Prompting for Code That Works

Context in, quality out

AI assistants can only work with what you tell them. A good coding prompt includes:

1. Goal: what should the code do?
2. Context: language, framework, versions, existing code it must fit into
3. Constraints: no new libraries, must handle empty input, performance limits
4. Interface: function name, inputs, outputs, examples
5. Definition of done: "include tests", "explain any trade-offs"

Weak

write a function to validate emails

Strong

Write a JavaScript function isValidEmail(email) that returns true/false.
- Must reject empty strings, missing @, and missing domain
- Don't use a regex longer than one line; readability matters
- Include 6 test cases using the format test("name", () => expect(...).toBe(...))

Iterate in small steps

Don't ask for a whole app in one prompt. Ask for one piece, run it, review it, then ask for the next. This is the same "small steps" habit from your first lesson.

When it's wrong

Give precise feedback: the failing input, the actual output, the expected output and the error message. Vague feedback ("it's broken") gets vague fixes.

Professional rules

• Never paste secrets (API keys, passwords, customer data) into AI tools, and follow your company's AI policy.
• Check licenses and sources for large generated code blocks.
• You own it: if you can't explain a line, don't commit it.

Practice: write a precise prompt for one of your exercises in any AI assistant, then compare its output to your own solution and run it against the tests in the Playground.