# Prompting

Seven habits that make every agent conversation sharper.

Great prompt writing equals great collaboration. LLMs are improving rapidly, so understanding where they're strongest improves your experience with Korey — and every AI agent. Keep these habits in mind next time you're chatting.

## 1. Keep conversations short

LLMs experience **context window degradation** — long conversations lead to worse output. Open new conversation windows often and keep prompts fairly self-contained and focused, rather than running long, open-ended threads.

## 2. Don't ask for too many things at once

Break work into smaller steps and ask for one thing at a time, rather than requesting groups of features or implementations at once. Open with "Start with…" or a similar phrase to signal there are multiple steps coming.

## 3. Confirm successful output

When an output looks good, say so — "that looks good" or "let's use that" — before moving on. This prevents unnecessary rework.

## 4. Avoid negative prompting

Focus on what the behavior *should* be rather than what it should *not* be. LLMs do better with positive instructions than negative ones, so describe your desired state.

## 5. Include examples

Supply examples that show the sort of output you want — links to documents, uploads, text, and so on.

## 6. Ask "why?"

Asking *why* the LLM chose its approach often yields better results, on top of what you learn. Models frequently reach realizations during reflection and improve from there.

## 7. Customize for your preferences

Ask Korey to remember your preferences and teach it to create in your preferred style and format. See [Quick start](/docs/getting-started/quick-start#2-save-a-memory-or-writing-style) for how to save a memory and customize your writing style.