Prompt engineering basics
🎯 Goal
Learn how to write better prompts to get the most out of ChatGPT and similar tools — for all kinds of tasks, not just coding.
🚀 Why This Matters
Most people write prompts like this:
“Explain X” “Write code for Y” “What’s a good name for a feature?”
That’s the equivalent of Googling "stuff".
Let’s go pro.
📦 Core Concepts of Prompting
1. Context is King
LLMs love context. The more you give, the better the output.
❌ “Write a blog post about AI.” ✅ “Write a 500-word blog post about how small dev agencies like MarsBased can use AI to automate documentation. Audience: CTOs. Tone: witty but professional.”
✅ “You're a senior backend dev at MarsBased. Given this GitHub issue, propose 3 implementation options with tradeoffs.”
2. Role Assignment
“You're a senior full-stack dev.” “Act like a sassy but accurate technical recruiter.” “You're a product manager writing internal release notes.”
Roles change everything. Try it.
3. Examples > Explanations
Few-shot prompting is powerful.
❌ “Write me an error message.” ✅ “Here are 2 examples of error messages from our system:\n\n1. 'We couldn’t connect to GitHub. Check your token.'\n2. 'Your session expired. Please log in again.'\n\nNow write a message for a failed API sync with Linear.”
4. Constraints Help
“Explain this like I’m five.” “Summarize this in 3 bullet points.” “Reply in markdown, use emojis.” “Write code with comments in Spanish.”
Constraints = better outputs.
5. Iterate, Don’t Expect Magic
You’re allowed to keep going:
“That’s too long. Can you shorten it?” “Add more sarcasm.” “Use TypeScript instead.” “Rewrite using a more optimistic tone.”
LLMs love feedback.
🔧 Dev Use Cases
🔍 Debugging: Paste an error and ask for causes + fixes.
📄 PR Summaries: “Summarize this PR in 3 sentences for a non-dev.”
🧠 Naming things: “Suggest 10 function names. Context:…”
📜 Commit messages: “Write a conventional commit for this diff.”
🤖 GitHub Copilot prompts: Use natural language to write comments that guide code generation.
💼 Non-Dev Use Cases
✍️ Docs: “Turn this bullet list into internal documentation.” “Write a changelog based on these commits.”
📣 Marketing: “Write a tweet thread from this blog post.” “Summarize this podcast transcript for LinkedIn.”
📅 Planning: “Create a rough roadmap based on these 4 Linear issues.”
🧑🏫 Hiring: “Generate 5 technical interview questions for a senior Vue dev.”
⚙️ Bonus: API-Level Prompting (1 Minute)
If using OpenAI's API:
{
"model": "gpt-4",
"temperature": 0.7,
"messages": [
{ "role": "system", "content": "You are a senior Rails engineer." },
{ "role": "user", "content": "Explain the pros and cons of Hotwire." }
]
}system: Sets the role/persona
user: The actual prompt
temperature: 0 = deterministic, 1 = creative/random
🎁 Cheat Code: Prompt Structure
[Persona or role]
[Clear task]
[Context]
[Constraints]
[Optional: examples]
[Optional: output formatting]
🔥
Last updated