AI assistants can draft functions, explain errors and suggest tests, but the quality of the answer depends heavily on the question. A vague request gets a generic answer. A clear one saves you time. This framework makes prompts repeatable.
The four parts of a good prompt #
- Role and context. Say what you are working on: language, framework version, and what the code is for.
- Task. State exactly what you want done, in one or two sentences.
- Constraints. List rules such as “no new dependencies”, “must work in Node 20” or “keep the public function signature”.
- Output format. Say how you want the answer: only the changed function, a diff, or code plus a short explanation.
A weak and a strong example #
Weak: “Fix my code.”
Strong: “I am using Python 3.12 and FastAPI. This endpoint returns a 500 error when the user list is empty. Find the cause and give me a corrected function. Keep the route and the response model unchanged. Explain the fix in two sentences.”
Give the model the evidence #
- Paste the exact error message and the relevant stack trace.
- Include the smallest piece of code that reproduces the problem.
- Say what you expected to happen and what happened instead.
Never paste secrets, customer data or proprietary code that your company does not allow you to share.
Work in small steps #
Ask for one change at a time and review each result. Large requests like “build my whole app” tend to produce code that looks right but hides errors. Smaller steps are easier to test.
Ask for tests and edge cases #
A good follow-up is: “Write unit tests for this function, including empty input, very large input and invalid types.” This often exposes bugs in the first answer.
Always verify #
AI-generated code can contain mistakes, outdated APIs or security flaws. Read it, run it, test it and review it as you would a teammate’s pull request. If an answer cites a library function, confirm that it exists in the documentation for your version.
Keep a prompt library #
When a prompt works well, save it as a template with blanks for the language, task and constraints. Over time you build a set of reliable starting points for debugging, refactoring, code review and documentation.
