it handle null, undefined, and empty inputs?
  • Are there any hardcoded values that should be configurable?
  • Does it import everything it uses?
  • Does it introduce any security risks?

If the generated code has issues, see Debugging with AI. For cleaning up structure and style, see Refactoring with AI. Once the code works, add tests with Writing Unit Tests with AI.