Getting good at prompt engineering isn’t about tricks

It’s about process, patience, & practice

Here’s what I’ve learned so far after hours of trial, error, & collaboration with AI models

Don’t read about prompting, do it

That’s right, stop reading this post & start prompting

Actually *doing it* will teach you more than any guide

Embrace the process

That’s how you learn

Iterate tirelessly

Your first prompt is usually no bueno & that’s totally fine

Outputs are experiments

Drill into the process

The more you refine, the closer you get to your goal

Clarity is everything

Write prompts as if you’re explaining to another human

Assume zero shared context, over-explain wherever needed

Be clear, detailed, precise

Ambiguity clouds results

Don’t trust the first output

AI is wrong all the time

If something feels off, it probably is

Use your expertise to correct hallucinations

Good outputs require guided correction

Leverage your expertise

Your domain knowledge is an unfair advantage

Use it to catch errors, enrich context, & get the model working toward your specific goals

Models aren’t experts, you are

Patience > shortcuts

There are no hacks

Curiosity, experimentation, & an open mind consistently outperform quick tricks

Explore, refine, & keep tweaking until you get it right

Ask the model to help you

AI is great at collaborating on writing better prompts

Have it interview you to refine what you're asking

If an output isn’t clicking, ask it to clarify or debug your request

AI isn’t just software, it’s a collaborator

Treat the interaction like a working relationship

Be honest, prescriptive, & clear with your goals

Expect effort, not magic

Final takeaway

There’s no single “right” way to prompt an AI, but here’s what’s universal:

- Test relentlessly
- Maintain clarity
- Never settle for the first attempt

Good prompts are engineered, not guessed

Break tasks into steps

AI struggles with vague requests

Spell it out

Systematically break down complex goals into smaller, manageable parts

It’s a conversation, not a magic spell

Speak to the model as you would a collaborator

Push back on vague responses

Ask follow-up questions

Probe for deeper insights