I've spent hundreds of hours at this point crafting prompts for various LLMs. Here's what I wish someone had told me on day one.
1. Specificity Beats Cleverness
Early on, I spent way too much time trying to craft the "perfect" prompt with clever framing and roleplay scenarios. Turns out, being specific and direct works better almost every time.
Instead of elaborate setups, just clearly state:
- What you want
- Why you want it
- What format the output should be in
- What constraints apply
2. Examples Are Worth a Thousand Words
If you want output in a specific style or format, show — don't tell. One concrete example does more than three paragraphs of description.
This is especially true for structured outputs, code style matching, and tone of voice.
3. Break Complex Tasks Into Steps
The single biggest improvement to my outputs came from stopping trying to do everything in one prompt. Instead:
- First, have the AI analyze the problem
- Then, ask it to propose an approach
- Then, have it implement step by step
- Finally, ask it to review its own work
Each step builds on the last, and you can course-correct along the way.
4. Context Window Management Is a Skill
Your conversation has a memory limit. Treat it like a resource:
- Front-load the most important context
- Summarize long exchanges periodically
- Start fresh conversations for genuinely new topics
- Reference files and docs rather than pasting everything inline
5. The Meta-Skill Is Knowing When NOT to Use AI
Not everything needs an AI solution. Sometimes the fastest path is:
- Reading the docs yourself
- Writing a quick script by hand
- Asking a human who's done it before
AI tools are force multipliers, not replacements for judgment. The best prompt engineers know when to close the chat and just do the work.
The Takeaway
Prompt engineering isn't magic — it's communication. The better you get at clearly expressing what you want, the better results you'll get. And that skill transfers to every other area of work, too.