🧠 Prompt Like a Pro: Mental Models for Coding with LLMs

27 Jul, 2025

We’ve all been there: typing a prompt into ChatGPT or GitHub Copilot and hoping for the best. Sometimes it works like magic. Other times, it’s spaghetti on your screen.

The difference? Mental models.

When you treat prompting like “talking to an intern,” things get vague fast. But when you think like a system designer, you shift from trial-and-error to reproducible workflows.

Let’s break it down.

🧱 The Bricklayer

“Give me the foundation, then I’ll build from it.”

This model is about structure-first thinking. You ask the LLM to define clean, composable blocks.

Example Prompt:

Create a class called `PhotoCleaner`. It should scan all images older than 90 days and return a list. Don’t delete anything yet.

Then you follow up with:

Add a method to delete the 10 largest files from that list.

Why it works:

• Keeps prompts atomic
• Encourages separation of concerns
• Easier to debug or modify

🎨 The Illustrator

“Rough first, refine later.”

This is vibe coding at its finest. You sketch something vague, then iterate with constraints.

Example Prompt:

Design a simple Android UI for cleaning old photos.

The model gives you a layout. You follow up with:

Add a toggle for preview mode and make the buttons stand out more.

Why it works:

• Mirrors how designers and frontend devs think
• Enables visual play before locking logic
• Keeps you in the flow

🧭 The Navigator

“You know the terrain, now steer.”

You’re not asking for code directly. You’re nudging the model to explore options, avoid traps, or optimize trade-offs.

Example Prompt:

Compare two approaches: deleting photos based on age vs. based on size. Which is safer for user trust?

Or:

Create a cleanup plan that uses minimal memory and works offline on Android.

Why it works:

• Great for architecture and planning
• Builds shared context before implementation
• Useful for agent-based and tool-augmented workflows

🧩 So What?

Prompting isn’t about asking nicely.
It’s about structuring intent.

When you pair your coding mindset with the right mental model:

• You reduce hallucination
• You improve maintainability
• You get closer to reliable AI tooling

🚀 Final Thought

Don’t just prompt harder. Prompt smarter.

Generative AI doesn’t need magical phrases — it needs clarity.
Your brain already has what it takes. Now you have the models too.

👉 For more on capturing progress and remembering AI sessions, check out Snapshots Are the Git Commits of AI—Here’s How to Use Them.

Tags: prompting, AI coding, LLMs, software engineering, AI hygiene