Prompt Engineering: The Art of Asking Questions in the AI Era

1 Introduction

With the AI wave sweeping the globe, the role of "Prompt Engineer" has become a buzzword. The author notes that AI enthusiasts share prompts, experiment with token control, and tweak temperature parameters. A famous Linus Torvalds quote "Talk is cheap. Show me the code." has evolved into "Code is cheap. Show me the prompts."

2 Do Your Homework, Don’t Be a Leech

Before asking, search documentation, read relevant material, and try to solve the problem yourself.

AI is not a personal assistant that can think for you. If you ask "Help me write a program" without clarifying the problem, the model can only guess.

A good prompt should reflect the effort you have already made, for example:

"I tried to fetch a website using Python's requests library but got a 403 error. I checked the network and added a User‑Agent header, yet the issue persists. Should I handle cookies or consider JS rendering? Here is the error log: ..."

This provides the model with enough context to pinpoint the issue and reduces wasted attempts.

3 Describe the Phenomenon, Not the Supposed Cause

Say "mouse cursor is distorted" instead of "graphics driver is broken".

Many people embed conclusions in prompts, such as "my code has a memory leak" or "this API definitely has a bug". The model lacks external context to verify these assumptions.

Provide observable facts, e.g.:

"After running for 10 minutes, memory usage climbs from 100 MB to 2 GB with no release. Here is the core code snippet and monitoring data."

Logs, runtime data, or screenshots for UI issues are the most useful observable facts.

4 Goal‑Oriented, Not Step‑Oriented

"I want to change a tire, not just ask how to use a jack."

People often fall into "path dependence"—insisting on a specific tool or method and forgetting the ultimate goal. State the goal first, then the obstacle.

Step‑oriented example: "How do I use VLOOKUP to match two columns?"

Goal‑oriented example: "I want to merge two tables by ID. I tried VLOOKUP but got #N/A even though both columns are text."

This allows the AI to suggest better alternatives such as XLOOKUP, Power Query, or even a Python solution.

5 Concise, Specific, Structured

Length ≠ information density.

A 500‑line script often conveys less useful information than a minimal, reproducible 10‑line example. Large models can handle long context, but noise weakens the effective signal.

Provide a minimal reproducible example.

Clearly state input, expected output, and actual output.

Organize information with code blocks or lists.

Example:

input: [1, 2, '3', 4]
expected: all integers → [1, 2, 3, 4]
actual: int() conversion raises an error
question: how to safely convert this mixed‑type list?

6 Seek Key Guidance, Not a Full Answer

Although AI can produce complete answers, the most efficient collaboration follows a "step‑by‑step guidance" pattern.

Traditional advice: ask "Where can I learn?" rather than "Teach me everything."

In the AI era, a phased interaction—first ask the model to outline possible solutions, then compare their pros and cons (performance, maintainability, complexity), and finally implement the chosen approach—reduces hallucinations and keeps humans in control.

Ask the AI to articulate the solution reasoning.

Request 2–3 feasible approaches.

Compare each approach’s strengths and weaknesses.

After human confirmation, proceed to concrete implementation.

7 Politeness + Closure = Long‑Term Win‑Win

A simple "thank you" makes experts more willing to help.

Even though AI lacks emotions, structured politeness and feedback improve interaction quality. Begin with a clear background, end with gratitude, and if the suggestion works, follow up: "I applied your advice, the issue is resolved—thanks!"

This feedback loop, while not triggering online learning for static models, simulates reinforcement learning from human feedback (RLHF) and reinforces effective communication.

8 Conclusion: Asking Is a Skill and a Form of Respect

The core of "The Wisdom of Asking Questions" is attitude: respect others' time, respect knowledge boundaries, and respect the complexity of the problem itself. In the AI era we have instant, powerful experts, but lazy questioning yields mediocre answers. A truly effective Prompt Engineer does not merely memorize prompts; they know how to co‑construct understanding with AI.

Good questions are a stimulus and a gift.

9 Further Reading

How To Ask Questions The Smart Way (original English text)

How to Report Bugs Effectively – Simon Tatham